Download from the Online Article on Must Be Supplied in an Editable Format (Word Or Excel)

An Ofﬁ cial Journal of the Society of Biological Psychiatry Biological Psychiatry: Cognitive Neuroscience and Neuroimaging

Volume 5, Number 2 February 2020

A journal of cognitive neuroscience, computation, ISSN 2451-9022 and neuroimaging in psychiatry www.sobp.org/BPCNNI

BBPSC_v5_i2_COVER.inddPSC_v5_i2_COVER.indd 1 114-12-20194-12-2019 15:27:0115:27:01 Aims and Scope Biological Psychiatry: Cognitive Neuroscience and Neuroimaging is an official journal Biological of the Society for Biological Psychiatry, whose purpose is to promote excellence in scientific research and education in fields that investigate the nature, causes, mechanisms, and treatments of disorders of thought, emotion, or behavior. In accord with this mission, this peer-reviewed, rapid-publication, international journal focuses on studies using the tools and constructs of cognitive neuroscience, including the full range of non-invasive neuroimaging and human extra- and intracranial physiological Psychiatry recording methodologies. It publishes both basic and clinical studies, including those that incorporate genetic data, pharmacological challenges, and computational modeling approaches. Cognitive Neuroscience The journal publishes novel results of original research which represent an important new lead or significant impact on the field. Reviews and commentaries that focus on CNNI and Neuroimaging topics of current research and interest are also encouraged.

Editor Editorial Board Cameron S. Carter, MD Kazufumi Akiyama, MD, PhD Steven J. Luck, PhD University of California, Davis Dokkyo Medical Univ, Tochigi, Japan UC Davis, Davis, CA Sacramento, California Amy F.T. Arnsten, PhD Angus MacDonald, PhD Yale Univ, New Haven, CT Univ of Minnesota, Minneapolis, MN Deputy Editors James Blair, PhD Richard J. Maddock, MD Anissa Abi-Dargham, MD NIMH, Bethesda, MD UC Davis, Sacramento, CA Stony Brook Univ, Stony Brook, NY David L. Braff, MD Stephen R. Marder, MD UC San Diego, La Jolla, CA UCLA, Los Angeles, CA Deanna M. Barch, PhD WUSTL, St. Louis, MO E. Sherwood Brown, MD, PhD Graeme F. Mason, PhD UT Southwestern, Dallas, TX Yale Univ, New Haven, CT Edward T. Bullmore, MB, PhD, FMedSci Univ of Cambridge, Cambridge, Vince D. Calhoun, PhD Daniel H. Mathalon, MD, PhD United Kingdom Univ of New Mexico, Albuquerque, NM UCSF, San Francisco, CA Quentin J.M. Huys, MD, PhD Edwin H. Cook Jr., MD Andreas Meyer-Lindenberg, MD, PhD Univ of Illinois Chicago, Chicago, IL Central Inst of Mental Health, Mannheim, Germany Univ College London, London, United Kingdom Zafiris Jeff Daskalakis, MD, PhD, FRCP(C) Rajesh Narendran, MD Murray B. Stein, MD, MPH Univ of Toronto, Toronto, Ontario, Canada Univ of Pittsburgh, Pittsburgh, PA UC San Diego, La Jolla, CA Adriana Di Martino, MD Thomas E. Nichols, PhD Child Mind Institute, New York, NY Univ of Oxford, Oxford, United Kingdom Editorial Committee Ronald S. Duman, PhD Martin P. Paulus, MD Carrie E. Bearden, PhD Yale Univ, New Haven, CT Laureate Institute, Tulsa, OK UCLA, Los Angeles, CA Klaus P. Ebmeier, MD Godfrey Pearlson, MD Dennis S. Charney, MD Univ of Oxford, Oxford, United Kingdom Yale Univ, Hartford, CT Mount Sinai, New York, NY Damien A. Fair, PA-C, PhD Mary L. Phillips, MD David Goldman, MD Oregon Health & Science Univ, Portland, OR Univ of Pittsburgh, Pittsburgh, PA NIAAA, Rockville, MD Catherine Fassbender, PhD Daniel S. Pine, MD UC Davis, Sacramento, CA NIMH, Bethesda, MD Raquel E. Gur, MD, PhD Univ of Pennsylvania, Philadelphia, PA Kate D. Fitzgerald, MD Diego A. Pizzagalli, PhD Univ of Michigan, Ann Arbor, MI Harvard Med School, Belmont, MA Ex Officio Alex Fornito, PhD J. Daniel Ragland, PhD Monash Univ, Clayton, Victoria, Australia UC Davis, Sacramento, CA Karen F. Berman, MD Karl J. Friston, MBBS, MA, MRCPsych Trevor W. Robbins, PhD NIMH, Bethesda, MD Univ College London, London, United Kingdom Univ of Cambridge, Cambridge, United Kingdom Mary L. Phillips, MD John D.E. Gabrieli, PhD Katya Rubia, PhD Univ of Pittsburgh, Pittsburgh, PA MIT, Cambridge, MA King’s College London, London, United Kingdom Scott L. Rauch, MD James M. Gold, PhD Gerard Sanacora, MD, PhD McLean Hospital, Belmont, MA Univ of Maryland, Baltimore, MD Yale Univ, New Haven, CT Anthony A. Grace, PhD Marjorie Solomon, PhD Editor-in-Chief, SOBP Publications Univ of Pittsburgh, Pittsburgh, PA UC Davis, Sacramento, CA John H. Krystal, MD Ralitza Gueorguieva, PhD Stephen M. Strakowski, MD Yale Univ, New Haven, CT Yale Univ, New Haven, CT Univ of Texas at Austin, Austin, TX Ahmad R. Hariri, PhD Stephan F. Taylor, MD Duke Univ, Durham, NC Univ of Michigan, Ann Arbor, MI Catherine J. Harmer, DPhil Bruce I. Turetsky, MD Univ of Oxford, Oxford, United Kingdom Univ of Pennsylvania, Philadelphia, PA Editorial Office Andrew D. Krystal, MD, MS Lucina Q. Uddin, PhD Phone: (214) 648-0880 UCSF, San Francisco, CA Univ of Miami, Coral Gables, FL [email protected] www.sobp.org/BPCNNI Scott A. Langenecker, PhD Sophia Vinogradov, MD Univ of Illinois Chicago, Chicago, IL Univ of Minnesota, Minneapolis, MN Administrative Editor Carol A. Tamminga, MD Ellen Leibenluft, MD Jong H. Yoon, MD Managing Editor Rhiannon M. Bugno NIMH, Bethesda, MD Stanford Univ, Stanford, CA Sr. Publication Coordinator Rosa Garces David A. Lewis, MD Carlos A. Zarate Jr., MD Publication Coordinator Glenda Stroud Univ of Pittsburgh, Pittsburgh, PA NIMH, Bethesda, MD Editorial Assistant Kelly Schappert Jeffrey A. Lieberman, MD Andrew Zalesky, PhD Editorial Assistant Elisha Cadena-Bowles Columbia Univ, New York, NY Univ of Melbourne, Melbourne, Australia Biological Psychiatry: Cognitive Neuroscience and Neuroimaging document delivery. Special rates are available for educational insti- (ISSN 2451-9022) is published 12 times a year by Elsevier Inc., 230 tutions that wish to make photocopies for non-profit educational Park Avenue, Suite 800, New York, NY 10169-0901, USA. classroom use.

Customer Service (orders, claims, online): Please visit our Support Permissions: May be sought directly from Elsevier’s Global Rights Hub page https://service.elsevier.com for assistance. Department in Oxford, UK: tel: (215) 239-3804 or +44 (0)1865 843830; fax: +44 (0)1865 853333, e-mail [email protected]. Yearly subscription rates: United States and possessions: Institu- Requests may also be completed online via the Elsevier homepage tional: USD 1296. All other countries: Institutional: USD 1296. (http://www.elsevier.com/permissions). In the USA, users may clear Advertising information: Advertising orders and inquiries can be sent permissions and make payments through the Copyright Clearance to: United States, Canada, and South America, Ken Senerth at Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; tel: (978) Therapeutic Solutions, PO Box 2083, Apopka, FL 32712, USA; 750-8400; fax: (978) 750-4744, and in the UK through the Copyright tel: (609) 577-0916; fax: (212) 633-3820. Classified advertising orders Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham and inquiries can be sent to Traci Peppers at Elsevier Inc., 230 Park Court Road, London W1P 0LP, UK; tel: (+44) 20 7631 5555; fax: (+44) Avenue, Suite 800, New York, NY 10169-0901, USA; tel: (212) 633- 20 7631 5500. Other countries may have a local reprographic rights 3766; fax: (212) 633-3820. Europe and the rest of the world: Robert agency for payments. Bayliss at Elsevier, 125 London Wall, London, EC2Y 5AS, UK; tel: (+44) Derivative works: Subscribers may reproduce tables of contents or 20 7424 4454; e-mail: [email protected]. prepare lists of articles including abstracts for internal circulation within their institutions. Permission of the Publisher is required for Author inquiries: For inquiries relating to the submission of articles, see Guide for Authors in this issue or contact the Journal’s Editorial resale or distribution outside the institution. Permission of the Office. You can track your submitted article at http://www.elsevier. Publisher is required for all other derivative works, including com/track-submission. Please also visit http://www.elsevier.com/ compilations and translations. trackarticle to track accepted articles and set up e-mail alerts to Electronic storage or usage: Permission of the Publisher is required to inform you of when an article’s status has changed, as well as obtain store or use electronically any material contained in this journal, detailed artwork guidelines, copyright information, frequently asked including any article or part of an article. Except as outlined above, no questions, and more. Contact details for questions arising after part of this publication may be reproduced, stored in a retrieval system acceptance of an article, especially those relating to proofs, are pro- or transmitted in any form or by any means, electronic, mechanical, vided after registration of an article for publication. You are also wel- photocopying, recording or otherwise, without prior written permission come to contact Customer Support via http://support.elsevier.com. of the Publisher. Address permissions requests to: Elsevier Rights Department, at the fax and e-mail addresses noted above. ª 2020 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. Notice: Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, This journal and the individual contributions contained in it are methods, compounds or experiments described herein. Because of protected under copyright by the Society of Biological Psychiatry, and rapid advances in the medical sciences, in particular, independent the following terms and conditions apply to their use: verification of diagnoses and drug dosages should be made. To the Photocopying: Single photocopies of single articles may be made for fullest extent of the law, no responsibility is assumed by the publisher personal use as allowed by national copyright laws. Permission of for any injury and/or damage to persons or property as a matter of the Publisher and payment of a fee is required for all other photo- products liability, negligence or otherwise, or from any use or operation copying, including multiple or systematic copying, copying for of any methods, products, instructions or ideas contained in the advertising or promotional purposes, resale, and all forms of material herein. Volume 5, Number 2, February 2020

IN THIS ISSUE -FEBRUARY 152 Neuroanatomical Dysconnectivity Underlying Cognitive Deﬁcits in Bipolar 131 A brief summary of the articles Disorder appearing in this issue of Biological Genevieve McPhilemy, Leila Nabulsi, Psychiatry: Cognitive Neuroscience Liam Kilmartin, Denis O’Hora, and Neuroimaging Stefani O’Donoghue, Giulia Tronchin, Laura Costello, Pablo Najt, Srinath Ambati, COMMENTARIES Gráinne Neilsen, Sarah Creighton, Fintan Byrne, James McLoughlin, Colm McDonald, 133 What Might Prediction Tell Us About the Brian Hallahan, and Dara M. Cannon Dopaminergic Mechanisms of Depression? Michael Browning 163 Machine Learning Identiﬁes Large-Scale » See corresponding article on page 163 Reward-Related Activity Modulated by Dopaminergic Enhancement in Major 135 Of Forests and Trees: Bridging the Gap Depression Between Neurobiology and Behavior in Yuelu Liu, Roee Admon, Monika S. Mellem, Posttraumatic Stress Disorder Emily L. Belleau, Roselinde H. Kaiser, Christopher L. Averill, Lynnette A. Averill, Rachel Clegg, Miranda Beltzer, Franziska Goer, Siyan Fan, and Chadi G. Abdallah Gordana Vitaliano, Parvez Ahammad, and » See corresponding article on page 203 Diego A. Pizzagalli » See commentary on page 133 138 Functional Connectivity and Cognitive Control in Late-Life Depression 173 Volatility Estimates Increase Choice Julie A. Dumas Switching and Relate to Prefrontal Activity » See corresponding article on page 213 in Schizophrenia Lorenz Deserno, Rebecca Boehme, Christoph Mathys, Teresa Katthagen, ARCHIVAL REPORTS Jakob Kaminski, Klaas Enno Stephan, 140 Frontolimbic, Frontoparietal, and Default Andreas Heinz, and Florian Schlagenhauf Mode Involvement in Functional Dysconnectivity in Psychotic Bipolar 184 Ultra-High-Resolution Imaging of Amygdala Disorder Subnuclei Structural Connectivity in Major Leila Nabulsi, Genevieve McPhilemy, Depressive Disorder Liam Kilmartin, Joseph R. Whittaker, Stephanie S.G. Brown, John W. Rutland, Fiona M. Martyn, Brian Hallahan, Gaurav Verma, Rebecca E. Feldman, Colm McDonald, Kevin Murphy, and Molly Schneider, Bradley N. Delman, Dara M. Cannon James M. Murrough, and Priti Balchandani 194 Elevated Amygdala Activity in Young Adults 222 Inhibition-Related Cortical With Familial Risk for Depression: A Hypoconnectivity as a Candidate Potential Marker of Low Resilience Vulnerability Marker for Obsessive- Tracy Barbour, Avram J. Holmes, Compulsive Disorder Amy H. Farabaugh, Stephanie N. DeCross, Adam Hampshire, Ana Zadel, Garth Coombs, Emily A. Boeke, Stefano Sandrone, Eyal Soreq, Rick P.F. Wolthusen, Maren Nyer, Naomi Fineberg, Edward T. Bullmore, Paola Pedrelli, Maurizio Fava, and Trevor W. Robbins, Barbara J. Sahakian, and Daphne J. Holt Samuel R. Chamberlain

203 Opponent Effects of Hyperarousal and 231 Duration of Untreated Psychosis Correlates Re-experiencing on Affective Habituation With Brain Connectivity and Morphology in Posttraumatic Stress Disorder in Medication-Naïve Patients With Katherine L. McCurry, B. Christopher Frueh, First-Episode Psychosis Pearl H. Chiu, and Brooks King-Casas Jose O. Maximo, Eric A. Nelson, » See commentary on page 135 William P. Armstrong, Nina V. Kraguljac, and Adrienne C. Lahti 213 Cognitive Control Network Homogeneity and Executive Functions in Late-Life 239 Regulation of Craving and Negative Depression Emotion in Alcohol Use Disorder Matteo Respino, Matthew J. Hoptman, Shosuke Suzuki, Maggie Mae Mell, Lindsay W. Victoria, George S. Alexopoulos, Stephanie S. O’Malley, John H. Krystal, Nili Solomonov, Aliza T. Stein, Maria Coluccio, Alan Anticevic, and Hedy Kober Sarah Shizuko Morimoto, Chloe J. Blau, Lila Abreu, Katherine E. Burdick, Conor Liston, and Faith M. Gunning » See commentary on page 138

The color-coded regions of interest highlighted on the cover from Hampshire et al.(inthis issue, pages 222e230), derived from activation maps of contrasts of interest, were used to conduct subsequent connectivity analyses. In this work, the authors found cortical hypoconnectivity during an inhibitory control task in patients with obsessive-compulsive disorder and their unaffected relatives, which may represent a marker of vulnerability for this disorder. See Figure 1 for full details. Visualization software developed by Eyal Soreq.

www.sobp.org/BPCNNI GUIDE FOR AUTHORS

ARTICLE TYPES Transferred Submission Archival Reports Revised Submission Priority Communications Revised Submission Files Reviews File Type Requirements Techniques and Methods Word Limits Correspondence In This Issue Feature Commentaries and Editorials Early Career Investigator Commentaries PEER REVIEW PROCESS

PREPARATION & FORMATTING REQUIREMENTS EDITORIAL POLICIES Cover Letter Authorship Manuscript Corresponding Author Title Page Disclosure of Biomedical Financial Interests and Potential Conﬂicts of Interest Abstract Ethical Considerations Main Text Clinical Trials Registration Acknowledgments Research and Data Reporting Guidelines Disclosures Materials and Genes References Repository Data Figure/Table Legends Preprint Policy Tables Figures ACCEPTANCE AND PUBLICATION Proofs Supplemental Information Publication Schedule Multimedia Content Press and Embargo Policy 3D Neuroimaging Fees Style Cover Art Psychopharmacology Nomenclature Article Sharing Gene/Protein Nomenclature NIH Public Access Policy and Other Funding Body Agreements Copyright SUBMISSION PROCESS New Submission New Submission Files QUESTIONS? CONTACT US Author Notiﬁcations Referee Suggestions

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging (BP:CNNI) is an official journal of the Society of Biological Psychiatry. The Journal focuses on studies using the tools and constructs of cognitive neuroscience, including the full range of non-invasive neuroimaging and human extra and intracranial physiological recording methodologies. It publishes both basic and clinical studies, including those that incorporate genetic data, pharmacological challenges, and computational modeling approaches. Except where explicitly stated otherwise, BP:CNNI conforms to the guidelines set forth by the International Committee of Medical Journal Editors (ICMJE) (see Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals (December 2018): Available from http://www. ICMJE.org). All new manuscripts must be submitted through the journal website: https://www.editorialmanager.com/bpsc. All correspondence should be directed to the Editorial Office at [email protected].

ARTICLE TYPES advance in the field and are not intended for publication of preliminary results. They are expected to be acceptable for publication in essentially the form submitted. Papers that require substantial revisions or do not fit the criteria will be considered Archival Reports as Archival Reports. See Archival Reports for structure, word length, and other Archival Reports are original research papers reporting novel results on a broad requirements. range of topics related to the pathophysiology and treatment of major neuropsychiatric disorders. Clear explication of methods and results is critical to facilitate review of papers and replicability of findings. Reviews Reviews are concise and focus on current aspects of interest and research. Word Limit: 4000 words in main body of text* Reviews should be novel and have sufficient supporting literature, which should be Abstract: 250 word limit; Structure as follows: Background, Methods, Results, integrated into a mechanistic model when applicable. Reviews should generally not Conclusions focus solely on the authors’ own work. Note that meta-analyses report original data Main Text: Structure as follows: Introduction, Methods and Materials, Results, and thus are not considered review papers; meta-analyses should be submitted as Discussion Archival Reports. Tables/Figures: No limit, as needed Word Limit: 4000 words in main body of text* References: No limit, as needed Abstract: 250 word limit; unstructured Supplement: Allowed, unlimited length Main Text: Structure with headings as needed Priority Communications Tables/Figures: Allowed to summarize or illustrate important points These are Archival Reports that clearly document novel experimental findings of References: 150 maximum unusual and timely significance. These papers should represent a conceptual Supplement: Allowed, unlimited length GUIDE FOR AUTHORS

Techniques and Methods Commentaries and Editorials These articles feature new, improved, or noteworthy comments about techniques These articles address points directly related to articles in the concurrent issue, and/or or methods relevant to basic or clinical research in, or treatment of, psychiatric focus on topics of current research and interest. These are generally invited, but disorders. interested contributors may contact the Editor. 3000 words in main body of text* Word Limit: Word Limit: 1500 words in main body of text* 150 word limit; unstructured Abstract: Abstract: Not permitted Structure as follows: Introduction, Methods and Materials, Results, Main Text: Main Text: Unstructured, headings are not permitted Discussion Tables/Figures: A single summarizing figure or table is encouraged Maximum of two Tables/Figures: References: 10 maximum No limit, as needed References: Supplement: Not permitted Supplement: Allowed, unlimited length Early Career Investigator Commentaries These articles provide publishing opportunities to early career investigators (ECI), Correspondence as part of a joint project between the Journal and the Education Committee of the These letters to the editor are directly related to methods, procedures or inter- Society of Biological Psychiatry. These are invited articles for which an ECI serves pretation of data presented in work recently published in our journal and uses new as the sole and corresponding author. Each ECI shall be 1) a current member of the analysis of data presented, the support of previously published work, and/or sci- Society of Biological Psychiatry, 2) no more than 10 years out from terminal degree, entific points to be addressed based on methodological issues. They may also and 3) not hold an academic faculty rank higher than Assistant Professor. A senior present a case report that clearly and unambiguously illustrates important new investigator mentors each ECI, acts as the content reviewer, and is recognized in principles that have not yet been demonstrated in clinical trials. When warranted, a the Acknowledgments section. reply from author(s) of the original work is solicited; in such cases, the editor does Word Limit: 1500 words in main body of text* not issue a final decision until both articles are submitted and the pair is then Abstract: Not permitted published together. Main Text: Unstructured, headings are not permitted Word Limit: 1000 words in main body of text* Tables/Figures: A single summarizing figure or table is encouraged Abstract: Not permitted References: 10 maximum Main Text: Unstructured Supplement: Not permitted Tables/Figures: Not encouraged, but 1-2 allowed if needed to illustrate important *Word limits include main text of the article only, e.g., for Archival Reports, the word points count includes the Introduction, Methods and Materials, Results, and Discussion References: No limit, as needed sections. When calculating word counts, exclude abstract, references, table/figure Supplement: Not permitted legends, acknowledgments, and disclosures.

PREPARATION & FORMATTING REQUIREMENTS permitted in abstracts. Avoid the use of abbreviations/acronyms that are not used at least three times. The basic elements of all submissions are as follows: Main Text Cover letter The text of papers should be double-spaced and structured according to the article Manuscript type. It should not exceed the word limits as detailed above. Articles reporting › Title page original research (Archival Reports, Priority Communications, Techniques and › Abstract Methods) should be structured with the following headings: Introduction, Methods › Main text of article and Materials, Results, Discussion. The introduction should provide a brief back- › Acknowledgments ground and state the objectives/hypotheses of the current work; it should not › Disclosures include the findings/results of the study. The Methods and Materials section should › References include sufficient detail to allow other investigators to replicate the work. It is not › Legends for tables and figures appropriate to move the entire text of the methods to the supplement to adhere to Tables the Journal’s word count limits. Manufacturer name and location should be Figures included at first mention, where applicable. Authors may reference other pub- Supplemental Information lications for methods that have previously been published in full detail elsewhere. Further details on each element are provided below, followed by guidance on style. Relevant ethics statements must be included; see Ethical Considerations section, below. The Results section should clearly present the experimental findings and Cover Letter test statistics in a logical order. The Discussion section should describe the results, Cover letters are optional, but suggested, for all submissions. A cover letter must interpret them in the context of prior literature, and discuss the implications and be uploaded as a separate file, as it is not made available to peer reviewers. significance of the finding(s). Limitations of the current work should also be discussed. Manuscript Manuscripts should contain the following sections: title page, abstract, main article The Journal supports efforts in the biomedical research community to improve text, acknowledgments, disclosures, references, footnotes, and table/figure leg- transparency and reproducibility in published research. Thus, we are pleased to ends. The manuscript may also include tables, in text format, at the end of the file. support the Resource Identification Initiative and therefore, strongly encourage the Begin all sections on separate pages. The manuscript file should be supplied in inclusion of RRID identifiers where applicable in the Methods section. RRIDs Word, not in PDF. provide persistent, unique identifiers to key study resources (antibodies, model organisms, cell lines, and tools including software and databases). Authors may Title Page search for RRIDs at https://scicrunch.org/resources. For example, an appropriate The title page should be the first page of the manuscript file and should include the RRID citation for a software tool would be as follows: “Data were processed using following elements: ImageJ (https://imagej.net/; RRID:SCR_003070).” Full article title, 200 characters or less; acronyms/abbreviations are prohibited Acknowledgments Full names of all authors, in order, and their affiliations This section should include detailed information regarding all sources of funding, Corresponding author’s complete mailing address, phone, and email including grant and other material or financial support. Specify granting agency, Short/running title, 55 characters or less (including spaces); standard grant number, and recipient for each funding source. The role of study sponsor(s), if acronyms are permitted any, should be stated. Identify any data that was published previously, in abstract/ Six keywords poster form or on a preprint server. This section may also be used to acknowledge non-author contributors/collaborators and individuals who provided personal and Abstract technical assistance. If a consortium/group is listed as an author, then the indi- Abstracts should be structured or unstructured according to the article type and vidual members must be named here. Authors should secure written permission should not exceed the word limits as detailed above. Structured abstracts should from all individuals named in this section. have the following sections: Background, Methods, Results, Conclusions. The Methods section should explicitly state the sample size and sex/species of sub- Disclosures jects, when applicable. For those manuscripts that require clinical trials registration This section must include the required financial disclosures and conflict of interest (see Clinical Trials Registration section, below), the registry name, URL, and reg- statements for each author. Even if every author has nothing to disclose, this must istration number should be included at the end of the abstract. References are not be explicitly stated. See section on Disclosure, below. GUIDE FOR AUTHORS

References individual elements of an image may not be adjusted, manipulated, or cropped in References should be numbered and listed by their order of appearance in the text. order to selectively highlight, obscure, delete, or otherwise misrepresent the image Refer to references in the text with the appropriate number in parentheses. Refer- or its interpretation. ences in tables and figures should also be numbered. List all authors; if there are more than seven authors, list the first six then et al. Periodical abbreviations should follow Supplemental Information those used by Index Medicus. It is not appropriate to reference papers that have not Supplemental information, relevant to the work but not critical to support the yet been published (i.e., are submitted or under review). The following are sample findings, is strongly encouraged by the Journal and is made available online via references for a published journal article (1), a book (2), and an edited book (3). links in the published article. All such material is peer-reviewed, but not typeset or proofed and so should be carefully prepared. Unlike other files, all supplemental 1. Krystal JH, Carter CS, Geschwind D, Manji HK, March JS, Nestler EJ, et information (including text, tables, and figures) should be uploaded in a single Word al. (2008): It is time to take a stand for medical research and against file whenever possible. Exceptions are large and/or lengthy tables, which may be terrorism targeting medical scientists. Biol Psychiatry 63: 725-727. submitted in Excel. Word documents will automatically be converted to PDF before 2. American Psychiatric Association (2013): Diagnostic and Statistical Man- being posted online. ual of Mental Disorders, 5th ed. Washington, DC: American Psychiatric Publishing. Do not number sections of text; rather, use textual headings to clearly differentiate 3. Martin JH (1985): Properties of cortical neurons, the EEG, and the sections. Supplementary figures and tables should appear with their titles/legends, mechanisms of epilepsy. In: Kandel ER, Schwartz JH, editors. Principles and be numbered consecutively (i.e., Figure S1, Figure S2, Table S1). References of Neural Science, 2nd ed. New York: Elsevier, pp 461-471. should be included as a separate list from those in the main manuscript; number beginning with (1) and include a reference list at the end of the supplemental The Journal also encourages the citation of underlying or relevant datasets in document. The CONSORT diagram for randomized controlled trials, when appli- manuscripts by citing them in the text and including a data reference in the ref- cable, will be published in the supplement. erence list. Data references should include the following elements: author name(s), dataset title, data repository, version (where available), year, and global persistent Multimedia Content identifier. Add [dataset] immediately before the reference so that it can be properly Multimedia content, in formats such as AVI or MPG, is encouraged and should be identified as a data reference. The [dataset] identifier does not appear in published uploaded as an “e-component” in the drop-down menu at the upload screen. articles. 3D Neuroimaging Figure/Table Legends The Journal also encourages enrichment of online articles via support of 3D Provide a brief title and legend for each figure and table. For multi-part figures, neuroimaging data visualizations. Authors may provide 3D neuroimaging data in describe each panel. Avoid duplicating information in the figure/table legends that either single (.nii) or dual (.hdr and .img) NIfTI file formats. Recommended size is already presented in the Methods and Materials or Results sections. of a single uncompressed dataset is 100 MB. Multiple datasets can be submitted. Each dataset should be zipped and uploaded separately as “3D Tables neuroimaging data” in the drop-down menu at the upload screen. Please Tables should be cited in the text and numbered consecutively (i.e., 1, 2, 3) in the provide a short informative description for each dataset by filling in the order of their mention. Each table should have a title, along with a brief description ‘Description’ field when uploading a dataset. If an article is accepted, the (legend). Do not duplicate information that is already presented in the text. Tables uploaded datasets will be available for download from the online article on must be supplied in an editable format (Word or Excel). They may either be ScienceDirect. For more information, see: https://www.elsevier.com/authors/ included at the end of the manuscript file, or uploaded individually, but not both. author-resources/data-visualization. Table footnotes should use superscript lowercase letters, rather than symbols or bold/italic text. Colored text or shading is not permitted in tables. Style Basic style points are as follows: Figures Basic figure instructions are provided here. Further details regarding electronic Layout › Double-space all text artwork quality and preparation can be found at https://www.elsevier.com/ authors/author-schemas/artwork-and-media-instructions. › Number each page › Line numbering is not necessary Key Requirements for Figures Spelling › Use American, as opposed to British, spellings File Formats › TIFF, PDF, PPT, or EPS are preferred; JPEG is acceptable Language › English Resolution › Halftone or combination art: 300-500 dpi Font › Any standard typeface is acceptable (e.g., Arial, Times › Line art: 1000 dpi or supply as vector image New Roman) Image Size › Single column width: 90 mm (255 pt) › Be consistent throughout (use the same typeface and size) › 1.5 column width: 140 mm (397 pt) Acronyms/ › Define at first use in the abstract › Double column (full page) width: 190 mm (539 pt) Abbreviations › Define again at first use in the text and also in each legend › Note: 72 points ¼ 1 inch › Avoid unnecessary/uncommon abbreviations Font › 8-12 point (minimum size variability) Nomenclature › See below › Standard typeface (e.g., Arial, Times New Roman) › Consistent throughout Our readership is diverse, and authors should consider that many readers are in specialty areas other than their own. It is important, therefore, to avoid jargon. Multi-Panel Figures › Label each panel/part with a capital letter (A, B, C,.) Manuscripts with the broadest appeal are focused and clearly written. In highly Figure Titles/ › Include in manuscript file, not in figure files specialized areas, the introduction should be a concise primer. Legends We encourage authors whose first language is not English to ask a native English File Naming › Use the figure number (Fig1.tif, Fig2.pdf, etc.) speaker to review their manuscript or to use a professional language editing service Upload › Supply as individual files (a single file for each figure) prior to submission. Accepted manuscripts are copyedited to conform to the AMA Manual of Style. Figures should be cited in the text, numbered consecutively (i.e., 1, 2, 3) in the order of their mention, and have brief legends. Each figure should be consistent in color, Psychopharmacology Nomenclature size, and font, and be designed proportionally so that it can later be sized as needed BP:CNNI supports the Neuroscience-based Nomenclature (NbN) project (http:// without loss of legibility or quality. Letters and numbers, in particular, should not nbnomenclature.org/), which aims to promote the use of mechanism-based vary greatly in size. RGB color mode is preferred over CMYK. High quality versions nomenclature that is pharmacologically-driven, rather than indication-based. The of each figure should be uploaded individually (i.e., two figures should be uploaded NbN system characterizes medications based on their pharmacological domain separately as Figure 1 and Figure 2). To reduce TIFF file size, flatten layers and then and mode(s) of action. Authors should use NbN’s glossary or official apps in order save with LZW compression before uploading. A minimum resolution of 300 dpi is to translate between the old and new nomenclature. required. Note that the quality of a low-resolution figure cannot be improved by artificially increasing the resolution in graphics software; figures must be initially Gene/Protein Nomenclature created with sufficient quality/resolution. Figure titles and legends should be Gene symbols should be italicized and differentiate by species. Human sym- included as editable text in the manuscript file and not in the figure files. bols should be all uppercase (DISC1), whereas symbols for rodents and other species should be lowercase using only an initial capital (Disc1). Protein Images should represent the original data and be minimally processed. Uniform products, regardless of species, are not italicized and use all uppercase letters adjustments (e.g., brightness, contrast) may be applied to an entire image, but (DISC1). GUIDE FOR AUTHORS

Authors should use approved nomenclature for gene symbols by consulting the listing multiple names separated by a slash, such as ‘Oct4/Pou5f1’. Use one appropriate public databases for correct gene names and symbols. Approved name throughout and include any alias(es) upon the ﬁrst reference. Authors human gene symbols are available from HUGO Gene Nomenclature Committee should submit proposed gene names that are not already approved to the (HGNC) at http://www.genenames.org/. Approved mouse symbols are provided appropriate nomenclature committees as soon as possible. It is the authors’ by The Jackson Laboratory at http://www.informatics.jax.org/marker/. Use responsibility to ensure these are deposited and approved before publication of symbols (e.g., SLC6A4, DISC1) as opposed to italicized full names, and avoid an article.

SUBMISSION PROCESS not have a conflict of interest in reviewing the manuscript. Affiliations of the suggested referees should all be different, and none should share an affiliation All manuscripts must be submitted in electronic form through the BPCNNI online with any of the authors. Editors are not appropriate to suggest as a reviewer. submission and review website (https://www.editorialmanager.com/bpsc). Sub- Authors are also permitted to identify reviewers who should be excluded from mission is a representation that all authors have personally reviewed and given final reviewing their work, but final peer reviewer selections remain at the editors’ approval of the version submitted, and neither the manuscript nor its data have discretion. been previously published (except in abstract or preprint form) or are currently under consideration for publication elsewhere. Transferred Submission Some authors may be offered the opportunity to directly transfer their papers from The has created checklists to assist authors in the efficient submission of Journal to . Upon acceptance of a transfer offer, the sub- both new and revised manuscripts. They are entirely optional and intended solely Biological Psychiatry BP:CNNI mission will be transmitted directly to the Editorial Office. Authors whose to help authors adhere to our submission guidelines and save time so BP:CNNI papers have not yet been peer reviewed need take no further action; the paper will that submissions do not need to be returned for correction. The checklists be assigned to an editor for handling. Authors whose papers have already been are available here: http://www.biologicalpsychiatrycnni.org/content/bpsc- peer reviewed at will then have the submission returned to submission-checklists. Biological Psychiatry them in order to revise the paper in accordance with the reviewers’ comments. In To ensure transparency, authors are expected to clearly declare other reports/ other words, although it will be a “new” submission at BP:CNNI, proceed as if publications of their own that have used the same dataset or sample. Authors must submitting a revision. The manuscript should be revised and a detailed response to also identify figures, tables, and/or data that have been published elsewhere. It is reviewers file must be included as part of the submission. Any other revised files the author’s responsibility to obtain permission from the copyright holder(s) to should also be updated/replaced as necessary. The revised paper will be returned reproduce or modify any previously published materials. to the original reviewers at the editor’s discretion. All other instructions remain applicable. The person designated in the system as the corresponding author must be one of the individuals named as a corresponding author on the title page. Upon finalizing the submission, the corresponding author will immediately receive an email noti- Revised Submission When submitting a revised manuscript, authors are asked to provide a detailed fication that the submission has been received by the Editorial Office. If such response to reviewers, which must be uploaded as a unique Word or PDF documentation has not been received, then a problem has occurred during the file (separate from the cover letter). Authors may upload a ‘tracked changes’ ver- submission process and should be investigated. Any manuscripts not conforming sion of their revision, but must always include a ‘clean’ non-marked version of the to these guidelines will be returned to the author for correction before the manu- manuscript. If revisions are a condition of publication, only two revised versions of script is processed. The manuscript status is available to the corresponding author the paper will be considered. Unsolicited revisions are not allowed. at all times by logging into the website. The submission will receive a manuscript number once it has been processed and assigned to an editor. Revised Submission Files All files (cover letter, response to reviewers, manuscript, figures, etc.) must be uploaded separately at revision, and should be labeled with appropriate and New Submission descriptive file names (e.g., SmithText.doc, Fig1.tif, Table1.doc). File format When submitting a new manuscript, authors will be asked to provide the following: requirements are specified in the below table. The system will build a single PDF of valid email addresses for all authors; the names, emails and affiliations of 6 indi- the submission from the uploaded files. Authors should be careful to replace all viduals who would be appropriate to review the work; and all submission files. files that have been updated since original submission and ensure all files are Further details are as follows. correctly labeled (particularly if figures and/or tables have been rearranged and subsequently renumbered).

New Submission Files To ease the burden of the submission process, we permit authors to upload the File Type Requirements entire submission (minus a cover letter) as a single file, with pages numbered, in Word or PDF. Tables and figures may either be placed within the body of the Cover Letter Word or PDF manuscript or presented separately at the end. Authors must ensure that all Detailed Response to Reviewers Word or PDF elements are clearly legible for editors and peer reviewers. Alternatively, authors may upload individual files (cover letter, manuscript, figures, etc.) separately. All Manuscript Word files should be labeled with appropriate and descriptive names (e.g., Smith- Tables Word or Excel Text.doc, Fig1.eps, Table1.doc). The system will build a single PDF of the Figures TIFF, PDF, PPT, or EPS submission from the uploaded files. Regardless of how files are uploaded at this Supplement Word stage, all essential components of a manuscript are still required. See Manu- script section, above. In This Issue Feature Word

Word Limits Author Notifications BP:CNNI strictly enforces its word limits when a revised manuscript is submitted. The Journal sends a notification to every individual named as an author upon Needing to address the reviewers’ concerns is not a sufficient reason for receipt of every new submission. This email provides details of the sub- exceeding the stated maximum word limits. We advise authors to critically mission, including the full author list and the text of both the acknowledg- evaluate their manuscripts to ensure that they are written as concisely and clearly ments and disclosures sections. This policy requires valid email addresses for as possible. Additionally, the Journal strongly encourages the use of Supple- all coauthors, which must be supplied at submission; institutional email mental Information. This can be text, tables, and/or figures that are relevant to addresses are strongly preferred. When a consortium/group is named as an the work but not critical to support the findings. Supplemental Information is author, this group must be entered as an author at the relevant screen. An published separately from the main text of the article and therefore does not email address for the primary contact/principal investigator of the consortium/ count against the word limits. group should be supplied. The named individual should be someone responsible for the consortium/group and must be a member of this group. In This Issue Feature The submission of revised manuscripts (except Commentaries and Corre- Referee Suggestions spondence) requires a new unique file with a brief non-technical summary of the For all new submissions (except Commentaries and Correspondence), authors article. The blurb should be uploaded as a text file, 50-75 words in length, and be are required to provide the full names and contact information (affiliation and written in laymen’s terms. Should the article be accepted for publication, this email) of 6 individuals who are especially qualified to referee the work and would summary will be used for the In This Issue feature when the article is published. GUIDE FOR AUTHORS

PEER REVIEW PROCESS Authors should be aware that manuscripts may be returned without outside review when the editors deem that the paper is of insufficient general interest for the broad All submissions (with the general exception of Editorials, Commentaries, and Corre- readership of BP:CNNI, or that the scientific priority is such that it is unlikely to spondence) will be subject to single-blind peer review. The actual selection of reviewers receive favorable reviews. Editorial rejection is done to speed up the editorial will be made by the editors. As a general rule, papers will be evaluated by three or more process and to allow the authors’ papers to be promptly submitted and reviewed independent reviewers and, on occasion, an additional review for statistical adequacy elsewhere. may also be obtained. The comments of the reviewers are generally communicated to the authors within 30-45 days of submission. BP:CNNI is a member of the Neuroscience Peer Review Consortium, an alliance of neuroscience journals that have agreed to accept manuscript reviews from each BP:CNNI excludes reviewers who work at the same institution as any author, or other. Authors may submit a revision of their manuscript to another Consortium those who have any other obvious conflict of interest. The identity of individual journal, and, at the author’s request, BP:CNNI will forward the peer reviews to that reviewers remains confidential to all parties except the Editorial Office. Reviewers are journal. Authors can find a list of Consortium journals and details about forwarding expected to treat manuscripts under peer review with the strictest confidentiality. reviews at http://nprc.incf.org.

EDITORIAL POLICIES Dr. Archimedes reported no biomedical ﬁnancial interests or potential conﬂicts of interest.

Authorship It is the responsibility of all authors to ensure that their conflicts of interest and To qualify for authorship, an individual must have participated sufficiently in the financial disclosures are included in the manuscript. Manuscripts that fail to include work to take public responsibility for all or part of the content, given final approval the complete statements of all authors upon submission will be returned to the of the submitted version, and made substantive intellectual contributions to the corresponding author and will delay the processing and evaluation of the submitted work in the form of: 1) conception and design, and/or acquisition of manuscript. data, and/or analysis of data; and 2) drafting the article, and/or revising it critically for important intellectual content. Authorship also requires agreement to be Ethical Considerations accountable for all aspects of the work in ensuring that questions related to the Authors should consider all ethical issues relevant to their research. In the Methods accuracy or integrity of any part of the work are appropriately investigated and and Materials section of the manuscript, authors should identify the institutional resolved. All individuals who meet criteria for authorship must be named as and/or licensing committee that approved the experiment(s) and confirm that all authors, and all individuals named as authors must meet all authorship criteria. If experiments were performed in accordance with relevant guidelines and regu- authorship is attributed to a group (either solely or in addition to 1 or more lations. Authors of reports on human studies should include detailed information on individual authors), all members of the group must meet the full criteria and the informed consent process, including the method(s) used to assess the sub- requirements for authorship as described above. Any changes in authorship after ject’s capacity to give informed consent, and safeguards included in the study initial submission (additions, deletions, reordering) must be approved in writing by design for protection of human subjects. When relevant patient follow-up data are all authors. available, this should also be reported. When reporting experiments on animals, authors should indicate that institutional and national guidelines for the care and The Journal permits shared/joint authorship in either the first or senior positions. use of laboratory animals were followed. Authors may denote on the title page which authors contributed equally and, BP:CNNI takes seriously its responsibility in ensuring scientific integrity, and will should the article be accepted for publication, a notation will be included in the pursue any allegations of misconduct, including but not limited to plagiarism, published paper. duplicate submission or publication, data fabrication or falsification, unethical treatment of research subjects, authorship disputes, falsified referee suggestions, Corresponding Author and undisclosed conflicts of interest. The Journal generally follows the guidelines By electing to approve and finalize the submission of a manuscript as the recommended by the Committee on Publication Ethics (https://publicationethics. org/), although we also reserve the right to take alternative action(s) as deemed corresponding author, BP:CNNI assumes the author’s acknowledgment and acceptance of the following responsibilities: 1) act as the sole correspondent necessary, including contacting the authors’ institution(s), funding agency, or other with the Editorial Office and the publisher, Elsevier, on all matters related to appropriate authority for investigation. Literature corrections, via errata or retrac- the submission, including review and correction of the typeset proof; 2) tions, are handled on a case-by-case basis. assurance that all individuals who meet the criteria for authorship are included as authors on the manuscript title page, and that the version submitted is the Clinical Trials Registration version that all authors have approved; and 3) assurance that written per- In concordance with the ICMJE, BP:CNNI requires the prospective registration of mission has been received from all individuals whose contributions to the work all clinical trials as a condition of consideration for publication. A clinical trial is are included in the Acknowledgments section of the manuscript. defined as any research study that prospectively assigns human participants or groups to one or more interventions to evaluate the effects of those interventions Although a single person must serve as the corresponding author and be on health-related biomedical or behavioral outcomes. Health-related interventions responsible for the manuscript from submission through acceptance, we do permit are those used to modify a biomedical or health-related outcome; examples two individuals to be named as contacts in the final, published version of a paper. include drugs, surgical procedures, devices, behavioral treatments, dietary inter- This may be noted on the title page of the paper and, should the article be ventions, educational programs and treatment/prevention/diagnostic strategies. accepted for publication, both individuals will be named in the published paper. Health outcomes include any biomedical or health-related measures obtained in patients or participants; examples include pharmacokinetic measures, adverse events, health-related behaviors, and changes to physiological, biological, psy- Disclosure of Biomedical Financial Interests and Potential Conflicts of chological, or neurodevelopmental parameters. Purely observational studies (those Interest in which the assignment of the medical intervention is not at the discretion of the BP:CNNI requires all authors to provide full disclosure of any and all biomedical investigator) will not require registration. financial interests. Further, we require all authors for all article types to specify the nature of potential conflicts of interest, financial or otherwise. This disclosure Trials must have been registered at or before the onset of patient enrollment. includes direct or indirect financial or personal relationships, interests, and affili- Retrospective registration (i.e., at the time of submission) is not acceptable. For all ations relevant to the subject matter of the manuscript that have occurred over the clinical trials and secondary analyses of original clinical trials, the trial name, URL, last two years, or that are expected in the foreseeable future. This disclosure and registration number should be included at the end of the abstract. Acceptable includes, but is not limited to, grants or funding, employment, affiliations, patents registries are ClinicalTrials.gov (https://www.clinicaltrials.gov) or any primary reg- (in preparation, filed, or granted), inventions, honoraria, consultancies, royalties, istries in the World Health Organization International Clinical Trials Registry Plat- stock options/ownership, or expert testimony. This policy of full disclosure is form (http://www.who.int/ictrp/network/primary/en/index.html). similar to the policies of the ICMJE and other such organizations. The conflict of interest statements should be included in the Financial Disclosures section of the Research and Data Reporting Guidelines manuscript at the time of submission for all article types. If an author has nothing to BP:CNNI supports initiatives aimed at improving the reporting of biomedical declare, this must be explicitly stated. Authors should contact the Editorial Office research. Checklists have been developed for a number of study designs, including with questions or concerns, but should err on the side of inclusion when in doubt. randomized controlled trials (CONSORT) and systematic reviews (PRISMA). A The following is a sample text: comprehensive list of reporting guidelines is available from the EQUATOR Network Dr. Einstein reports having received lecture fees from EMC Laboratories, and Library (http://www.equator-network.org). Authors should make use of the research funding from Quantum Enterprises. Dr. Curie disclosed consulting fees appropriate guidelines when drafting their papers. Peer reviewers are asked to refer from RA Inc. Dr. Newton reported his patent on “Newtonian physics”. to these checklists when evaluating these studies. GUIDE FOR AUTHORS

BP:CNNI requires the inclusion of the CONSORT materials (flow diagram and BP:CNNI that employ repository data and/or biomaterials must be in full com- checklist) at submission for all randomized controlled trials. Authors of other pliance with the rules developed by the respective repository governing the correct study designs are encouraged, but not required, to include the relevant citation of the repository, funding agencies, and investigators who contributed to checklists at submission. All such materials will be published as supplemental the repository. Any other stipulation by the repository governing publications using information. repository data and/or biomaterials must also be followed. Authors must provide sufficient information in the manuscript for the Editor and reviewers to determine Materials and Genes that these conditions have been met and that the repository has been established Upon publication, it is expected that authors willingly distribute to qualified and maintained according to current ethical standards. The Editors may require academic researchers any materials (such as viruses, organisms, antibodies, authors to provide additional documentation regarding the repository during the nucleic acids and cell lines) that were utilized in the course of the research and that review process. are not commercially available. GenBank/EMBL accession numbers for primary nucleotide and amino acid sequence data should be included in the manuscript at the end of the Methods and Preprint Policy Materials section. All microarray data (proteomic, expression arrays, chromatin BP:CNNI permits the submission of manuscripts that have been posted on preprint arrays, etc.) must be deposited in the appropriate public database and must be servers, including bioRxiv. However, we request that authors do not update the accessible without restriction from the date of publication. An entry name or posted article to include changes made in response to the reviewers’ comments. accession number must be included in the Methods and Materials section. Authors should disclose that the article has been posted on a preprint server in the Repository Data Acknowledgments/Disclosures section of the paper. If the article is accepted for A growing number of private and public repositories are accumulating demo- publication, authors must be able to transfer copyright to the Society of Biological graphic and clinical data, genetic and genetic analysis data, DNA, and other bio- Psychiatry, or agree to the terms of and pay the associated fee for an open-access materials for use in medical research. Manuscripts submitted for publication in license.

ACCEPTANCE AND PUBLICATION media. For an extra charge, paper offprints can be ordered via the offprint order form which is sent once the article is accepted for publication. Authors may order Proofs offprints at any time via Elsevier’s WebShop (https://webshop.elsevier.com). The corresponding author will receive proofs by email generally within 3-5 weeks of acceptance, which must be corrected and returned within 48 hours of receipt. NIH Public Access Policy and Other Funding Body Agreements Authors are responsible for carefully reviewing and proofreading the entire article As a service to our authors, our publisher, Elsevier, will deposit peer-reviewed for accuracy. Once a corrected proof is published online, additional corrections manuscripts to PubMed Central that have reported research funded by the cannot be made without an erratum. National Institutes of Health (NIH). To initiate this process, the corresponding author must indicate that the study received NIH funding when completing the Publication Schedule Publishing Agreement Form, which is sent to the corresponding author via email Accepted articles are published online, prior to copyediting, within one week of after acceptance. Elsevier has also established agreements and policies with final acceptance. They will be immediately citable, with an assigned digital object multiple other funding bodies, including Wellcome Trust, to help authors comply identifier (DOI) number. Corrected proofs are published online approximately 28 with manuscript archiving requirements. Please see the full details at https://www. business days from final acceptance. Articles generally appear in their final pub- elsevier.com/about/open-science/open-access/agreements. lished form in an issue of the journal within 3-6 months of acceptance. Copyright Press and Embargo Policy Upon acceptance of an article by the Journal, the corresponding author will be The Journal does not typically embargo articles, but can do so in instances where asked to transfer copyright to the Society of Biological Psychiatry on behalf of all authors or their institutions wish to coordinate a press release. Authors should authors. This transfer will ensure the widest possible dissemination of information contact the Editorial Office immediately after notification of an acceptance if they under U.S. Copyright Law. Once accepted, a paper may not be published else- would like an embargo set for their article. where, including electronically, in the same form, in English or in any other lan- Fees guage, without the written consent of the copyright holder. All copies, paper or BP:CNNI does not have publication charges. However, authors may choose to electronic, or other use of information must include an indication of the Elsevier Inc. make their article open-access, for which a fee of $3000 (US dollars) applies. and Society of Biological Psychiatry copyright and full citation of the journal Open-access articles will be made available to all (including non-subscribers) via source. All requests for other uses will be handled through Elsevier Inc. the ScienceDirect platform. Authors of accepted articles who wish to take advantage of this option should complete and submit the online order form sent Authors retain the following rights: 1) Patent and trademark rights and rights to after acceptance. any process or procedure described in the article. 2) The right to photocopy or make single electronic copies of the article for their own personal use, including Cover Art for their own classroom use, or for the personal use of colleagues, provided the BP:CNNI generally selects cover art relevant to an article appearing in that issue. copies are not offered for sale and are not distributed in a systematic way The Journal encourages the submission of scientifically and visually interesting outside of their employing institution (e.g. via an email list or public file server). images that do not appear in the paper, but that would be suitable for cover art, Posting of the article on a secure network (not accessible to the public) within particularly those that summarize or represent the article’s findings. Authors may the author’s institute is permitted. However, if a prior version of this work upload images to be considered for the cover during the submission process, or (normally a preprint) has been posted to an electronic public server, the email them separately to the Editorial Office. Any such images must be the property author(s) agree not to update and/or replace this prior version on the server in of the submitting authors. Figures that appear in the paper are automatically order to make it identical in content to the final published version, and further considered for covers. that posting of the article as published on a public server can only be done with Article Sharing Elsevier’s written permission. 3) The right, subsequent to publication, to use the The Society of Biological Psychiatry and Elsevier support responsible sharing. The article or any part thereof free of charge in a printed compilation of works of corresponding author will, at no cost, receive a customized Share Link (https:// their own, such as collected writings or lecture notes, in a thesis, or to expand www.elsevier.com/about/policies/sharing) providing 50 days free access to the the article into book-length form for publication. Please see the Journal Pub- final published version of the article on ScienceDirect. The Share Link can be used lishing Agreement, sent to the corresponding author via email after acceptance, for sharing the article via any communication channel, including email and social for full details.

QUESTIONS? CONTACT US For questions about the submission or review process, please contact the Editorial Office at [email protected], or by phone at (214) 648-0880. There is also a help menu, accessible from all screens during the submission process.

Updated 6/11/19 Biological Psychiatry: In This Issue CNNI

Volume 5, Number 2, February 2020 A brief summary of the articles appearing in this issue of Biological Psychiatry: Cognitive Neuroscience and Neuroimaging.

Graph Theory Studies of Dysconnectivity in Bipolar and task-based fMRI to investigate the cognitive processes Disorder and neural activity underlying this deficit. The authors found Bipolar disorder (BD) is associated with affective dysregulation that PSZ, compared with healthy volunteers, inferred the and widespread alterations in structural and functional con- environment as overly volatile, a finding that was replicated in a nectivity, but our network-level understanding of brain orga- second, independent sample. Additionally, PSZ showed higher nization in BD remains limited. Using a graph theory approach, activity in the dorsolateral prefrontal cortex related to their Nabulsi et al. (pages 140–151) report that relative to control beliefs about environmental instability. These data provide subjects, euthymic patients with BD exhibited subnetwork insight into the increased choice switching that is often present dysconnectivity involving frontolimbic, frontotemporal, and in PSZ. posterior-occipital functional connections, preserved whole- brain functional connectivity, and comparable structural- Amygdala Subnuclei Structural Connectivity at 7T in functional relationships among whole-brain and edge-based Depression connections. These changes are suggestive of trait-like fea- The amygdala is implicated in mood disorders, but limited tures, which may be necessary to maintain a remitted clinical resolution in human neuroimaging studies has restricted study state in BD. of this diverse structure, often analyzed as a single entity. Cognitive deficits and neuroanatomical network alterations Using ultra-high-resolution 7T diffusion MRI, Brown et al. are present in BD, although knowledge of the relationship (pages 184–193) found increased connection density in 3 of between the two is lacking. Here, McPhilemy et al. (pages the right amygdala nuclei in patients with MDD, compared with 152–162) used anatomical brain network representations, healthy control subjects. These increases were driven by the derived from diffusion-weighted magnetic resonance imaging, uncinate fasciculus and stria terminalis. Decreased connec- to assess patterns of connectivity that were associated with tivity was found in the left medial nucleus. These data identified cognitive processes in adults with BD and healthy individuals. MDD-specific differential changes in the amygdala lateral, The authors found that executive function deficits are related basal, central, and medial substructures, suggesting that to altered global topology and not to highly interconnected hub future studies of MDD should consider separating amygdala regions. Specific subnetwork connectivity patterns supported subregions. spatial memory similarly in both groups and did not relate to deficits in episodic memory, short-term memory, or social Amygdala Activity and Risk for Depression cognition. This study provides insight into specific network- Elevated amygdala activity has been detected in patients with level brain–cognition relationships in BD. depression but also in unaffected relatives with familial risk for depression. It remains unclear whether these amygdala Neural Mechanisms of Altered Reward Processing changes in unaffected relatives reflect the presence of sub- Major depressive disorder (MDD) is associated with altered threshold symptoms or an increased vulnerability to depres- reward processing, which can be restored via enhanced sion. Using fMRI in nondepressed young adults, Barbour et al. dopaminergic signaling, yet how such pharmacological in- (pages 194–202) found that the family history–positive group, terventions affect the underlying neural mechanisms is not compared with the family history–negative group, showed clear. Using a machine learning approach with whole-brain increased amygdala responsivity to moving faces, which functional magnetic resonance imaging (fMRI) data, Liu et al. correlated with low resilience levels, but not subthreshold (pages 163–172) identified reliable brain-wide features that symptoms of depression or anxiety. These data provide a link differentiated unmedicated individuals with MDD and control between elevated amygdala activity and poor resilience in in- subjects with high accuracy. These features, which included dividuals at high familial risk for depression. reward-related striatal activation and connectivity, were largely normalized after a pharmacological challenge that increases Cognition, Behavior, and Brain Connectivity dopaminergic signaling. These findings provide new insights Posttraumatic stress disorder (PTSD) is associated with aber- into the pathophysiology of depression and the mechanism rant emotion processing and altered neural reactivity. In this through which antidepressants may exert their effects at the fMRI study of combat veterans with varying symptom clusters system level. of PTSD, McCurry et al. (pages 203–212) found that hyper- Patients with schizophrenia (PSZ) show increased switching arousal severity is related to decreased habituation to negative between options when making reward-based decisions. Here, relative to neutral images in a widespread neural network, while Deserno et al. (pages 173–183) used computational modeling re-experiencing severity is related to increased habituation.

ª 2019 Society of Biological Psychiatry. 131 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:131–132 www.sobp.org/BPCNNI Biological Psychiatry: CNNI In This Issue

Moreover, greater re-experiencing severity was associated with Untreated Psychosis and Brain Connectivity stronger amygdalar functional connectivity with prefrontal re- Psychosis often goes untreated initially, and this is associated gions for negative relative to neutral images. These data high- with poorer clinical outcomes, but the specific effects on the light the heterogeneous contributions of specific PTSD brain of the duration of untreated psychosis (DUP) is not clear. symptoms to trauma-related emotion dysregulation. Maximo et al. (pages 231–238) conducted a longitudinal Depressed older adults show deficits in executive functions, multimodal imaging study to examine the effect of DUP in including difficulties engaging in goal-directed behaviors and antipsychotic-naïve patients with first-episode psychosis. Re- inhibiting irrelevant stimuli. Respino et al. (pages 213–221) sults revealed that longer DUP is associated with poorer treat- examined the relationship between measures of executive ment response; reduced functional connectivity in default function and homogeneity in the cognitive control network, mode, salience, and executive networks; reduced surface area measured via resting-state fMRI. Relative to healthy control in salience and executive networks; and increased cortical subjects, older adults with late-life depression showed greater thickness in default mode and salience networks. These data regional homogeneity in the bilateral dorsal anterior cingulate provide a neurobiological link between prolonged DUP and cortex and the right middle temporal gyrus. Increased regional poorer clinical outcomes. homogeneity in the dorsal anterior cingulate cortex correlated with better executive function, suggesting that this may serve as a compensatory mechanism for depression-related execu- Regulation of Craving and Negative Emotion in tive dysfunction. Alcohol Use Disorder Obsessive-compulsive disorder (OCD) is associated with Alcohol use disorder has been linked to impairments in pre- poor inhibitory control, which may be related to dysfunction in frontal brain regions that are associated with cognitive control. inhibitory control brain circuitry. Using fMRI during an inhibitory Here, Suzuki et al. (pages 239–250) demonstrate that in- control task, Hampshire et al. (pages 222–230) report that dividuals with alcohol use disorder can effectively use cogni- patients with OCD and their unaffected relatives show reduced tive strategies to regulate craving and negative emotions, and connectivity between different brain regions, including frontal recruit these prefrontal brain regions while doing so. The re- cortical and cerebellar regions, compared with control partic- sults further suggest the presence of both common and ipants. These findings suggest that hypoconnectivity during distinct pathways that are involved in regulating craving and inhibitory control may be a marker of vulnerability for OCD. negative emotions.

132 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:131–132 www.sobp.org/BPCNNI Biological Psychiatry: Commentary CNNI

What Might Prediction Tell Us About the Dopaminergic Mechanisms of Depression?

Michael Browning

We often ask mechanistic questions in the field of biological the D2 receptor is found both pre- and postsynaptically, with a psychiatry. For example, we might investigate the neural single dose of amisulpride acting primarily on the presynaptic mechanisms of a particular treatment by testing whether it autoreceptors, thus enhancing dopaminergic signaling. The alters activity in a specific region of the brain. In studies of this authors use this to test whether, by releasing control of sort, inferential statistics are generally used to determine dopaminergic signaling, amisulpride acts to rescue the pre- whether observed differences, perhaps between treated and sumptive hypoactivation of the dopaminergic system in nontreated participants, are more pronounced than those ex- depressed patients. pected by chance. Of course, not all questions concern The analytic approach used in the study is novel and mechanism; we may also be interested in prediction—for requires some scrutiny. Three classifiers were developed to example, whether a patient’s functional magnetic resonance distinguish, pairwise, between the 3 participant groups. The imaging (fMRI) signal predicts his or her response to a treat- features used in the classifiers were mainly regression ment. A different statistical framework is used in studies of this weights from analysis of the fMRI data, averaged within sort: generally a classifier is developed linking a set of pre- Automated Anatomical Labeling–defined anatomical re- dictors (e.g., fMRI data) to an outcome (e.g., treatment gions. (A few other features derived from individually response), with the performance of the classifier defined as its determined anatomical masks and previous connectivity accuracy when predicting response in a separate group of analyses were used, but these did not substantially influ- participants (1). Despite the distinct statistical approaches, ence the results.) The “ground truth” the classifiers were there is a straightforward link between questions of mecha- predicting was the group membership of individual partici- nism and prediction; by definition, the mechanism of a treat- pants (e.g., was a patient in the MDD-amisulpride vs. the ment leads to its response, so if we understand (and can MDD-placebo group). The feature selection process used measure) a treatment’s mechanism we are likely to be able to for the classifiers involved several phases: first, an elastic predict its response. In this issue of Biological Psychiatry: net procedure (which discards the less-predictive features) Cognitive Neuroscience and Neuroimaging, Liu et al. (2) sug- was used. Next, the features from the elastic net analysis gest that we can also use this link to draw inferences about the were rank ordered by their weight (how influential they were mechanism of a treatment. Specifically, they suggest that the in prediction). Last, a final classifier was selected using the effects of a treatment on pathological processes can be n most informative features, where n was a number that inferred by comparing the performance of a series of predictive produced the highest overall classification accuracy while classifiers trained to distinguish different groups of patients or being less than the number of participants included in the control subjects. analysis. The focus of the Liu et al. study (2) is the role of the dopa- The basic logic underpinning the analysis is conceptually minergic system in depression, a topical and somewhat similar to that used for representational analysis in cognitive controversial area. Building on a well-developed preclinical neuroscience (7); for a brain region to be useful in discrimi- literature that has described the role of central dopaminergic nating between 2 groups of participants, something in the signaling in learning about rewarding outcomes (3), there has activity of that region (or the interaction between that and other been interest in whether symptoms of depression generally, and regions) must differ between the groups. The slightly unusual anhedonia in particular, may be linked to dysfunction of the aspect of the current analysis is that it is based on the relative dopaminergic system (4). However, evidence in favor of this importance of the features used in the classification, rather proposal has been mixed, and there have been a number of than on classification accuracy itself (7). As the features recent negative studies (5,6). On this background, Liu et al. (2) themselves were derived from anatomically defined brain re- reanalyzed data from 2 studies in which participants completed gions, the authors used the ranking of feature importance to the monetary incentive delay task, a well-validated model of draw conclusions about the specific regions involved in reward responsivity, while fMRI data were collected. Their treatment response. The results of the studies and the authors’ analysis involved 3 groups of participants: patients with major interpretations are summarized in Table 1. depressive disorder (MDD) who were receiving a single dose of Overall, the authors suggest that their analysis demon- placebo (MDD-placebo), patients with MDD receiving a single strates that amisulpride normalizes reward-related activity in dose of the D2/D3 antagonist amisulpride (MDD-amisulpride), depressed patients; that is, it makes the brain activity of and healthy control subjects who received placebo. The depressed patients more like the brain activity of control psychopharmacological aspect of the study is sophisticated; subjects.

SEE CORRESPONDING ARTICLE ON PAGE 163 https://doi.org/10.1016/j.bpsc.2019.12.009 ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. 133 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:133–134 www.sobp.org/BPCNNI Biological Psychiatry: CNNI Commentary

Table 1. Summary of Key Study Findings and Interpretation Result Description Interpretation A The classifiers trained to distinguish the Amisulpride acts to change activity in the same MDD-amisulpride from the MDD-placebo regions of the brain where activity differs groups and the MDD-placebo from the HC- between nondepressed and depressed people placebo groups selected similar features B The classifier trained to distinguish The effect of amisulpride in depressed patients is MDD-amisulpride from HC-placebo groups to render the activity of the regions it influences selected a largely different set of features (i.e., the set of regions defined in A above) similar than the other 2 classifiers to that seen in nondepressed HC subjects (as these regions are no longer useful in distinguishing between groups) C The sign of the classifiers’ weights was The quantitative difference between depressed patients similar for the 2 classifiers that involved taking placebo vs. amisulpride is similar to the quantitative the MDD-placebo group difference between depressed patients and HC subjects. The effect of amisulpride is quantitatively similar to the effect that would be expected if a depressed patient stopped being depressed HC, healthy control; MDD, major depressive disorder.

The article illustrates an interesting avenue by which pre- Acknowledgments and Disclosures dictive analyses might be leveraged to draw mechanistic This work was supported by Clinician Scientist Fellowship (Grant No. MR/ inference from between-subject psychopharmacological N008103/1) from the Medical Research Council and by the National Institute studies. A potential advantage of this approach is that multi- for Health Research Oxford Health Biomedical Research Centre. variate classifiers are able capture complex interactions within MB has received travel expenses from Lundbeck for attending confer- data that are difficult to detect using traditional massed uni- ences and has acted as a consultant for Janssen Research and the Centre for Human Drug Research. variate approaches (7). Many clinically relevant neural processes, including reward processing, are distributed across brain regions and thus may be more accurately measured Article Information using multivariate metrics. Having said that, several open From the Department of Psychiatry, University of Oxford, and the Oxford questions remain regarding this approach to analysis. First, it is Health National Health Service Trust, Warneford Hospital, Oxford, United not clear how to quantify the evidence provided by the relative Kingdom. ranking of features in a predictive classifier. As illustrated in Address correspondence to Michael Browning, D.Phil., M.B.B.S., Department of Psychiatry, University of Oxford, Warneford Hospital, Oxford Supplemental Figure S1 in Liu et al. (2), the benefit to classi- OX3 7JX, United Kingdom; E-mail: [email protected]. fication accuracy of adding predictors asymptotes quickly— Received Dec 13, 2019; accepted Dec 13, 2019. meaning that a small change in a feature, such as that produced by measurement noise, may lead to a large change in its ranking. Given this, it may be useful to validate the approach References used in this article in a population with a well-characterized 1. Bzdok D, Ioannidis JPA (2019): Exploration, inference, and prediction in cerebral pathology. For example, an analysis of patients with neuroscience and biomedicine. Trends Neurosci 42:251–262. Parkinson’s disease who are on and off dopaminergic 2. Liu Y, Admon R, Mellem MS, Belleau EL, Kaiser RH, Clegg R, et al. medication might be expected to show a similar pattern of (2020): Machine learning identifies large-scale reward-related activity results, with the effects driven by the signal from striatal modulated by dopaminergic enhancement in major depression. Biol Psychiatry Cogn Neurosci Neuroimaging 5:163–172. regions. Second, the benefit of multivariate approaches to 3. Schultz W, Dayan P, Montague PR (1997): A neural substrate of pre- analysis in terms of capturing complex interactions tends to be diction and reward. Science 275:1593–1599. offset by costs in terms of interpretability. Thus, it is difficult to 4. Pizzagalli DA (2014): Depression, stress, and anhedonia: draw conclusions about the role played in reward processing Toward a synthesis and integrated model. Annu Rev Clin Psychol or depressive symptomatology by the regions identified in the 10:393–423. current analysis. It is thus likely to be necessary to follow up 5. Schneier FR, Slifstein M, Whitton AE, Pizzagalli DA, Reinen J, McGrath PJ, et al. (2018): Dopamine release in antidepressant-naive multivariate analyses such as this with detailed, hypothesis-led major depressive disorder: A multimodal [11C]-(1)-PHNO positron studies of the neural systems identified. emission tomography and functional magnetic resonance imaging In conclusion, while inferential and predictive analyses ask study. Biol Psychiatry 84:563–573. distinct questions, they can provide complementary informa- 6. Rutledge RB, Moutoussis M, Smittenaar P, Zeidman P, Taylor T, tion on mechanistic processes. Innovative analytical ap- Hrynkiewicz L, et al. (2017): Association of neural and emotional proaches that weave both techniques together, as described impacts of reward prediction errors with major depression. JAMA Psychiatry 74:790–797. by Liu et al. (2), may allow us to better understand the funda- 7. Kriegeskorte N, Kievit RA (2013): Representational geometry: mental mechanisms relevant to psychiatry and thus guide the Integrating cognition, computation, and the brain. Trends Cogn Sci development of much-needed novel treatments. 17:401–412.

134 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:133–134 www.sobp.org/BPCNNI Biological Psychiatry: Commentary CNNI

Of Forests and Trees: Bridging the Gap Between Neurobiology and Behavior in Posttraumatic Stress Disorder

Christopher L. Averill, Lynnette A. Averill, Siyan Fan, and Chadi G. Abdallah

The puzzles that constitute normal science exist only because no The authors evaluated the relationship between PTSD and paradigm that provides a basis for scientific research ever habituation to negative affective stimuli in a large sample of completely resolves all its problems. combat-exposed veterans with and without PTSD and also in a –Thomas Kuhn civilian comparison group. Though finding no significant difference in habituation as an effect of categorical criteria (PTSD diagnosis, overall PTSD severity, combat exposure, or veter- All models are wrong, but some are useful. ans vs. civilians), the authors did find significant trans- –George Box diagnostic effects when evaluating symptom clusters Evidence has linked chronic or traumatic stressors with independently from PTSD diagnosis or overall severity. Hy- impaired synaptic plasticity and strength, structural alterations, perarousal and re-experiencing symptoms were related, in and the disturbance of intrinsic functional connectivity patterns opposing directions, to habituation to aversive stimuli in a in networks thought to be related to the regulation of cognitive network of prefrontal and limbic brain regions. They found that and emotional attention/control (1,2). Like purported neural hyperarousal severity drove sensitization while re-experiencing correlates of severe stress exposure, the associated psycho- drove neural habituation, and more specifically, the pattern of social complications are quite broad, negatively impacting disparity between the two clusters drove habituation/sensiti- almost every aspect of functioning. Posttraumatic stress dis- zation (4). order (PTSD) is a diagnosis that is intended to encapsulate this Earlier work using a similar task reported significant diag- extreme heterogeneity under a single construct. The DSM-5 nostic and dimensional effects of PTSD and sometimes clusters 20 behaviorally based symptoms under four, often recruited participants based on elevated physiological and intercorrelated criteria with high comorbidity and that overlap affective responses. The current null findings highlight poten- other psychiatric disorders (1,2). As research examining the tial issues in the heterogeneity of phenotypes that may result psychoneurobiology of chronic stress pathology and PTSD from broad recruitment of participants with PTSD rather than has grown, and despite well-designed and well-executed strategic recruitment of relevant phenotypic groups. The au- pioneering work characterizing the neurobiopsychosocial un- thors also suggest that negative image content/context may derpinnings of the disorder, the field still lacks reproducible be a crucial consideration (4). Specifically, neural activation biomarkers, which leaves us with mechanistic and clinical based on habituation or other cognitive–emotional states is confusion (1,2). likely to vary significantly, such that negative images that are Though variability in methodology and study design not personally relevant may be less provocational than those contribute to the lack of consistent findings, it is likely that the that are specifically selected to be trauma- or threat-specific. larger culprit is the amorphous nature of the disorder and its Another interesting finding from McCurry et al. (4)is lack of a “one-size-fits-all” presentation. With more than increased functional connectivity of the amygdala with 636,000 symptom combinations that meet DSM-5 PTSD emotion suppression regions in relation to re-experiencing classification criteria (3), it is not surprising that we struggle to symptoms. This finding mirrors another study in which over- reproduce findings from previous studies. The odds of inde- all PTSD severity was related to anterior hippocampus and pendent samples of individuals having similar PTSD symptom dorsal amygdala abnormalities, but re-experiencing was pri- and neurobiological phenotypes through random recruitment marily related to the shape of the amygdala and hyperarousal are infinitesimal. It is possible that important findings repre- related to hippocampal shape (5). Though a structural rather senting specific phenotypes or profiles of chronic stress pa- than functional imaging analysis, the study by Averill et al. thology, or PTSD pathology, may be lost in the averages. We differentially associated PTSD and depression symptom se- may be so focused on the big picture that we fail to notice the verities with specific subfields of the hippocampus (6). fine brush stokes that comprise it. Together, such findings suggest a pattern not only of func- In this issue of Biological Psychiatry: Cognitive Neurosci- tional connectivity but also of altered anatomy related specif- ence and Neuroimaging, exciting work from McCurry et al. (4) ically to these phenotypes. When combined with McCurry exemplifies these challenges well and, though not their intent, et al.’s results (or lack thereof), their use of symptom clusters, highlights ways to improve study design and methods that and the way they created an amygdala seed (almost creating may increase the likelihood of discovering robust biomarkers. their own “subfield” using significant voxels within an

SEE CORRESPONDING ARTICLE ON PAGE 203 https://doi.org/10.1016/j.bpsc.2019.12.010 ª 2019 Published by Elsevier Inc on behalf of Society of Biological Psychiatry. 135 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:135–137 www.sobp.org/BPCNNI Biological Psychiatry: CNNI Commentary

amygdala mask), much of the recent work in the field highlights As the field has matured, the questions we ask have potential new value in diving deeper and breaking down increased in complexity, and the way we have classified monolithic classifications of behavior and biology into more groups is now perhaps too gross and nonspecific to be use- multidimensional, modular, precise subunits. fully informative; this has led us into a reproducibility crisis, McCurry et al. (4) used symptom cluster scores to shed light which may well require not only a transdiagnostic paradigm on a relationship that was masked when investigating the shift but also the proliferation of new methods and tools [such higher order factor, PTSD. This is far from the first paper to as hierarchical modularity atlases (9)] that will allow us to investigate PTSD at the symptom cluster level; however, the design studies that evaluate data at different levels of simultaneous presentation of diagnostic and transdiagnostic complexity and according to the needs of the question at approaches presents an opportunity to reflect on the need to hand. When considering modularity, composition, separation, evolve our practice of research and highlights the potential and feasibility, for example, while we do not want the value in considering models of chronic stress pathology that complexity of more than 600,000 symptom profiles, useful turn traditional classification criteria on their head, leveraging biomarkers for treatment and research may be buried in some biological models to examine behavior, rather than the trans- combination of PTSD phenotypes. The DSM-IV split PTSD into diagnostic behavior-to-biology approaches that have domi- 3 symptom clusters, and the DSM-5 further split arousal into nated the literature to date (1,2). This biology-first approach anxious and dysphoric arousal clusters. Other investigators has significant potential to support the revolution of precision have suggested a potential 7-factor model of PTSD (10)inan medicine, supporting informed individualized treatment plan- attempt to further increase the granularity of our approach and ning based on biological markers, the development of novel understanding. As with subcortical subfields/nuclei (6) and with therapeutics, and the identification of individuals who may 1) intrinsic connectivity networks (2,9), there is the possibility of a be at risk for the onset or worsening of certain symptom hierarchy of symptom clusters, and the level of complexity constellations or 2) be more or less likely to respond to a given required to evaluate the relevant signal may depend on the treatment/intervention based on their neurobiological profile. research question. One way or another, future neuro- The traditional approaches may unintentionally create sce- biomarkers will likely be found neither by total scores, nor by narios where signals of interest are washed out by the comparing thousands of phenotypes, but by adapting to new confound of differing phenotypes averaged across the sample. paradigms, optimizing signal for the research question, and In addition, unidimensional study designs in PTSD research thinking hierarchically about both clinical and biological phe- may sometimes be fundamentally flawed by not focusing on notypes and methods. We have spent a lot of time as a field the subset of symptoms that contain the signal of interest. looking at the big picture, and we have done excellent work Evidence-based psychotherapy and pharmacotherapy for exploring the circuitry issues that have been so central to un- PTSD vary in effectiveness based on nuances in clinical and derstanding the forest, so to speak—now we must spend more neurobiological presentations that are not yet fully elucidated. time looking for detail in the trees so we can begin to develop McCurry et al. (4) note that exaggerated hyperarousal is biology-first, transdiagnostic models for the enhancement of associated with nonresponse to treatment and less longitudi- precision medicine. nal symptom improvement (7). The relationship between hyperarousal and re-experiencing may have the capacity to improve treatment outcomes through increased potential for Acknowledgments and Disclosures habituation in those with increased re-experiencing (4). We thank the Department of Veterans Affairs, the National Center for Abdallah et al. (8) recently reported that two evidence-based Posttraumatic Stress Disorder, the National Institute of Mental Health, the treatments, cognitive processing therapy and present- Association for Suicide Prevention, the Brain and Behavior Foundation/ centered therapy, were associated with differing patterns of National Alliance for Research on Schizophrenia and Depression, and the neural alteration as a result of similarly successful treatment. Robert E. Leet and Clara M. Guthrie Patterson Trust. Cognitive processing therapy, which focuses significantly on CGA has served as a consultant, speaker, and/or advisory board member for Genentech, Janssen, Lundbeck, and FSV7; is the editor of addressing maladaptive cognitions, impaired decision making, Chronic Stress for Sage Publications, Inc.; and has filed a patent for using and trauma interpretations, decreases symptom severity and mammalian target of rapamycin complex 1 inhibitors to augment the effects increases functional connectivity of the central executive of antidepressants (filed Aug 20, 2018). CLA, LAA, and SF declare no network. Present-centered therapy focuses patients on the biomedical financial interests or potential conflicts of interest. here-and-now and demonstrated reduced functional connectivity of the salience network, perhaps by reducing the salience of traumatic memories and worry about the future (8). Etkin Article Information et al. (2) used functional magnetic resonance imaging to From the Clinical Neurosciences Division (CLA, LAA, SF, CGA), National identify a neurobehavioral phenotype of PTSD characterized Center for Posttraumatic Stress Disorder, Department of Veterans Affairs, by impaired verbal memory recall and resting-state functional West Haven, and the Department of Psychiatry (CLA, LAA, SF, CGA), Yale connectivity in the ventral affective network. Mechanistically School of Medicine, New Haven, Connecticut. and clinically meaningful neurobiological phenotypes in PTSD CLA and LAA contributed equally to this work. may be elucidated by strategically sampling based on the Address correspondence to Christopher L. Averill, B.S., Clinical Neuro- sciences Division, National Center for Posttraumatic Stress Disorder, specific phenotypes or characteristics implicated in our hy- Department of Veterans Affairs, 950 Campbell Avenue, 151E, West Haven, potheses rather than simply recruiting patients with PTSD CT 06501; E-mail: [email protected]. regardless of phenotype. Received Dec 13, 2019; accepted Dec 13, 2019.

136 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:135–137 www.sobp.org/BPCNNI Biological Psychiatry: Commentary CNNI

References 6. Averill CL, Satodiya RM, Scott JC, Wrocklage KM, Schweinsburg B, Averill LA, (2017): Posttraumatic stress disorder and depression 1. AbdallahCG,AverillLA,AkikiTJ,RazaM,AverillCL,GomaaH, et al. symptom severities are differentially associated with hippocampal (2019): The neurobiology and pharmacotherapy of post- et al. subfield volume loss in combat veterans [published online ahead of traumatic stress disorder. Annu Rev Pharmacol Toxicol 59: print Dec 13]. Chronic Stress (Thousand Oaks). 171–189. 7. Stein NR, Dickstein BD, Schuster J, Litz BT, Resick PA (2012): Tra- 2. EtkinA,Maron-KatzA,WuW,FonzoGA,HuemerJ,VertesPE, jectories of response to treatment for posttraumatic stress disorder. et al. (2019): Using fMRI connectivity to define a treatment-resistant Behav Ther 43:790–800. form of post-traumatic stress disorder. Sci Transl Med 11: 8. Abdallah CG, Averill CL, Ramage AE, Averill LA, Alkin E, Nemati S, eaal3236. et al. (2019): Reduced salience and enhanced central executive con- 3. Galatzer-Levy IR, Bryant RA (2013): 636,120 ways to have post- nectivity following PTSD treatment [published online ahead of print Apr traumatic stress disorder. Perspect Psychol Sci 8:651–662. 15]. Chronic Stress (Thousand Oaks). 4. McCurry KL, Frueh BC, Chiu PH, King-Casas B (2020): Opponent ef- 9. Akiki TJ, Abdallah CG (2019): Determining the hierarchical architecture fects of hyperarousal and re-experiencing on affective habituation in of the human brain using subject-level clustering of functional net- posttraumatic stress disorder. Biol Psychiatry Cogn Neurosci Neuro- works. Sci Rep 9:19290. imaging 5:203–212. 10. Pietrzak RH, Tsai J, Armour C, Mota N, Harpaz-Rotem I, 5. Akiki TJ, Averill CL, Wrocklage KM, Schweinsburg B, Scott JC, Southwick SM (2015): Functional significance of a novel 7-factor Martini B, et al. (2017): The association of PTSD symptom severity with model of DSM-5 PTSD symptoms: Results from the National Health localized hippocampus and amygdala abnormalities [published online and Resilience in Veterans study. J Affect Disord 174: ahead of print Aug 8]. Chronic Stress (Thousand Oaks). 522–526.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:135–137 www.sobp.org/BPCNNI 137 Biological Psychiatry: CNNI Commentary

Functional Connectivity and Cognitive Control in Late-Life Depression

Julie A. Dumas

One of the most exciting advances in neuroscience in the last analysis, while other study designs are better examined with a decade is the potential to use magnetic resonance imaging to data-driven approach similar to independent component understand brain function and dysfunction. As the science has analysis. Given the number of ways that resting-state data can advanced from examining single activated brain areas to be examined, it is no surprise that studies find different pat- interacting networks reaching across wide brain regions, so terns of connectivity depending on the questions being has the complexity of analysis and interpretation. Creating a examined and how the data are probed. As a result, the clinical coherent picture of normal brain functioning and of disordered utility of resting-state data and functional connectivity data brain functioning is a difficult task that investigators work to needs additional investigation. disentangle to help explain the neurobiology of psychiatric In the current issue of Biological Psychiatry: Cognitive disease. One example of applying a state-of-the-art analysis Neuroscience and Neuroimaging, Respino et al. (7) make some approach to understanding the neurobiology of a psychiatric important progress in attempting to detail the relationships disease can be seen in studies of late-life depression (LLD)— between connectivity patterns and cognition in older adults major depressive disorder that occurs in adults $60 years of with and without LLD. They aimed to move the literature for- age. Many older adults with LLD may also have cognitive ward by comparing two different measures of functional con- impairment in executive functions in general and in cognitive nectivity—regional homogeneity (ReHo) and network control in particular. Older adults with LLD and cognitive homogeneity (NeHo)—and by using both hypothesis-driven control impairments often respond differentially to treatment and data-driven analyses. They described their intent to (1). Even if the depression is treated, older adults often still assess the benefits of each method of analysis while mini- have the cognitive impairment (1). It would be beneficial to the mizing the pitfalls. For example, defining hypothesis-driven field and to individuals with LLD to understand the neurobi- regions of interest still has the potential to miss important ology underlying LLD and the associated cognitive changes connectivity relationships that have not yet been observed, that may eventually have implications for individualized treat- while using a purely data-driven approach ignores what is ment recommendations. However, the field is still in search of a already known about specific depression-related and age- biomarker with good predictive power regarding tailoring related brain network functions. Even with this variety of ap- treatments to individuals based on functional brain imaging proaches to the analysis, the results produced a consistent measures. pattern that further detailed the underlying relationships be- Recent studies of older adults with LLD have examined tween altered connectivity and cognition in older adults with resting-state functional connectivity using functional magnetic and without LLD (7). resonance imaging. Assessing the blood oxygen level– The results showed that the ReHo measure differentiated dependent signal during rest allows for an examination of the patients with depression from adults without depression with intrinsic functional connectivity of the brain when it is not greater ReHo in the bilateral dorsal anterior cingulate cortex performing an explicit cognitive task (2). However, even during (dACC) and the middle temporal gyrus (MTG). ReHo is a task performance, studies have shown that the intrinsic measure of temporal synchronization of the blood oxygen network activity is maintained (3). Thus, investigators have level–dependent time series of every voxel and its nearest attempted to use patterns of functional connectivity as bio- neighbor and provides an assessment of local homogeneity markers of LLD and other psychiatric disorders. Studies have within a network. Patients with depression had greater ReHo examined network abnormalities in adults with LLD compared compared with participants without depression. The relation- with healthy older adults (4) to detail the effects of depression ship between ReHo in the dACC and MTG and performance on on a variety of network metrics. Additional studies have the cognitive tasks that showed group differences were then examined the effects of different medications and treatment for examined within each group separately. Relationships were depression (5,6). What complicates the process is that there observed between ReHo in the dACC and the Trail-Making are a number of methods of analysis of functional connectivity; Test B and Digit Span Test for the patients with depression. thus far, no method appears to be superior to others in terms No relationships were observed between cognition and the of reliability from study to study with different participants or in MTG region for either group, and no relationships were its relationship to cognitive measures. In addition, it is possible observed for the participants without depression. Given that to explore connectivity by looking within a network or between these correlations were positive, the authors hypothesized that networks across much of the brain. Some investigators argue the increased ReHo may be a compensatory response for the for hypothesis-driven region of interest–based connectivity cognitive control difficulties related to LLD. Next, in an

SEE CORRESPONDING ARTICLE ON PAGE 213

138 ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.bpsc.2019.12.014 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:138–139 www.sobp.org/BPCNNI ISSN: 2451-9022 Biological Psychiatry: Commentary CNNI

exploratory analysis, they used the dACC and MTG as seeds future studies. The neuroscience of LLD is advancing in a way to examine connectivity with the rest of the brain and found that is exciting for neuroscientists exploring the effect of LLD that the patients with depression had decreased connectivity on brain function and has the potential to reveal biomarkers between the dACC and the bilateral precuneus. Overall, the that may soon be useful for making treatment decisions. This ReHo findings were interpreted as increased ReHo in the also represents an advance in the neuroscience of LLD that is dACC as a compensatory response to the cognitive challenges hopeful for those with the illness. of depression while the decreased connectivity between the dACC and precuneus revealed the impact of depression on long-range brain connectivity. Both the dACC and the pre- Acknowledgments and Disclosures cueus are parts of the cognitive control network that have been This work was supported by National Institute on Aging Grant Nos. R01 AG050716, R56 AG062105, and R01 AG066159. demonstrated to be negatively impacted by LLD. Interestingly, The author reports no biomedical financial interests or potential conflicts while there were ReHo differences between the groups in the of interest. MTG, there were no relationships between ReHo in this region and cognitive performance or connectivity with other brain regions. Overall, these patterns help refine the specificity of the Article Information cortical signature of LLD as measured by these resting-state From the Department of Psychiatry, Larner College of Medicine, University metrics. of Vermont, Burlington, Vermont. NeHo is a measure of connectivity between a voxel and Address correspondence to Julie A. Dumas, Ph.D., Department of Psy- every other voxel in the same network, and it assesses within- chiatry, Larner College of Medicine, University of Vermont, 1 South Prospect St., Burlington, VT 05401; E-mail: [email protected]. network connectivity in an unbiased way. Respino (7) et al. Received Dec 18, 2019; accepted Dec 19, 2019. found no differences between the patients with LLD and the participants without LLD in the NeHo measure. While examining NeHo across all voxels in a network allows for an unbiased assessment of connectivity, they interpreted this result to References indicate that averaging across the whole network may have 1. Alexopoulos GS, Kiosses DN, Heo M, Murphy CF, Shanmugham B, masked subtle differences that may exist between the two Gunning-Dixon F (2005): Executive dysfunction and the course of groups in this study. geriatric depression. Biol Psychiatry 58:204–210. Overall, these data showed patterns of connectivity that 2. Damoiseaux JS, Rombouts SA, Barkhof F, Scheltens P, Stam CJ, Smith SM, Beckmann CF (2006): Consistent resting-state networks were specific to the patients with LLD and that were not seen across healthy subjects. Proc Natl Acad Sci U S A 103:13848–13853. in the older adults without LLD. Thus, the findings are not 3. Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE, et al. simply an effect of aging on the brain and appear to be the (2009): Correspondence of the brain’s functional architecture during effects of depression on brain functioning in older adults. The activation and rest. Proc Natl Acad Sci U S A 106:13040–13045. results tell a concise story that is a result of the ancillary an- 4. Tadayonnejad R, Ajilore O (2014): Brain network dysfunction in late-life alyses where Respino (7) showed that the results are depression: A literature review. J Geriatr Psychiatry Neurol 27:5–12. et al. 5. Alexopoulos GS, Manning K, Kanellopoulos D, McGovern A, Seirup JK, specific to the patients with LLD and the correlations with Banerjee S, Gunning F (2015): Cognitive control, reward-related deci- cognition did not appear on cognitive tests that did not sion making and outcomes of late-life depression treated with an an- differentiate the two groups. Thus, the secondary analyses tidepressant. Psychol Med 45:3111–3120. give the reader confidence that the patterns observed are 6. Andreescu C, Tudorascu DL, Butters MA, Tamburo E, Patel M, Price J, specific to patients with LLD. While the current study presents et al. (2013): Resting state functional connectivity and treatment a consistent picture of the examination of these resting-state response in late-life depression. Psychiatry Res 214:313–321. 7. Respino M, Hoptman MJ, Victoria LW, Alexopoulos GS, Solomonov N, metrics used to differentiate patients with LLD from partici- Stein AT, et al. (2020): Cognitive contol network homogeneity and ex- pants without LLD, confidence in the further application of ecutive functions in late-life depression. Biol Psychiatry Cogn Neurosci these findings will be increased as the results are replicated in Neuroimaging 5:213–221.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:138–139 www.sobp.org/BPCNNI 139 Biological Psychiatry: CNNI Archival Report

Frontolimbic, Frontoparietal, and Default Mode Involvement in Functional Dysconnectivity in Psychotic Bipolar Disorder

Leila Nabulsi, Genevieve McPhilemy, Liam Kilmartin, Joseph R. Whittaker, Fiona M. Martyn, Brian Hallahan, Colm McDonald, Kevin Murphy, and Dara M. Cannon

ABSTRACT BACKGROUND: Functional abnormalities, mostly involving functionally specialized subsystems, have been associated with disorders of emotion regulation such as bipolar disorder (BD). Understanding how independent functional subsystems integrate globally and how they relate with anatomical cortical and subcortical networks is key to understanding how the human brain’s architecture constrains functional interactions and underpins abnormalities of mood and emotion, particularly in BD. METHODS: Resting-state functional magnetic resonance time series were averaged to obtain individual functional connectivity matrices (using AFNI software); individual structural connectivity matrices were derived using deterministic non-tensor-based tractography (using ExploreDTI, version 4.8.6), weighted by streamline count and fractional anisotropy. Structural and functional nodes were defined using a subject-specific cortico-subcortical mapping (using Desikan-Killiany Atlas, FreeSurfer, version 5.3). Whole-brain connectivity alongside a permutation-based statistical approach and structure–function coupling were employed to investigate topological variance in individuals with predominantly euthymic BD relative to psychiatrically healthy control subjects. RESULTS: Patients with BD (n = 41) exhibited decreased (synchronous) connectivity in a subnetwork encompassing frontolimbic and posterior-occipital functional connections (T . 3, p = .048), alongside increased (antisynchronous) connectivity within a frontotemporal subnetwork (T . 3, p = .014); all relative to control subjects (n = 56). Preserved whole-brain functional connectivity and comparable structure-function coupling among whole-brain and edge-class connections were observed in patients with BD relative to control subjects. CONCLUSIONS: This study presents a functional map of BD dysconnectivity that differentially involves communication within nodes belonging to functionally specialized subsystems—default mode, frontoparietal, and frontolimbic systems; these changes do not extend to be detected globally and may be necessary to maintain a remitted clinical state of BD. Preserved structure–function coupling in BD despite evidence of regional anatomical and functional deficits suggests a dynamic interplay between structural and functional subnetworks. Keywords: Bipolar disorder, Functional connectivity, Graph theory, Psychosis, rs-fMRI, Structure–function coupling https://doi.org/10.1016/j.bpsc.2019.10.015

Bipolar disorder (BD) is a major psychiatric condition associ- (4). However, a priori investigations reported local patterns of ated with widespread dysconnectivity thought to arise from functional dysconnectivity within the amygdala and the pre- changes in integration and segregation within its brain net- frontal and cingulate cortices in euthymic BD (4) and in female works (1). Neuroimaging work including diffusion and func- patients with BD (5). Additionally, instabilities within the DMN tional magnetic resonance imaging (fMRI) has provided are present among individuals with BD with a positive history evidence of altered patterns of neuroanatomical and functional of psychosis, and these changes may persist in patients in connectivity, collectively suggesting that affective dysregula- remission (4). Furthermore, opposing spatiotemporal patterns tion associated with BD may arise from both structural and observed within the DMN and frontoparietal network may un- functional changes primarily involving neural circuitries derpin depressive and manic episodes of BD (6). Moreover, responsible for emotion regulation, cognitive control, and ex- along with evidence of dysconnectivity within the limbic sys- ecutive functions (2,3). tem there is evidence involving reward system–related Functional features of euthymic BD include preserved structures (7–9). Additionally, abnormalities within regions whole-brain functional connectivity of the default mode anatomically connecting with the limbic system such as the network (DMN), frontoparietal network, and salience network prefrontal cortex have been linked to features of emotional and

140 ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI ISSN: 2451-9022 Biological Psychiatry: Subnetwork Functional Dysconnectivity in BD CNNI

cognitive control in BD (9,10). Although these a priori in- subnetwork functional connectivity as they relate to the illness- vestigations may have been led by morphological study find- affective dysregulation. We anticipated preserved whole-brain ings and may be task and mood dependent, they suggest that functional connectivity and aberrant connectivity patterns functional impairments of BD are not confined to the DMN but involving nodes belonging to limbic and DMN systems, as pre- rather extend to involve emotion regulatory centers. These viously highlighted by the literature (4), in patients with BD relative localized functional changes may constitute a compensatory to control subjects. Furthermore, we explored the relationship mechanism of neural activity that underlies the general stability between functional interactions and previously identified struc- observed across resting-state fMRI networks in euthymic tural (white matter) abnormalities in patients with BD relative to subjects with BD; these changes may be necessary to sustain control subjects (14), at the whole-brain and subnetwork levels. a remitted clinical state of the illness. Findings from neuroimaging investigations demonstrate that BD is unlikely to arise from changes involving brain regions METHODS AND MATERIALS in isolation; rather, the clinical syndrome that we currently refer Participants to as BD may originate from a disruption of the brain’s struc- Patients included in this study overlap by 88% ( = 36 sub- tural and functional neurocircuitries (2) and perhaps the rela- n tionship dynamics between these neurocircuitries. The jects) with those in a previously presented structural connectivity analysis from our research group (15). Participants 18 to application of graph theory methods to neuroscience has allowed a network-level understanding of the cortico- 65 years of age were recruited from the western regions of Ireland’s Health Services via referral (outpatients) or public subcortical organization of the brain, specifically how independent functional subsystems integrate in global processing advertisement (patients and control subjects). A diagnosis of BD was confirmed using the DSM-IV-TR Structured Clinical streams, particularly in disorders of mood and emotions. In- vestigations of structural and functional dysconnectivity have Interview for DSM Disorders (American Psychiatric Associa- tion, 1994 version) conducted by an experienced psychiatrist. been increasingly implemented; however, a paucity of studies to date have applied graph theoretical tools to understand the Euthymia was defined during medical screening and at MRI scanning using the 21-item Hamilton Depression Rating Scale functional organization of BD in a network-like fashion (Table 1). Collectively, these studies report weak global effects (HDRS-21), score ,8; and Young Mania Rating Scale (YMRS), score ,7. Anxiety signs and symptoms were defined using the (not surviving multiple comparisons) but localized changes involving fronto-temporo-parietal and limbic nodes. These ef- Hamilton Anxiety Rating Scale (HARS), score ,18. Subjects were excluded if they had neurological disorder, learning fects somewhat overlap with previous structural and functional non–graph theory observations in patients with BD (4) and in disability, comorbid misuse of substances/alcohol, other Axis 1 disorder, or history of head injury resulting in loss of con- unaffected siblings at high risk of BD (11–13). A more definite interpretation of the network-level understanding of the func- sciousness for .5 minutes, or a history of oral steroid use in the previous 3 months. Healthy control subjects needed to tional organization of BD is limited by the paucity of graph- theoretical studies and clinical samples investigated, different have no personal history of a psychiatric illness or history among first-degree relatives, defined using the Structured methodological approaches, and the variety of network met- Clinical Interview for DSM-IV Nonpatient edition (American rics employed (Table 1). However, a trend of functional network-level changes of BD appears to involve specific Psychiatric Association, 1994 version). To ensure that significant findings were features of BD type I, analyses were functional subsystems that predominantly encompass DMN and limbic centers, as opposed to being widespread. repeated removing subjects with BD type II; furthermore, to ensure that significant findings were features of euthymic BD, The temporal sequence of events or primarily anatomical antecedents of the observed abnormalities in patients with BD analyses were repeated removing subjects not meeting criteria for euthymia. Ethical approval was received by the University remain speculative and will emerge subsequent to future longitudinal population-based studies. Abnormalities in brain function College Hospital Galway Research Ethics Committee, and participants gave written fully informed consent before may reflect abnormalities in the underlying brain’s wiring patterns; considering that the interpretation as to what abnormal participating. neuroanatomical deficits might represent in relation to aberrant function is somewhat speculative, structure–function relations MRI Acquisition are a promising way forward to determine the nature of functional MRI data were obtained on a 3T Achieva scanner (Philips, abnormalities in anatomical networks shown to be abnormal in Best, The Netherlands) at the Welcome Trust Health Research brain disease and specifically in BD. Therefore, despite the evi- Board National Centre for Advanced Medical Imaging at dence presented to date, disentangling whether these abnor- St. James’s Hospital Dublin, Ireland. High-resolution malities are intertwined and relate to BD affective dysregulation is 3-dimensional T1-weighted turbo field echo magnetization- unclear, and the field may benefit from the application of prepared rapid gradient-echo sequence was acquired using network-based analyses and structure–function integration ap- an 8-channel head coil (repetition/echo times = 8.5/3.046 ms, 1 proaches. This study built from previous studies on structural and mm3 voxel size). Diffusion-weighted images were acquired at functional brain connectivity and brain network organization in b = 1200 s/mm2 along with a single nondiffusion-weighted individuals with BD and aimed to further clarify unresolved image (b = 0), using high angular resolution diffusion imaging questions related to the neurobiology of BD. The aim of this study involving 61 diffusion gradient directions, 1.8 3 1.8 3 1.9 mm was to investigate functional changes in individuals with euthy- voxel dimension, and field of view 198 3 259 3 125 mm. mic BD and specifically to describe features of whole-brain and Resting-state fMRI data were acquired using a single-shot

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI 141 Psychiatry: CNNI Biological 142

Table 1. Overview of Connectivity Network Findings of Today’s Functional Connectivity Graph Theory Studies ilgclPyhar:CgiieNuocec n eriaigFbur 00 5:140 2020; February Neuroimaging and Neuroscience Cognitive Psychiatry: Biological Reference Sample Methodology Network Measures Investigated Findings (BD vs. HC) Zhang et al., 2019 (37) 57 young (13–28 years) Functional network (SPM8 and Structural: global strength, network- Structural: preserved global (FA) strength; euthymic BD, 42 HCs DPABI) based statistics, rich-club NBS: Y fronto-parietal-temporal connectivity Structural network (AAL-90 and connectivity (FA); [ frontal cortex and subcortical regions whole-brain DTI tractography, Functional: modularity (FA) FSL) Structure–function coupling Functional: Y intramodular connectivity; (marginally) [ intermodular connectivity Y Structure–function coupling whole-brain and involving intrahemispheric connections Dvorak et al., 2019 (24) 20 euthymic BD (13 w/ BDI and AAL-90, DPARSF Network-based statistics, Global NBS: [ (synchronous) FCS in rh/lh temporal 7 w/ BDII), 15 MDD, 30 HCs (CC, CPL, EGlobal), nodal (CC, regions, w/ Y (synchronous) FCS specifically CPL, degree, betweenness between lh angular gyrus–lh temporal pole centrality) and rh parietal gyrus–rh hippocampus Preserved global connectivity: [ CC (not surviving FDR correction), unchanged CPL and EGlobal CPL: Y rh olfactory cortex, rh hippocampus, rh middle temporal and lh fusiform gyrus, rh/lh caudate, rh putamen Degree: [ rh middle frontal gyrus Wang et al., 2017 (42) (Unmedicated) 48 BDII w/ DPARSF/SPM8 FCS of short-range (,75 mm) Long fibers: [ FCS rh MTG and cerebellum depression, 48 MDD, 51 HCs Gray matter probability map (SPM8) and long-range (.75 mm) Short fibers:[ FCS lh/rh thalamus and lh/rh Voxelwise whole-brain functional fibers cerebellum, Y lh/rh precuneus network analysis Wang et al., 2017 (27) (Unmedicated) 31 BDII w/ GRETNA (SPM8) Network-based statistics NBS: no difference in FCS. Nodes: DMN depression, 32 (unmedicated) AAL-90 random parcellation into Measures: FCS, CC, CPL, EGlobal, (69%), and FPN (20%); edges: DMN-DMN UD, 43 HCs 1024 ROIs, Zalesky et al., normalized CPL, normalized CC, (59%), DMN-FPN (14%), FPN-FPN (10%) 2010 (43) small-world, ELocal, modularity [ Path length, Y EGlobal—not surviving

multiple correction BD in Dysconnectivity Functional Subnetwork Y ELocal in DMN, limbic system and cerebellum; YStrength: lh/rh precuneus and

– SFG, lh middle cingulum, rh temporal pole; 151 lh posterior cerebellar lobe. Disrupted

www.sobp.org/BPCNNI intramodular connectivity within DMN and limbic system Roberts et al., 2017 (11) 49 young (16–30 years) BD w/ SPM8; AAL-90 random parcellation Network-based statistics on NBS: Y FCS between lh IFG and mild depression (28 w/ BDI; into 513 ROIs, Zalesky et al., 2010 the IFG, CC, PI, CPL frontotemporal regions (lh insula, lh 21 w/ BDII), 71 at risk, 80 HCs (43) putamen, lh/rh STG, lh/rh VLPFC, lh/rh mPFC Y CC in IFG (no effect of lithium/antipsych/ antidepr on CC); no change in CPL Table 1. Continued BD in Dysconnectivity Functional Subnetwork Reference Sample Methodology Network Measures Investigated Findings (BD vs. HC) Douchet et al., 2017 (12) 78 BDI, 64 unaffected siblings, AAL-90 random parcellation into 620 CPL, EGlobal, CC, small-world, No difference in CPL, EGlobal, and CC 41 HCs ROIs, Crossley et al., 2013 (44), nodal degree, PI, modularity; [ Degree in supplementary motor area, MFG, Zalesky et al., 2010 (43) network-based statistics supramarginal gyrus, MedialFG, ITG; Y degree in precentral lobule, PostCG ilgclPyhar:CgiieNuocec n eriaigFbur 00 5:140 2020; February Neuroimaging and Neuroscience Cognitive Psychiatry: Biological Modularity: Y FCS in sensorimotor network Y PI vmPFC, hippocampus NBS: Ysensorimotor and visual networks Zhao et al., 2017 (25) 20 euthymic BDII, 38 HCs DPARSF/SPM8/REST; MNI FCS between any pair of symmetric FCS: Y MFG and preCG; No [ recorded template; VMHC interhemispheric voxels; Global VMHC Spielberg et al., 2016 (26) (Unmedicated) total of 60 BDI Time series extracted using a Network-based statistics NBS: [ connectivity mostly involving and BDII: 30 w/ depression, 30 181-ROI random parcellation, no Global: Global efﬁciency, CPL, amygdala (42%). A subnetwork involving hypomanic cerebellum, Craddock et al., assortativity, CC in signiﬁcant FrontalG was associated w/ YMRS; 2012 (45) NBS ROIs subnetwork involving OrbitofrontalG w/ HDRS YEGlobal; YCC rh amygdala Wang et al., 2016 (46) (Unmedicated) 37 BDII GRETNA (SPM8) FCS FCS:Y DMN; [ ParahippocampalG, amygdala, w/ depression, 37 HCs ACC, STG, LingG, cerebellum

Wang et al., 2015 (47) 36 BDII w/ depression—66.7% DPARSF/SPM8/REST; MNI FCS between any pair of symmetric FCS: Y fusiform/lingual gyrus and cerebellum unmedicated template; VMHC interhemispheric voxels 32 UDs during a depressive episode—71.8% unmedicated, 40 HCs He et al., 2015 (48) (Unmedicated) 13 BDI and BDII SPM8; Connectivity analysis: group FCS, nodal CC, ELocal [ CC in DLPFC, VLPFC (Y w/ depression), w/ depression, 40 MDD, 33 HCs ICA—48 ICNs; graph theory SFG, ACC analysis

AAL-90 (1024), Automated Anatomical Labeling Atlas; ACC, anterior cingulate cortex; antidepr, antidepressant; antipsych, antipsychotic; BD, bipolar disorder; BDI, bipolar disorder type I; BDII, bipolar disorder type II; CC, clustering coefficient; CPL, characteristic path length; DLPFC, dorsolateral prefrontal cortex; DMN, default mode network; DPABI, Data Processing and Analysis for Brain Imaging toolbox; DPARSF, Data Processing Assistant for Resting-State fMRI; DTI, diffusion tensor imaging; EGlobal, global efficiency; ELocal, local efficiency; FA, fractional anisotropy; FCS, functional connectivity strength; FDR, false discovery rate; FPN, frontoparietal network; FrontalG, frontal gyrus; GRETNA, graph theoretical network analysis; HC, healthy control subject; HDRS, Hamilton Depression Rating Scale; ICA, independent component analysis; ICN, intrinsic connectivity networks; IFG, inferior frontal gyrus; ITG, inferior temporal gyrus; lh, left hemisphere; LingG, lingual gyrus; MDD, major depressive disorder; MedialFG, medial frontal gyrus; MFG, middle frontal gyrus; MNI, Montreal Neurological Institute; MTG, middle temporal gyrus; mPFC, medial prefrontal cortex; NBS, network-based statistics; OrbitofrontalG, orbitofrontal gyrus; ParahippocampalG, parahippocampal gyrus; –

151 PI, participation index; PostCG, post cingulate gyrus; preCG, precingulate gyrus; REST, rs-fMRI data analysis toolkit; rh, right hemisphere; ROI, region of interest; SFG, superior frontal gyrus; SPM, statistical parametric mapping; STG, superior temporal gyrus; UD, unipolar depression; VMHC, voxel-mirrored homotopic connectivity (voxelwise whole brain analysis); www.sobp.org/BPCNNI VLPFC, ventrolateral prefrontal cortex; vmPFC, ventromedial prefrontal cortex; w/, with; YMRS, Young Mania Rating Scale; Y, reduced; [, increased. 143 CNNI Psychiatry: Biological Biological Psychiatry: CNNI Subnetwork Functional Dysconnectivity in BD

gradient echo planar imaging (EPI) sequence and involved Whole-Brain Functional Connectivity Measures whole-brain acquisition of 180 volumes (repetition/echo Global parameters summarizing whole-brain connectivity times = 2000/28 ms, flip angle 90 , field of view 240 3 240 3 properties of BD functional networks included characteristic 3 133 mm, 3 mm resolution, 80 3 80 matrix size, and 38 axial path length, global efficiency, global (positive/negative) slices of 3.2 mm each); each subject was asked to lie still in the strength, and clustering coefficient and were calculated as the scanner with their eyes open and fixed on a crosshair on the mean of the 86 regional estimates. Furthermore, a global screen for the entire duration of the scan. Structural and measure of influence and centrality, betweenness centrality functional MR images were visually inspected before/after and network resilience, and assortativity were investigated processing for accuracy of segmentation/parcellation, regis- (Brain Connectivity Toolbox version 1.52, MATLAB) (19). tration, motion, and outliers. Functional connectivity matrices were thresholded (r . 0) to retain only positive weights due to computational difficulties and the trivial interpretability introduced by negative edges particularly for network measures that depend on shortest MRI Data Analysis paths. Negative correlations were set to 0 for all measures, Diffusion MRI data processing details can be found in the with the exception of whole-brain functional strength, Supplement. fMRI data processing was optimized to reduce excluding an average of 16% of the connections (16% of motion and physiological noise as much as possible. Individual control subjects, 16% of patients). Recently, a test–retest structural T1-weighted scans underwent cortico-subcortical reliability of functional connectivity measures reported that segmentation/parcellation (FreeSurfer version 5.3; http:// weighted whole-brain network metrics are more reliable than surfer.nmr.mgh.harvard.edu/). EPI images underwent despik- binarized ones (20); therefore, we obtained whole-brain func- ing to remove spikes of activation (i.e., outliers) across time tional measures using threshold-weighted matrices. Statistical series (3dDespike, AFNI version 18.1.19; http://afni.nimh.nih. analyses were performed with diagnosis as fixed factor co- gov/afni), motion correction (3dvolreg, AFNI), and registration varying for age and gender using a multivariate analysis of to the FreeSurfer skull-stripped structural image (FLIRT, FSL covariance statistical approach (p , .05; 2-tailed) (SPSS version 5.0.4; https://fsl.fmrib.ox.ac.uk/fsl/fslwiki), followed by version 23, IBM Corp., Armonk, NY). nuisance regression of 6 movement parameters (3dDe- convolve, AFNI), and slice timing (3dTshift, Fourier’s trans- Permutation-Based Analysis of the Functional formation, AFNI). The global signal was not regressed out to Connectome reduce the likelihood of introducing spurious negative activation measures in the subsequent analyses (15). ANATICOR (16) A nonparametric statistical analysis, the Network-Based Sta- bandpass filtering (0.01–0.1 Hz) and motion scrubbing were tistics (NBS version 1.2) (21), was employed to perform mass performed in a single step (3dTproject, AFNI). The EPI data univariate hypothesis testing at every functional connection underwent motion correction as well as censoring during composing the graph to identify a differently functionally preprocessing. Similar to previous research (17), in-scanner connected subgraph component, meanwhile controlling for the motion-induced corruption of EPI volumes was defined as familywise error rate. Independent Student's t tests and ana- framewise displacement .0.5 mm (Euclidean normalization); lyses of covariance, with fixed factor diagnosis (covarying for subjects with .30 corrupted volumes were excluded during age and gender), were computed to test for group functional connectivity strength differences ( = 5000, , .05; 1-tailed). preprocessing (n = 3 patients with BD; n = 0 healthy control M p subjects), ensuring at least 5 minutes of resting-state fMRI Connections were thresholded (T = 1.5–3.5) to obtain a set of data for all subjects. More details on movement effects can be suprathreshold connections that were tested for main effects found in Supplemental Figure S2. The Desikan-Killiany Atlas of diagnosis, gender, and gender-by-diagnosis interaction. The (18), containing information about the cortico-subcortical choice of primary threshold is a user-determined parameter; mapping, was used to define spatial regions of interest—34 however, familywise error rate control is applied regardless of cortical and 9 subcortical brain regions bilaterally including the threshold choice (21). NBS was run on functional con- cerebellum, for a total of 86 nodes, for each individual sub- nectivity matrices including both positive and negative ject’s T1-weighted scan. Thus, normalization of the anatomical weights. Furthermore, to assess the separate contribution of registered functional data to a standard space was not synchronous and antisynchronous components to BD dys- necessary, and spatial smoothing of functional data was not connectivity, post hoc investigations were carried out retaining performed. positive or negative weights only, underpinning synchronous Regional time series were extracted for each region of in- or antisynchronous connectivity, respectively. terest by averaging the time series of all voxels within each node. The pairwise Pearson’s and partial correlations of neural Structure–Function Coupling Analysis times series between 2 nodes in the network were carried out We conducted network structural–functional connectivity ana- to generate individual (86 3 86) weighted undirected functional lyses in line with previous research (22). In brief, structural connectivity matrices (MATLAB release 2017b; The Math- matrices were derived within a graph-theoretical structural con- Works, Inc., Natick, MA) (Figure 1). Functional connectivity nectome analysis using a subject-specific cortico-subcortical matrices and Pearson’s/partial coefficients distribution were non-tensor-based connectome approach (14) (see details in visually inspected, screening for widespread and inflated the Supplement). Nonzero structural connectivity weights—the positive or widely distributed correlation values within and number of streamlines (NOS) weights—were isolated and across subjects. mapped using an inverse Gaussian transformation (22,23)to

144 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI Biological Psychiatry: Subnetwork Functional Dysconnectivity in BD CNNI FPO = web 4C Figure 1. Construction of functional resting-state connectivity matrices. The average time series were correlated between cortico-subcortical regions to obtain undirected weighted functional connectivity (FC) matrices (Pearson’s coefﬁcient, top image, and partial coefﬁcient, bottom image). FC matrices were used for 1) whole-brain network analysis (MATLAB release 2017b) and 2) permutation-based analysis (NBS version 1.2), and 3) were combined with structural connectivity (SC) matrices in a structure–function coupling analysis (MATLAB release 2017b). BD, bipolar disorder; BOLD, blood oxygen level–dependent; fMRI, functional MRI; HC, healthy control subject; MRI, magnetic resonance imaging.

achieve a structural weight distribution that was practically Correlations With Clinical Variables bounded within a [14, 24] range (i.e., 64s of an N(0,1) distri- Significant network measures were correlated with symptom bution). Prior to normalization, all NOS values below set severity as rated using the HDRS, HARS, and YMRS clinical thresholds were set to 0 (i.e., noise floor thresholding); group mood scales; age of onset; and illness duration. differences were carried out at the primary threshold TNOS =1 (i.e., no noise thresholding) and confirmed across different thresholds ( =2–10). The thresholding effects on network TNOS RESULTS density are depicted in Supplemental Figure S1. The res corresponding functional connectivity (Pearson’s) values for each Participants' Clinical and Demographic individual (22). Furthermore, fractional anisotropy (FA)-weighted Characteristics matrices were also investigated by the present analysis (FA We investigated 41 participants with BD and 56 healthy vol- weights were not rescaled). Group-level comparisons (multi- unteers who were matched for age, gender, and education variate analysis of variance, with fixed factor diagnosis, p , .05) level. Patients and control subjects did not differ in age across (MATLAB release 2017b) were performed at the whole con- diagnosis-by-gender subgroups (F3,96 = 1.25, p = .298) nectome level and within hubs, feeder (hub-to-nonhub) con- (Table 2). We included 33 participants who were BD type I and nections, and local (nonhub-to-nonhub) connections (22). 8 who were type II in the analyses. The vast majority of the Specifically, a single Pearson’s r value was extracted for each participants with BD were euthymic at MRI scanning (68%), subject representing their whole-brain structure–function mea- with only 13 displaying mild manic (YMRS score .7) or sure of integration; additionally, a Pearson’s r value was depressive (HDRS-21 score .8) and anxiety (HARS score extracted for each subject for each connection class (hubs, .18) signs and symptoms. Significant findings were confirmed feeder, local) (Figure 1). Hubs were defined for patients with BD post hoc for removing subjects with BD who were not meeting and control subjects within a previous structural rich-club criteria for euthymia (HDRS-21 score .8, YMRS score .7), as analysis performed in an overlapping clinical sample of pa- well as those with HARS score .18 and those with BD type II. tients (n = 37, 93% overlap) and control subjects (n = 45) at the statistically different subnetwork (k . 30; z = 3.78, p , 2.2 3 10216) in patients with BD relative to control subjects (14); Whole-Brain Measures of Integration furthermore, the structure–function relationship was assessed The BD group’s whole-brain organization did not differ from that for the statistically different (FA-weighted) subnetwork between of control subjects for path length, global efficiency, global patients with BD and control subjects (T . 1.5, p = .031) defined (positive/negative) strength, clustering coefficient, betweenness in a previous subnetwork analysis that included limbic system centrality, and assortativity (F7,87 = 1.684, p = .123) (Table 3). and basal ganglia nodes (14). We employed the same nodal Analysis of whole-brain connectivity using partial correlation parcellation scheme for both structural and functional networks coefficients confirmed stability of whole-brain resting-state to allow direct comparison between these two networks. networks in patients with BD relative to control subjects.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI 145 Biological Psychiatry: CNNI Subnetwork Functional Dysconnectivity in BD

Table 2. Clinical and Sociodemographic Details of Patients With BD and HCs Statistical Comparison Sample HCs BD Diagnostic Groups Participants, n 56 41 –

Age, Years, Mean 6 SD 40.63 6 13.52 43.59 6 12.71 t95 = 1.09, p = .277 Male 41.84 6 13.31 40.50 6 13.51 Female 39.65 6 13.83 46.52 6 11.44 2 Gender, Male/Female, n 25/31 21/20 c 1 = 0.16, p = .686 Level of Education (SES Scale) 2 Median 6 5 c 5 = 10.58, p = .060 Range 2–72–7 Age of Onset, Years, Mean 6 SD – 26.6 6 10.0 – Illness Duration, Years, Mean 6 SD – 16.4 6 10.7 – HDRS Mean 6 SD 1.13 6 1.7 7.0 6 7.4 U = 1.82, p , .001a Range 0–70–28 Median 0 5 YMRS Mean 6 SD 0.8 6 1.5 1.9 6 2.6 U = 1.45, p = .010a Range 0–60–10 Median 0 1 HARS Mean 6 SD 0.7 6 1.6 5.0 6 6.4 U = 1.75, p , .001a Range 0–80–27 Median 0 3 Mood Stabilizers, n Medication naïve – 3 – Lithium only (0.4–1.2 g/day) – 5 – Sodium valproate only (0.3–1.4 g/day) – 3 – Lamotrigine only (0.05–0.45 g/day) – 8 – Combination – 8 – Antidepressants, n SNRI/SSRI/NaSSA – 7/7/2 – Antipsychotics, n Atypical/typical – 29/1 – Benzodiazepine, n – 2 – Other psychotropics,b n – 9 – Participants were age and gender matched across groups. BD, bipolar disorder; HARS, Hamilton Anxiety Rating Scale; HC, healthy control subject; HDRS, Hamilton Depression Rating Scale; NaSSA, nonadrenergic and speciﬁc serotonergic antidepressant; SES, socioeconomic status; SNRI, serotonin and norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; YMRS, Young Mania Rating Scale. ap , .05. n = 14 with HDRS.8; n = 8 subjects were BD type II. bOther psychotropics included the hypnotics zopiclone and zolpidem and the anticonvulsant carbamazepine.

Permutation-Based Subnetwork Analysis comparable (synchronous/antisynchronous) dysconnected We identiﬁed a functionally dysconnected subnetwork for pa- subnetwork in patients with BD compared with control subjects tients with BD relative to control subjects comprising parietal, (Pearson’s: T . 3, p = .039; not using partial). A small subnet- cingulate, and frontotemporal (synchronous/antisynchronous) work of stronger frontotemporal (antisynchronous) components was observed in patients with BD relative to control subjects functional connections (t test: Pearson’s: T . 3, p = .039; par- (post hoc test: Pearson’s: . 3, = .014) (Figure 2B, C, tial: T . 3.5, p = .037) (Figure 2A, Supplemental Table S1). A t T p subnetwork of comparable dysconnectivity (77% overlap) was Supplemental Table S2). observed using positive correlations only (post hoc test: t Structure–Function Coupling Analysis Pearson’s: T . 3, p = .048) (Figure 2B, C, Supplemental Table S2). We did not detect increased functional connectivity Structure–function association analysis was performed on 38 in patients with BD relative to control subjects. When covarying patients with BD and 45 control subjects matched for age 2 for age and gender (analysis of covariance), we observed a (U = 1.042, p = .087) and gender (c 1 = 0.092, p = .762). There

146 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI Biological Psychiatry: Subnetwork Functional Dysconnectivity in BD CNNI

Table 3. Whole-Brain Functional Connectivity Network Measures Statistical Comparison HCs, BD, Diagnosis, Measure Mean 6 SD Mean 6 SD F, p Value Positive Strength 15.44 6 3.70 15.93 6 4.06 0.43, .51 Negative Strength 2.90 6 1.24 3.02 6 1.06 0.15, .70 Betweenness Centrality 244.77 6 19.64 251.11 6 22.16 2.67, .11 Global Efficiency 0.27 6 0.04 0.27 6 0.04 0.68, .41 Characteristic Path Length 4.42 6 0.54 4.38 6 0.53 0.17, .68 Clustering Coefficient 0.17 6 0.05 0.18 6 0.05 0.90, .35 Assortativity 0.05 6 0.06 0.05 6 0.06 1.4, .24 The BD group global network organization did not differ from that of the HC group across all whole-brain network measures. Specifically, there was no main effect of diagnosis (F7,87 = 1.684, p = .123) in patients with BD relative to HCs. Multivariate analysis of covariance statistical approach, covarying for age and gender, p , .05. BD, bipolar disorder; HC, healthy control subject. was no main effect of diagnosis (F8,72 = 0.927, p = .500) abnormalities in BD do not extend to be detected globally, but (Figure 3) across the primary threshold (T = 1) using FA (range rather may be localized to specific subnetworks. of Pearson’s coefficients: BD group: r = 2.06 to 2.11; healthy control group: r = 2.06 to 2.12) (Figure 3) or NOS weights (range of Pearson’s coefficients: BD group: r = .20 to .28; healthy control group: r = .19 to .27) (Figure 3). These were Permutation-Based Analysis of the Functional confirmed across all thresholds (T =2–10). Additionally, no Connectivity Matrices difference was observed when increasing the rich-club nodal Although there was no difference in whole-brain functional definition at 70% of connections common to participants. connectivity in BD, changes were observed at the subnetwork level. Nodes thought to be associated with the DMN such as the Correlations With Clinical Variables prefrontal, caudal middle frontal, and posterior cingulate gyri Illness duration and age of onset did not relate to the signifi- were implicated in the significantly reduced (synchronous) sub- cant network measures (age of onset: r = 2.051 to .152, p = network in BD (Figure 2). This suggests reduced functional in- .363–.762; illness duration: r = 2.149 to .133, p = .371–.644). teractions in BD between structures that play key roles in self- referential thinking and emotion-regulation processing and cognitive control—features that are known to be functionally DISCUSSION altered in BD (28). Post hoc observations into BD functional We examined the functional organization of networks in patients network organization corroborated and expanded the reduced with BD, relative to that in control subjects, both globally and functional (synchronous) connectivity observed between fron- locally and reported preserved whole-brain functional connec- tolimbic and parieto-occipital nodes to reveal increased (anti- tivity with localized differences involving nodes belonging to synchronous) connectivity in a subnetwork encompassing default mode, frontotemporal, and limbic systems. These find- fronto-temporo-parietal nodes in patients with BD relative to ings appeared to be trait features of BD type I. Furthermore, the control subjects (Supplemental Table S2). These abnormalities present study provides preliminary evidence of preserved collectively present a functional map of BD dysconnectivity that structure–function relationships globally and within edge class in differentially involves communication between regions spanning euthymic individuals with BD relative to control subjects. across multiple brain subsystems. This may underpin a compensatory mechanism of neural activity underlying whole- Whole-Brain Measures of Connectivity brain functional stability in BD that may be necessary to sustain We did not detect changes in whole-brain functional connec- a remitted clinical state of the illness and may provide flexibility in tivity in patients with BD relative to control subjects; this is in the ability to switch between segregated and integrated states. line with previous studies conducted in individuals with Although one may infer that antisynchronous activity underpin- euthymic BD (4,12,24,25) and in young individuals with BD ning anticorrelations between brain regions may represent inhi- who also have (mild) depression (11). Changes in whole-brain bition or competitive functional interactions, these should be functional connectivity have been reported in subjects with cautiously interpreted considering that the underlying physio- BD who are actively depressed or unmedicated, relative to logical basis of this type of connectivity remains rather unclear. control subjects, and these changes are defined by longer Recent evidence suggests that antisynchronous activity paths and lower global efficiency (26,27). Therefore, functional observed in fMRI is not an analytic artifact nor does it necessarily large-scale changes may be characteristic of patients with BD represent a direct, antagonistic relationship between coherent who are unmedicated and symptomatic, and our findings of networks; rather, antisynchronous activity may embody different preserved whole-brain functional connectivity may be dynamic configurations of these networks around the same considered a trait feature, rather than state, of this illness. This anatomical skeleton (29). The investigation of dysconnectivity in was supported by preserved whole-brain (-weighted) structural BD via means of dynamic fMRI may further clarify the underlying connectivity in an overlapping clinical sample relative to control physiological basis of the antisynchronous subnetwork dys- subjects (14) and collectively suggests that network-level connectivity observed in BD.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI 147 Biological Psychiatry: CNNI Subnetwork Functional Dysconnectivity in BD

synchronous and anti-synchronous connections synchronous connections only anti-synchronous connections only C FPO = web 4C Figure 2. Permutation-based subnetwork analysis of the functional connectivity matrices. (A) Significant original subnetwork of synchronous/antisynchronous components (Pearson’s t test, T . 3.5, p = .037, in orange) with patients showing a differential subnetwork of (synchronous/antisynchronous) components relative to those of control subjects. (B) Post hoc significant subnetworks of synchronous (in yellow) and antisynchronous (in red) components, as well as the original significant (synchronous/antisynchronous) subnetwork (in orange) plotted together. Patients with bipolar disorder, relative to control subjects, showed a weaker subnetwork of functional (synchronous, Pearson’s t test, T . 3, p = .048) components and a stronger subnetwork of functional (antisynchronous, Pearson’s t test, T . 3, p = .014) components. Synchronous components explain most functional connections within the original significant subnetwork. (C) Ringlike visualization of both original and post hoc subnetworks of connected components. Images were obtained using NeuroMArVL software (http://immersive.erc.monash.edu.au/ neuromarvl/). A, anterior; Amy, amygdala; banksstts, posterior banks of the superior temporal sulcus; CauACG, caudal anterior cingulate gyrus; CaudMFG, caudal middle frontal gyrus; Enth, entorhinal cortex; Hipp, hippocampus; IPG, inferior parietal gyrus; IsthCingG, isthmus of cingulate gyrus; ITG, inferior temporal gyrus; L, left; LingG, lingual gyrus; MOFG, medialorbitofrontal gyrus; P, posterior; ParaCentralG, paracentral gyrus; ParsOperc, pars opercularis; ParsTriang, pars trian- gularis; PostCentralG, postcentral gyrus; PreCun, precuneus; R, right; RostralMFG, rostral middle frontal gyrus; SFG, superior frontal gyrus; STG, superior temporal gyrus; TempPole, temporal pole; Thal, thalamus.

148 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI Biological Psychiatry: Subnetwork Functional Dysconnectivity in BD CNNI

NOS-weighted Figure 3. Structure–function coupling analysis 0.6 across whole-brain and edge-class connections. Plots are shown for Pearson’s correlation values between functional connectivity (FC) matrices and 0.4 fractional anisotropy (FA)-weighted (top) and number of streamlines (NOS)-weighted (bottom, T =1) structural connectivity (SC) matrices (median 6 SD). 0.2 Across all measures, there was no main effect of SC-FC diagnosis (F8,72 = 0.927, p = .500). These ﬁndings were conﬁrmed across the other considered 0.0

Pearson's Coefficient thresholds (NOS weights: T =2–10). “Whole-brain” category represents structure–function coupling of all nodes in the network; “Hubs” category repre- -0.2 sents structure–function coupling of rich-club (RC) HC BD HC BD HC BD HC BD hubs; “Feeder” category represents structure– Whole-brain Rich-club Hubs RC Feeder Local function coupling of nodes connecting to RC hubs; “NBS nodes” category represents structure– function coupling of nodes implicated in the statis- FA-weighted tically different network-based statistics (NBS) 0.6 subnetwork between patients with bipolar disorder (BD) and healthy control (HC) subjects; “Nodes 0.4 feeder” category represents structure–function coupling of connecting to the NBS nodes; and “Local” category represents structure–function 0.2 coupling of nodes composing the remaining of the network. SC-FC 0.0

Pearson's Coefficient -0.2

-0.4 HC BD HC BD HC BD HC BD

Whole-brain NBS nodes NBS feeder Local

HC (n=56) BD (n=41)

Resting-state fMRI investigations of the brain have largely emotion processing (2,3). Additionally, the frontoparietal func- focused on the synchronous activation of regions of the DMN; tional system participates in mechanisms of attention, memory, these have been shown to be particularly relevant to psychiatric and cognitive control (35)—cognitive features altered function- disorders and a robust feature of BD functional dysconnectivity ally in BD (36). Considering that regions belonging to the impli- (4,6,30,31)(Table 1). Interestingly, increased functional connec- cated functional subsystems have been shown to be tivity within DMN nodes has been reported in unaffected siblings at anatomically in continuity with each other (33), they may depend high risk compared with those with the illness (12)thatmaybe on the synergistic functioning of each region to optimally un- considered a biological feature underlying resilience to BD (32). derpin higher-order cognitive functioning specifically in BD. Therefore, a specific subset of regions emerges as being vulnerable to functional changes in BD that may be considered a viable target for future interventions to ameliorate symptoms. Structure–Function Relationship Although the DMN is composed of selective regions that are To the best of our knowledge, this is the first study to explore thought to execute functions that are categorically different from structure–function connectome coalescence in an adult sample of those of other networks, there is evidence that this system may euthymic individuals with BD. Previous investigations include a not act independently but rather may be on a continuum with young euthymic BD group (37) and young offspring of patients other networks (33). This global organization was further sup- with BD (38). These independent studies reported decreased ported by the observed high correspondence between cortical structure–function coupling in individuals with BD (37)and gradients of functional connectivity and myelin density across increased coupling in offspring (38), with structure–function most cortical areas (34). Thus, it is important to understand the breakdown involving intrahemispheric and whole-brain connec- connectivity relationship between nodes belonging to this tivity (37) or long-range connections (38). We failed to detect any functional subsystem with other existing networks to further difference in structure–function associations in individuals with appreciate the DMN’s functional role specifically in the context BD relative to control subjects globally and within connection of mood regulation in BD. Furthermore, the frontolimbic system classes. The discrepant findings could be explained by the has also been heavily implicated both structurally and func- different structural and functional network reconstruction tionally in the pathophysiology of BD due to its key role in methods employed and the different clinical characteristics of

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI 149 Biological Psychiatry: CNNI Subnetwork Functional Dysconnectivity in BD

these cohorts; furthermore, in BD, significant changes in “dysconnection syndrome” provides an intuitive explanation structure–function coupling may be occurring and thus may be for the vast heterogeneity in symptomatology that character- detectable at the onset of the illness, rather than at later stages of izes the illness, as the field is shifting away from localizing the disease when the critical period of brain network development specific symptoms to specific gray matter and white matter has passed. regions or functional seeds, and instead moving toward the Structural networks are thought to place significant physical examination of abnormal interaction between brain regions constraints on functional connectivity both globally and locally and considering the brain’s network as a whole. Current un- (39) so that a change in the relationship between these two derstanding of the neurobiological basis of BD is limited. This measures would be suggestive of illness expression (22,23). represents a limit to accurate diagnosis and pharmacological Crucially, while anatomical connections give rise to and shape intervention that importantly impacts the quality life of affected functional connections, it is likely that there are several possible individuals. Our findings support the clinical construction of BD spatiotemporal reconfigurations of functional connectivity as a dysconnection syndrome. Furthermore, the observed expressed around the same anatomical skeleton, even within structural and functional dysconnectivity in BD highlights the a short time scale (29,40). This implies that brain function is need to examine network abnormalities at both anatomical and not static but instead dynamic, constantly switching between physiological levels, as well as to incorporate multimodal im- large-scale metastable wave patterns (41). Herein, preserved aging for a more meaningful understanding of dysconnectivity whole-brain structure–function coupling corroborates intact in psychotic illness such as BD. whole-brain structural (14) and functional connectivity in BD. However, in patients with BD relative to control subjects, we observed preserved structure–function coupling within connec- ACKNOWLEDGMENTS AND DISCLOSURES tion classes despite evidence of regional structural and func- This research is supported by the Irish Research Council Postgraduate Scholarship, Ireland (to LN) and by the Health Research Board Grant No. tional deficits, suggesting that more complex, perhaps dynamic, HRA-POR-324 (to DMC). interactions may be occurring between structure and function at We gratefully acknowledge the participants and the support of the the subnetwork level. Additionally, while anatomical connectivity Wellcome-Trust Health Research Board Clinical Research Facility and the may inform functional interactions, it is not per se a sufficient Centre for Advanced Medical Imaging, St. James Hospital, Dublin, Ireland. description of connectivity, and optimal models should be The authors report no biomedical financial interests or potential conflicts identified to examine structure–function relationships (40). of interest. We did not detect an effect of lithium on the significant functional connectivity measures; however, we may have been ARTICLE INFORMATION underpowered to investigate this outcome (BD subjects on From the Centre for Neuroimaging and Cognitive Genomics (NICOG) (LN, lithium, = 14; BD subjects off lithium, = 27). However, all but 3 GMcP, FMM, BH, CMcD, DMC), Clinical Neuroimaging Laboratory, NCBES n n Galway Neuroscience Centre, College of Medicine, Nursing, and Health BD subjects on lithium were taking other medications as well. Sciences, National University of Ireland Galway; College of Engineering and Furthermore, our findings can be considered trait features of BD Informatics (LK), National University of Ireland Galway, Galway, Ireland; and type I as these were confirmed post hoc when removing sub- Cardiff University Brain Research Imaging Centre (JRW, KM), Cardiff, United jects with BD type II or those not meeting criteria for euthymia. Kingdom. Collectively, BD is associated with localized functional Address correspondence to Leila Nabulsi, Ph.D., Centre for Neuro- dysconnectivity that does not extend to be detected globally, imaging and Cognitive Genomics (NICOG), Clinical Neuroimaging Labora- tory, NCBES Galway Neuroscience Centre, College of Medicine Nursing and which is suggestive of reduced regional functional interactions Health Sciences, National University of Ireland Galway, H91 TK33 Galway, in individuals with a history of psychotic and depressive epi- Ireland; E-mail: [email protected] or [email protected]. sodes. These changes predominantly involve regions Received Aug 15, 2019; revised Oct 29, 2019; accepted Oct 30, 2019. belonging to the DMN and limbic system, both of which play Supplementary material cited in this article is available online at https:// key roles in several cognitive domains known to be functionally doi.org/10.1016/j.bpsc.2019.10.015. altered in BD. Although we observed structural deficits between and within basal ganglia and limbic system connections REFERENCES (14), basal ganglia did not appear to be involved in the func- 1. O’Donoghue S, Holleran L, Cannon DM, McDonald C (2017): tional dysconnected subnetwork. We conclude that these Anatomical dysconnectivity in bipolar disorder compared with localized changes are suggestive of traitlike features of sub- schizophrenia: A selective review of structural network analyses using jects with euthymic BD that may be necessary to maintain a diffusion MRI. J Affect Disord 209:217–228. remitted clinical state of the illness. The applications of 2. Perry A, Roberts G, Mitchell PB, Breakspear M (2018): Connectomics of different conceptualizations of how information can flow bipolar disorder: A critical review, and evidence for dynamic instabilities around the human connectome, such as the dynamic repre- within interoceptive networks. Mol Psychiatry 24:1296–1318. 3. Caseras X, Murphy K, Lawrence NS, Fuentes-Claramonte P, Watts J, sentations of these functional systems, may be used to more Jones DK, Phillips ML (2015): Emotion regulation deficits in euthymic comprehensively describe intermittent behaviors characteristic bipolar I versus bipolar II disorder: A functional and diffusion-tensor of neuropsychiatric disorders such as BD. imaging study. Bipolar Disord 17:461–470. Despite striking evidence of cognitive deficits in BD and its 4. Syan SK, Smith M, Frey BN, Remtulla R, Kapczinski F, Hall GBC, social and personal burden, there is no pharmacological Minuzzi L (2018): Resting-state functional connectivity in individuals treatment that is specific to the management of core symp- with bipolar disorder during clinical remission: A systematic review. J Psychiatry Neurosci 43:170175. toms of BD, possibly made challenging by the cyclic nature of 5. Syan SK, Minuzzi L, Smith M, Allega OR, Hall GBC, Frey BN (2017): BD illness and the wide array of symptoms and cognitive Resting state functional connectivity in women with bipolar disorder deficits individuals with BD experience. The theory that BD is a during clinical remission. Bipolar Disord 19:97–106.

150 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI Biological Psychiatry: Subnetwork Functional Dysconnectivity in BD CNNI

6. Martino M, Magioncalda P, Huang Z, Conio B, Piaggio N, Duncan NW, 26. Spielberg JM, Beall EB, Hulvershorn LA, Altinay M, Karne H, Anand A et al. (2016): Contrasting variability patterns in the default mode and (2016): Resting state brain network disturbances related to hypomania sensorimotor networks balance in bipolar depression and mania. Proc and depression in medication-free bipolar disorder. Neuro- Natl Acad Sci U S A 113:4824–4829. psychopharmacology 41:3016–3024. 7. Strakowski SM, Adler CM, Almeida J, Altshuler LL, Blumberg HP, 27. Wang Y, Wang J, Jia Y, Zhong S, Zhong M, Sun Y, et al. (2017): Topologically Chang KD, et al. (2012): The functional neuroanatomy of bipolar dis- convergent and divergent functional connectivity patterns in unmedicated order: A consensus model. Bipolar Disord 14:313–325. unipolar depression and bipolar disorder. Transl Psychiatry 7:e1165. 8. Gruber SA, Rogowska J, Yurgelun-Todd DA (2004): Decreased acti- 28. Townsend J, Altshuler LL (2012): Emotion processing and regulation in vation of the anterior cingulate in bipolar patients: An fMRI study. bipolar disorder: A review. Bipolar Disord 14:326–339. J Affect Disord 82:191–201. 29. Deco G, Jirsa VK, McIntosh AR (2011): Emerging concepts for the 9. Rey G, Piguet C, Benders A, Favre S, Eickhoff SB, Aubry JM, dynamical organization of resting-state activity in the brain. Nat Rev Vuilleumier P (2016): Resting-state functional connectivity of emotion Neurosci 12:43–56. regulation networks in euthymic and non-euthymic bipolar disorder 30. Öngür D, Lundy M, Greenhouse I, Shinn AK, Menon V, Cohen BM, patients. Eur Psychiatry 34:56–63. Renshaw PF (2010): Default mode network abnormalities in bipolar 10. Chase HW, Phillips ML (2016): Elucidating neural network functional disorder and schizophrenia. Psychiatry Res 183:59–68. connectivity abnormalities in bipolar disorder: Toward a harmonized 31. Calhoun VD, Maciejewski PK, Pearlson GD, Kiehl KA (2008): Temporal methodological approach. Biol Psychiatry Cogn Neurosci Neuro- lobe and “default” hemodynamic brain modes discriminate between imaging 1:288–298. schizophrenia and bipolar disorder. Hum Brain Mapp 29:1265–1275. 11. Roberts G, Lord A, Frankland A, Wright A, Lau P, Levy F, et al. (2017): 32. Frangou S (2012): Brain structural and functional correlates of resil- Functional dysconnection of the inferior frontal gyrus in young people ience to bipolar disorder. Front Hum Neurosci 5:184. with bipolar disorder or at genetic high risk. Biol Psychiatry 81: 33. Margulies DS, Ghosh SS, Goulas A, Falkiewicz M, Huntenburg JM, 718–727. Langs G, et al. (2016): Situating the default-mode network along a 12. Doucet GE, Bassett DS, Yao N, Glahn DC, Frangou S (2017): The role principal gradient of macroscale cortical organization. Proc Natl Acad of intrinsic brain functional connectivity in vulnerability and resilience Sci U S A 113:12574–12579. to bipolar disorder. Am J Psychiatry 174:1214–1222. 34. Huntenburg JM, Bazin PL, Margulies DS (2018): Large-scale gradients 13. Cattarinussi G, Di Giorgio A, Wolf RC, Balestrieri M, Sambataro F (2019): in human cortical organization. Trends Cogn Sci 22:21–31. Neural signatures of the risk for bipolar disorder: A meta-analysis of struc- 35. Laird AR, Fox PM, Eickhoff SB, Turner JA, Ray KL, McKay DR, et al. tural and functional neuroimaging studies. Bipolar Disord 21:215–227. (2011): Behavioral interpretations of intrinsic connectivity networks. 14. Nabulsi L, McPhilemy G, Kilmartin L, O’Hora D, O’Donoghue S, J Cogn Neurosci 23:4022–4037. Forcellini G, et al. (2019): Bipolar disorder and gender are associated 36. Altshuler L, Bookheimer S, Townsend J, Ma P, Sabb F, Mintz J (2008): with frontolimbic and basal ganglia dysconnectivity: A study of topo- Regional brain changes in bipolar I depression: A functional magnetic logical variance using network analysis. Brain Connect 9:745–759. resonance imaging study. Bipolar Disord 10:708–717. 15. Murphy K, Fox MD (2017): Towards a consensus regarding global 37. Zhang R, Shao R, Xu G, Lu W, Zheng W, Miao Q, et al. (2019): Aberrant signal regression for resting state functional connectivity MRI. Neu- brain structural–functional connectivity coupling in euthymic bipolar roimage 154:169–173. disorder. Hum Brain Mapp 40:3452–3463. 16. Jo HJ, Gotts SJ, Reynolds RC, Bandettini PA, Martin A, Cox RW, 38. Collin G, Scholtens LH, Kahn RS, Hillegers MHJ, van den Heuvel MP Saad ZS (2013): Effective preprocessing procedures virtually eliminate (2017): Affected anatomical rich club and structural–functional distance-dependent motion artifacts in resting state FMRI. J Appl coupling in young offspring of schizophrenia and bipolar disorder Math 2013:935154. patients. Biol Psychiatry 82:746–755. 17. Whittaker JR, Foley SF, Ackling E, Murphy K, Caseras X (2018): The 39. Honey CJ, Thivierge J-P, Sporns O (2010): Can structure predict functional connectivity between nucleus accumbens and the ventro- function in the human brain? Neuroimage 52:766–776. medial prefrontal cortex as an endophenotype for bipolar disorder. Biol 40. Friston KJ (2011): Functional and effective connectivity: A review. Psychiatry 84:803–809. Brain Connect 1:13–36. 18. Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, 41. Roberts JA, Gollo LL, Abeysuriya RG, Roberts G, Mitchell PB, et al. (2006): An automated labeling system for subdividing the human Woolrich MW, Breakspear M (2019): Metastable brain waves. Nat cerebral cortex on MRI scans into gyral based regions of interest. Commun 10:1056. Neuroimage 31:968–980. 42. Wang Y, Wang J, Jia Y, Zhong S, Niu M, Sun Y, et al. (2017): Shared 19. Rubinov M, Sporns O (2010): Complex network measures of brain and speciﬁc intrinsic functional connectivity patterns in unmedicated connectivity: Uses and interpretations. Neuroimage 52:1059–1069. bipolar disorder and major depressive disorder. Sci Rep 7:3570. 20. Wang JH, Zuo XN, Gohel S, Milham MP, Biswal BB, He Y (2011): 43. Zalesky A, Fornito A, Harding IH, Cocchi L, Yücel M, Pantelis C, Graph theoretical analysis of functional brain networks: Test-retest Bullmore ET (2010): Whole-brain anatomical networks: Does the evaluation on short- and long-term resting-state functional MRI data. choice of nodes matter? Neuroimage 50:970–983. PLoS One 6:e21976. 44. Crossley NA, Mechelli A, Vértes PE, Winton-Brown TT, Patel AX, 21. Zalesky A, Fornito A, Bullmore ET (2010): Network-based statistic: Ginestet CE, et al. (2013): Cognitive relevance of the community Identifying differences in brain networks. Neuroimage 53:1197–1207. structure of the human brain functional coactivation network. Proc Natl 22. Hearne LJ, Lin HY, Sanz-Leon P, Tseng WI, Gau SS, Roberts JA, Acad Sci USA 110:11583–11588. Cocchi L (2019): ADHD symptoms map onto noise-driven structure- 45. Craddock RC, James GA, Holtzheimer PE, Hu XP, Mayberg HS (2012): function decoupling between hub and peripheral brain regions [pub- A whole brain fMRI atlas generated via spatially constrained spectral lished online ahead of print Oct 31]. Mol Psychiatry. clustering. Hum Brain Mapping 33:1914–1928. 23. Van Den Heuvel MP, Sporns O, Collin G, Scheewe T, Mandl RCW, 46. Wang Y, Zhong S, Jia Y, Sun Y, Wang B, Liu T, et al. (2016): Disrupted Cahn W, et al. (2013): Abnormal rich club organization and functional resting-state functional connectivity in nonmedicated bipolar disorder. brain dynamics in schizophrenia. JAMA Psychiatry 70:783–792. Radiology 280:529–536. 24. Dvorak J, Hilke M, Trettin M, Wenzler S, Hagen M, Ghirmai N, et al. 47. Wang Y, Zhong S, Jia Y, Zhou Z, Wang B, Pan J, Huang L (2015): (2019): Aberrant brain network topology in fronto-limbic circuitry dif- Interhemispheric resting state functional connectivity abnormalities in ferentiates euthymic bipolar disorder from recurrent major depressive unipolar depression and bipolar depression. Bipolar Disord 17:486–495. disorder. Brain Behav 9:e01257. 48. He H, Yu Q, Du Y, Vergara V, Victor TA, Drevets WC, et al. (2016): 25. Zhao L, Wang Y, Jia Y, Zhong S, Sun Y, Qi Z, et al. (2017): Altered Resting-state functional network connectivity in prefrontal regions interhemispheric functional connectivity in remitted bipolar disorder: A differs between unmedicated patients with bipolar and major resting state fMRI study. Sci Rep 7:4698. depressive disorders. J Affect Disord 190:483–493.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:140–151 www.sobp.org/BPCNNI 151 Biological Psychiatry: CNNI Archival Report

Neuroanatomical Dysconnectivity Underlying Cognitive Deﬁcits in Bipolar Disorder

Genevieve McPhilemy, Leila Nabulsi, Liam Kilmartin, Denis O’Hora, Stefani O’Donoghue, Giulia Tronchin, Laura Costello, Pablo Najt, Srinath Ambati, Gráinne Neilsen, Sarah Creighton, Fintan Byrne, James McLoughlin, Colm McDonald, Brian Hallahan, and Dara M. Cannon

ABSTRACT BACKGROUND: Graph theory applied to brain networks is an emerging approach to understanding the brain’s topological associations with human cognitive ability. Despite well-documented cognitive impairments in bipolar disorder (BD) and recent reports of altered anatomical network organization, the association between connectivity and cognitive impairments in BD remains unclear. METHODS: We examined the role of anatomical network connectivity derived from T1- and diffusion-weighted magnetic resonance imaging in impaired cognitive performance in individuals with BD (n = 32) compared with healthy control individuals (n = 38). Fractional anisotropy- and number of streamlines-weighted anatomical brain networks were generated by mapping constrained spherical deconvolution-reconstructed white matter among 86 cortical/subcortical bilateral brain regions delineated in the individual’s own coordinate space. Intelligence and executive function were investigated as distributed functions using measures of global, rich-club, and interhemispheric connectivity, while memory and social cognition were examined in relation to subnetwork connectivity. RESULTS: Lower executive functioning related to higher global clustering coefficient in participants with BD, and lower IQ performance may present with a differential relationship between global and interhemispheric efficiency in individuals with BD relative to control individuals. Spatial recognition memory accuracy and response times were similar between diagnostic groups and associated with basal ganglia and thalamus interconnectivity and connectivity within extended anatomical subnetworks in all participants. No anatomical subnetworks related to episodic memory, short-term memory, or social cognition generally or differently in BD. CONCLUSIONS: Results demonstrate selective influence of subnetwork patterns of connectivity in underlying cognitive performance generally and abnormal global topology underlying discrete cognitive impairments in BD. Keywords: Bipolar disorder, Cognition, Diffusion magnetic resonance imaging, Graph theory, Network analysis, Rich club https://doi.org/10.1016/j.bpsc.2019.09.004

Bipolar disorder (BD) is a psychiatric illness associated with interactions to better capture the integration between distinct cognitive impairment, including executive function, memory, neural systems that underlies cognitive functioning (15–19). and social cognition deficits (1–4). Prevalent in approximately Network investigations find that the brain is topologically 40% to 60% of individuals with BD (5), cognitive impairments configured to enable higher cognitive processing; a combina- are not accounted for by residual mood symptoms (1)or tion of high clustering and short path length supports both medication use (6–8) and are associated with a poorer quality local segregation and global integration while minimizing cost of life (9). Structural and diffusion magnetic resonance im- (20,21), modular structure facilitates functional specialization aging studies have found widespread structural brain ab- (22,23), and hub rich-club regions integrate information glob- normalities in BD (10–14), with consistent reports of reduced ally between modules (24,25). Emerging reports show that hippocampus, amygdala, and thalamus volume; reduced global efficiency and rich-club connectivity of anatomical prefrontal, temporal, and parietal cortical thickness (12,13); networks is associated with intelligence and executive function and altered white matter organization in temporoparietal and and that interhemispheric connectivity is associated with in- limbic tracts (10,11,14). However, the relationship between telligence in healthy individuals (26–30). Given that global neuroanatomical alterations and cognitive deficits remains efficiency, rich-club connectivity, and interhemispheric con- unknown. nectivity may be altered in BD (31–35), their investigation in Network analysis incorporates not only the anatomy of relation to intelligence and executive function deficits in BD is certain brain areas, but also multiple brain regions and their warranted.

152 ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI ISSN: 2451-9022 Biological Psychiatry: Neuroanatomical Dysconnectivity and Cognition in BD CNNI

Several gray and white matter regions have been implicated disorders, learning disability, comorbid substance or alcohol in BD to date, but no study has examined patterns of con- abuse, history of head injury resulting in a loss of conscious- nectivity underlying such impairments. Lower IQ in BD ness (.5 minutes), or any other illness potentially affecting (1,36–39) was associated with reduced magnetization transfer cognitive function. For mood rating, the Hamilton Rating Scale ratio, a measure of dendritic density and neuronal size and for Depression (56) and Young Mania Rating Scale (57) were number, in the superior temporal gyrus, uncus, and para- used on the day of scanning and cognitive testing. Euthymia hippocampal gyrus (40) and with reduced prefrontal cortical was defined as scores of ,8 and ,7, respectively. All par- folding (41). Executive functioning impairments (1–3,42) were ticipants provided fully informed written consent, and the associated with reduced prefrontal cortex volumes (43) and study was approved by the clinical research ethics commit- widespread white matter disorganization (44,45) regionally in tees of University College Hospital Galway and St. James’s the internal capsule (46) and anterior thalamic radiation (47). Hospital Dublin. Given that intelligence and executive function rely on distributed neural networks, including the frontal and parietal cortices, thalamus, basal ganglia, and cerebellum (48–51), Cognitive Assessment network measures that represent the capacity for global Selected subtests of the Wechsler Adult Intelligence Scale network integration may more optimally capture the basis for (vocabulary, similarities, block design, and matrix reasoning) their disruption. We hypothesized that IQ and executive func- were combined to obtain full-scale IQ (58). The Cambridge tioning would be associated with measures of global, rich- Neuropsychological Test Automated Battery was used to club, and interhemispheric connectivity and that disruption of measure executive function (Intra/Extra Dimensional Shift), these network features in BD would relate to intelligence and episodic memory (Paired Associates Learning), short-term executive function deficits. memory (Delayed Match to Sample) and spatial recognition Memory impairments in BD (1–3) have been associated memory (Spatial Recognition Memory) (59). The “Reading the with reduced amygdala volume (52) and altered diffusivity Mind in the Eyes” Test assessed social cognition (60). A values in the superior corona radiata and corticospinal tract multivariate analysis of covariance with age and gender as (53). However, no anatomical subnetwork connectivity covariates or nonparametric Mann-Whitney U was used to investigation has been conducted. This is despite evidence compare cognitive performance between groups. that variance in the pattern of brain structural connectivity underlies variance in healthy human performance of such tasks, in particular a temporal lobe subnetwork that includes Image Acquisition and Processing the hippocampus, temporal cortex, and insula (54). Theory Magnetic resonance imaging was performed on a 3T Philips of mind or social cognition is impaired in BD (4)andwas Achieva scanner (Philips, Best, the Netherlands) at the Centre found to positively associate with anatomical connectivity for Advanced Medical Imaging, St. James’s Hospital, Dublin, between default-mode regions in a recent healthy human Ireland. T1-weighted images were acquired using a three- network investigation (29); however, this has yet to be dimensional turbo field echo sequence (repetition time/echo investigated in BD. time = 8.5/3.9 ms; 1 mm3 isotropic voxel size). Diffusion- Here we investigated shared or differential cognition–brain weighted images were obtained using high angular diffusion network relationships in individuals with BD compared with imaging consisting of 1 non-diffusion-weighted image and 61 control individuals using novel anatomical network approaches diffusion gradient directions with b = 1200 s/mm2 (repetition across a wide range of cognitive domains to enhance under- time/echo time = 514/59 ms; SENSE parallel imaging factor = standing of the distinct brain network basis of cognitive 2.5; field of view = 200 3 257 3 125 mm; reconstructed 1.8 3 impairment in BD. We assessed relationships between vari- 1.8 3 1.9 mm3 voxel size; acquired 2.1-mm slice thickness; ance in global, rich-club, and interhemispheric connectivity in-plane resolution = 0.8 mm2). Diffusion images were cor- patterns and the global cognitive processes of intelligence and rected for eddy current distortions, motion artifacts, sus- executive function and in regional subnetworks underlying ceptibility effects, and rotations of the b-matrix for motion memory and social cognition, all commonly affected and and registered (nonlinear) to the T1-parcellation space playing a role in impaired quality of life experienced by in- (ExploreDTI version 4.8.6) (61). Quality assessment involved dividuals with BD (9). careful visual inspection for geometric distortions, large signal dropouts, abnormal model residuals (62), and regis- METHODS AND MATERIALS tration accuracy and resulted in the removal of 12 cases (7 HC individuals and 5 individuals with BD). A deterministic Participants non-tensor-based constrained spherical deconvolution al- We recruited individuals with a diagnosis of BD or healthy gorithm was applied to corrected diffusion-weighted data control (HC) individuals between 18 and 65 years of age and included recursive calibration of the response function through mental health services of the western region of (ExploreDTI version 4.8.6) (63,64). This estimates multiple fi- Ireland. The DSM-IV-TR criteria for BD were confirmed by a ber orientations within each voxel through the fiber orienta- psychiatrist using the Structured Clinical Interview for DSM- tion distribution function, allowing more accurate diffusion IV-TR (55). Healthy volunteers had no personal history of profiles and streamline reconstructions in the extensive areas psychiatric illness confirmed using the Structured Clinical of brain in which there are complex fiber arrangements Interview for DSM-IV nonpatient edition and had no first- compared with the single-fiber orientation per voxel afforded degree family history. Exclusion criteria included neurological by diffusion-tensor-based algorithms (65).

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI 153 Biological Psychiatry: CNNI Neuroanatomical Dysconnectivity and Cognition in BD

Network Reconstruction Anatomical Subnetwork Connectivity A total of 86 regions (34 cortical, 8 subcortical and cerebellar We investigated main effects of episodic memory, short-term hemispheres bilaterally) were defined, inspected, and corrected memory, spatial recognition memory, and social cognition and in a subject-specific manner (FreeSurfer version 5.3.0) (66,67). interactions between these cognitive performance measures For each participant, an 86 3 86 connectivity matrix was ob- and diagnosis on anatomical subnetwork connectivity using tained (ExploreDTI version 4.8.6), whereby one or more recon- cluster-based statistical methods that control for the familywise structed streamlines terminating in a pair of regions deemed error rate (Network-Based Statistic version 1.2). A t statistic them structurally connected. Connections were represented by representing the main effect of cognitive performance or inter- 1 or 0 to indicate the presence or absence of connections in the action between cognitive performance and diagnosis for each binary case and by fractional anisotropy over all connecting connection was calculated using a general linear model (Pear- streamlines or total number of connecting streamlines in the son’s correlation equivalent) while covarying for age, gender, weighted case. Subsequent analysis used binary, fractional and diagnosis. A primary t-statistic threshold of 2 corresponding anisotropy–weighted, and number of connecting streamlines– to p , .025 was applied, and 5000 permutations were used to weighted network measures to investigate network correlates calculate familywise error rate–corrected p values (pFWE) at .05 of cognitive performance in each diagnostic group as indicated. for every remaining connected component against a null distribution of maximum component size. Owing to the arbitrary Global, Rich-Club, and Interhemispheric choice of threshold, we searched for anatomical subnetworks at Connectivity additional thresholds of 1.5 and 2.5 and the statistical package default of 3 (71). Results were investigated post hoc by corre- Full-scale IQ and executive function were investigated in lating the average strength of significant subnetworks with relation to measures of global connectivity and topology, rich- cognitive measures in each diagnostic group. While IQ and ex- club connectivity, and interhemispheric connectivity as hy- ecutive function were not hypothesized as being related to pothesized using partial correlation covarying for age and distinct anatomical subnetworks, exploratory analysis investi- gender. Uncorrected p values are presented for this analysis. gated main effects of these facets and interactions with diag- Fisher r-to-z transformation compared relationships between nosis on anatomical subnetwork connectivity. groups. Measures of global connectivity and topology included density, global strength, global efficiency, and global clustering coefficient (Supplemental Table S1)(68). Rich-club organiza- RESULTS tion within weighted networks was established using the Participants weighted rich-club coefficient BW ðkÞ (69), whereby the total connection weight for the group of brain regions with greater In total, 32 individuals with BD and 38 psychiatrically healthy individuals balanced for age and gender but not for years of than k connections (W,k) is divided by the total connection weight of the same number of strongest connections within the education were included (Table 1). Of the BD group, 27 in- ranked dividuals met the DSM-IV diagnosis for BD I (13 men and 14 network (obtained by w Þ. The formula is as follows: women; mean age = 43 6 14 years) and 5 for BD II (2 men and W, 3 women; mean age = 43 6 13 years). At cognitive testing, all BW ðkÞ¼ P k ; E, ranked except 3 participants with BD were taking medication: 18 k w l21 l taking mood stabilizers (9 lithium), 19 taking antipsychotic medications (18 atypical antipsychotics), 10 taking antide- where E, is the subset of connections between regions with k pressant medications, 1 taking benzodiazepine, 6 taking other more than k connections. Normalized rich-club coefficients were calculated to determine the presence of rich-club orga- psychotropic medications, and 2 taking antiepileptic mood nization; observed rich-club coefficients were divided by the stabilizing medication (Supplemental Table S2). Formal average rich-club coefficient from 500 reference networks assessment of mood state at the time of scanning revealed obtained by randomly rewiring edges to retain degree distri- that 9 individuals with BD did not reach the threshold for bution (70). We obtained the top 10–ranking brain regions in euthymia (28%; Hamilton Rating Scale for Depression mean = terms of degree in each diagnostic group, and brain regions 16.44, SD = 4.93, range = 11–26; Young Mania Rating Scale common to both groups were defined as rich-club regions. mean = 0.87, SD = 1.59, range = 0–10). Removing individuals Connections were divided into rich club, those interconnecting taking lithium or those not meeting criteria for euthymia did not rich-club regions; local, those interconnecting non-rich-club change results presented hereafter. Mood scores did not regions; and feeder, those connecting rich-club and non- significantly differ between the day of scanning and cognitive rich-club regions. The total connection weight represented testing for the Hamilton Rating Scale for Depression (t = 21.45, connectivity in each class. To ensure that effects were not p = .16) or Young Mania Rating Scale (t = 21.06. p = .30). Time limited to this rich-club definition, rich-club regions were also between scanning and cognitive testing did not significantly defined post hoc using the top 12–ranking and top 15–ranking differ between control individuals and individuals with BD brain regions common to both groups. Interhemispheric con- (t = 20.10, p = .92). nectivity was calculated as the average inverse shortest path length for pairs of brain regions in contralateral hemispheres Comparison of Cognitive Performance Between (32). Relationships were not expected between global, rich- Diagnostic Groups club, or interhemispheric connectivity and memory or social The BD group had significantly worse performance in full-scale cognition, and they were examined as exploratory. IQ (F = 4.92, p = .03), executive function (U = 430.00, p = .04),

154 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI Biological Psychiatry: Neuroanatomical Dysconnectivity and Cognition in BD CNNI

Table 1. Demographic and Clinical Characteristics Statistical Comparison Control Group Bipolar Disorder Test Statistic (n = 38) Group (n = 32) (t or c2) p Value Age, Mean (SD) 39 (14) 43 (13) 21.07 .29 Gender, Male/Female, n 17/21 15/17 0.03 .86 Level of Education, na 12.58 .03b Junior high school 1 1 Some high school 0 2 High school graduate 3 5 Some college or technical school, at least 1 year 6 8 College graduate 12 14 Graduate training 15 2 HAM-D Score, Mean (SD), Range 1.08 (1.82), 0–7 6.50 (7.10), 0–26 24.5 2 3 1025b YMRS Score, Mean (SD), Range 0.94 (1.66), 0–6 1.53 (2.27), 0–10 21.4 0.18 Mood scores provided are from the day of scanning. HAM-D, Hamilton Depression Rating Scale; YMRS, Young Mania Rating Scale. aData missing for 1 healthy control subject; n = 69. bSignificant difference at p , .05. episodic memory (F = 7.37, p = .01), short-term memory (F = correction for multiple comparison. These were also seen 4.55, p = .04) and theory of mind (F = 6.44, p = .01) and when defining the rich club as the top 15–ranking brain regions had similar performance in terms of response times (F = 0.31, common to both groups. Visualizations of rich-club regions p = .58) and accuracy (F = 3.31, p = .07) in spatial recognition included at these thresholds are shown in Supplemental memory compared with the control group (Table 2 and Figure S2. Executive function was not associated with any Supplemental Figure S1). Removal of outliers did not change measures of global connectivity and topology (Figure 1B)or results for executive function (1 HC individual and 2 individuals rich-club connectivity (Figure 2) in our control group. Neither IQ with BD, U = 399.50, p = .05) or episodic memory (3 individuals nor executive function was associated with interhemispheric with BD, F = 4.50, p = .04). connectivity in control individuals (range r = 2.15 to .32, range p = .06 to .98) (Supplemental Table S3). We found no distinct Network Properties Related to Intelligence and anatomical subnetworks associated with IQ or executive function during exploratory post hoc investigation. Executive Function Higher full-scale IQ was not significantly associated with Network Properties Related to Episodic Memory, higher global efficiency in control individuals (r = .32, p = .06) (Figure 1A). The presence of rich-club organization was Short-Term Memory, Spatial Recognition Memory, confirmed in this sample. No associations were found between and Theory of Mind full-scale IQ and rich-club connectivity using our primary rich- We investigated episodic, short-term, and spatial recognition club definition (Figure 2). Post hoc investigation defining the memory and social cognition in relation to anatomical sub- rich club as the top 12–ranking brain regions common to both networks. Greater connectivity within overlapping networks diagnostic groups identified positive correlations between IQ involving the basal ganglia and thalamus was associated with and rich-club connectivity (r = .38, p = .02) and feeder con- faster response times (with the hippocampus, amygdala, and nectivity (r = .44, p = .01) in control individuals, not surviving frontal cortex, t = 2.0, pFWE = .02) (Figure 3), slower response

Table 2. Cognitive Performance: Control Group Versus Bipolar Disorder Group Statistical Comparison Control Group Bipolar Group Test Statistic Effect Size Task Outcome Measure (n = 38), Mean 6 SD (n = 32), Mean 6 SD (F or U) p Value (Cohen’s d) Full-Scale IQ IQ score 116.52 613.86 105.72 6 19.12 4.92 .03a 0.65 Intra/Extra Dimensional Shift Total errors adjusted 27.45 6 32.32 40.69 6 42.99 430.00 .04a 0.35 Paired Associates Learning First trial memory scoreb 20.26 6 3.47 18.57 6 4.67 2.17 .15 0.41 Total errors adjustedb 11.00 6 9.80 26.73 6 32.08 7.37 .01a 0.66 Delayed Match to Sample Percentage correct 91.58 6 6.67 87.19 6 8.32 4.55 .04a 0.58 Spatial Recognition Memory Percentage correct 80.26 6 11.15 74.22 6 11.92 3.31 .07 0.52 Mean correct latency (ms) 2639.95 6 807.05 2795.71 6 880.17 0.31 .58 0.18 Reading the Mind in the Eyes Total correct 26.63 6 3.87 23.53 6 4.57 6.44 .01a 0.73 aSigniﬁcant difference at p , .05. bData missing for 2 participants with bipolar disorder; n = 30.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI 155 Biological Psychiatry: CNNI Neuroanatomical Dysconnectivity and Cognition in BD

Figure 1. Relationships between (A) full-scale IQ and global efficiency (GEFA)(z = 1.94, p = .05), (B) errors on the executive function task and global clustering coefficient (CCbinary)(z = 21.60, p = .11), and (C) spatial recognition memory percentage correct and global efficiency (GEbinary)(z = 22.24, p = .03) across diagnostic groups. Asterisk represents significant relationship at p , .05. Healthy control (HC) individuals are represented by black open circles (B), regression line, and dashed line confidence intervals, while individuals with bipolar disorder (BD) are represented by gray closed circles (•), regression line, and dashed line confidence intervals. FA, fractional anisotropy; WAIS-III, Wechsler Adult Intelligence Scale.

times (with the cerebellum and left parietal cortex, t = 2.0, p = .41; z = 1.94, p = .05) (Figure 1A). To ensure that the pFWE = .02) (Figure 4), and lower accuracy (with the basal altered relationship between IQ and global efficiency was not ganglia and thalamus alone, t = 2.5, pFWE = .04) (Figure 5)in driven by differences in IQ, we divided the whole sample into spatial recognition memory in the whole cohort. These sub- low- and high-IQ groups, with 35 people in each, and tested networks were not seen at the additional thresholds tested. No for relationships with global efficiency. No relationships were anatomical subnetworks were associated with episodic found in either group (high IQ: r = .11, p = .54; low IQ: r = .09, memory, short-term memory, or social cognition. p = .61; z = 0.08, p = .94), supporting this as a diagnostic As expected, no significant associations were found be- effect. tween episodic, short-term, and spatial recognition memory or Lower executive functioning in the BD group was accom- social cognition and measures of global, rich-club, or inter- panied by a positive association between executive function hemispheric connectivity during exploratory post hoc and global clustering coefficient that was not significantly investigation. different from the relationship seen in the control group (HC individuals: r = .06, p = .70; individuals with BD: r = .44, p = .02; Group Differences z = 21.60, p = .11) (Figure 1B). When false discovery rate was Lower IQ performance in the BD group was accompanied by a corrected at 5% for 12 comparisons, relationships between dissociation between IQ and full-scale IQ and global efficiency global measures and intelligence and executive function were (HC individuals: r = .32, p = .06; individuals with BD: r = 2.16, no longer statistically significant. FPO = web 4C Figure 2. (A) Rich-club organization within structural brain networks. Rich-club regions were defined as the common top 10–ranking brain regions by nodal degree in each diagnostic group. Brain regions are scaled by nodal degree (size of spheres) and colored to indicate whether they represent rich-club (red) or non-rich-club (gray) regions. Connections between rich-club regions are represented in red. (B) Correlations between cognitive performance and rich-club edge levels. Intelligence and executive function were measured using the composite Wechsler Adult Intelligence Scale and intra/extra dimensional shift total errors adjusted scores, respectively. L, left; R, right.

156 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI Biological Psychiatry: Neuroanatomical Dysconnectivity and Cognition in BD CNNI

Figure 3. A single number of streamlines (NOS)- weighted anatomical subnetwork was negatively correlated with spatial recognition memory mean correct latency over all participants, covarying for age, gender, and diagnosis (t = 2.0, p =.02), while no subnetwork differentially related to spatial recognition memory mean correct latency between diagnostic groups. (A) Visualization of signiﬁcant anatomical subnetwork. (B) Relation- ship between average strength of this anatomical subnetwork and spatial recognition memory mean correct latency score separated by diagnostic group. Note that relationships are not signiﬁcantly different between groups. Partial correlations included age and gender as covariates. Asterisk

FPO represents significant relationship at p , .05. = Healthy control (HC) individuals are represented by black open circles (B), regression line, and dashed line confidence intervals, while individuals with bipolar disorder (BD) are represented by gray closed circles ( web 4C •), regression line, and dashed line confidenceintervals.R,right.

No differential relationships were seen between IQ and ex- Despite detecting anticipated deficits in social cognition in BD (4), ecutive function and either rich-club or interhemispheric con- no anatomical subnetwork was found as relating to social nectivity in the BD group relative to the control group except IQ cognitive ability in the population or differentially in BD. and interhemispheric efficiency (HC individuals: r = .32, p = .06; individuals with BD: r = 2.15, p = .45; z = 1.92, p = .05) (Supplemental Table S3). Network Features of IQ and Executive Function Subnetwork relationships with spatial recognition memory We detected IQ deficits, possibly due to residual mood symp- response times and accuracy detailed above were not signif- toms or lower levels of education (1,73), and executive func- icantly different in the BD group compared with the control tioning deficits in individuals with BD relative to control group (Figures 3, 4, and 5). Exploratory post hoc investigation individuals, both consistent with recent meta-analyses (1,3). found that spatial recognition memory accuracy was positively Matching study cohorts for IQ may account for less consistent associated with global efficiency in individuals with BD (r = .39, reporting of IQ deficits in BD literature (74). Our network findings p = .03), a relationship not present in control individuals are not inconsistent with previous work establishing the rele- (r = 2.15, p = .39; z = 22.24, p = .03) (Figure 1C). vance of global anatomical network efficiency for intelligence in No anatomical subnetworks were found that related to healthy individuals (26,75–77), and suggest a dissociation be- episodic memory, short-term memory, or social cognition tween IQ and this network feature in BD, which may relate to differently in individuals with BD compared with control in- previously observed reductions in global and interhemispheric dividuals. The same was true for IQ and executive function efficiency that reflect abnormal widespread network integration during exploratory post hoc investigation. (30,32–34). In light of studies adopting anatomically localized approaches implicating the temporal lobe (40) and prefrontal cortex (41) in IQ impairments in BD, future investigations DISCUSSION determining the extent to which local network changes influence Consistent with a substantial body of work to date, we detected altered global network support of IQ are warranted. deficits in cognition associated with BD that incorporate pro- We detected no relationship between intelligence and rich- cesses expected to be global in their anatomical underpinnings, club connectivity in healthy participants, consistent with including intelligence and executive function and those relying on several reports (31,78), or in participants with BD despite the more anatomically specificnetworkssuchassocialcognitionand reliance of this facet on global integration that is thought to forms of memory. We found that IQ may have a differential rela- emerge from the interconnectivity of the rich club (72). Studies tionship with global efficiency in individuals with BD compared we are at variance with have used general cognitive ability with control individuals and that IQ performance was not (27,77) or a perceptual reasoning index (79) compared with our explained in either group by the highly interconnected rich-club measure of full-scale IQ and have included frontal cortex re- subnetwork expected to underlie core cognitive integration (72). gions known to support both IQ and executive functioning Executive functioning deficits in BD related to increased segre- (49,51). Examining these regions in the current data at addi- gation globally, while neither global nor rich-club connectivity tional rich-club thresholds corroborates previous relationships explained executive function performance generally. Basal with intelligence (27,77,79). Results presented here suggest an ganglia and thalamus interconnections appeared to be important absence of the established relationship between IQ and global for spatial recognition memory accuracy, and their concomitant anatomical network efficiency in individuals with BD and no facilitatory and inhibitory connections with other brain regions relationship between IQ performance and rich-club connec- related to response times. This complex subnetwork relationship tivity in HC individuals or individuals with BD. with spatial memory did not generalize to episodic or short-term We found that increased segregation globally may relate memory and did not explain deficits in the latter pair evident in BD. to worse executive function performance in BD, while global,

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI 157 Biological Psychiatry: CNNI Neuroanatomical Dysconnectivity and Cognition in BD

Figure 4. A single number of streamlines (NOS)- weighted anatomical subnetwork was positively correlated with spatial recognition memory mean correct latency score over all participants, covarying for age, gender, and diagnosis (t =2.0, p = .02), while no subnetwork differentially related to spatial recognition memory mean correct latency between diagnostic groups. (A) Visualization of anatomical subnetwork. (B) Relationship between average strength of this anatomical subnetwork and spatial recognition memory mean correct latency score separated by diagnostic group. Partial correlations included age and gender as covariates. Asterisk represents signiﬁ-

FPO cant relationship at p , .05. Healthy control (HC) = individuals are represented by black open circles (B), regression line, and dashed line conﬁdence

web 4C intervals, while individuals with bipolar disorder (BD) are represented by gray closed circles (•), regression line, and dashed line conﬁdence intervals. L, left; R, right.

rich-club, and interhemispheric connectivity do not explain Network Features of Memory and Social Cognition executive function performance generally. Both this study We detected previously reported episodic and short-term and Ajilore et al. (80) detected relationships between memory deficits in BD (1,3) and no spatial recognition memory increased anatomical network segregation and worse exec- deficits, mirroring several investigations (84–86) and contrasting utive functioning in BD, in our case globally and in the pre- one report (87) in which mixed/manic mood state may account vious study locally within the lateral orbitofrontal cortex. This for poorer BD performance (73). Basal ganglia and thalamus remains consistent with previous anatomically localized interconnections were associated with spatial memory accu- findings implicating widespread white matter alterations in racy, while their connections to frontal cortex and limbic areas executive functioning deficits (44–47). Larger cohorts may be and the parietal cortex and cerebellum related to faster and required to detect relationships between executive func- slower response times, respectively. These results lend support tioning and global efficiency in healthy individuals (29,81,82) to the basal ganglia as a point of integration between cognitive and subtle differential relationships in BD. However, we and motor systems (18) that are involved in opposing processes detected no relationship between executive functioning and depending on coupled regions (88). Considering the resolution rich-club connectivity (83), in contrast to a similarly powered limits of diffusion magnetic resonance imaging approaches, we investigation that excluded subcortical connections cannot speak to the underlying inhibitory or excitatory nature of from brain networks (27). We note that the majority of studies the connections involved. The exclusively right hemisphere we are at variance with consider multiple domains consti- subnetwork related to faster response times is consistent with a tuting executive functioning (27,29,81) and that control right hemisphere bias for spatial encoding and retrieval (89,90) individuals here showed low variance in executive func- and extends reports of striatum, caudate nucleus (91), and tioning due to high performance, both of which could amygdala (92) involvement in memory processes to the contribute to discrepancies. Overall, our findings suggest anatomical subnetwork level. This complex subnetwork rela- that increased global segregation, and not rich-club or tionship did not hold for episodic or short-term memory and did interhemispheric connectivity, relates to BD executive func- not explain deficits in these forms of memory in BD. Applying tioning deficits. brain-wide correction for multiple comparisons can lead to false

Figure 5. A single fractional anisotropy (FA)- weighted anatomical subnetwork was negatively correlated with spatial recognition memory percentage correct score over all participants, covarying for age, gender, and diagnosis (t =2.5, p = .04), while no subnetwork differentially related to spatial recognition memory percentage correct between diagnostic groups. (A) Visualization of anatomical subnetwork. (B) Relationship between average strength of this anatomical subnetwork and spatial recognition memory percentage correct score separated by diagnostic group. Note that relationships are not signiﬁcantly different between groups. Partial correlations included age and gender as covariates. Asterisk represents

FPO significant relationship at p , .05. Healthy control = (HC) individuals are represented by black open circles (B), regression line, and dashed line confidence intervals, while individuals with bipolar disorder (BD) are represented by gray closed circles ( web 4C •), regression line, and dashed line confidence intervals. L, left; R, right.

158 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI Biological Psychiatry: Neuroanatomical Dysconnectivity and Cognition in BD CNNI

negatives, and larger homogeneous cohorts may be required to false positives. There also remains interindividual differences overcome this (54). However, this does remain potentially that we would not be able to detect with the current design and consistent with previous anatomically localized investigations approach (98). Furthermore, altered functional connectivity implicating the amygdala (52) and corticospinal tract (53)in may exist in the absence of currently detectable architectural delayed memory deficits and the corona radiata (53) in short- perturbations (18,99,100), and future work on functional term memory deficits in BD that this cluster-based network network dynamics can delve deeper into the influence of study might not have been able to detect. altered network connectivity on cognitive function in BD (101). We detected an expected deficit in social cognition in BD (4) and no anatomical subnetworks related to this facet in the Conclusions population or differently in BD. Research points to abnormal The potential of graph theory to understand the brain’s topo- limbic activation in BD during social cognition tasks (93), and logical associations with human cognitive ability was demon- we provide evidence suggesting that anatomical subnetwork strated. Here, we detected selective influence of subnetwork connectivity, which forms the basis for dynamic functional patterns of connectivity in underlying cognitive performance interactions, does not explain social cognition deficits in BD. generally and abnormal global topology in underlying discrete Expected deficits in executive function in BD were associ- cognitive impairments in BD. ated with measures of global anatomical network segregation but not the more anatomically limited and highly interconnected rich-club subnetwork. This suggests that alter- ACKNOWLEDGMENTS AND DISCLOSURES ations in global topology, but not rich-club topology, in BD This research was funded by the Health Research Board (Grant No. HRA- demonstrated by others (33,34) and previously reported in the POR-324 [to DMC]) and the Irish Research Council Government of Ireland current sample (94) may contribute to cognitive deficits. Postgraduate Scholarship (to GM). Complex subnetwork relationships with spatial recognition We gratefully acknowledge the participants and the support of the memory, not impaired in BD, were found generally but did not Welcome Trust-HRB Clinical Research Facility and the Centre for Advanced Medical Imaging at St. James’s Hospital Dublin. We also thank Andrew explain anticipated episodic or short-term memory deficits or Hoopes (MGH/HST Martinos Center for Biomedical Imaging) for FreeSurfer social cognition deficits in BD. Cognitive deficits have impor- software support, Christopher Grogan for his contribution to data pro- tant implications for quality of life and functional outcomes in cessing, and Jenna Pittman and Fiona Martyn for their contribution to data BD (95), and a better understanding of the brain basis that handling. accompanies difficulties with these processes could provide a The authors report no biomedical financial interests or potential conflicts foundation for treatments targeting these as part of a wider of interest. treatment approach. The only previous study applying a network-based approach to address this found relationships ARTICLE INFORMATION between reduced interhemispheric connectivity and both From the Centre for Neuroimaging & Cognitive Genomics (GM, LN, SO, GT, processing speed and working memory deficits in BD (80). LC, PN, SA, GN, SC, FB, JM, CM, BH, DMC), Clinical Neuroimaging Lab, Future network-based studies broadening this literature can NCBES Galway Neuroscience Centre, College of Medicine, Nursing, and clarify the network alterations important for distinct cognitive Health Sciences; College of Science and Engineering (LK); and School of impairments, which so far appear to be features of global Psychology (DO), National University of Ireland Galway, Galway, Republic of integration, segregation, and interhemispheric connectivity. Ireland. Address correspondence to Genevieve McPhilemy, B.Sc., Clinical Neu- roimaging Laboratory, 1023 Human Biology Building, University Road, National University of Ireland Galway, Galway, Ireland; E-mail: g.mcphilemy1@ Strengths, Limitations and Future Directions nuigalway.ie. This study used network analysis to examine relationships Received Apr 2, 2019; revised Sep 6, 2019; accepted Sep 7, 2019. between cognition and neural structure in participants with BD Supplementary material cited in this article is available online at https:// doi.org/10.1016/j.bpsc.2019.09.004. and healthy participants, addressing the multivariate pattern of integration that underlies complex cognitive processing. Similar relationships suggest that variations in network struc- REFERENCES ture have similar implications in terms of cognitive perfor- 1. Bourne C, Aydemir O, Balanzá-Martínez V, Bora E, Brissos S, mance, while altered relationships suggest a breakdown in the Cavanagh JTO, et al. (2013): Neuropsychological testing of cognitive extent to which network structure is providing support for impairment in euthymic bipolar disorder: An individual patient data cognitive functions in BD. The application of non-tensor-based meta-analysis. Acta Psychiatr Scand 128:149–162. tractography in combination with subject-specific cortical and 2. Torres IJ, Boudreau VG, Yatham LN (2007): Neuropsychological subcortical brain region definition produces more accurate functioning in euthymic bipolar disorder: A meta-analysis. Acta Psychiatr Scand 116:17–26. network reconstructions and increases the anatomical sensi- 3. Mann-Wrobel MC, Carreno JT, Dickinson D (2011): Meta-analysis tivity of our findings (96,97). Despite capitalizing on the largest of neuropsychological functioning in euthymic bipolar disorder: cohort to date investigating these network–behavior relation- An update and investigation of moderator variables. Bipolar Dis- ships, we had limited sensitivity to detect more subtle effects. ord 13:334–342. In addition, the effect sizes of our global analyses are moderate 4. Bora E, Bartholomeusz C, Pantelis C (2016): Meta-analysis of theory of and would not have survived false discovery rate correction, in mind (ToM) impairment in bipolar disorder. Psychol Med 46:253–264. 5. Cullen B, Ward J, Graham NA, Deary IJ, Pell JP, Smith DJ, Evans JJ part owing to the breadth of cognitive assessments used (2016): Prevalence and correlates of cognitive impairment in euthymic across multiple network measures that, while representing a adults with bipolar disorder: A systematic review. J Affect Disord strength of the current work, somewhat increases our risk of 205:165–181.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI 159 Biological Psychiatry: CNNI Neuroanatomical Dysconnectivity and Cognition in BD

6. Goswami U, Sharma A, Varma A, Gulrajani C, Ferrier IN, Young AH, connectome structure in bipolar I disorder. Hum Brain Mapp 37:122– et al. (2009): The neurocognitive performance of drug-free and 134. medicated euthymic bipolar patients do not differ. Acta Psychiatr 31. Collin G, de Nijs J, Hulshoff Pol HE, Cahn W, van den Heuvel MP Scand 120:456–463. (2016): Connectome organization is related to longitudinal changes in 7. Wingo AP, Wingo TS, Harvey PD, Baldessarini RJ (2009): Effects of general functioning, symptoms and IQ in chronic schizophrenia. lithium on cognitive performance. J Clin Psychiatry 70:1588–1597. Schizophr Res 173:166–173. 8. Sabater A, García-Blanco AC, Verdet HM, Sierra P, Ribes J, Villar I, 32. Leow A, Ajilore O, Zhan L, Arienzo D, Gadelkarim J, Zhang A, et al. (2017): Comparative neurocognitive effects of lithium and an- et al. (2013): Impaired inter-hemisphericintegrationinbipolar ticonvulsants in long-term stable bipolar patients. J Affect Disord disorder revealed with brain network analyses. Biol Psychiatry 190:34–40. 73:183–193. 9. Baune BT, Malhi GS (2015): A review on the impact of cognitive 33. O’Donoghue S, Kilmartin L, O’Hora D, Emsell L, Langan C, dysfunction on social, occupational, and general functional out- McInerney S, et al. (2017): Anatomical integration and rich-club comes in bipolar disorder. Bipolar Disord 17(suppl 2):41–55. connectivity in euthymic bipolar disorder. Psychol Med 47:1609– 10. Vederine FE, Wessa M, Leboyer M, Houenou J (2011): A meta-analysis 1623. of whole-brain diffusion tensor imaging studies in bipolar disorder. 34. Wang Y, Deng F, Jia Y, Wang J, Zhong S, Huang H, et al. (2019): Prog Neuropsychopharmacol Biol Psychiatry 35:1820–1826. Disrupted rich club organization and structural brain connectome in 11. Nortje G, Stein DJ, Radua J, Mataix-Cols D, Horn N (2013): Sys- unmedicated bipolar disorder. Psychol Med 49:510–518. tematic review and voxel-based meta-analysis of diffusion tensor 35. Perry A, Roberts G, Mitchell PB, Breakspear M (2019): Connectomics imaging studies in bipolar disorder. J Affect Disord 150:192–200. of bipolar disorder: A critical review, and evidence for dynamic in- 12. Hibar DP, Westlye LT, Van Erp TGM, Rasmussen J, Leonardo CD, stabilities within interoceptive networks. Mol Psychiatry 24:1296– Faskowitz J, et al. (2016): Subcortical volumetric abnormalities in 1318. bipolar disorder. Mol Psychiatry 21:1710–1716. 36. Stefanopoulou E, Manoharan A, Landau S, Geddes JR, Goodwin G, 13. Hibar DP, Westlye LT, Doan NT, Jahanshad N, Cheung JW, Frangou S (2009): Cognitive functioning in patients with affective Ching CRK, et al. (2017): Cortical abnormalities in bipolar disorder: An disorders and schizophrenia: A meta-analysis. Int Rev Psychiatry MRI analysis of 6503 individuals from the ENIGMA Bipolar Disorder 21:336–356. Working Group. Mol Psychiatry 23:932–942. 37. Trotta A, Murray RM, Maccabe JH (2015): Do premorbid and post- 14. Emsell L, Leemans A, Langan C, Van Hecke W, Barker GJ, onset cognitive functioning differ between schizophrenia and bipo- McCarthy P, et al. (2013): Limbic and callosal white matter changes in lar disorder? A systematic review and meta-analysis. Psychol Med euthymic bipolar I disorder: An advanced diffusion magnetic reso- 45:381–394. nance imaging tractography study. Biol Psychiatry 73:194–201. 38. Bora E, Yucel M, Pantelis C (2009): Cognitive endophenotypes of 15. Bressler SL, Menon V (2010): Large-scale brain networks in cognition: bipolar disorder: A meta-analysis of neuropsychological deficits in Emerging methods and principles. Trends Cogn Sci 14:277–290. euthymic patients and their first-degree relatives. J Affect Disord 16. Mcintosh AR (1999): Mapping cognition to the brain through neural 113:1–20. interactions. Memory 7:523–548. 39. Tsitsipa E, Fountoulakis KN (2015): The neurocognitive functioning in 17. Petersen SE, Sporns O (2015): Brain networks and cognitive archi- bipolar disorder: A systematic review of data. Ann Gen Psychiatry tectures. Neuron 88:207–219. 14:42. 18. Medaglia JD, Lynall M-E, Bassett DS (2015): Cognitive network 40. Bruno SD, Papadopoulou K, CercignaniI M, Cipolotti L, Ron MA neuroscience. J Cogn Neurosci 27:1471–1491. (2006): Structural brain correlates of IQ changes in bipolar disorder. 19. Sporns O (2014): Contributions and challenges for network models in Psychol Med 36:609–618. cognitive neuroscience. Nat Neurosci 17:652–660. 41. McIntosh AM, TWJ Moorhead, McKirdy J, Hall J, Sussmann JED, 20. Bullmore E, Sporns O (2012): The economy of brain network orga- Stanfield AC, et al. (2009): Prefrontal gyral folding and its cognitive nization. Nat Rev Neurosci 13:336–349. correlates in bipolar disorder and schizophrenia. Acta Psychiatr 21. Bassett DS, Bullmore ET (2017): Small-world brain networks revis- Scand 119:192–198. ited. Neuroscientist 23:499–516. 42. Robinson LJ, Thompson JM, Gallagher P, Goswami U, Young AH, 22. Espinosa-Soto C, Wagner A (2010): Specialization can drive the Ferrier IN, Moore PB (2006): A meta-analysis of cognitive deficits in evolution of modularity. PLoS Comput Biol 6:e1000719. euthymic patients with bipolar disorder. J Affect Disord 93:105– 23. Sporns O, Betzel RF (2016): Modular brain networks. Annu Rev 115. Psychol 67:613–640. 43. Haldane M, Cunningham G, Androutsos C, Frangou S (2008): 24. Van den Heuvel MP, Kahn RS, Goni J, Sporns O (2012): High-cost, Structural brain correlates of response inhibition in bipolar disorder I. high-capacity backbone for global brain communication. Proc Natl J Psychopharmacol 22:138–143. Acad Sci. U S A 109:11372–11377. 44. Poletti S, Bollettini I, Mazza E, Locatelli C, Radaelli D, Vai B, et al. 25. van den Heuvel MP, Sporns O (2013): Network hubs in the human (2015): Cognitive performances associate with measures of white brain. Trends Cogn Sci 17:683–696. matter integrity in bipolar disorder. J Affect Disord 174:342–352. 26. Li Y, Liu Y, Li J, Qin W, Li K, Yu C, Jiang T (2009): Brain anatomical 45. Magioncalda P, Martino M, Conio B, Piaggio N, Teodorescu R, network and intelligence. PLoS Comput Biol 5:e1000395. Escelsior A, et al. (2016): Patterns of microstructural white matter 27. Baggio HC, Segura B, Junque C, de Reus MA, Sala-Llonch R, Van abnormalities and their impact on cognitive dysfunction in the den Heuvel MP (2015): Rich club organization and cognitive perfor- various phases of type I bipolar disorder. J Affect Disord 193:39– mance in healthy older participants. J Cogn Neurosci 27:1801–1810. 50. 28. Koenis MMG, Brouwer RM, Swagerman SC, van Soelen ILC, 46. Linke J, King AV, Poupon C, Hennerici MG, Gass A, Wessa M (2013): Boomsma DI, Hulshoff Pol HE (2018): Association between structural Impaired anatomical connectivity and related executive functions: brain network efficiency and intelligence increases during adoles- Differentiating vulnerability and disease marker in bipolar disorder. cence. Hum Brain Mapp 39:822–836. Biol Psychiatry 74:908–916. 29. Baum GL, Ciric R, Roalf DR, Betzel RF, Moore TM, Shinohara RT, 47. Oertel-Knöchel V, Reinke B, Alves G, Jurcoane A, Wenzler S, et al. (2017): Modular segregation of structural brain networks sup- Prvulovic D, et al. (2014): Frontal white matter alterations are asso- ports the development of executive function in youth. Curr Biol ciated with executive cognitive function in euthymic bipolar patients. 27:1561–1572.e8. J Affect Disord 155:223–233. 30. Collin G, van den Heuvel MP, Abramovic L, Vreeker A, de Reus MA, 48. Colom R, Karama S, Jung RE, Haier RJ (2010): Human intelligence van Haren NEM, et al. (2016): Brain network analysis reveals affected and brain networks. Dialogues Clin Neurosci 12:489–501.

160 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI Biological Psychiatry: Neuroanatomical Dysconnectivity and Cognition in BD CNNI

49. Jung RE, Haier RJ (2007): The parieto-frontal integration theory (P- 72. van den Heuvel MP, Sporns O (2013): An anatomical substrate for FIT) of intelligence: Converging neuroimaging evidence. Behav Brain integration among functional networks in human cortex. J Neurosci Sci 30:135–154. 33:14489–14500. 50. Rabinovici GD, Stephens ML, Possin KL (2015): Executive dysfunc- 73. Vrabie M, Marinescu V, Talas¸ man A, Tautu O, Drima E, Miclut¸ia I tion. Continuum (Minneap Minn) 21:646–659. (2015): Cognitive impairment in manic bipolar patients: Important, 51. Niendam TA, Laird AR, Ray KL, Dean YM, Glahn DC, Carter CS understated, significant aspects. Ann Gen Psychiatry 14:41. (2012): Meta-analytic evidence for a superordinate cognitive control 74. Torres I, Sole B, Vieta E, Martinez-Aran A (2012): Neurocognitive network subserving diverse executive functions. Cogn Affect Behav impairment in the bipolar spectrum [review]. Neuropsychiatry 2:43–55. Neurosci 12:241–268. 75. Koenis MMG, Brouwer RM, van den Heuvel MP, Mandl RCW, van 52. Killgore WDS, Rosso IM, Gruber SA, Yurgelun-Todd DA (2009): Soelen ILC, Kahn RS, et al. (2015): Development of the brain’s Amygdala volume and verbal memory performance in schizophrenia structural network efficiency in early adolescence: A longitudinal DTI and bipolar disorder. Cogn Behav Neurol 22:28–37. twin study. Hum Brain Mapp 36:4938–4953. 53. Bauer IE, Ouyang A, Mwangi B, Sanches M, Zunta-Soares GB, 76. Zalesky A, Fornito A, Seal ML, Cocchi L, Westin CF, Bullmore ET, Keefe RSE, et al. (2015): Reduced white matter integrity and verbal et al. (2011): Disrupted axonal fiber connectivity in schizophrenia. Biol fluency impairment in young adults with bipolar disorder: A diffusion Psychiatry 69:80–89. tensor imaging study. J Psychiatr Res 62:115–122. 77. Yeo RA, Ryman SG, van den Heuvel MP, de Reus MA, Jung RE, 54. Coynel D, Gschwind L, Fastenrath M, Freytag V, Milnik A, Spalek K, Pommy J, et al. (2016): Graph metrics of structural brain networks in et al. (2017): Picture free recall performance linked to the brain’s individuals with schizophrenia and healthy controls: Group differ- structural connectome. Brain Behav 7:e721. ences, relationships with intelligence, and genetics. J Int Neuro- 55. American Psychiatric Association (2013): Diagnostic and Statistical psychol Soc 22:240–249. Manual of Mental Disorders, 5th Edition: DSM-5. Washington, DC: 78. Ma J, Kang HJ, Kim JY, Jeong HS, Im JJ, Namgung E, et al. (2017): American Psychiatric Press. Network attributes underlying intellectual giftedness in the devel- 56. Hamilton M (1959): The assessment of anxiety states by rating. Br J oping brain. Sci Rep 7:11321. Med Psychol 32:50–55. 79. Kim DJ, Davis EP, Sandman CA, Sporns O, O’Donnell BF, Buss C, 57. Young RC, Biggs JT, Ziegler VE, Meyer DA (1978): A rating scale for Hetrick WP (2016): Children’s intellectual ability is associated with mania: Reliability, validity and sensitivity. Br J Psychiatry 133:429– structural network integrity. NeuroImage 124:550–556. 435. 80. Ajilore O, Vizueta N, Walshaw P, Zhan L, Leow A, Altshuler LL (2015): 58. Wechsler D (1997): WAIS-III: Administration and Scoring Manual: Connectome signatures of neurocognitive abnormalities in euthymic Wechsler Adult Intelligence Scale. San Antonio, TX: Psychological bipolar I disorder. J Psychiatr Res 68:37–44. Corporation. 81. Wen W, Zhu W, He Y, Kochan NA, Reppermund S, Slavin MJ, et al. 59. Cambridge Cognition (2018): CANTAB [cognitive assessment soft- (2011): Discrete neuroanatomical networks are associated with ware]. Cambridge, UK: Cambridge Cognition. specific cognitive abilities in old age. J Neurosci 31:1204–1212. 60. Baron-Cohen S, Wheelwright S, Hill J, Raste Y, Plumb I (2001): The 82. Xiao M, Ge H, Khundrakpam BS, Xu J, Bezgin G, Leng Y, et al. (2016): “Reading the Mind in the Eyes” Test revised version: A study with Attention performance measured by attention network test is corre- normal adults, and adults with Asperger syndrome or high- lated with global and regional efficiency of structural brain networks. functioning autism. J Child Psychol Psychiatry 42:241–251. Front Behav Neurosci 10:194. 61. Leemans A, Jeurissen B, Sijbers J, Jones D (2009): ExploreDTI: A 83. Mai N, Zhong X, Chen B, Peng Q, Wu Z, Zhang W, et al. (2017): graphical toolbox for processing, analyzing, and visualizing diffusion Weight rich-club analysis in the white matter network of late-life MR data. Proceedings of the 17th International Society of Magnetic depression with memory deficits. Front Aging Neurosci 9:279. Resonance in Medicine, April 18–24, Honolulu, Hawaii, p. 2527. 84. Roiser JP, Cannon DM, Gandhi SK, Tavares JT, Erickson K, Wood S, 62. Tournier J, Mori S, Leemans A (2011): Diffusion tensor imaging and et al. (2009): Hot and cold cognition in unmedicated depressed beyond. Magn Reson Med 1556:1532–1556. subjects with bipolar disorder. Bipolar Disord 11:178–189. 63. Tournier JD, Yeh CH, Calamante F, Cho KH, Connelly A, Lin CP 85. Tournikioti K, Ferentinos P, Michopoulos I, Alevizaki M, Soldatos CR, (2008): Resolving crossing fibres using constrained spherical Dikeos D, Douzenis A (2017): Clinical and treatment-related pre- deconvolution: Validation using diffusion-weighted imaging phantom dictors of cognition in bipolar disorder: Focus on visual paired data. NeuroImage 42:617–625. associative learning. Eur Arch Psychiatry Clin Neurosci 267:661–669. 64. Jeurissen B, Leemans A, Jones DK, Tournier J-D, Sijbers J (2011): 86. Bauer IE, Wu MJ, Frazier TW, Mwangi B, Spiker D, Zunta-Soares GB, Probabilistic fiber tracking using the residual bootstrap with con- Soares JC (2016): Neurocognitive functioning in individuals with bi- strained spherical deconvolution. Hum Brain Mapp 32:461–479. polar disorder and their healthy siblings: A preliminary study. J Affect 65. Tournier JD, Calamante F, Gadian DG, Connelly A (2004): Direct Disord 201:51–56. estimation of the fiber orientation density function from diffusion- 87. Sweeney JA, Kmiec JA, Kupfer DJ (2000): Neuropsychologic im- weighted MRI data using spherical deconvolution. NeuroImage pairments in bipolar and unipolar mood disorders on the CANTAB 23:1176–1185. neurocognitive battery. Biol Psychiatry 48:674–684. 66. Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, 88. Nagano-Saito A, Martinu K, Monchi O (2014): Function of basal Blacker D, et al. (2006): An automated labeling system for subdividing ganglia in bridging cognitive and motor modules to perform an ac- the human cerebral cortex on MRI scans into gyral based regions of tion. Front Neurosci 8:187. interest. NeuroImage 31:968–980. 89. Bellgowan PSF, Buffalo EA, Bodurka J, Martin A (2009): Lateralized 67. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, Haselgrove C, et al. spatial and object memory encoding in entorhinal and perirhinal (2002): Whole brain segmentation: Automated labeling of neuroana- cortices. Learn Mem 16:433–438. tomical structures in the human brain. Neuron 33:341–355. 90. Postma A, Kessels RPC, van Asselen M (2008): How the brain re- 68. Rubinov M, Sporns O (2010): Complex network measures of brain members and forgets where things are: The neurocognition of object- connectivity: Uses and interpretations. NeuroImage 52:1059–1069. location memory. Neurosci Biobehav Rev 32:1339–1345. 69. Opsahl T, Colizza V, Panzarasa P, Ramasco JJ (2008): Prominence 91. Méndez-Couz M, Conejo NM, Gonzá Lez-Pardo H, Arias JL (2015): and control: The weighted rich-club effect. Phys Rev Lett Functional interactions between dentate gyrus, striatum and ante- 101:168702. rior thalamic nuclei on spatial memory retrieval. Brain Res 70. Maslov S, Sneppen K (2002): Specificity and stability in topology of 1605:59–69. protein networks. Science 296:910–913. 92. Inman CS, Manns JR, Bijanki KR, Bass DI, Hamann S, Drane DL, et al. 71. Zalesky A, Fornito A, Bullmore ET (2010): Network-based statistic: (2018): Direct electrical stimulation of the amygdala enhances Identifying differences in brain networks. NeuroImage 53:1197–1207. declarative memory in humans. Proc Natl Acad Sci U S A 115:98–103.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI 161 Biological Psychiatry: CNNI Neuroanatomical Dysconnectivity and Cognition in BD

93. Turchi F, Amodeo G, Favaretto E, Righini S, Mellina E, La Mela C, quantiﬁcation of brain subcortical volume. Brain Imaging Behav Fagiolini A (2016): Neural basis of social cognition in bipolar disorder 12:1678–1695. [Italian]. Riv Psichiatr 51:177–189. 98. Wolfers T, Doan NT, Kaufmann T, Alnæs D, Moberget T, Agartz I, 94. Nabulsi L, McPhilemy G, Kilmartin L, O’Hora D, O’Donoghue S, et al. (2018): Mapping the heterogeneous phenotype of schizophrenia Forcellini G, et al. (2019): Bipolar disorder and gender are associated and bipolar disorder using normative models. JAMA Psychiatry with fronto-limbic and basal ganglia dysconnectivity: A study of to- 75:1146–1155. pological variance using network analysis [published online ahead of 99. Honey CJ, Honey CJ, Sporns O, Sporns O, Cammoun L, print Oct 8]. Brain Connectivity. Cammoun L, et al. (2009): Predicting human resting-state functional 95. Depp CA, Mausbach BT, Harmell AL, Savla GN, Bowie CR, connectivity from structural connectivity. Proc Natl Acad Sci U S A Harvey PD, Patterson TL (2012): Meta-analysis of the association 106:2035–2040. between cognitive abilities and everyday functioning in bipolar dis- 100. Abdelnour F, Voss HU, Raj A (2014): Network diffusion accurately order. Bipolar Disord 14:217–226. models the relationship between structural and functional brain 96. Sarwar T, Ramamohanarao K, Zalesky A (2019): Mapping con- connectivity networks. NeuroImage 90:335–347. nectomes with diffusion MRI: Deterministic or probabilistic tractog- 101. Chase HW, Phillips ML (2016): Elucidating neural network functional raphy? Magn Reson Med 81:1368–1384. connectivity abnormalities in bipolar disorder: Toward a harmonized 97. Akudjedu TN, Nabulsi L, Makelyte M, Scanlon C, Hehir S, Casey H, methodological approach. Biol Psychiatry Cogn Neurosci Neuro- et al. (2018): A comparative study of segmentation techniques for the imaging 1:288–298.

162 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:152–162 www.sobp.org/BPCNNI Biological Psychiatry: Archival Report CNNI

Machine Learning Identiﬁes Large-Scale Reward-Related Activity Modulated by Dopaminergic Enhancement in Major Depression

Yuelu Liu, Roee Admon, Monika S. Mellem, Emily L. Belleau, Roselinde H. Kaiser, Rachel Clegg, Miranda Beltzer, Franziska Goer, Gordana Vitaliano, Parvez Ahammad, and Diego A. Pizzagalli

ABSTRACT BACKGROUND: Theoretical models have emphasized systems-level abnormalities in major depressive disorder (MDD). For unbiased yet rigorous evaluations of pathophysiological mechanisms underlying MDD, it is critically important to develop data-driven approaches that harness whole-brain data to classify MDD and evaluate possible normalizing effects of targeted interventions. Here, using an experimental therapeutics approach coupled with machine learning, we investigated the effect of a pharmacological challenge aiming to enhance dopaminergic signaling on whole-brain response to reward-related stimuli in MDD. METHODS: Using a double-blind, placebo-controlled design, we analyzed functional magnetic resonance imaging data from 31 unmedicated MDD participants receiving a single dose of 50 mg amisulpride (MDDAmisulpride), 26 MDD participants receiving placebo (MDDPlacebo), and 28 healthy control subjects receiving placebo (HCPlacebo) recruited through two independent studies. An importance-guided machine learning technique for model selection was used on whole-brain functional magnetic resonance imaging data probing reward anticipation and

consumption to identify features linked to MDD (MDDPlacebo vs. HCPlacebo) and dopaminergic enhancement (MDDAmisulpride vs. MDDPlacebo). RESULTS: Highly predictive classification models emerged that distinguished MDDPlacebo from HCPlacebo (area under the curve = 0.87) and MDDPlacebo from MDDAmisulpride (area under the curve = 0.89). Although reward-related striatal activation and connectivity were among the most predictive features, the best truncated models based on whole- brain features were significantly better relative to models trained using striatal features only. CONCLUSIONS: Results indicate that in MDD, enhanced dopaminergic signaling restores abnormal activation and connectivity in a widespread network of regions. These findings provide new insights into the pathophysiology of MDD and pharmacological mechanism of antidepressants at the system level in addressing reward processing deficits among depressed individuals. Keywords: Biomarker, Biotypes, Depression, Dopamine, fMRI, Machine learning https://doi.org/10.1016/j.bpsc.2019.10.002

Major depressive disorder (MDD) is a debilitating disorder, often dopaminergic signaling might modulate large-scale whole- characterized by anhedonia (1), which is poorly addressed by brain activation and functional coordination in MDD. Besides current treatments (1,2). Converging evidence across species the striatum, other brain regions, including the orbitofrontal suggests that mesocorticolimbic dopaminergic pathways cortex, amygdala, and anterior cingulate cortex (ACC), have involving the striatum are essential for reward processing (3–5). been implicated in reward processing (11–14). Given that an- Dysfunction in this circuit has been associated with deficits in tidepressant treatments aiming to increase dopaminergic reward processing across psychiatric diseases (6). In MDD, signaling might have faster therapeutic onsets (15,16), it is neuroimaging studies have documented decreased striatal important to investigate the effects of dopaminergic activation and reduced functional connectivity between the enhancement to better understand the potential neural striatum and other nodes of the brain reward system in mechanism through which these interventions may address response to reward-related stimuli (7–9). Notably, some of these reward processing deficits in MDD. Thus, we identified several abnormalities were found to be restored in the short term by needs to address in this study, including developing and pharmacologically induced dopaminergic enhancement (10). evaluating 1) a robust, data-driven, multivariate approach to Despite advancements in our understanding of the patho- analyze whole-brain data to probe the purported distributed physiology of MDD, an unresolved issue is how enhanced nature of the reward system, 2) an approach to assess

SEE COMMENTARY ON PAGE 133

ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. 163 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI Biological Psychiatry: CNNI Machine Learning in Depression

MDD-related abnormalities and putative normalization of those and 2) transient dopamine enhancement would rescue such abnormalities, and 3) comparisons between a multivariate abnormalities. We further compared whole-brain and approach and a hypothesis-driven approach to evaluate hypothesis-driven approaches. whether a broad set of regions beyond the striatum does indeed better highlight reward-related abnormalities. METHODS AND MATERIALS Toward these goals, we used a machine learning–based approach to analyze whole-brain functional magnetic reso- Participants nance imaging (fMRI) data collected from a double-blind, pla- Participants were recruited by the Center for Depression, cebo-controlled study, in which unmedicated individuals with Anxiety and Stress Research at McLean Hospital using online MDD and healthy control subjects (HCs) performed a monetary advertisements, mailing, and flyers within the Boston metro- incentive delay (MID) task after being randomized to either a politan areas for two independent studies using identical pro- single low dose of amisulpride (50 mg) or placebo. Ami- cedures that each enrolled individuals with MDD and HCs. sulpride, a selective dopamine D2/D3 receptor antagonist, was Across the first (ClinicalTrials.gov identifier: NCT01253421) and selected because of its high affinity to block presynaptic second (NCT01701258) studies, 62 unmedicated individuals with autoreceptors at lower doses, thereby increasing dopamine MDD (34 randomized to amisulpride [MDDAmisulpride], 28 random- release (17). In a first step, to identify the effects of enhanced ized to placebo [MDDPlacebo]) and 63 demographically matched dopaminergic transmission on reward-related brain activity, HCs (HCPlacebo: n = 30; HCAmisulpride: n = 33) were run in the imaging whole-brain fMRI data were entered into an importance-guided session. For the current analyses, we focused on analyses aiming model selection procedure (based on the logistic regression at classifying case versus controls (MDDPlacebo vs. HCPlacebo with elastic net regularization) (Figure 1) to identify brain re- model) and classifying the potential normalizing effects of dopa- gions in which reward-related metrics were most predictive of minergic enhancement (MDDPlacebo vs. MDDAmisulpride model); differences between the MDD individuals receiving amisulpride thus, 92 participants were considered. Among these 92, 85 had versus placebo. Next, to investigate the potential normalizing useable fMRI data (participants included in final analysis included effect of enhanced dopaminergic transmission on MDD- MDDAmisulpride: n = 31; MDDPlacebo: n =26;HCPlacebo: n = 28). A related abnormalities, brain regions from the previous step subset of participants (46 MDD, 23 randomized to amisulpride, 23 were compared with those most predictive of differences be- to placebo; 20 HC controls randomized to placebo) were included tween MDD and HC group receiving placebo. The regions with in a recent study that used a region-of-interest (ROI) approach to MDD-related abnormalities that also demonstrated an MDD probe the effects of MDD and amisulpride on striatal activation and amisulpride effect constitute a potential multivariate signature functional connectivity (10). Groups were matched for age, gender, that we used to assess amisulpride-induced blood oxygen ethnicity, and years of education (Table 1). General inclusion level–dependent (BOLD) normalization in subjects with MDD. criteria were right-handedness, age between 18 and 45 years, no Based on prior findings (7,10,18–22), we hypothesized that 1) MRI contraindications, no lifetime substance dependence, no under placebo, MDD would be associated with widespread past-year substance abuse, and no serious medical conditions. reward-related abnormalities along the brain’s reward pathway For the MDD groups, a diagnosis of MDD according to the

Figure 1. (A) An illustration of the importance- Full Model A Truncated guided sequential model selection procedure used Model k to ﬁnd the optimal set of features. First, a full model Feature Truncated Top 1 w Truncated Model 2 including all features is trained using logistic 1 Model 1 Feature Top 1 w regression with elastic net regularization to deter- Feature 1 Feature Top w 1 w Top mine relative importance of individual features. Next, 2 2 Feature 1 1 w1 … Feature w … Top 2 2 a series of truncated models were trained based on … … Feature w 2 2 a progressively increasing set of top features rank … … Feature w Top ordered by the full model. The set of features in the 238 238 Feature w best truncated model on the evaluation set were k k Logistic regression with deemed as the optimal feature set. (B) An illustration elastic net regularization of the nested cross-validation procedure used to ⇒ Feature importance train, validate, and test the models. A grid search procedure with threefold cross-validation was B implemented on the developmental set to determine All Participants the best model parameters. The resulting model was Repeat 100 times: further tested on the evaluation set, which contained each iteration with an independent set of participants not used in a random partition training and validation. The entire procedure was of the data repeated on 100 different random partitioning of the web 4C/FPO Developmental Set (80% of data): Evaluation Set data to allow for stable model performance. w refers Training + Validation (20% of data): Testing i to regression weight for feature in the model. 3 fold grid search i cross-validation

Validation Training

164 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI Biological Psychiatry: Machine Learning in Depression CNNI

Table 1. Clinical and Demographic Characteristics of the Participants

MDDAmisulpride (n = 31) MDDPlacebo (n = 26) HCPlacebo (n = 28) Characteristic Mean SD Mean SD Mean SD Age, Years 27.2 7.7 25.6 5.0 25.1 6.1 Education, Years 15.4 2.2 16.8 3.0 15.2 2.9 Beck Depression Inventory-II 26.3 7.9 26.7 7.9 1.8 2.7 Hamilton Depression Rating Scale 15.6 3.7 16.7 5.3 1.0 1.2 Mood and Anxiety Symptom Questionnaire Total score 168.5 22.9 174.1 21.7 91.5 13.3 General distress anxiety subscore 23.6 5.1 25.4 6.6 12.3 1.2 General distress depression subscore 37.9 9.4 39.0 9.3 13.9 2.0 Anxious arousal subscore 24.0 6.0 25.6 6.4 18.4 2.0 Anhedonic depression subscore 82.9 11.2 84.1 9.1 47.0 11.3 Snaith-Hamilton Pleasure Scale 31.7 4.7 31.4 7.0 22.8 6.7 Duration of Current Major Depressive Episode, Months 17.3 20.0 17.6 31.9 N/A N/A Number of Past Depressive Episodes 3.2 2.6 3.3 3.2 N/A N/A

n % n % n % Female 28 90.3 19 73.1 22 81.5 Caucasian 20 64.5 13 50.0 13 48.1 Current Comorbid Anxiety Disorders 10 32.3 11 42.3 N/A N/A Past Comorbid Anxiety Disorders 13 41.9 12 46.2 N/A N/A Groups were matched for age, gender, race, and years of education (one-way analysis of variance; c2 test). All participants were right-handed.

Between the major depressive disorder (MDD) groups administered amisulpride (MDDAmisulpride) and placebo (MDDPlacebo), participants were matched for current and past comorbid anxiety disorders, as well as clinical scale measures (c2 test; two-sample t test). HC, healthy control subjects; N/A, not applicable.

Structured Clinical Interview for DSM-IV-TR Axis I Disorders (2) penalties), which robustly recruit mesocorticolimbic regions was required, and exclusion criteria included psychotropic medi- (12,13) and have been used to uncover reward-related ab- cation in the past 2 weeks (6 weeks for fluoxetine, 6 months for normalities in both magnitude of activation and functional dopaminergic drugs or antipsychotics) and any other Axis I disconnectivity in MDD (7,9,10,22,24). orders (however, social anxiety disorder, simple phobia, or generalized anxiety disorder were allowed if secondary to MDD). Data Acquisition and Preprocessing For HCs, exclusion criteria were any medication in the last 3 For both studies, MRI data were acquired at the McLean Im- weeks, current or past psychiatric illnesses (Structured Clinical aging Center using a Siemens Tim Trio 3T MR scanner Interview for DSM-IV-TR Axis I Disorders), and first-degree familial equipped with a 32-channel head coil (Siemens Medical So- psychiatric illness. Participants received $15/hour in addition to lutions USA, Inc., Malvern, PA). Data collection for the two earnings in the fMRI task. The two protocols were approved by studies overlapped in time. See Supplemental Methods for Partners Human Research Committee, and all participants pro- acquisition parameters and preprocessing. vided written informed consent. Feature Extraction Procedure The features used in our classifiers consisted of coefficients The two studies followed identical procedures, pharmacolog- from the single-subject-level general linear models averaged ical challenge, and MRI acquisition. In the first session, a according to the Automated Anatomical Labeling template Ph.D.- or master’s-level clinician administered the Structured (25). To obtain these features, for each participant, we first Clinical Interview for DSM-IV-TR Axis I Disorders to determine fitted a general linear model to the fMRI data during the MID eligibility, and participants filled out self-report scales (Table 1) task [see (10) for more details]. Next, for each regressor in the (Supplement). In the second session, participants performed general linear models, the estimated coefficients were aver- the MID task during fMRI scanning after receiving a single dose aged according to the Automated Anatomical Labeling tem- of amisulpride or placebo. The MID task was started 1 hour plate, producing one averaged coefficient for each ROI. ROIs after pill administration based on pharmacokinetic data indi- for the left and right nucleus accumbens (NAcc) were further cating that plasma concentration of amisulpride has a first extracted according to a manually segmented MNI-152 brain peak approximately 1 to 1.5 hours after administration (17). (26) and added to the existing Automated Anatomical Labeling ROIs, resulting in 118 ROIs. The following BOLD contrasts fMRI Task were included as features in our classification models to The MID has been described in detail (10,23). Briefly, the task represent reward anticipation and consumption, respectively: includes anticipation and receipt of monetary rewards (and 1) reward cue minus neutral cue and 2) reward outcome minus

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI 165 Biological Psychiatry: CNNI Machine Learning in Depression

no-change outcome following reward cue. In addition, two After identifying the best truncated models for the classifiers, striatal connectivity features emerging from our previous work we compared the feature sets—both the selected regions and the (10) were included in our classification models, representing regression weight signs (positive or negative), as they indicated the psychophysiological interaction under the reward outcome the direction of the BOLD difference (greater for one class over condition between 1) caudate and dorsal ACC and 2) NAcc another). Based on how we set up the classifiers, those regions

and midcingulate cortex (MCC). In total, 238 features (118 shared by the MDDPlacebo versus HCPlacebo and MDDPlacebo ROIs 3 2 contrasts 1 2 psychophysiological interactions) were versus MDDAmisulpride classifiers with convergent regression included in the classification models. Modeling was also done signs constitute a potential multivariate signature that we can use without the psychophysiological interaction regressors to to assess amisulpride-induced BOLD normalization in MDD establish if they brought any additional predictive information subjects. We calculated signed BOLD sum scores by summing (see Supplement). All features were standardized to zero mean up the BOLD values of the convergent features multiplied by the and unit variance before being entered into the models. regression weight sign to assess normalization. The convergent features should largely be absent in the set of highly differenti- Classification and Importance-Guided Sequential ating features of the MDDAmisulpride versus HCPlacebo classifier if Model Selection they have been normalized with amisulpride.

Two main classifiers were built to classify 1) MDDPlacebo versus HCPlacebo and 2) MDDPlacebo versus MDDAmisulpride.Thesewere Statistical Analysis designed to capture features linked to 1) MDD and 2) the effect of The significance of the models’ performances against chance short-term dopaminergic enhancement on whole-brain BOLD level was tested using a random permutation test scheme in activation in individuals with MDD. To further test the hypothesis which the truncated model based on the optimal feature set were that dopaminergic enhancement transiently normalized reward- retrained on label shuffled training data (28). The entire test pro- related abnormalities in MDD, a third classifier was built to cedure was iterated 1000 times to empirically construct the null classify MDD versus HC . Across analyses, we Amisulpride Placebo distribution of test AUCs. The p values were obtained by used logistic regression with elastic net regularization (27)for comparing the AUC from the best truncated model based on classification. The elastic net regularization is well suited for unshuffled data against the empirical null distribution. The per- problems where the number of features is much greater than the formances between models were statistically compared via number of observations (27). The models were trained and tested Mann-Whitney U tests. Effect sizes between two distributions via the following nested cross-validation procedure. First, we were calculated using Cohen’s d. performed model training on a development set containing 80% of the participants via a threefold grid search cross-validation procedure (stratified using class labels) (Figure 1B). Then, the RESULTS model with the best regularization parameters was further tested on the evaluation set containing an independent set of 20% Classification Performances participants, which the model had not seen during the training The best truncated models selected by the importance-guided and validation phases. The above procedure was repeated 100 model selection procedure (Figure 1) based on most predictive times to ensure that stable performance was obtained on a large features from whole-brain BOLD activations and striatal con- number of development-evaluation splits. The area under the nectivity achieved high predictive performances (Table 2)(see receiver operating characteristics curve (AUC) was selected as Supplemental Figure S1 for model performance as a function of

the metric to quantify model performance, and reported AUCs are top features). For both MDDPlacebo versus HCPlacebo and only from testing on the independent evaluation set. MDDPlacebo versus MDDAmisulpride, the AUCs of the best trun- To identify the set of most predictive features for each cated models were significantly above chance level (MDDPlacebo classifier (i.e., MDDPlacebo vs. HCPlacebo and MDDPlacebo vs. vs. HCPlacebo: mean AUC = 0.87, permutation testing p =.004; MDDAmisulpride), we adopted the following importance-guided MDDPlacebo vs. MDDAmisulpride:meanAUC=0.89,p =.002) sequential model selection procedure (Figure 1A). Specif- (Figure 2A, B)(Supplemental Figure S2). Predictive features ically, we first rank-ordered the features using the mean model displayed some collinearity, but collinearity did not account for weights across 100 implementations as a measure of pre- the diminishing AUC returns of the lower-ranked predictive dictability. Then, we built a series of truncated models such features (see Supplemental Results and Supplemental that each model only took the top k most predictive features as Figures S3 and S4). Compared with models trained using inputs to perform the classification tasks, with k varying from striatal features only (Supplemental Methods), the performances the top 1 most predictive feature to the number of participants of the best truncated models based on whole-brain features involved in a given classifier. Imposing the number of partici- were significantly better for both contrasts (both p , .001, pants as the upper limit was to ensure that models’ perfor- Mann-Whitney U test). The histograms of sum scores created mance was not mainly driven by the regularization term. All by summing up the top feature values while taking into account truncated models underwent the nested cross-validation pro- the sign of the corresponding model weights demonstrated high

cedure described above, and the test performance from each separability between MDDPlacebo and HCPlacebo as well as be- truncated model on the independent evaluation set was ob- tween MDDPlacebo and MDDAmisulpride (Figure 2C, D). Overall, tained. The set of features used by the truncated model these results indicate that our models were able to extract achieving the highest AUC on the evaluation set was deemed highly predictive information embedded in the whole-brain as the optimal feature set. BOLD signal.

166 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI Biological Psychiatry: Machine Learning in Depression CNNI

Table 2. Classiﬁcation Performance for the Best Truncated Models Striatum Only

MDDPlacebo vs. MDDPlacebo vs. MDDPlacebo vs. MDDPlacebo vs. HCPlacebo MDDAmisulpride HCPlacebo MDDAmisulpride Mean SD Mean SD Mean SD Mean SD AUC 0.87 0.12 0.89 0.09 0.59 0.14 0.61 0.17 Accuracy 0.77 0.12 0.80 0.10 0.59 0.13 0.59 0.13 Sensitivity 0.84 0.18 0.89 0.11 0.58 0.25 0.65 0.19 Speciﬁcity 0.72 0.22 0.67 0.24 0.59 0.22 0.50 0.28 Number of Features 48 44 6 11 AUC, area under the curve; HC, healthy control subjects; MDD, major depressive disorder.

Brain Regions Specific to Reward Anticipation HCPlacebo classification included the thalamus, supplementary motor area, and ventromedial prefrontal cortex. Again, these Positive Model Weights. The best truncated model for regions were not among the top features in the MDDAmisulpride MDDPlacebo versus MDDAmisulpride identified the lateral orbitofrontal cortex (lOFC), visual cortex, ACC, dorsomedial pre- versus HCPlacebo model (Supplemental Figure S5), suggesting frontal cortex, MCC, and precuneus as most predictive that amisulpride mitigated the hyperactivation in these regions features with positive weights during reward anticipation within the MDD group. (Figure 3A and Supplemental Table S1). This indicates that within the MDD group, BOLD activation in these regions Negative Model Weights. Regions selected by the best related to the contrast of reward cue minus neutral cue was MDDPlacebo versus MDDAmisulpride model with negative model reduced following administration of amisulpride compared with weights included the putamen, pallidum, amygdala, posterior placebo. Critically, the lOFC, visual cortex, and MCC were also parietal cortex (PPC), and temporal cortex (Figure 3A and Supplemental Table S1). The negative weights observed in the selected by the best MDDPlacebo versus HCPlacebo model as top features having positive weights (Figure 3B and Supplemental putamen and pallidum were consistent with the hypothesis Table S2), and at the same time these regions, except a right that amisulpride might have increased dopaminergic signaling occipital region, were not among the most predictive features in the basal ganglia in MDD (10,14). This effect is rather pronounced as the MDD versus HC model showed in the MDDAmisulpride versus HCPlacebo model (Supplemental Amisulpride Placebo Figure S5). Collectively, these findings indicate that within the that the contrast of reward cue minus neutral cue evoked MDD group, amisulpride largely normalized the heightened higher activation in the putamen in the MDDAmisulpride group BOLD activation in these regions toward reward cues. Other even compared with the HCPlacebo group (Supplemental Figure S5). Within the MDD group, reduced activation regions with positive weights in the MDDPlacebo versus Placebo FPO = web 4C Figure 2. Comparing classification performance between the data-driven models based on features selected from the whole-brain and the hypothesis- driven models based only on striatal features for (A) MDDPlacebo vs. HCPlacebo and (B) MDDPlacebo vs. MDDAmisulpride classifications. Asterisks denote significantly different median area under the receiver operating characteristic (ROC) curve measures between the data-driven and hypothesis-driven models as assessed by the Mann-Whitney U test. The black markers denote outliers falling outside the 61.5 interquartile range. The histogram of the signed sum score from the model-identified most predictive brain regions show high separability between (C) MDDPlacebo vs. HCPlacebo and (D) MDDPlacebo vs. MDDAmisulpride. HC, healthy control subjects; MDD, major depressive disorder.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI 167 Biological Psychiatry: CNNI Machine Learning in Depression

MDD vs. MDD A Placebo Amisulpride B MDDPlacebo vs. HCPlacebo z = -15 z = 5 x = -3 z = -15 z = 10 x = -5 lOFC SMA lOFC MCC Precu dmPFC L L Cu Thal MCC

19 22

Vermis vmPFC ACC 0 0 OC OC OC Cal z = 0 Cal z = -8 z = 40 y = -12 x = 10 y = 1 L Ins dmPFC L Put L L Oper

TC 0 0 Pal FPO =

Amyg -19 Hipp/ -16 PPC PHG web 4C Figure 3. Weight maps showing the most predictive brain regions for the contrast of the reward minus neutral cue conditions. (A) Weight map for the MDDPlacebo vs. MDDAmisulpride model. Positive weights indicate higher blood oxygen level–dependent values in the MDDPlacebo group relative to the MDDAmisulpride group and negative weights indicate the opposite direction. (B) Weight map for the MDDPlacebo vs. HCPlacebo model, with positive weights indicating higher blood oxygen level–dependent values in the MDDPlacebo group relative to the HCPlacebo group and vice versa. ACC, anterior cingulate cortex; Amyg, amygdala; Cal, calcarine sulcus; Cu, cuneus; dmPFC, dorsomedial prefrontal cortex; HC, healthy control subjects; Hipp, hippocampus;Ins, insula; L, left; lOFC, lateral orbitofrontal cortex; MCC, midcingulate cortex; MDD, major depressive disorder; OC, occipital cortex; Oper, operculum; Pal, pallidum; PHG, parahippocampal gyrus; PPC, posterior parietal cortex; Precu, precuneus; Put, putamen; SMA, supplementary motor area; TC, temporal cortex; Thal, thalamus; vmPFC, ventromedial prefrontal cortex.

in the operculum, hippocampus, parahippocampal gyrus versus HCPlacebo model included the inferior frontal gyrus, (PHG), and dorsomedial prefrontal cortex was observed rela- PPC, precuneus, and MCC. The lack of predictability from

tive to HCs during reward anticipation (features in the these regions between MDDAmisulpride and HCPlacebo MDDPlacebo vs. HCPlacebo model with negative weights) (Supplemental Figure S5) again suggests a mitigating effect of (Figure 3B and Supplemental Table S2). The reduced activa- amisulpride on the hyperactivation in these regions. tion in the hippocampus and operculum persisted in the MDD versus HC model (Supplemental Amisulpride Placebo Negative Model Weights. The most predictive regions Figure S5), indicating that amisulpride had limited effects in from the contrast of reward minus no-change outcomes with these regions. negative weights in the MDDPlacebo versus MDDAmisulpride model included the putamen, NAcc, PHG, and temporal pole Brain Regions Speciﬁc to Reward Consumption (Figure 4A and Supplemental Table S3), as well as the con- Positive Model Weights. Examining features selected nectivity between the NAcc and MCC (Figure 4B). This sug- from the contrast of reward minus no change outcomes in the gests that within the MDD group, amisulpride increased BOLD

MDDPlacebo versus MDDAmisulpride model revealed that the activation and corticostriatal connectivity to reward feedback lOFC, posterior parietal cortex (PPC), superior frontal gyrus, in these regions. Highlighting again convergence, the NAcc, and the pre- and postcentral gyrus were selected as most PHG, temporal pole, and the NAcc-MCC connectivity were predictive features with positive weights (Figure 4A and also selected as most predictive features having negative

Supplemental Table S3). This indicates reduced activation in weights in the MDDPlacebo versus HCPlacebo classiﬁcation these regions during reward consumption in MDDAmisulpride (Figure 4C, D and Supplemental Table S4), and none of these compared with MDDPlacebo. Of note, the lOFC and PPC regions was selected as among the top predictive features in emerged as among the most predictive features with positive the MDDAmisulpride versus HCPlacebo model (Supplemental weights in the MDDPlacebo versus HCPlacebo model (Figure 4C Figure S5). Thus, in MDD, amisulpride normalized both hypo- and Supplemental Table S4). Additionally, while the lOFC activation and hypoconnectivity in response to rewards in

hyperactivation was still observed in the MDDAmisulpride versus these regions. Other most predictive features with negative HCPlacebo model, the PPC was not identiﬁed as a predictive weights in the MDDPlacebo versus HCPlacebo model included the feature (Supplemental Figure S5). Overall, these results sug- visual cortex, inferior temporal cortex, operculum, ACC, and gest that under placebo, the MDD group was characterized by the connectivity between the caudate and dorsal ACC. These increased BOLD activity in these regions during reward con- features, except the caudate-dorsal ACC connectivity, were

sumption relative to HCs and that the hyperactivation was not identiﬁed as among the top features in the MDDAmisulpride reduced by amisulpride. Other brain regions identiﬁed as most versus HCPlacebo model (Supplemental Figure S5), indicating predictive features with positive weights in the MDDPlacebo increased activation to rewards in these regions following

168 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI Biological Psychiatry: Machine Learning in Depression CNNI

MDD vs. MDD A Placebo Amisulpride B MDDPlacebo vs. MDDAmisulpride lOFC z = -15 z = 45 SFG y = -5 y = 12 y = 15 x = 6 MCC L PreCG/ PostCG L L

Put Put 20 0

NAcc 0 NAcc PPC PHG -17 TP

MDD vs. HC C Placebo Placebo D MDDPlacebo vs. HCPlacebo lOFC z = -10 z = 30 x = -6 MCC x = -16 x = 6 L IFG dACC -10.55 Cau PCC/ Precu 14

PPC PPC 0

z = -8 z = 15 y = 15 x = 6 x = 7 y = 12 Oper MCC ACC L L ACC L

PHG 0

NAcc NAcc NAcc web 4C/FPO ITC -14 TP OC OC

Figure 4. Weight maps showing the most predictive brain regions and/or connectivity for the contrast of reward minus no-change outcomes. (A) Weight map for the MDDPlacebo vs. MDDAmisulpride model, with positive weights indicating higher blood oxygen level–dependent values in the MDDPlacebo group relative to the MDDAmisulpride group and vice versa. (B) Negative weight assigned to the nucleus accumbens (NAcc)-midcingulate cortex (MCC) connectivity in the MDDPlacebo vs. MDDAmisulpride model. (C) Weight map for the MDDPlacebo vs. HCPlacebo model. Positive weights indicate higher blood oxygen level–dependent values in the MDDPlacebo group relative to the MDDAmisulpride group and vice versa. (D) Negative weights assigned to the caudate (Cau)-dorsal anterior cingulate cortex (dACC) and NAcc-MCC connectivity features by the MDDPlacebo vs. HCPlacebo model. CG, central gyrus; HC, healthy control subjects; IFG, inferior frontal gyrus; ITC, inferior temporal cortex; L, left; lOFC, lateral orbitofrontal cortex; MDD, major depressive disorder; OC, occipital cortex; Oper, operculum; PCC, posterior cingulate cortex; PHG, parahippocampal gyrus; PPC, posterior parietal cortex; Precu, precuneus; Put, putamen; SFG, superior frontal gyrus; TP, temporal pole.

amisulpride administration in the MDD group. The fact that DISCUSSION amisulpride did not normalize the hypoconnectivity between This study used a machine learning–based approach to iden- caudate and dorsal ACC in the MDD group is consistent with tify reliable brain-wide features that delineated MDD-related previously published ROI-based results obtained on a subset abnormalities as well as features linked to their normalization of the participants (10). after an acute dopaminergic pharmacological challenge. In

addition to increased striatal activation in the MDDAmisulpride A Multivariate Signature of Normalization relative to MDDPlacebo group [which is consistent with ROI- The signed BOLD sum scores calculated from the convergent based conventional analyses of a smaller subset of the par-

features across the MDDPlacebo versus HCPlacebo and MDDPlacebo ticipants included here (10)], the classification model also versus MDDAmisulpride classifiers showed that the multivariate identified an extensive set of reward-related brain regions neural signature is significantly greater in the MDDPlacebo than in differentiating these groups, which provided additional pre- either MDDAmisulpride or HCPlacebo groups (Figure 5)(allp , .001, dictive power over striatal regions alone. Converging of fea- Mann-Whitney U test), while the latter two groups were statisti- tures between the MDDPlacebo versus MDDAmisulpride model and cally equivalent based on equivalence testing (p =.01)(seethe the MDDPlacebo versus HCPlacebo model suggested that ami- Supplement for more information). Taken together, these results sulpride had a bidirectionally normalizing effect on reward- suggest that amisulpride normalized MDD-related abnormalities. related activation and functional connectivity of brain regions

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI 169 Biological Psychiatry: CNNI Machine Learning in Depression

administration of amisulpride [see (29) for conceptually similar

imaging findings using a single dose of the novel D2 antagonist lurasidone]. This supports the validity of the importance- guided model selection procedure and fits the view that lower doses of amisulpride enhance dopaminergic signaling in the striatum (17). Among regions outside the striatum, one notable finding was that increased lOFC activation during reward anticipation in MDD was reduced after administration of amisulpride. Neurophysiological evidence has shown that subpopulations of neurons in the lOFC respond to nonreward or unpleasant events and maintain elevated firing rate after such events (30). This led to the theory implicating overly reactive and prolonged activation of the lOFC nonreward circuit as a potential mechanism underlying depression (31). Previous studies have documented increased lOFC activation in MDD (32), and our

result ﬁts this theoretical view. In the MDDAmisulpride group, reduced lOFC activation suggests that amisulpride may normalize reward processing by decreasing lOFC hyperactivation, consistent with previous reports that improvements in depressive symptoms were accompanied by reduced lOFC activation (33) and that electrical stimulation of the lOFC acutely improved depressive symptoms (34). In addition to effects in frontostriatal circuitry, amisulpride restored hypoactivation in the PHG and temporal pole in MDD. The hippocampus and parahippocampal complex connect

FPO with the medial OFC and are hypothesized to facilitate the = formation of episodic memory regarding reward (35). Decreased hippocampal activation has emerged in MDD, and

web 4C prolonged or repeated depressive episodes have been linked Figure 5. Multivariate signatures across groups demonstrated to reduced hippocampal volume (36,37). These abnormalities amisulpride-based brain normalization. The signed blood oxygen level– have been linked to dysfunctions in both memory encoding dependent (BOLD) sum scores calculated across the convergent features and retrieval characteristic of MDD, even after treatment of MDDPlacebo vs. MDDAmisulpride and MDDPlacebo vs. HCPlacebo models suggest normalization of MDD-related abnormalities following amisulpride (38,39). The fact that amisulpride restored parahippocampal

administration. MDDPlacebo subjects had overall greater multivariate neural and temporal pole activation suggests that interventions aim- signatures compared to HCPlacebo or MDDAmisulpride (***p , .001 for both ing to increase dopaminergic signaling might improve encod- tests). Equivalence testing demonstrated that HCPlacebo and MDDAmisulpride ing and retrieval of positive memories in MDD. However, it had statistically equivalent (denoted using “e” in plot) scores (p = .01). HC, should be noted that hippocampal activation did not differen- healthy control subjects; MDD, major depressive disorder. tiate between the MDDAmisulpride and MDDPlacebo groups, suggesting that the effects on memory might be limited following a single acute pharmacological challenge. spanning the lOFC, NAcc, PHG, MCC, PPC, and areas of the Hyperactivation in the MCC toward the reward cue was also visual cortex among depressed individuals. Taken together, reduced among depressed individuals after amisulpride. these results highlight the unique contribution of machine Moreover, amisulpride also reduced reward cue-evoked acti- learning–based approaches to examine brain-wide circuit vations in adjacent ACC and dorsomedial prefrontal cortex. engagement and potential normalization after a single dose. The supracallosal part of the cingulate cortex receives Such mechanistic evidence can help evaluate novel com- neuronal projections from the lOFC and is thought to also pounds before pursuing longer efficacy-oriented clinical trials encode nonreward and punishing events such as physical and with a compound. Overall, this study provided novel evidence social pain (40,41). A recent study has identified a nociceptive for the mechanism through which (transient) dopaminergic pathway between the MCC and posterior insula responsible for enhancement might restore system-level activity during reward generating a hypersensitive state for pain, providing a mech- processing among individuals with MDD. anism for the increased pain sensitivity by psychosocial factors Amisulpride appeared to have bidirectional normalizing ef- (42). The reduced hyperactivation in these regions following fects on brain activation and functional coordination among amisulpride administration may indicate decreased sensitivity depressed individuals. Within the striatum, consistent with to negative affective states among individuals with MDD and previous ROI-based analyses based on a subset of the par- therefore priming or biasing them toward reward. ticipants used here (10), results from our classification models In MDD, amygdalar activation evoked by reward cues was showed that while striatal–basal ganglia activation and corti- enhanced following administration of amisulpride. Reduced costriatal connectivity were initially decreased among amygdalar response to positive and rewarding stimulus, depressed individuals, they were enhanced following acute coupled with heightened amygdalar activation toward

170 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI Biological Psychiatry: Machine Learning in Depression CNNI

negative stimulus, are well-documented findings in MDD, Psychiatry (ELB, GV, DAP), Harvard Medical School, Boston, Massachusetts; which highlights an imbalanced reactivity toward emotionally and Department of Psychology and Neuroscience (RHK), University of Colo- salient cues (43). Antidepressant treatment has been shown rado, Boulder, Colorado. PA and DAP contributed equally to this work. to address this imbalance by partially normalizing the bidi- Address correspondence to Diego A. Pizzagalli, Ph.D., Center for rectional abnormal amygdalar activation (43,44). These find- Depression, Anxiety and Stress Research, McLean Hospital, 115 Mill Street, ings were further bolstered by the recent report that enhanced Belmont, MA 02478; E-mail: [email protected]. amygdalar response toward positive memories through real- Received Sep 3, 2019; revised Sep 30, 2019; accepted Oct 1, 2019. time fMRI neurofeedback was associated with reduction in Supplementary material cited in this article is available online at https:// depressive symptoms (45). The increased amygdalar activa- doi.org/10.1016/j.bpsc.2019.10.002. tion evoked by reward cues is consistent with these studies and implicates improved sensitivity toward reward following acute dopaminergic enhancement. REFERENCES While several regions showed predictive power following 1. Calabrese JR, Fava M, Garibaldi G, Grunze H, Krystal AD, Laughren T, the administration of amisulpride, it is difficult to assess et al. (2014): Methodological approaches and magnitude of the clinical whether changes in these regions reflected a direct modula- unmet need associated with amotivation in mood disorders. J Affect Disorders 168:439–451. tion resulting from the enhanced dopaminergic signaling or 2. American Psychiatric Association (2000): Diagnostic and Statistical alternatively reflected secondary responses through network Manual of Mental Disorders, 4th ed, Text Revision: DSM-IV-TR. interactions. Future studies could utilize network analysis and/ Washington, DC: American Psychiatric Publishing. or neural perturbation methods to further dissociate direct 3. Wise RA (2004): Dopamine, learning and motivation. Nat Rev Neurosci versus indirect effects (34). In addition, amisulpride also has 5:483–494. 5-HT (5-hydroxytryptamine receptor 7) antagonism (46), 4. Berridge KC, Kringelbach ML (2015): Pleasure systems in the brain. 7 Neuron 86:646–664. which has been hypothesized to contribute to its antide- 5. Der-Avakian A, Markou A (2012): The neurobiology of anhedonia and pressant property. While we cannot rule out that the effects other reward-related deficits. Trends Neurosci 35:68–77. observed here may be partially caused by this off-target 6. Husain M, Roiser JP (2018): Neuroscience of apathy and anhedonia: A mechanism, additional research is needed to distinguish the transdiagnostic approach. Nat Rev Neurosci 19:470–484. 7. Pizzagalli DA, Holmes AJ, Dillon DG, Goetz EL, Birk JL, Bogdan R, effect of dopaminergic enhancement versus 5-HT7 antagonism of amisulpride. Lastly, we only focused on et al. (2009): Reduced caudate and nucleus accumbens response to rewards in unmedicated individuals with major depressive disorder. investigating the effects of dopaminergic enhancement on Am J Psychiatry 166:702–710. reward processing among depressed individuals. Future 8. Pizzagalli DA (2014): Depression, stress, and anhedonia: Toward studies could seek to examine the effect of enhanced a synthesis and integrated model. Annu Rev Clin Psychol dopamine on whole-brain fMRI activity in depression under 10:393–423. additional conditions; additionally, based on hypotheses of 9. Admon R, Nickerson LD, Dillon DG, Holmes AJ, Bogdan R, Kumar P, shared mesocorticolimbic dopaminergic abnormalities, this et al. (2014): Dissociable cortico-striatal connectivity abnormalities in major depression in response to monetary gains and penalties. Psy- molecule could be tested in other disorders such as addiction chol Med 45:121–131. or schizophrenia [e.g., (47,48)]. 10. Admon R, Kaiser RH, Dillon DG, Beltzer M, Goer F, Olson DP, et al. (2017): Dopaminergic enhancement of striatal response to reward in major depression. Am J Psychiatry 174:378–386. ACKNOWLEDGMENTS AND DISCLOSURES 11. McClure SM, York MK, Montague RP (2004): The neural substrates of This project was supported by the National Institute of Mental Health reward processing in humans: the modern role of fMRI. Neuroscientist (Grants Nos. R01 MH068376, R37 MH068376, and R01MH095809 [to 10:260–268. DAP]). The content is solely the responsibility of the authors and does not 12. Oldham S, Murawski C, Fornito A, Youssef G, Yücel M, Lorenzetti V necessarily represent the official views of the National Institutes of Health. (2018): The anticipation and outcome phases of reward and loss DAP designed both studies and obtained funding for both; RC, MB, FG, processing: A neuroimaging meta-analysis of the monetary incentive and GV collected data; YL, MM, RA, ELB, RHK performed the analyses; YL, delay task. Hum Brain Mapp 39:3398–3418. DAP, RA, MM, and PA wrote the manuscript. All authors approved the 13. Wilson RP, Colizzi M, Bossong MG, Allen P, Kempton M, MTAC, et al. manuscript. (2018): The neural substrate of reward anticipation in health: A meta- Data are available at the NIMH Data Archive (https://nda.nih.gov/). analysis of fMRI findings in the monetary incentive delay task. Neu- Analysis scripts are available upon request. ropsychol Rev 28:496–506. YL, MM, and PA are current or previous full-time employees at Black- 14. Schott BH, Minuzzi L, Krebs RM, Elmenborst D, Lang M, Winz OH, thorn Therapeutics Inc. Over the past 3 years, DAP has received consulting et al. (2008): Mesolimbic functional magnetic resonance imaging ac- fees from Akili Interactive Labs, BlackThorn Therapeutics, Boehringer tivations during reward anticipation correlate with reward-related Ingelheim, Posit Science, and Takeda Pharmaceuticals and an honorarium ventral striatal dopamine release. J Neurosci 28:14311–14319. from Alkermes for activities unrelated to the current review. All other authors 15. Amore M, Jori M (2001): Faster response on amisulpride 50 mg versus report no biomedical financial interests or potential conflicts of interest. sertraline 50-100 mg in patients with dysthymia or double depression: ClinicalTrials.gov: The Effects of Dopamine on Reward Processing; A randomized, double-blind, parallel group study. Int Clin Psycho- http:/// https://clinicaltrials.gov/ct2/show/NCT01253421; NCT01253421; pharmacol 16:317–324. and An Investigation of Early Life Stress and Depression; http:// https:// 16. Cassano G, Jori M (2002): Efficacy and safety of amisulpride 50 mg clinicaltrials.gov/ct2/show/NCT01701258; NCT01701258. versus paroxetine 20 mg in major depression: A randomized, double-blind, parallel group study. Int Clin Psychopharmacol 17:27–32. ARTICLE INFORMATION 17. Rosenzweig P, Canal M, Patat A, Bergougnan L, Zieleniuk I, From the BlackThorn Therapeutics (YL, MSM, PA), San Francisco, California; Bianchetti G (2002): A review of the pharmacokinetics, tolerability and Department of Psychology (RA), University of Haifa, Haifa, Israel; McLean pharmacodynamics of amisulpride in healthy volunteers. Hum Psy- Hospital (ELB, RC, MB, FG, GV, DAP), Belmont, Massachusetts; Department of chopharmacol 17:1–13.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI 171 Biological Psychiatry: CNNI Machine Learning in Depression

18. Viviani R, Graf H, Wiegers M, Abler B (2013): Effects of amisulpride on improvement in major depressive disorder. Biol Psychiatry 50:171– human resting cerebral perfusion. Psychopharmacology (Berl) 229:95–103. 178. 19. Metzger CD, Wiegers M, Walter M, Abler B, Graf H (2016): Local and 34. Rao VR, Sellers KK, Wallace DL, Lee MB, Bijanzadeh M, Sani OG, et al. global resting state activity in the noradrenergic and dopaminergic (2018): Direct electrical stimulation of lateral orbitofrontal cortex pathway modulated by reboxetine and amisulpride in healthy subjects. acutely improves mood in individuals with symptoms of depression. Int J Neuropsychoph 19:1–9. Curr Biol 28:3893–3902. 20. Forbes EE, Hariri AR, Martin SL, Silk JS, Moyles DL, Fisher PM, et al. 35. Suzuki WA, Naya Y (2014): The perirhinal cortex. Annu Rev Neurosci (2009): Altered striatal activation predicting real-world positive affect in 37:39–53. adolescent major depressive disorder. Am J Psychiatry 166:64–73. 36. Milne A, MacQueen GM, Hall GBC (2012): Abnormal hippocampal 21. Kumar P, Waiter G, Ahearn T, Milders M, Reid I, Steele JD (2008): activation in patients with extensive history of major depression: an Abnormal temporal difference reward-learning signals in major fMRI study. J Psychiatry Neurosci 37:28–36. depression. Brain 131:2084–2093. 37. McKinnon MC, Yucel K, Nazarov A, MacQueen GM (2009): A meta- 22. Stoy M, Schlagenhauf F, Sterzer P, Bermpohl F, Hägele C, analysis examining clinical predictors of hippocampal volume in pa- Suchotzki K, et al. (2012): Hyporeactivity of ventral striatum towards tients with major depressive disorder. J Psychiatry Neurosci 34:41–54. incentive stimuli in unmedicated depressed patients normalizes after 38. Dillon DG (2015): The neuroscience of positive memory deficits in treatment with escitalopram. J Psychopharmacol 26:677–688. depression. Front Psychol 6:1295. 23. Knutson B, Westdorp A, Kaiser E, Hommer D (2000): FMRI visualiza- 39. Dillon DG, Pizzagalli DA (2018): Mechanisms of memory disruption in tion of brain activity during a monetary incentive delay task. Neuro- depression. Trends Neurosci 41:137–149. image 12:20–27. 40. Grabenhorst F, Rolls ET (2011): Value, pleasure and choice in the 24. Knutson B, Bhanji JP, Cooney RE, Atlas LY, Gotlib IH (2008): Neural ventral prefrontal cortex. Trends Cogn Sci 15:56–67. responses to monetary incentives in major depression. Biol Psychiatry 41. Rotge JY, Lemogne C, Hinfray S, Huguet P, Grynszpan O, Tartour E, 63:686–692. et al. (2015): A meta-analysis of the anterior cingulate contribution to 25. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, social pain. Soc Cogn Affect Neursci 10:19–27. Delcroix N, et al. (2002): Automated anatomical labeling of activations 42. Tan L, Pelzer P, Heinl C, Tang W, Gangadharan V, Flor H, et al. (2017): in SPM using a macroscopic anatomical parcellation of the MNI MRI A pathway from midcingulate cortex to posterior insula gates noci- single-subject brain. Neuroimage 15:273–289. ceptive hypersensitivity. Nat Neurosci 20:1591–1601. 26. AdmonR,HolsenLM,AizleyH,RemingtonA,Whitfield-Gabrieli S, 43. Victor TA, Furey ML, Fromm SJ, Öhman A, Drevets WC (2010): Rela- Goldstein JM, et al. (2015): Striatal hypersensitivity during stress in remitted tionship between amygdala responses to masked faces and mood individuals with recurrent depression. Biol Psychiatry 78:67–76. state and treatment in major depressive disorder. Arch Gen Psychiatry 27. Zou H, Hastie T (2005): Regularization and variable selection via the 67:1128–1138. elastic net. J R Stat Soc Series B Stat Methodol 67:301–320. 44. Fu CH, Williams SCR, Cleare AJ, Brammer MF, Walsh ND, Kim J, et al. 28. Ojala M, Garriga GC (2010): Permutation tests for studying classifier (2004): Attenuation of the neural response to sad faces in major depres- performance. J Mach Learn Res 11:1833–1863. sion by antidepressant treatment: a prospective, event-related functional 29. Wolke SA, Mehta MA, O’Daly O, Zelaya F, Zahreddine N, Keren H, magnetic resonance imaging study. Arch Gen Psychiatry 61:877–889. et al. (2019): Modulation of anterior cingulate cortex reward and pen- 45. Young KD, Siegle GJ, Zotev V, Phillips R, Misaki M, Yuan H, alty signalling in medication-naive young-adult subjects with depres- et al. (2017): Randomized clinical trial of real-time fMRI amygdala sive symptoms following acute dose lurasidone. Psychol Med neurofeedback for major depressive disorder: Effects on symp- 49:1365–1377. toms and autobiographical memory recall. Am J Psychiatry 30. Thorpe S, Rolls ET, Maddison S (1983): The orbitofrontal cortex: 174:748–755. neuronal activity in the behaving monkey. Exp Brain Res 49:93–115. 46. Abbas AI, Hedlund PB, Huang XP, Tran TB, Meltzer HY, Roth BL 31. Rolls ET (2016): A non-reward attractor theory of depression. Neurosci (2009): Amisulpride is a potent 5-HT7 antagonist: relevance for anti- Biobehav Rev 68:47–58. depressant actions in vivo. Psychopharmacology 205:119–128. 32. Cheng W, Rolls ET, Qiu J, Liu W, Tang Y, Huang C, et al. (2016): Medial 47. Robison AJ, Nestler EJ (2011): Transcriptional and epigenetic mech- reward and lateral non-reward orbitofrontal cortex circuits change in anisms of addiction. Nat Rev Neurosci 12:623. opposite directions in depression. Brain 139:3296–3309. 48. Laviolette SR (2007): Dopamine modulation of emotional processing in 33. Brody AL, Saxena S, Mandelkern MA, Fairbanks LA, Ho ML, Baxter LR cortical and subcortical neural circuits: Evidence for a final common (2001): Brain metabolic changes associated with symptom factor pathway in schizophrenia? Schizophr Bull 33:971–981.

172 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:163–172 www.sobp.org/BPCNNI Biological Psychiatry: Archival Report CNNI

Volatility Estimates Increase Choice Switching and Relate to Prefrontal Activity in Schizophrenia

Lorenz Deserno, Rebecca Boehme, Christoph Mathys, Teresa Katthagen, Jakob Kaminski, Klaas Enno Stephan, Andreas Heinz, and Florian Schlagenhauf

ABSTRACT BACKGROUND: Reward-based decision making is impaired in patients with schizophrenia (PSZ), as reflected by increased choice switching. The underlying cognitive and motivational processes as well as associated neural signatures remain unknown. Reinforcement learning and hierarchical Bayesian learning account for choice switching in different ways. We hypothesized that enhanced choice switching, as seen in PSZ during reward-based decision making, relates to higher-order beliefs about environmental volatility, and we examined the associated neural activity. METHODS: In total, 46 medicated PSZ and 43 healthy control subjects performed a reward-based decision-making task requiring flexible responses to changing action–outcome contingencies during functional magnetic resonance imaging. Detailed computational modeling of choice data was performed, including reinforcement learning and the hierarchical Gaussian filter. Trajectories of learning from computational modeling informed the analysis of functional magnetic resonance imaging data. RESULTS: A 3-level hierarchical Gaussian filter accounted best for the observed choice data. This model revealed a heightened initial belief about environmental volatility and a stronger influence of volatility on lower-level learning of action–outcome contingencies in PSZ as compared with healthy control subjects. This was replicated in an independent sample of nonmedicated PSZ. Beliefs about environmental volatility were reflected by higher activity in dorsolateral prefrontal cortex of PSZ as compared with healthy control subjects. CONCLUSIONS: Our study suggests that PSZ inferred the environment as overly volatile, which may explain increased choice switching. In PSZ, activity in dorsolateral prefrontal cortex was more strongly related to beliefs about environmental volatility. Our computational phenotyping approach may provide useful information to dissect clinical heterogeneity and could improve prediction of outcome. Keywords: Bayesian learning, Computational psychiatry, Neuroimaging, Psychosis, Reinforcement learning, Schizophrenia https://doi.org/10.1016/j.bpsc.2019.10.007

Cognitive and motivational deﬁcits are important characteris- with phasic dopamine (20,21). Considering enhanced presyn- tics of patients with schizophrenia (PSZ) associated with clin- aptic dopamine synthesis capacity in striatum of PSZ (22,23), ical and social outcomes (1–5). Reward-based learning and this could translate into enhanced phasic dopamine in PSZ, decision making require the integration of cognition and which in turn might result in increased learning rates (24). This motivation and are impaired in PSZ (6,7). These impairments could theoretically account for unstable behavior in PSZ, but are present at the onset of the disorder, are independent of increased learning rates were not found [for reviews, see lower general IQ, remain stable over time (8,9), and have been (18,24,25)]. proposed as neurocognitive markers with potential clinical Theories of predictive coding (26) and hierarchical Bayesian utility (10). However, the mechanisms and associated neural inference hypothesize that symptoms of PSZ (27–29) are a signatures remain to be identiﬁed. consequence of false inference about the world due to altered Flexible reward-based learning and decision making can be precision attributed to beliefs at different hierarchical levels. probed via variants of reversal learning [e.g., (11)]. In such Dysfunction at higher levels, which are thought to extract and tasks, PSZ show increased switching between choice options represent general and stable features of the environment, (8,12–17). The mechanisms of this unstable behavior remain might lead to experiencing the world as being more or less unknown but can be targeted by computational modeling (18). volatile. With regard to positive symptoms (30), this is sup- In reinforcement learning (RL) (19), choices are selected based ported by empirical evidence [e.g., (31)]. When applying this on expected values, which are learned by weighting reward framework to reward-based decision making, beliefs about the prediction errors (RPEs) with a learning rate. RPEs closely align probability of rewards are formed at lower levels but are also

ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. 173 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI Biological Psychiatry: CNNI Volatility and Choice Switching in Schizophrenia

determined by learning about the volatility of reward proba- Task bilities (32). This environmental volatility is related to learning Participants performed a task requiring flexible decision mak- from lower-level RPEs in that it scales the belief update. Thus, ing during fMRI (42–44). The task had 160 trials, each with a a belief in high environmental volatility can induce rapid up- choice between two cards (Figure 1A). The selected card dates of lower-level beliefs about reward probabilities and resulted in a monetary win or a monetary loss of 10 Eurocents. promote enhanced choice switching in PSZ. One card was initially assigned with a reward probability of Striatal and prefrontal activity is reduced during reward 80% and a loss probability of 20% (vice versa for the other anticipation and receipt in PSZ (33–35). Reduced striatal RPE card). The task had a simple higher-order structure (Figure 1B): activity was observed in nonmedicated PSZ (17) but not in an anticorrelation between the reward probabilities; whenever medicated PSZ (15,36). Neural correlates of hierarchical one card was associated with a probability of 80%, the other Bayesian learning were demonstrated in functional magnetic card would be associated with a probability of 20%. Reward resonance imaging (fMRI) studies in healthy individuals (37,38), contingencies were stable for the first 55 trials (pre-reversal) linking volatility and uncertainty to activity in frontostriatal cir- and for the last 35 trials (post-reversal). During the reversal cuits (32,39). While neural correlates of hierarchical Bayesian phase, contingencies changed four times after 15 or 20 trials. learning were successfully used to distinguish between in- For more details, see Supplement. dividuals with and without hallucinations and PSZ with and without psychosis (31), this has not yet facilitated an under- Analysis of Choice Behavior standing of the cognitive and motivational processes underlying impairments in flexible reward-based decision making. Performance was quantified by correct choices of the stimuli Here, we used a reward-based reversal learning task during with high (80%) reward probability and was analyzed using fMRI in PSZ and healthy control subjects (HCs). Computational repeated-measures analysis of variance (ANOVA) with the modeling was applied to the behavioral data by comparing RL between-subject factor group and the within-subject factor and a hierarchical Bayesian learning model, the hierarchical phase (pre-reversal, reversal, or post-reversal). Repeated- Gaussian filter (HGF) (40,41). We hypothesized that enhanced measures ANOVA was used to test the effect of feedback on choice switching in PSZ relates to higher-order beliefs about subsequent choices (win–stay and lose–stay). the volatility of the environment and examined the associated neural activity as measured by fMRI. Computational Models of Learning In RL, the difference between received rewards and expecta- METHODS AND MATERIALS tions, the RPE, is used to update expectations for the chosen stimulus (weighted by the learning rate a). For comparison with Participants and Instruments previous work (17,42–44), we included RL with separate In total, 46 medicated PSZ and 43 HCs were included (see learning rates for reward and loss trials (RL1 and RL2). Supplemental Table S1). Measures used to characterize par- The HGF describes learning as a process of inductive ticipants are summarized in Supplemental Table S1 and the inference under uncertainty. It considers hierarchically orga- Supplement. Written informed consent was obtained from all nized states in which learning at a higher-level state de- participants. The study was performed in accordance with the termines learning at a lower-level state by dynamically Declaration of Helsinki and was approved by the local ethics adjusting the lower level’s learning rate. In our case, the top committee of Charité Universitätsmedizin. level represents environmental volatility (how likely a change in

Figure 1. (A) Trial sequence from the decision- making task. (B) Reward probabilities of both choice options were perfectly anticorrelated and were stable for the ﬁrst 55 trials (pre-reversal); changed 4 times, after 15 or 20 trials, during the reversal phase; and remained stable for the last 35 trials (post-reversal). (C) Percentage choices of the stimulus with 80% reward probability were signiﬁ- cantly lower in the patients with schizophrenia (PSZ) group (main effect of group, F = 14.52, p , .001). (D) PSZ were less likely to repeat the previous action web 4C/FPO independent of feedback received in the previous trial (main effect of group, F = 27.77, p , .001; feedback 3 group interaction, F = 0.02, p = .89). HCs, healthy control subjects; ITI, intertrial interval; RT, reaction time.

174 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI Biological Psychiatry: Volatility and Choice Switching in Schizophrenia CNNI

action–outcome contingencies is to occur). This top-level es- between win and loss trials. We captured this by estimating timate is dynamically coupled with learning at the lower level parameters for win and loss trials that reﬂect this difference in

(see Figure 2). Trial-by-trial updates of posterior means at each choice perseveration (rwin and rloss). Models that included in- level are proportional to the prediction error (PE) from the level verse decision noise b as a free parameter had lower evidence below weighted by a precision ratio. See Supplement for (see Supplement) than models where this was ﬁxed to unity equations. We were particularly interested in environmental (b = 1). We also tested the possibility that volatility is directly

volatility (m3) and its coupling with the lower level (k) and thus linked to choice probabilities by letting third-level trial-by-trial inferred subject-specific parameters. We included a 2-level volatility (HGF3) serve as the inverse decision noise (see variant (HGF2) to test whether the representation of volatility Supplement). Because this introduced a volatility scale in the 3-level HGF3 made it superior in explaining behavior. anchored in observed behavior (switching or staying), it HGF and RL provide different ways to learn expectations allowed for estimating the mean of the subjective a priori belief ð0Þ about rewards, and both update expectations of the chosen about initial volatility at the third level, m3 , as a parameter of card only (single update [SU]). Based on the anticorrelated task HGF3. This cannot be applied to RL and HGF2 because they structure, we implemented a variant of each model updating do not feature inference on volatility. This led to 2 additional values (RL) or posterior means (HGF) of the unchosen card models (HGF3-SU-V and HGF3-DU-V), resulting in a total of 10 simultaneously; that is, an increase of the chosen card implies models. For model fitting, see Supplement. a decrease of the unchosen card (double update [DU]). For equations, see Supplement. SU and DU variants of each model (RL1, RL2, HGF2, and HGF3) were fit to the choice data Model Selection (Table 1). The negative variational free energy (an approximation to the log model evidence) was used for random-effects Bayesian Decision Models model selection (45). The protected exceedance probability Values (RL) or posterior means (HGF) were transformed to (PXP) governed our model selection, which protects against choice probabilities by using the softmax (logistic sigmoid) the null possibility that there are no differences in the likelihood function (see Supplement). In binary choice tasks with anti- of models across the population (46). We also examined correlated reward probabilities such as ours, there is choice whether the models explained the data better than chance perseveration independent of learning or inference that differs (17,47). A subject was classified as not fit better than chance if the log likelihood of the data relative to the number of trials did not significantly differ from chance (see Supplemental Methods). Simulations of the task were run using the inferred parameters to reproduce the observed data.

Model Parameters Parameters of the winning model were compared between groups using t test or the nonparametric Mann-Whitney U test if assumptions of normality were violated (Kolmogorov- Smirnov test). Bonferroni correction was applied according to the number of parameters.

FPO Statistical Analysis of fMRI Data = Using the general linear model approach in SPM8, an event- related analysis was applied. On the first level, 1 regressor web 4C spanned the entire trial from cue to outcome as in a previous Figure 2. Model graph. The hierarchical Gaussian filter deploys hierarchically organized states in which learning about environmental volatility at a study (38). We added the following 5 modeling-based trajectories as parametric modulators (not orthogonalized) to best higher-level state x3 determines lower-level learning about reward probabilities x2. The lowest level, x1, is binary and corresponds to a choice being capture different aspects of the hierarchical inference process: rewarded ( = 1) or not ( = 0) at a given trial. The probability of a choice x1 x1 second- and third-level precision-weighted PEs (ε2 and ε3), being rewarded is a logistic sigmoid function of x2: p(x1 =1)=s(x2). y rep- which were time locked to the outcome, precision weights (j2 resents the response of the subject. Shaded quantities are observed. Solid and j ), and the third-level volatility (m ). All regressors span- lines indicate dependence in the generative model. Dashed lines indicate 3 3 ned the entire trial and changed at outcome accordingly to PE dependence on inferred quantity (the generative model for y depends on m2 updates identical to Iglesias (38). Regressors were and m3, the inferred values of x2 and x3, respectively). The constant step size et al. u3 is the evolution rate of environmental volatility. k reflects the coupling convolved with the canonical hemodynamic response function between the levels. The best-fitting model was a three-level implementation in SPM8 and its temporal derivative (see Supplemental (HGF3-DU-V) with double updating (not illustrated) together with a decision Methods). For second-level analysis, a random-effects model capturing choice repetition separately after rewards and losses (r), ANOVA model, including contrast images of the 5 modeling- third-level environmental volatility determining decision noise, and the initial based trajectories (precision-weighted PEs [ε2 and ε3], preci- belief about environmental volatility m3 as an additional parameter inferred from the data. DU, double update; HGF, hierarchical Gaussian filter; V, sion weights [j2 and j3], and the third-level volatility [m3]) and environmental volatility. the factor group, was estimated.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI 175 Biological Psychiatry: CNNI Volatility and Choice Switching in Schizophrenia

Table 1. Bayesian Model Selection RL1-SU RL1-DU RL2-SU RL2-DU HGF2-SU HGF2-DU HGF3-SU HGF3-DU HGF3-SU-V HGF3-DU-V HCs 1 PSZ All (n = 89) PP 9.4 5.1 7.4 3.0 3.1 7.0 3.3 7.7 5.3 48.8 XP00000000 0 100 PXP00000000 0 99.9 HCs All (n = 43) PP 4.4 4.7 4.4 3.8 2.7 5.1 2.7 5.3 5.8 61.1 XP00000000 0 100 PXP00000000 0 100 PSZ All (n = 46) PP 11.8 7.1 11.6 4.75 8.3 10.0 9.3 10.9 4.9 21.4 XP 6.9 0.7 6.21 0.1 1.4 3.4 2.5 4.6 0.2 74.0 PXP 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.1 HCs 1 PSZ Fit (n = 73) PP 9.4 5.1 7.4 3.0 3.2 6.9 3.3 7.7 5.3 48.8 XP00000000 0 100 PXP00000000 0 100 HCs Fit (n = 42) PP 4.5 4.5 4.5 3.8 2.8 5.3 2.8 5.5 5.8 60.5 XP00000000 0 100 PXP 000000 0 100 PSZ Fit (n = 31) PP 11.6 8.2 11.3 4.9 7.9 9.7 8.9 10.4 4.9 22.3 XP 5.5 1.1 4.8 0.1 1.0 2.5 1.7 3.3 0.1 80.0 PXP 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.1 Bayesian model selection was governed by protected exceedance probabilities (PXPs) to protect against the risk that differences in model evidence occur by chance. In this table, we also report exceedance probabilities (XPs) and expected posterior probabilities (PPs); also see the Supplement. XP describes the probability of a model exceeding all other models in the comparison set, the probability that expected PPs differ. DU, double update; HCs, healthy control subjects; HGF, hierarchical Gaussian ﬁlter with 2 or 3 levels; HGF3-**-V, 3-level HGF with environmental volatility linked to decision noise with either SU or DU; PSZ, patients with schizophrenia; RL, reinforcement learning with one learning rate (RL1) or separate learning rates for rewards and losses (RL2); SU, single update.

RESULTS In total, 15 PSZ and 1 HCs were not fit better than chance by any model (Figure 3A). Neither when considering all PSZ Behavioral Data (PSZ-fit 1 PSZ-nofit) nor when considering PSZ-fit alone did Repeated-measures ANOVA on correct choices showed that Bayesian model selection reveal a clearly superior model (both performance differed between phases, dropping during the times Bayes omnibus risk = 1) (Table 1). However, the identi- reversal phase (main effect of phase, F = 23.74, p , .001). PSZ fication of PSZ-fit ensures that individuals included in further chose the better card less frequently irrespective of task modeling-based analyses are fit better than chance by every phases (main effect of group, F = 14.52, p , .001, and phase 3 model (i.e., equally good models instead of equally poor group interaction, F = 1.87, p = .16) (Figure 1C). The factor models). phase was dropped from further analyses. Repeated-measures ANOVA on the probability of choice Revisiting Behavioral Data repetition showed that all participants stayed more with the Based on this heterogeneity in PSZ regarding absolute model previous action after rewards compared with losses (main ef- fit, we revisited choice data with respect to 3 groups (HCs, fect of feedback, F = 369.80, p , .001) and that PSZ switched PSZ-fit, and PSZ-nofit). There was a main effect of group on more independent of feedback from the previous trial (main correct choices (F = 32.63, p , .001) (Figure 3B). PSZ-nofit effect of group, F = 27.77, p , .001, and feedback 3 group showed performance around chance levels (HCs vs. PSZ- interaction, F = 0.02, p = .89) (Figure 1D). nofit, t = 7.04, p , .001; PSZ-fit vs. PSZ-nofit, t = 6.90, p , .001) (Figure 3B), while PSZ-fit had performance comparable to HCs (t = 1.51, p = .14) (Figure 3B). The analysis of win–stay and Computational Modeling: Model Selection lose–stay behavior revealed a group 3 feedback interaction Random-effects Bayesian model selection revealed a 3-level (F = 20.68, p , .001). This resulted from a pronounced HGF with double updating and third-level environmental reduction of win–stay behavior in PSZ-nofit only (PSZ-fit vs. volatility linked to decision noise (Figure 2) as the most plau- PSZ-nofit, t = 10.74, p , .001) (Figure 3C), while reduced lose– sible model (HGF3-DU-V, PXP = 99.5%; for PXPs of all stay was not significantly different between PSZ-fit and PSZ- models, see Table 1). This model (HGF3-DU-V) was superior in nofit(t = 0.01, p = .99) (Figure 3D). A group 3 feedback HCs (PXP = 100%, Bayes omnibus risk = 0) and remained first interaction was also significant when comparing only HCs and ranking in PSZ (posterior probabilities = 21.4%, PXP = 74.0%). PSZ-fit(F = 6.79, p = .01) as well as significant main effects of In PSZ, there was no convincing evidence that models per- feedback (F = 636.30, p , .001) and group (F = 10.22, p = .01). formed differently from each other (Bayes omnibus risk = 1, all This difference between HCs and PSZ-fit was driven by PXPs = 10.0%). switching after loss (Figure 3C, D).

176 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI Biological Psychiatry: Volatility and Choice Switching in Schizophrenia CNNI

Figure 3. (A, B) Classification above (black dots) and beIow (red crosses) chance (A) and its influence on overall choice performance (B). There was a main effect of group (F = 32.63, p , .001). Patients with schizophrenia (PSZ)-nofit (red) showed overall poor performance (healthy control subjects [HCs] vs. PSZ-nofit, t = 7.04, p , .001; PSZ-fit vs. PSZ-nofit, t = 6.90, p , .001), while PSZ-fit (green) performed comparably to HCs (blue) (t = 1.51, p , .14) (B). (C, D) Analysis of win–stay (C) and lose–stay (D) behavior showed a group 3 feedback interaction (F = 20.68, p , .001). There was a pronounced

web 4C/FPO reduction of win-stay behavior in PSZ-nofit only (PSZ-fit vs. PSZ-nofit, t = 10.74, p , .001) (C), while reduced lose-stay behavior characterized both groups of PSZ (PSZ-fit vs. PSZ-nofit, t = 0.01, p = .99) (D). This group 3 feedback interaction was also significant when comparing only HCs and PSZ-fit.

In an exploratory analysis of 6 cognitive tests, only mea- Computational Modeling: Replication in sures of verbal memory and working memory differed between Nonmedicated PSZ the 2 groups of PSZ, that is, were more impaired in PSZ-nofit We tested our model (HGF3-DU-V) in an independent sample compared with PSZ-fit (see Supplemental Results). This sug- of nonmedicated PSZ (n = 24) and HCs (n = 24), who per- gests that PSZ-fit and PSZ-nofit mapped on distinct cognitive formed another reversal-learning task (17). For statistics, see profiles. Because poor fit hinders the interpretation of Table 2. This replicated between-group findings and remained modeling-based behavioral and neuroimaging analyses in the significant when excluding participants not fit better than PSZ-nofit subgroup, all subsequent modeling-based results chance (23 HCs, 13 PSZ; not reported). are reported based on HCs (n = 42) and PSZ-fit(n = 31) only.

Computational Modeling: Parameters Relation to Symptoms Comparison of parameters of HGF3-DU-V (Table 2 and We explored the relation between the 2 parameters that Figure 4) revealed that the estimated mean of the a priori belief differed between groups with different measures of cognition ð0Þ (n = 6) and clinical measures (n = 7) within PSZ (Supplemental about initial environmental volatility m3 was higher in PSZ (z = Table S1) applying Bonferroni correction (p , .0083). In PSZ, a 3.15, p , .01) (Figure 4A). Trial-by-trial environmental volatility ð0Þ was more strongly coupled with lower-level updating, as higher initial belief about volatility m3 was associated with reduced executive functioning and cognitive speed (Trail demonstrated by higher k in PSZ (z = 2.51, p , .01) (Figure 4B). Making Test B: = 2.56, , .001; Digit Symbol Substitution The evolution rate of environmental volatility u3 did not differ r p Test: = 2.56, , .001) (Supplemental Table S3 and significantly between groups (z = 0.73, p = .47). To illustrate the r p effects of differences in parameters on behavior after losses Supplemental Figure S3). These correlations were not present (when PSZ-fit showed increased switching), we analyzed the in the HC group. For all explorative correlations, see Supplemental Table S3. trajectory of m3 in a mixed-effects regression model with group and feedback as predictors. This revealed higher m3 in PSZ-fit overall. Across groups, m3 was higher after losses compared fMRI Task Effects (Pooled Across Groups) with rewards, which was more pronounced in PSZ (resulting Activity related to ε2 (a precision-weighted RPE) peaked in from enhanced coupling between higher and lower levels of k). bilateral ventral striatum and ventromedial prefrontal cortex For statistics, see Supplemental Results and Supplemental (PFC) among other regions (p-FWEwholebrain , .05, where FWE Figure S2. is familywise error) (Figure 5A and Supplemental Table S4),

including the midbrain (p-FWEmidbrain-voi , .05, where voi is Computational Modeling: Reproducing Observed volume of interest) (Supplemental Table S9), a well-known Behavior network associated with RPEs. In contrast, third-level preci-

Simulating data based on the inferred parameters of HGF3- sion-weighted PE (ε3) was associated with activity in prefrontal, DU-V (42 HCs, 31 PSZ-ﬁt, 10 simulations per subject) parietal, and left insular regions (Figure 5A and Supplemental

showed that PSZ-ﬁt switched more than HCs (main effect of Table S5). Environmental volatility (m3) covaried with activa- group, F = 11.17, p , .001) and showed a pronounced ten- tion in bilateral insula, cingulate cortex, parietal cortex, middle dency to switch after losses (group 3 feedback, F = 7.68, p = temporal gyrus, globus pallidus, and thalamus as well as su- .01). Between-group ﬁndings on behavioral data were fully perior, middle, and inferior frontal gyrus (Figure 6A, reproduced, which yields an important validation of the Supplemental Figure S5, and Supplemental Table S8). For model’s ability to capture the observed data. more details on group-level fMRI effects, including activity

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI 177 Biological Psychiatry: CNNI Volatility and Choice Switching in Schizophrenia

Table 2. Between-Group Comparisons of Model Parameters Using t Tests or the Nonparametric Mann-Whitney U Test If Assumptions of Normality Were Violated Models HCs (n = 42) PSZ (n = 31) Test Statistic Learning Model

m3 20.84 6 0.49 20.47 6 0.48 z = 3.15, p , .01 k 0.87 6 0.60 1.35 6 1.12 z = 2.51, p , .01

u3 26.00 6 0.02 25.99 6 0.05 z = 0.73, p = .47 Decision Model

rwin 0.97 6 0.57 1.08 6 0.49 z = 0.88, p = .38

rloss 0.08 6 0.31 20.12 6 0.44 z = 2.37, p = .02 Replication Samplea HCs (n = 24) Nonmedicated PSZ (n = 24) Test Statistic Learning Model

m3 21.17 6 0.61 20.43 6 0.70 z = 3.88, p , .01 k 0.61 6 0.58 1.56 6 1.21 z = 3.46, p , .01

u3 26.09 6 0.02 25.99 6 0.06 z = 1.35, p = .18 Decision Model

rwin 0.70 6 0.74 0.43 6 0.66 z = 1.35, p = .18

rloss 20.07 6 0.59 20.28 6 0.77 z = 1.05, p = .30 Bonferroni correction was applied according to the number of parameters (5) (p , .01). HCs, healthy control subjects; PSZ, patients with schizophrenia. aSchlagenhauf et al.(17).

specific for each group outside the effect of each regressor during reward-based decision making. We present two main combined for HCs and PSZ, see Supplemental Results. findings. First, our modeling suggests that medicated PSZ acted under an a priori enhanced higher-level belief about initial envi- fMRI Between-Group Effects ronmental volatility and increased coupling between higher and We conducted between-group comparison of the covariance lower levels of learning, which leads to enhanced lower-level between the modeling regressors derived from the best-fitting belief updating about action–outcome contingencies. This pro- model and blood oxygen level–dependent response within vides a computational account of choice switching, as was previously observed in PSZ (8,12–17). Using parameters of the SPM. For the regressor of environmental volatility m3, a group difference between HCs and PSZ was found in right dorso- winning model to simulate new data, we fully reproduced observed patterns in the behavioral data, and we replicated our lateral PFC (DLPFC) (F contrast, using a mask representing the average effect of m over all participants for correction of findings on parameters in an independent cohort of non- 3 medicated PSZ. Second, medicated PSZ displayed higher multiple comparison [x = 34, y = 44, z = 24], F = 19.89, z = 4.24, DLPFC activity related to environmental volatility, which points pFWE = .04) (Figure 6B). Post hoc analysis revealed stronger activity related to volatility in DLPFC of PSZ compared with toward a prominent role of this region in promoting unstable behavior in PSZ. HCs (t = 4.46, z =4.4,pFWE = .02) (Figure 6C). There was no significant difference between groups for any other PSZ show enhanced choice switching (8,12–17), and we regressor. demonstrate a possible underlying mechanism: an enhanced initial belief about the environmental volatility and a stronger coupling of volatility and lower-level learning of action–outcome DISCUSSION contingencies. This has two consequences. First, PSZ had an To the best of our knowledge, this is the first study to apply overall stronger tendency to switch (enhanced initial belief about hierarchical Bayesian learning to choice and fMRI data of PSZ volatility). Second, lower-level beliefs fluctuated more strongly and

Figure 4. (A) The initial estimate of environmental volatility is significantly higher in patients with schizophrenia (PSZ)-fit(n = 31) as compared with healthy control subjects (HCs). (B) The coupling between the third level (environmental volatility) and the second level is significantly stronger in PSZ-fit (n = 31) compared with HCs. web 4C/FPO

178 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI Biological Psychiatry: Volatility and Choice Switching in Schizophrenia CNNI FPO = web 4C Figure 5. Blood oxygen level–dependent signal across all participants related to precision-weighted prediction errors from the second (red) and third (blue) levels (A) and precision weight from the second (red) and third (green) levels (B), with overlap in yellow (both at p-FWEwholebrain , .05, k = 10). FWE, familywise error.

led to increased choice switching, particularly after (irrelevant) heightened neural representation of volatility may generate more losses. Thus, PSZ inferred more contingency changes in this dy- stochastic behavior, although in our correlational and cross- namic task environment that are putatively signaled through los- sectional study we cannot ascertain a causal link. We found a ses. Enhanced estimates of changes in context probabilities negative correlation of the initial belief about volatility with in- were alsofound in PSZinthenonreward domain(48). In contrast to dependent neuropsychological measures of executive func- our finding of enhanced initial belief about volatility, Powers et al. tioning and cognitive speed in PSZ, thereby emphasizing its (31) probed conditioned hallucinations and reported dysfunctional character, while these associations were not stronger lower-level priors about perceptual inputs combined with observed in the HC group. reduced evolution of volatility in hallucinating participants. A PSZ’s heightened belief about volatility may render a system possible explanation may be that alterations of volatility (hyper)sensitive to any new input (51,52), thereby impeding the estimates differ with regard to the investigated functional domain, detection of regularities in probabilistic environments and leading potentially related to different symptom dimensions. As sug- to (hyper)flexible updating in response to (irrelevant) information. gested by our finding, cognitive beliefs about the structure of the Meta-analyses showed reduced prefrontal activity in PSZ environment appear to be more unstable, while perceptual beliefs compared with HCs for (relevant) task versus (irrelevant) condi- about sensory inputs appear to be overly stable and not appro- tions across multiple cognitive measures (53,54). On the one hand, priately adjusted following changes in the environment (31). prefrontal dysfunction in PSZ may contribute to an enhanced Because our model revealed consistent results in medicated higher-level belief about volatility (e.g., by impairing the detection and nonmedicated PSZ, elevated beliefs about environmental of higher-level regularities), and such beliefs about volatility might volatility may represent an important mechanism of impaired be assigned with enhanced precision (potentially in a compen- flexible decision making. In a similar vein, an inability to stabilize satory manner). On the other hand, lower-level beliefs may be behavior according to an internal model of action–outcome more unstable and presumably assigned with lower precision, contingencies was found after administration of ketamine in leading to distinct aberrant experiences and behaviors depending healthy control subjects (49), in line with the assumption that on the tested domain, with most evidence so far coming from reduced (prefrontal) NMDA receptor functioning (27,50)maylead perceptual processing (55). to aberrant cortical information processing (51). In line with this Aberrant cortical processing was theorized, at least in non- idea, we found a stronger association in PSZ than in HCs of medicated PSZ, to increase striatal dopamine turnover (27,50), beliefs about volatility with blood oxygen level–dependent ac- which might interfere with striatal and midbrain RPE signals. tivity in DLPFC. However, in our fMRI study, we cannot infer Indeed, in nonmedicated PSZ, striatal RPE activity was found to about involved neurochemical systems. On the behavioral level, be reduced (17,56). In RL accounts (24), enhanced spontaneous beliefs about higher volatility, in our model directly linked to phasic dopamine transients could highlight irrelevant stimuli and decision noise, can lead to more stochastic behavior (in our task disturb the signaling of (relevant) RPEs. In our medicated sample overall more choice switching). We therefore suggest that of PSZ, no significant differences in striatal activations to RPEs

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI 179 Biological Psychiatry: CNNI Volatility and Choice Switching in Schizophrenia FPO = web 4C Figure 6. (A) Across all participants, blood oxygen level–dependent signal related to volatility estimate from the third level at p-FWE , .05 for the whole brain, k = 10. (B) m3-related blood oxygen level–dependent signal in the dorsolateral prefrontal cortex (DLPFC) differs between patients with schizophrenia (PSZ)-ﬁt(n = 31) and healthy control subjects (HCs) (F contrast displayed at p , .001 uncorrected; corrected for main effect of m3 over all participants [x = 34, y = 44, z = 24], F = 19.89, z = 4.24, p = .038). (C) A post hoc analysis of regression parameter estimates at the peak of the group difference (x = 34, y = 44, z = 24) showed that this was driven by heightened activation related to environmental volatility in DLPFC of PSZ-ﬁt(n = 31) compared with HCs (t = 4.46, z = 4.4, p = .019). FWE, familywise error.

were observed, in line with reports of absent differences in striatal generalization of results. However, 2 subgroups were identiﬁed RPEs in medicated PSZ (14,15). This suggests medication status with (task-independent) different cognitive proﬁles, which as an important factor relating to striatal RPE signaling similar to contributes to disentangling heterogeneity across PSZ—a medication effects on striatal reward anticipation in PSZ fundamental challenge for psychiatric research (59). By con- (35,57,58). In hierarchical Bayesian learning, representation of trolling whether subjects’ performance can actually be inter- lower-level PEs may be similar in patients and control subjects, preted as assumed by the theories, we eliminated a potential but potentially reduced precision of lower-level beliefs might key confound. Nevertheless, it remains problematic if clinical highlight irrelevant inputs (e.g., resulting in choice switching). This groups differ in how well they are described by models of in- could, at least in theory, result from a common aberrant terest, as observed in our study because parameters are prefrontal process, as discussed above, but also from an effect of conditional on the model. This impedes the interpretation of

antipsychotic D2 receptor antagonists in the striatum (29). differences in parameters across groups [for a discussion of However, we did not observe group differences in midbrain and this problem, see (60)]. In principle, Bayesian model averaging striatum for precision weights and precision-weighted PEs. While (61) can help, but this is not established for non-nested models our data indicate a disrupted higher-level process with evidence as used here. Future studies might implement tasks with from behavioral modeling and fMRI, disturbed lower-level adaptive difﬁculty to reduce the number of patients whose processes are supported by our behavioral modeling but not by behavior cannot be explained by any model, which may the presented fMRI data. potentially be a result of excessive cognitive demands. Furthermore, the relation between different Bayesian modeling Limitations approaches such as the HGF and active inference models First, the lack of clear superiority of any model for the data should be explored (62). from PSZ, as well as the substantial number of PSZ in which Second, we suggest that overestimating volatility is one no model ﬁtted better than chance, needs to be considered. possible explanation for choice switching. However, in our Excluding these patients from further modeling-based ana- model, volatility partly determines decision noise. This limits lyses can be considered restrictive and may reduce the the differentiation between the concepts of volatility and

180 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI Biological Psychiatry: Volatility and Choice Switching in Schizophrenia CNNI

exploration and limits interpretability of volatility estimates to Berlin, Berlin; Bernstein Center for Computational Neuroscience (AH, FS), some extent. The finding that this model performs best is an Berlin; Max Planck Institute for Human Cognitive and Brain Sciences (LD, indication that our data do not fully support a strong distinction FS), Leipzig; and Max Planck Institute for Metabolism Research (KES), Cologne, Germany; Max Planck UCL Centre for Computational Psychiatry between volatility and exploration. Additional task-based ma- and Ageing Research (LD, CM) and Wellcome Centre for Human Neuro- nipulations would be required to overcome this (37), which imaging (LD, CM, KES), Institute of Neurology, University College London, would most likely involve longer tasks than our patient-friendly London, United Kingdom; Center for Social and Affective Neuroscience fMRI task (,15 minutes, 160 trials). The formulation of our (RB), Linköping University, Linköping, Sweden; Scuola Internazionale response model suggests that results are still informative. We Superiore di Studi Avanzati (CM), Trieste, Italy; Translational Neuromodeling control for overall differences in stickiness with 2 parameters Unit (CM, KES), Institute for Biomedical Engineering, University of Zurich and ETH Zurich, Zurich, Switzerland. that change the shape of the decision function differently for LD and RB contributed equally to this work. wins and losses and exert a bias toward repeating or switching Address correspondence to Lorenz Deserno, M.D., Max Planck UCL responses irrespective of learned expectations (see Centre for Computational Psychiatry and Ageing Research, London, 10-12 Supplement). Therefore, decision noise is determined not only Russell Square, London WC1B 5EH, United Kingdom; E-mail: l.deserno@ by trialwise volatility but also by subject-specific traits that are ucl.ac.uk. expressed in a condition-specific fashion. This disentangles Received Jul 24, 2019; revised Sep 11, 2019; accepted Oct 6, 2019. Supplementary material cited in this article is available online at https:// volatility and decision noise to some degree. doi.org/10.1016/j.bpsc.2019.10.007. Third, this is a case-control study that fundamentally limits the inferences that can be drawn from the results, for example, the development of inappropriately high initial beliefs about REFERENCES volatility over the course of illness, its stability over time, and its relation to broader cognitive deficits consistently found in PSZ. 1. Bowie CR, Leung WW, Reichenberg A, McClure MM, Patterson TL, Heaton RK, et al. (2008): Predicting schizophrenia patients’ real-world In summary, we present a computational mechanism behavior with specific neuropsychological and functional capacity putatively underlying unstable behavior in PSZ: a stronger measures. Biol Psychiatry 63:505–511. coupling of heightened beliefs about environmental volatility 2. Green MF (1996): What are the functional consequences of neuro- with lower-level learning, which was present in medicated and cognitive deficits in schizophrenia? Am J Psychiatry 153:321–330. nonmedicated PSZ. In medicated PSZ, this was accompanied 3. Nuechterlein KH, Subotnik KL, Green MF, Ventura J, Asarnow RF, by enhanced activity related to environmental volatility in Gitlin MJ, et al. (2011): Neurocognitive predictors of work outcome in recent-onset schizophrenia. Schizophr Bull 37(suppl 2):S33–S40. DLPFC. Future studies should aim to test specificity of the 4. Gonzalez-Ortega I, de Los Mozos V, Echeburua E, Mezo M, Besga A, presented results for PSZ and overcome the limitation of the Ruiz de Azua S, et al. (2013): Working memory as a predictor of lack of longitudinal clinical data. Computational modeling may negative symptoms and functional outcome in first episode psychosis. aid in the identification of subgroups of PSZ (63) and poten- Psychiatry Res 206:8–16. tially inform the prediction of treatment response to antipsy- 5. Dominguez Mde G, Viechtbauer W, Simons CJ, van Os J, chotic drugs by aiming to dissect the important biological Krabbendam L (2009): Are psychotic psychopathology and neurocognition orthogonal? A systematic review of their associations. heterogeneity and interindividual differences among patients Psychol Bull 135:157–171. (64). These steps toward clinically useful procedures will 6. Gold JM, Waltz JA, Prentice KJ, Morris SE, Heerey EA (2008): Reward require carefully designed prospective studies in the frame- processing in schizophrenia: A deficit in the representation of value. work of computational psychiatry (29,65–67). Schizophr Bull 34:835–847. 7. Barch DM, Dowd EC (2010): Goal representations and motivational drive in schizophrenia: The role of prefrontal-striatal interactions. ACKNOWLEDGMENTS AND DISCLOSURES Schizophr Bull 36:919–934. This study was supported by the Max Planck Society and grants from the 8. Murray GK, Cheng F, Clark L, Barnett JH, Blackwell AD, Fletcher PC, German Research Foundation (Grant Nos. DFG SCHL1969/1-2, DFG SCHL et al. (2008): Reinforcement and reversal learning in first-episode 1969/3-1, and DFG SCHL1969/4-1 [to FS]). JK is supported by the Charité psychosis. Schizophr Bull 34:848–855. Clinician-Scientist Program of the Berlin Institute of Health. KES acknowl- 9. Leeson VC, Robbins TW, Matheson E, Hutton SB, Ron MA, Barnes TR, edges support by the René and Susanne Braginsky Foundation and the et al. (2009): Discrimination learning, reversal, and set-shifting in first- University of Zurich. episode schizophrenia: Stability over six years and specific associa- LD, CM, KES, and FS designed the study. LD, RB, TK, and JK performed tions with medication type and disorganization syndrome. Biol research. LD and RB analyzed the data. CM and FS supervised data anal- Psychiatry 66:586–593. ysis. LD and RB wrote the initial version of the manuscript. LD, RB, TK, JK, 10. Ragland JD, Cohen NJ, Cools R, Frank MJ, Hannula DE, Ranganath C CM, KES, AH, and FS read and corrected versions of the manuscript. (2012): CNTRICS imaging biomarkers final task selection: Long-term Data from this study were presented at the following conferences: 71st memory and reinforcement learning. Schizophr Bull 38:62–72. annual meeting of the Society for Biological Psychiatry, May 12–14, 2016, 11. Cools R, Clark L, Owen AM, Robbins TW (2002): Defining the neural Atlanta, Georgia; 6th biennial Schizophrenia International Research Society mechanisms of probabilistic reversal learning using event-related Conference, April 4–8, 2018, Florence, Italy; and 11th Forum of Neurosci- functional magnetic resonance imaging. J Neurosci 22:4563–4567. ence, July 7–11, 2018, Berlin, Germany. 12. Waltz JA, Gold JM (2007): Probabilistic reversal learning impairments All authors report no biomedical financial interests or potential conflicts in schizophrenia: Further evidence of orbitofrontal dysfunction. of interest. Schizophr Res 93:296–303. 13. Waltz JA, Kasanova Z, Ross TJ, Salmeron BJ, McMahon RP, Gold JM, et al. (2013): The roles of reward, default, and executive control net- ARTICLE INFORMATION works in set-shifting impairments in schizophrenia. PLoS One 8: From the Department of Psychiatry and Psychotherapy (LD, RB, TK, JK, AH, e57257. FS), Charité Universitätsmedizin Berlin, corporate member of Freie Uni- 14. Culbreth AJ, Gold JM, Cools R, Barch DM (2016): Impaired activation versität Berlin, Humboldt Universität zu Berlin, and Berlin Institute of Health, in cognitive control regions predicts reversal learning in schizophrenia. Berlin; Cluster of Excellence NeuroCure (AH), Charité Universitätsmedizin Schizophr Bull 42:484–493.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI 181 Biological Psychiatry: CNNI Volatility and Choice Switching in Schizophrenia

15. Culbreth AJ, Westbrook A, Xu Z, Barch DM, Waltz JA (2016): Intact 38. Iglesias S, Mathys C, Brodersen KH, Kasper L, Piccirelli M, den ventral striatal prediction error signaling in medicated schizo- Ouden HE, et al. (2013): Hierarchical prediction errors in midbrain and phrenia patients. Biol Psychiatry Cogn Neurosci Neuroimaging basal forebrain during sensory learning. Neuron 80:519–530. 1:474–483. 39. Rushworth MF, Noonan MP, Boorman ED, Walton ME, Behrens TE 16. Reddy LF, Waltz JA, Green MF, Wynn JK, Horan WP (2016): Proba- (2011): Frontal cortex and reward-guided learning and decision- bilistic reversal learning in schizophrenia: Stability of deficits and po- making. Neuron 70:1054–1069. tential causal mechanisms. Schizophr Bull 42:942–951. 40. Mathys C, Daunizeau J, Friston KJ, Stephan KE (2011): A Bayesian 17. Schlagenhauf F, Huys QJ, Deserno L, Rapp MA, Beck A, Heinze HJ, foundation for individual learning under uncertainty. Front Hum Neu- et al. (2014): Striatal dysfunction during reversal learning in unmedi- rosci 5:39. cated schizophrenia patients. NeuroImage 89:171–180. 41. Mathys CD, Lomakina EI, Daunizeau J, Iglesias S, Brodersen KH, 18. Deserno L, Boehme R, Heinz A, Schlagenhauf F (2013): Reinforcement Friston KJ, et al. (2014): Uncertainty in perception and the hierarchical learning and dopamine in schizophrenia: Dimensions of symptoms or Gaussian filter. Front Hum Neurosci 8:825. specific features of a disease group? Front Psychiatry 4:172. 42. Boehme R, Deserno L, Gleich T, Katthagen T, Pankow A, Behr J, et al. 19. Sutton RS, Barto AG (1998): Reinforcement Learning: An Introduction. (2015): Aberrant salience is related to reduced reinforcement learning Cambridge, MA: MIT Press. signals and elevated dopamine synthesis capacity in healthy adults. 20. Schultz W, Dayan P, Montague PR (1997): A neural substrate of pre- J Neurosci 35:10103–10111. diction and reward. Science 275:1593–1599. 43. Reiter AMF, Deserno L, Kallert T, Heinze HJ, Heinz A, Schlagenhauf F 21. Steinberg EE, Keiflin R, Boivin JR, Witten IB, Deisseroth K, Janak PH (2016): Behavioral and neural signatures of reduced updating of (2013): A causal link between prediction errors, dopamine neurons and alternative options in alcohol-dependent patients during flexible de- learning. Nat Neurosci 16:966–973. cision-making. J Neurosci 36:10935–10948. 22. Howes OD, Kambeitz J, Kim E, Stahl D, Slifstein M, Abi-Dargham A, 44. Reiter AMF, Heinze HJ, Schlagenhauf F, Deserno L (2017): Impaired et al. (2012): The nature of dopamine dysfunction in schizophrenia and flexible reward-based decision-making in binge eating disorder: Evi- what this means for treatment. Arch Gen Psychiatry 69:776–786. dence from computational modeling and functional neuroimaging. 23. Heinz A (2002): Dopaminergic dysfunction in alcoholism and Neuropsychopharmacology 42:628–637. schizophrenia—Psychopathological and behavioral correlates. Eur 45. Stephan KE, Penny WD, Daunizeau J, Moran RJ, Friston KJ (2009): Psychiatry 17:9–16. Bayesian model selection for group studies. NeuroImage 46:1004– 24. Maia TV, Frank MJ (2017): An integrative perspective on the role of 1017. dopamine in schizophrenia. Biol Psychiatry 81:52–66. 46. Rigoux L, Stephan KE, Friston KJ, Daunizeau J (2014): Bayesian model 25. Deserno L, Heinz A, Schlagenhauf F (2017): Computational ap- selection for group studies—revisited. NeuroImage 84:971–985. proaches to schizophrenia: A perspective on negative symptoms. 47. Stephan KE, Schlagenhauf F, Huys QJ, Raman S, Aponte EA, Schizophr Res 186:46–54. Brodersen KH, et al. (2017): Computational neuroimaging strategies 26. Rao RP, Ballard DH (1999): Predictive coding in the visual cortex: A for single patient predictions. NeuroImage 145:180–199. functional interpretation of some extra-classical receptive-field effects. 48. Kaplan CM, Saha D, Molina JL, Hockeimer WD, Postell EM, Apud JA, Nat Neurosci 2:79–87. et al. (2016): Estimating changing contexts in schizophrenia. Brain 27. Stephan KE, Friston KJ, Frith CD (2009): Dysconnection in schizo- 139:2082–2095. phrenia: From abnormal synaptic plasticity to failures of self- 49. Vinckier F, Gaillard R, Palminteri S, Rigoux L, Salvador A, Fornito A, monitoring. Schizophr Bull 35:509–527. et al. (2016): Confidence and psychosis: A neuro-computational ac- 28. Corlett PR, Frith CD, Fletcher PC (2009): From drugs to deprivation: A count of contingency learning disruption by NMDA blockade. Mol Bayesian framework for understanding models of psychosis. Psychiatry 21:946–955. Psychopharmacology (Berl) 206:515–530. 50. Krystal JH, Perry EB Jr, Gueorguieva R, Belger A, Madonick SH, Abi- 29. Adams RA, Huys QJ, Roiser JP (2016): Computational psychiatry: Dargham A, et al. (2005): Comparative and interactive human psy- Towards a mathematically informed understanding of mental illness. chopharmacologic effects of ketamine and amphetamine: Implications J Neurol Neurosurg Psychiatry 87:53–63. for glutamatergic and dopaminergic model psychoses and cognitive 30. Fletcher PC, Frith CD (2009): Perceiving is believing: A Bayesian function. Arch Gen Psychiatry 62:985–994. approach to explaining the positive symptoms of schizophrenia. Nat 51. Lewis DA, Gonzalez-Burgos G (2006): Pathophysiologically based Rev Neurosci 10:48–58. treatment interventions in schizophrenia. Nat Med 12:1016–1022. 31. Powers AR, Mathys C, Corlett PR (2017): Pavlovian conditioning- 52. Durstewitz D, Seamans JK (2008): The dual-state theory of prefrontal induced hallucinations result from overweighting of perceptual cortex dopamine function with relevance to catechol-o-methyl- priors. Science 357:596–600. transferase genotypes and schizophrenia. Biol Psychiatry 64:739–749. 32. Rushworth MF, Behrens TE (2008): Choice, uncertainty and value in 53. McTeague LM, Huemer J, Carreon DM, Jiang Y, Eickhoff SB, Etkin A prefrontal and cingulate cortex. Nat Neurosci 11:389–397. (2017): Identification of common neural circuit disruptions in cognitive 33. Juckel G, Schlagenhauf F, Koslowski M, Wustenberg T, Villringer A, control across psychiatric disorders. Am J Psychiatry 174:676–685. Knutson B, et al. (2006): Dysfunction of ventral striatal reward pre- 54. Glahn DC, Ragland JD, Abramoff A, Barrett J, Laird AR, Bearden CE, diction in schizophrenia. NeuroImage 29:409–416. et al. (2005): Beyond hypofrontality: A quantitative meta-analysis of 34. Schlagenhauf F, Sterzer P, Schmack K, Ballmaier M, Rapp M, Wrase J, functional neuroimaging studies of working memory in schizophrenia. et al. (2009): Reward feedback alterations in unmedicated schizo- Hum Brain Mapp 25:60–69. phrenia patients: Relevance for delusions. Biol Psychiatry 65:1032– 55. Sterzer P, Adams RA, Fletcher P, Frith C, Lawrie SM, Muckli L, et al. 1039. (2018): The predictive coding account of psychosis. Biol Psychiatry 35. Radua J, Schmidt A, Borgwardt S, Heinz A, Schlagenhauf F, 84:634–643. McGuire P, Fusar-Poli P (2015): Ventral striatal activation during 56. Reinen JM, Van Snellenberg JX, Horga G, Abi-Dargham A, Daw ND, reward processing in psychosis: A neurofunctional meta-analysis. Shohamy D (2016): Motivational context modulates prediction error JAMA Psychiatry 72:1243–1251. response in schizophrenia. Schizophr Bull 42:1467–1475. 36. Dowd EC, Frank MJ, Collins A, Gold JM, Barch DM (2016): Probabi- 57. Nielsen MO, Rostrup E, Wulff S, Bak N, Broberg BV, Lublin H, et al. listic reinforcement learning in patients with schizophrenia: Relation- (2012): Improvement of brain reward abnormalities by antipsychotic ships to anhedonia and avolition. Biol Psychiatry Cogn Neurosci monotherapy in schizophrenia. Arch Gen Psychiatry 69:1195–1204. Neuroimaging 1:460–473. 58. Schlagenhauf F, Juckel G, Koslowski M, Kahnt T, Knutson B, 37. Behrens TE, Woolrich MW, Walton ME, Rushworth MF (2007): Dembler T, et al. (2008): Reward system activation in schizophrenic Learning the value of information in an uncertain world. Nat Neurosci patients switched from typical neuroleptics to olanzapine. Psycho- 10:1214–1221. pharmacology (Berl) 196:673–684.

182 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI Biological Psychiatry: Volatility and Choice Switching in Schizophrenia CNNI

59. Stephan KE, Bach DR, Fletcher PC, Flint J, Frank MJ, Friston KJ, 63. Brodersen KH, Deserno L, Schlagenhauf F, Lin Z, Penny WD, et al. (2016): Charting the landscape of priority problems in psy- Buhmann JM, et al. (2014): Dissecting psychiatric spectrum disorders chiatry, part 1: Classiﬁcation and diagnosis. Lancet Psychiatry by generative embedding. NeuroImage Clin 4:98–111. 3:77–83. 64. Wolfers T, Doan NT, Kaufmann T, Alnaes D, Moberget T, Agartz I, et al. 60. Stephan KE, Penny WD, Moran RJ, den Ouden HE, Daunizeau J, (2018): Mapping the heterogeneous phenotype of schizophrenia and bi- Friston KJ (2010): Ten simple rules for dynamic causal modeling. polar disorder using normative models. JAMA Psychiatry 75:1146–1155. NeuroImage 49:3099–3109. 65. Huys QJ, Maia TV, Frank MJ (2016): Computational psychiatry as a 61. Penny WD, Stephan KE, Daunizeau J, Rosa MJ, Friston KJ, bridge from neuroscience to clinical applications. Nat Neurosci Schoﬁeld TM, et al. (2010): Comparing families of dynamic causal 19:404–413. models. PLoS Comput Biol 6:e1000709. 66. Stephan KE, Mathys C (2014): Computational approaches to psychi- 62. Cullen M, Davey B, Friston KJ, Moran RJ (2018): Active inference in atry. Curr Opin Neurobiol 25:85–92. OpenAI Gym: A paradigm for computational investigations into psy- 67. Heinz A (2017): A New Understanding of Mental Disorders: Compu- chiatric illness. Biol Psychiatry Cogn Neurosci Neuroimaging tational Models for Dimensional Psychiatry. Cambridge, MA: MIT 3:809–818. Press.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:173–183 www.sobp.org/BPCNNI 183 Biological Psychiatry: CNNI Archival Report

Ultra-High-Resolution Imaging of Amygdala Subnuclei Structural Connectivity in Major Depressive Disorder

Stephanie S.G. Brown, John W. Rutland, Gaurav Verma, Rebecca E. Feldman, Molly Schneider, Bradley N. Delman, James M. Murrough, and Priti Balchandani

ABSTRACT BACKGROUND: Major depressive disorder (MDD) is an increasingly common and disabling illness. As the amygdala has been reported to have pathological involvement in mood disorders, we aimed to investigate for the first time potential changes to structural connectivity of individual amygdala subnuclei in MDD using ultra-high-field 7T diffusion magnetic resonance imaging. METHODS: Twenty-four patients with MDD (11 women) and 24 age-matched healthy control participants (7 women) underwent diffusion-weighted imaging with a 1.05-mm isotropic resolution at 7T. Amygdala nuclei regions of interest were obtained through automated segmentation of 0.69-mm resolution T1-weighted images and 0.35-mm resolution T2-weighted images. Probabilistic tractography was performed on all subjects, with random seeding at each amygdala nucleus. RESULTS: The right lateral, basal, central, and centrocortical amygdala nuclei exhibited significantly increased connection density to the rest of the brain, whereas the left medial nucleus demonstrated significantly lower connection density (false discovery rate p , .05). Increased connection density in the right lateral and basal nuclei was driven by the stria terminalis, and the significant difference in the right central nucleus was driven by the uncinate fasciculus. Decreased connection density at the left medial nucleus did not appear to be driven by any individual white matter tract. CONCLUSIONS: By exploiting ultra-high-resolution magnetic resonance imaging, structural hyperconnectivity was demonstrated involving the amygdaloid nuclei in the right hemisphere in MDD. To a lesser extent, impairment of subnuclei connectivity was shown in the left hemisphere. Keywords: 7T, Amygdala, Diffusion MRI, Major depressive disorder, Neuroimaging, Tractography, Ultra-high-field https://doi.org/10.1016/j.bpsc.2019.07.010

Major depressive disorder (MDD) is a debilitating condition anteromedial temporal lobe, passing through the basolateral with significant worldwide prevalence and substantial mortality and central amygdala nuclei toward the midline (8,9). (1–3). An emerging body of literature implicates the amygdala Ascending branches of the amygdalofugal pathway cross as a particular brain region of interest (ROI) in the pathology of through the nucleus accumbens and conclude in the area of MDD (4,5). The amygdala plays an important functional role in the septal nuclei. Descending amygdalofugal white matter fi- emotional processing, fear conditioning and extinction, moti- bers project toward nuclei of the hypothalamus, and medial vation, social salience, and affective state, aberrant regulation fibers reach the basal forebrain and olfactory areas. Similarly, of which forms the depressive phenotype (6). the stria terminalis also connects the amygdala to the hypo- Concordant to its functional role, the amygdala has exten- thalamus (7). Functionally, each efferent projection is consid- sive intrinsic and extrinsic structural connections within the ered to have a distinct purpose. The anterior commissure plays brain. Classically, the 3 main efferent white matter tracts from an important role in coupling the amygdalae and temporal lobe the amygdala are the anterior commissure, stria terminalis, and hemispheres and is involved in emotion, olfaction, instinctual amygdalofugal pathway (7). The anterior commissure crosses behavior, and memory consolidation (10,11). The amygdalo- the midline of the brain, connecting the anterior temporal lobes fugal tract is thought to exert downstream control over the bilaterally (7). The stria terminalis originates from the central hypothalamus and septal nuclei, influencing threat reactivity nucleus of the amygdala, projecting to a terminus dorsal to the (12). Moreover, reports of amnesia in patients with damage to anterior commissure via the lateral side of the fornix. The the amygdalofugal tract suggest an additional role in memory amygdalofugal pathway is thought to originate in the (13). The stria terminalis is theorized to regulate social and

184 ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI ISSN: 2451-9022 Biological Psychiatry: Amygdala Subnuclei Structural Connectivity in MDD CNNI

motivational conduct (14). Amino acid afferent neurotrans- Table 1. Participant Demographics and Summary Statistics mission from the amygdala in nonhuman primates has been MDD Group Control Group described to reach the frontal, insular, cingulate, and temporal Gender, Female, % 45 29 cortices. Additionally, hippocampal cornu ammonis 1 and the Age, Years, Median 42.5 (43.2); [24–63] 37.3 (33.5); entorhinal cortex give rise to significant amygdala input (15). (Mean); [Range] [22–55] Intrinsically, the lateral nucleus of the amygdala appears to Age, Years, First Quarter/Third 31/47.5 29/51.25 project strongly to the basal nucleus (15). Additionally, there is Quarter evidence for a direct connection from the visual pulvinar of the Ever Treated for a Psychiatric 87 4 thalamus to the amygdala, possibly enabling emotional Illness, % salience attribution to simplistic visual information relayed from Talk Therapy Only, % 25 0 the visual cortex to facilitate swift and unconscious selective Ever Hospitalized for Mental 40 attention switching (16). In fact, this subcortical pathway has Health Condition, % been shown to originate in the superior colliculus in macaques Received ECT, % 0 0 (17) and has been reported to continue efferently from the Age of Onset, Years, Median 15 (16); [7–47] – amygdala to visual cortices 2 and 4 (18). (Mean); [Range] Given the broad and extensive structural connectivity profile Age of Onset, Years, First 10.75/21.5 – and functional relevance to emotional pathology of the amygdala, Quarter/Third Quarter it is not surprising that it is a structure frequently associated with Recurrent MDD, % 79 0 MDD pathology. Seed-based resting-state functional magnetic Duration of Current Episode, 48 (69.6); [2–252] – resonance imaging (MRI) analyses have revealed significantly Months, Median (Mean); [Range] reduced amygdala connectivity with the cerebellum, occipital Duration of Current Episode, 5/156 – cortex, caudate, superior and middle temporal lobes, and insula in Months, First Quarter/Third patients with MDD. Additionally, increased amygdala–temporal Quarter pole functional connectivity was negatively correlated with Marijuana Smoked in Past Month, % 4 0 depression symptom severity and anxiety ratings (19).Inadoles- MADRS Rating of MDD Symptom 25.5 (14.25); – cents with depression, the amygdala has been reported to exhibit Severity, Median (Mean); [Range] [13–43] lower positive functional coupling between the amygdala and MADRS Rating of MDD Symptom 27.5/34.5 – hippocampus and the parahippocampus and brainstem and Severity, First Quarter/Third increased connectivity between the amygdala and precuneus. Quarter Where functional connectivity was significantly increased ECT, electroconvulsive therapy; MADRS, Montgomery–Åsberg compared with control subjects, blood oxygenation level– Depression Rating Scale; MDD, major depressive disorder. dependent response was inversely correlated with general depression, lassitude, and dysphoria scores (20). Longitudinal assessment of resilient individuals at high risk for MDD revealed significantly increased functional connectivity between the through the Mood and Anxiety Disorders Program at the Icahn amygdala and the orbitofrontal cortex, which correlated with School of Medicine at Mount Sinai. Also recruited were 24 age- positive life events (21). Connectivity between the prefrontal matched control subjects (mean [SD] age = 39.8 [12.3] years, 7 cortices and the limbic system has been cited as frequently altered women). All participants were English-speaking and between in MDD (22), and uncinate fasciculus volume, as a proxy for 18 and 65 years of age. Age was not significantly different frontotemporal structural connectivity known to involve the between groups (p = 1.0). Eligible patients with MDD had no amygdala, has been shown to correlate with both amygdala vol- psychotic features and were assessed by the Structured ume and trait anxiety (23). Clinical Interview for DSM-IV Axis I Disorders (29) or the Owing to past constraints of MRI spatial resolution, imaging Structured Clinical Interview for DSM-5 Research Version (30). time, and the availability of technology, studies considering the Patients were antidepressant-free for at least 4 weeks before amygdala in depression have often treated the region as a single, study participation and were currently experiencing a major unified structure in analyses. However, both animal and human depressive episode. No depressed participants had previously studies have shown that the amygdala comprises at least 13 undergone electroconvulsive therapy. Healthy control subjects anatomically separate and discrete subnuclei (24,25). Several had no current or lifetime psychiatric disorder as determined studies support the notion that these amygdala subregions are by the Structured Clinical Interview for DSM-IV Axis I Disorders functionally distinct during emotional processing (26–28).We (29) or the Structured Clinical Interview for DSM-5 Research therefore set out to use a recently developed automated seg- Version (30). Participants with a current diagnosis of mentation technique and ultra-high-resolution diffusion MRI to obsessive-compulsive disorder; alcohol or substance abuse in investigate the role of amygdala subnuclei connectivity in MDD at the previous year; or lifetime history of a psychotic illness, bi- a high level of granularity for the first time. polar disorder, or neurological disease were excluded. In the MDD group, 33% had comorbid social phobia, 16% had METHODS AND MATERIALS generalized anxiety, 4% had binge-eating disorder, 4% had panic disorder, and 4% had posttraumatic stress disorder. Participants Participants with MRI contraindications, unstable medical Twenty-four participants (mean [SD] age = 38.7 [9.8] years, 11 conditions, or positive urine toxicology on the day of the scan women) with a primary diagnosis of MDD were recruited were also excluded. The mean (SD) age of illness onset in the

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI 185 Biological Psychiatry: CNNI Amygdala Subnuclei Structural Connectivity in MDD FPO = web 4C Figure 1. Axial, coronal, and sagittal views of probabilistic tractography seeded at all amygdala nuclei combined. Reconstructed white matter tracts include the stria terminalis, the uncinate fasciculus, a subcortical visual pathway passing along the inferior fronto-occipital fasciculus, the amygdalofugal tract, and the anterior commissure. The color of streamlines is determined by their directionality.

MDD group was 17.5 (10.4) years, and the mean (SD) duration number of gradient directions was 64, with 5 b = 0 s/mm2. of depressive episode was 75.0 (84.0) months. All participants The 5 b = 0 acquisitions were interleaved during the acqui- gave fully informed written consent before investigation. This sition to correct for artifacts at time points 0.0 seconds, 115.2 protocol was approved by the local institutional review board. seconds, 223.2 seconds, 338.4 seconds, and 453.6 seconds. Participant demographics and summary statistics are given in The b value for the sequence was 1500 s/mm2. Two diffusion Table 1. MRI reverse-direction scans were acquired to correct gradient distortions. MRI Acquisition MRI data were acquired for all participants on a 7T whole-body Structural Data Processing scanner (Magnetom; Siemens Healthcare, Erlangen, Germany). T1-weighted images were preprocessed using the FreeSurfer A SC72CD gradient coil was used with a single coil transmit (http://freesurfer.net) version 6.0 recon-all pipeline, nonpara- and a 32-channel receive head coil (Nova Medical, Wilmington, metric nonuniform intensity correction, intensity normalization, MA). A T1-weighted magnetization-prepared 2 rapid gradient skull stripping and neck removal, automatic segmentation, and echoes sequence (31) was performed on each participant, with parcellation steps (32). Multispectral amygdala segmentation a 0.7 mm 3 0.7 mm 3 0.7 mm voxel resolution. Field of view was carried out in FreeSurfer development version 6.0 using was 225 3 183 mm, acquisition matrix was 320 3 260 mm, the T1-weighted and T2-weighted high-resolution images, orientation of scan was coronal, repetition time was 6000 ms, producing masks for the basal, lateral, accessory basal, cen- and echo time was 3.62 ms. Number of slices for a single slab tral, medial, and cortical nuclei (33). FreeSurfer automated was 240. A coronal-oblique T2-weighted turbo spin echo amygdala segmentation was developed using an ex vivo sequence was also obtained for all participants, with a 0.43 dataset comprising 10 autopsied human brain hemispheres mm 3 0.43 mm 3 2.0 mm voxel resolution. Number of ac- and 7T field strength MRI scanning with 0.1-mm isotropic quired slices was 66. Acquisition matrix was 512 3 408 mm, resolution. Amygdala nuclei were verified by a neuroanatomist, field of view was 222 3 177, repetition time was 9000 ms, and and the automated atlas was built using an algorithm based on echo time was 69 ms. Bayesian inference. Validation using publicly available datasets A diffusion-weighted imaging sequence was performed in showed that the amygdala automatic atlas significantly out- the same scanning session for all participants. Echo time was performed estimations of amygdala volume as a whole by 67.6 ms, and repetition time was 7200 ms. Field of view was FreeSurfer version 5.1 and performed with 84% accuracy in 210 3 210 mm, acquisition matrix was 200 3 200 mm, Alzheimer’s disease discrimination from age-matched control number of slices was 132, flip angle was 90, spatial reso- subjects (33). The nuclei were also grouped into a basolateral lution of the diffusion data was 1.05 mm isotropic, and complex (accessory basal, basal, and lateral nuclei) and a

186 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI Biological Psychiatry: Amygdala Subnuclei Structural Connectivity in MDD CNNI

centrocortical complex (central, cortical, and medial nuclei) by summating individual nucleus metrics. All outputs were manually inspected to ensure quality of pre-processing and accuracy of segmentation.

Diffusion Data Processing Diffusion data were denoised using MRtrix 2-shell phase- reversed processing (https://mrtrix.readthedocs.io/en/latest/ index.html) (34,35). FreeSurfer segmented and parcellated images were used for whole-brain masking in the image pro-

cessing (32).B1 ﬁeld inhomogeneity correction was performed for the diffusion images (36). Fiber orientation distributions were estimated from the diffusion data using spherical deconvolution (36), and diffusion tensor estimation was carried out using iteratively reweighted linear least squares estimator (37). Amygdala nuclei ROIs were coregistered to diffusion space

using nearest neighbor interpolation in SPM12, with the B0 image as the reference image, the T2-weighted image as the source image, and the amygdala nuclei ROIs as additional images. MRtrix software was used to carry out probabilistic tractography (38), performed using each amygdala nucleus mask as an individual seed region for 250,000 random seeds. Streamlines were thresholded using a ﬁber orientation distribution amplitude cutoff of 0.15 and at a maximum angle between successive streamline generation steps of 60. The spherical deconvolution (SIFT2) algorithm was applied to all tracts to eliminate streamlines that were unlikely to be biologically accurate and allow ground truth accuracy of streamline count (39,40). Whole-brain fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AD) maps were created, and using MRtrix tcksample, microstructural measurements of FA, MD, RD, and AD were extracted from every voxel along every streamline. Mean measurements per streamline were produced and averaged to output 1 mean FA, MD, RD, and AD value per amygdala nucleus ROI per participant. Microstructural order is used to refer to the degree of diffusivity and anisotropy with regard to the tissue structure, as factors such as axon diameter, packing density, and membrane permeability affect diffusion metrics without being necessarily a correlate of structural integrity (41). Streamline count was used as extracted, using the MRtrix tckstat com- mand, applying SIFT2 weightings to improve accuracy to ground truth white matter and to eliminate spurious streamlines. The resultant metric is referred to as connection density. A secondary analysis was carried out with the aim of elucidating which particular amygdalae-associated white matter tracts contributed to signiﬁcant differences in connection density in MDD at a subnuclei level. ROIs were manually drawn around the previously produced tracts of the stria terminalis, uncinate fasciculus, amygdalofugal tract, inferior FPO

= fronto-occipital fasciculus, and anterior commissure for each participant. Additional tractography was then performed using the amygdala nuclei that exhibited significant between-group web 4C differences in streamline density as seeding regions and the Figure 2. Significantly increased streamline density of tracts seeded at the right lateral nucleus, right basal nucleus, right central nucleus, and right cen- individual tract ROIs as inclusion criteria for streamlines. Again, trocortical complex in the major depressive disorder (MDD) group compared with 250,000 random seeds were used per amygdala nucleus, the control group. streamlines were thresholded using a fiber orientation distri-

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI 187 Biological Psychiatry: CNNI Amygdala Subnuclei Structural Connectivity in MDD

probabilistic tractography, previous literature is suggestive that these tracts may represent reciprocal connectivity with the visual areas (17,18). The uncinate fasciculus was also tracked. Our analyses did not consider discrete end point ROIs, and anatomical connections were not investigated in a quantitative manner. Qualitatively, however, virtual rendering of the amygdala subnuclei structural connectivity shows that 1) nuclei exhibit differing proﬁles of streamlines and 2) the basal and lateral nuclei exhibit shorter range connections than the medial, central, and cortical nuclei.

Tissue Microstructure Several differences in the MDD group compared with the control group indicated an increase in microstructural order. In

FPO the left hemisphere, there was evidence of decreased micro- = structural order in the MDD group. However, when nonparametric between-group testing was performed with an

web 4C additional adjustment for age, no significant differences in Figure 3. Significant decrease in streamline density in the major tissue microstructure in the MDD group compared with the depressive disorder (MDD) group compared with the control group in control group survived. These findings are presented in the streamlines emanating from the left medial nucleus. Supplement.

Connection Density bution amplitude cutoff of 0.15, and the maximum angle be- Streamline count was signiﬁcantly increased in the MDD group tween successive streamline generation steps was 60. All compared the control group in tractography seeded at the right outputs were visually quality checked to ensure that stream- lateral nucleus ( = .03), the right basal nucleus ( = .03), lines accurately tracked the correct white matter bundles. pFDR pFDR the right central nucleus ( = .02), and the right cen- Streamline counts of the amygdala nuclei individual tracts were pFDR trocortical complex ( = .03) (Figure 2). In the left hemi- extracted with applied SIFT2 weightings. pFDR sphere, there was a signiﬁcant decrease in streamline count

originating from the left medial nucleus (pFDR = .03) in the MDD Statistical Analysis group (Figure 3). All results remained significant with statistical Microstructural and streamline count metrics underwent adjustment for age. between-group analyses using nonparametric Mann-Whitney U testing of the null hypothesis. Testing was carried out both with Connection Density of Individual White Matter and without age adjustment. Nonparametric regressions using Bundles the R package rfit were carried out on the data for the MDD Our findings of significant differences in connection density in group only, to minimize bias caused by between-group differ- white matter tracts seeded at the right lateral, right basal, right ences (42). Nonparametric statistical approaches were used to central, and left medial nuclei in MDD were robust to both account for nonnormal data distributions. Tractography metrics statistical adjustment for age and correction for multiple were regressed against age of illness onset and duration of comparisons. For this reason, a secondary analysis was car- depressive episode. Statistical results were corrected for mul- ried out on these nuclei to determine which individual white tiple comparisons using false discovery rate (FDR). All statistical matter tracts contributed to depression-related changes in analyses were carried out in R version 3.3.3 (43), and a statistical structural connectivity. The stria terminalis, anterior commis- significance threshold of .05 was used. sure, inferior fronto-occipital fasciculus, uncinate fasciculus, and amygdalofugal pathway were isolated for each nucleus in RESULTS each participant (Figure 4). FDR-corrected nonparametric between-group testing with adjustment for age revealed that Qualitative Description of Anatomical Connections the stria terminalis seeded from both the right lateral and the of the Amygdala Nuclei basal nuclei had significantly increased connection density in

The present ultra-high-resolution probabilistic tractography the MDD group (pFDR = .01 and pFDR = .02, respectively) analysis revealed 5 distinct white matter tracts connecting with (Figure 5). The anterior commissure, inferior fronto-occipital the amygdala (Figure 1). In keeping with prior reports of fasciculus, uncinate, and amygdalofugal pathway showed no amygdala structural connectivity, we identified the 3 main significant differences in the MDD group when seeded from the efferent pathways of the amygdala in all participants: stria right lateral and right basal nuclei. Significantly increased terminalis, anterior commissure, and amygdalofugal tract. A general connection density in streamlines seeded at the right subcortical-visual pathway was also seen in the tractog- central nucleus in MDD was driven by the uncinate fasciculus

raphies, appearing to pass along the inferior fronto-occipital (pFDR = .01) (Figure 5). No other white matter tract showed fasciculus. Whereas directionality cannot be assessed using signiﬁcant between-group differences when seeded at the

188 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI Biological Psychiatry: Amygdala Subnuclei Structural Connectivity in MDD CNNI FPO = web 4C Figure 4. Isolated white matter tracts seeded from the amygdala nuclei. (A) Stria terminalis. (B) Anterior commissure. (C) Amygdalofugal tract. (D) Inferior fronto-occipital fasciculus. (E) Uncinate fasciculus. The color of streamlines is determined by their directionality.

right central nucleus. Lower general connection strength in anatomically, reports suggest that more than 13 distinct sub- streamlines seeded at the left medial nucleus did not appear to nuclei make up the amygdala (24). We present here a study be driven by any of the single white matter tracts isolated in that uses 7T field strength, ultra-high-resolution imaging, and a this study, with no significant differences identified in the MDD recently developed segmentation technique to carry out an group in the stria, uncinate, anterior commissure, amygdalo- analysis more closely resembling ground truth biology. We fugal pathway, or inferior fronto-occipital fasciculus. show that 3 of the right amygdala nuclei display an increased connection density, suggesting structural hyperconnectivity of Association With Clinical Variables the right amygdala. Furthermore, we show that significantly Nonparametric regression in the MDD group with covariation increased connection density in MDD is driven by the stria for age revealed that RD of streamlines was positively corre- terminalis in tracts seeded at the right lateral and basal nuclei lated with age of onset at the left basolateral nucleus (p = .01) and the uncinate fasciculus in tracts seeded at the right central and left accessory basal nucleus (p = .01). MD of streamlines nucleus. Left medial amygdala nucleus tracts showed changes seeded at the left basolateral complex were also significantly in count suggestive of hypoconnectivity. Importantly, we show positively associated with age of illness onset (p = .01). How- differential changes in the amygdala lateral, basal, central, and ever, none of these associations survived correction for mul- medial substructures that are specific to MDD, indicating that tiple comparisons. No association was identified between the amygdala nuclei do not react uniformly to MDD status. diffusion imaging metrics and duration of depressive episode. Furthermore, we provide here both a recommendation for the Similarly, no significant relationship was identified using consideration of the amygdala nuclei as distinct entities and nonparametric age-controlled regression between connection further characterization of structural brain changes in density metrics of the individual white matter tracts and depression. depression onset age or episode duration. Our findings of increasing microstructural order without age adjustment in streamlines seeded at the right basal nucleus, right lateral nucleus, right central nucleus, right DISCUSSION centrocortical complex, and left lateral nucleus, combined In prior analyses of limbic structures in MDD, the amygdala has with statistically robust increases in streamline count, commonly been treated as a unified structure at conventional suggest that these nuclei are hyperconnected with the rest MRI field strengths (44–47). However, histologically and of the brain in MDD. Previous studies of the structural

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI 189 Biological Psychiatry: CNNI Amygdala Subnuclei Structural Connectivity in MDD

connectivity of the amygdala in depression are concordant with these results (48,49). Decreased MD and increased FA withintheleftamygdalawereidentified in individuals with remitted depression, suggesting a greater cell density and number of fibers. Additional tractography analyses in the same cohort revealed an increase in structural connectivity between the left amygdala and the hippocampus, cerebellum, and brainstem (48). Increased correlation strength between gray matter volume of the amygdala and angular gyrus was also identified in patients with MDD compared with control subjects (49). In adolescents with depression, significantly lower FA and increased RD were reported in the bilateral uncinate fasciculus, a major white matter tract connecting the amygdala to frontal regions (50). Further- more, healthy adolescents with high familial risk for MDD revealed significantly reduced FA in a tract-based spatial statistics analysis in the uncinate and inferior fronto- occipital fasciculi, both of which structurally involve the amygdalae (51). A meta-analysis of 7 voxel-based diffusion imaging studies confirmed FA in the inferior fronto-occipital fasciculus in patients with MDD alterations compared with control subjects (52). In nonhuman primates, the amygdala nuclei are structurally connected to a larger number of regions in juveniles, and connectivity is pruned and refined over time (53). Develop- mental white matter changes are particularly protracted in the uncinate fasciculus (53), and perhaps for this reason, the uncinate is particularly vulnerable to psychiatric illness status. Functional MRI studies have shown replicable alterations in temporal lobe–prefrontal coupling in major depression (54), the principal structural correlate of which is the uncinate fasciculus (55). Uncinate microstructure has previously been associated with apathy in humans and is considered to be important in episodic memory, social ability, and emotional function (56). The central nucleus is a major site for efferent projections from the amygdala, and it is highly involved in the mediation of behavioral responses to stress (57). Our results show for the first time evidence of central subnucleus-specific involvement of the uncinate fasciculus in MDD. The stria terminalis is another major amygdala efferent, projecting dorsally from multiple subnuclei and branching to the hypothalamic and septal nuclei (58). Given its terminus at the hypothalamus, it is unsurprising that the stria is involved in motivation and reactivity to stress, both of which are foci of dysregulation in MDD (7). The present methodology allows us to implicate connection density of the stria terminalis streamlines seeded specifically at the right basal and lateral amygdala nuclei in depression, which is an unprecedented level of detail in humans to date. Interestingly, the basal, lateral, and central nuclei are the 3 amygdala subregions that have been implicated in significant developmental pruning of their connections

FPO (53). It is possible that the maturation and plasticity required at = these nuclei during development make them liable to de- viations that associate with depression status.

web 4C In agreement with a portion of previously published results Figure 5. Significantly increased connection density of the stria terminalis in MDD (19,46), we also report a finding of significantly in streamlines seeded at the right lateral and right basal nuclei and signifi- decreased connection density in depressed patients, localized cantly increased connection density of the uncinate fasciculus seeded from the right central nucleus. MDD, major depressive disorder. in streamlines seeded at the medial nucleus. This finding did

190 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI Biological Psychiatry: Amygdala Subnuclei Structural Connectivity in MDD CNNI

not appear to be driven by any individual white matter tract but Department of Radiology, Icahn School of Medicine at Mount Sinai, and was an effect observed in medial nucleus streamlines generally Siemens Healthcare (to PB). to the whole brain. Left lateral nucleus MD, left basolateral PB (principal investigator in this study) is a named inventor on patents relating to MRI and radiofrequency pulse design that have been complex RD, and left lateral nucleus RD along emanating licensed to GE Healthcare, Siemens AG, and Philips International and streamlines were also significantly increased when analysis receives royalty payments relating to these patents. In the past 5 years, was carried out without age adjustment, suggesting a possible JMM has provided consultation services to Boehreinger Ingelheim, loss in microstructural order. Taken together, the present re- Sage Therapeutics, FSV7, LLC, Novartis, Allergan, Fortress Biotech, sults illustrate an important feature concerning amygdala Janssen Research and Development, MedAvante-ProPhase, and Global connectivity in MDD: the structural connectivity of the amyg- Medical Education (GME) and has received research support from Avanir Pharmaceuticals, Inc. The other authors report no biomedical dala appears to be differentially enhanced or weakened based financial interests or potential conflicts of interest. on the particular nuclei and hemisphere. Interestingly, we also reveal a significant positive association with 3 microstructural measures and age at illness onset, although the results did not survive correction for multiple comparisons. The range of ARTICLE INFORMATION onset ages in the sample may therefore be a potential con- From the Translational and Molecular Imaging Institute (SSGB, JWR, GV, founding factor in the dataset, showing that our finding of REF, PB), Icahn School of Medicine at Mount Sinai; Depression and Anxiety significantly decreased structural connectivity in MDD is sen- Disorders Centre for Discovery and Treatment (MS, JMM), Department of sitive to the age at which individuals develop MDD and as such Psychiatry, Icahn School of Medicine at Mount Sinai; Department of Radi- should be interpreted with caution. ology (BND), Icahn School of Medicine at Mount Sinai; and Department of Neuroscience (JMM), Icahn School of Medicine at Mount Sinai, New York, Our study is a promising demonstration of how ultra-high- New York. field MRI may enable differential interrogation of connectivity JMM and PB contributed equally to this work. among amygdala subnuclei in MDD; however, there are several Address correspondence to Stephanie S.G. Brown, Ph.D., Translational limitations to the study that should be noted. First, duration of and Molecular Imaging Institute, 1470 Madison Avenue, New York, NY illness, participant age, and age at onset of the first episode of 10029; E-mail: [email protected]. MDD were varied within the sample, which may be considered Received Jul 2, 2019; accepted Jul 30, 2019. Supplementary material cited in this article is available online at https:// confounding factors. Second, although a sizable sample for a 7T doi.org/10.1016/j.bpsc.2019.07.010. investigation, the sample size used here was relatively small, owing in part to more cautious MRI screening protocols at 7T than at conventional field strengths. A significant limitation of the methodology is also that probabilistic tractography cannot REFERENCES differentiate between fibers emanating from each nucleus and 1. Conwell Y, Duberstein PR, Cox C, Herrmann JH, Forbes NT, Caine ED fibers passing through the nuclei. Given that the amygdala (1996): Relationships of age and Axis I diagnoses in victims of subnuclei are extensively interconnected in a manner well completed suicide: A psychological autopsy study. Am J Psychiatry 153:1001–1008. defined by the nonhuman primate literature (55,59), efferent, 2. Cuijpers P, Smit F (2002): Excess mortality in depression: A meta- afferent, or origination status of streamlines cannot be accu- analysis of community studies. J Affect Disord 72:227–236. rately identified in the present study. Additionally, longitudinal 3. Kessler RC, Birnbaum HG, Shahly V, Bromet E, Hwang I, study of the participants was not carried out, and therefore the McLaughlin KA, et al. (2010): Age differences in the prevalence and significant differences in structural connectivity in MDD could co-morbidity of DSM-IV major depressive episodes: Results from either preexist onset of depression and represent a possible the WHO World Mental Health Survey Initiative. Depress Anxiety 27:351–364. vulnerability, or be a consequence of the illness. Nevertheless, 4. Tang Y, Wang F, Xie G, Liu J, Li L, Su L, et al. (2007): Reduced the presented findings are a useful pilot demonstration of the ventral anterior cingulate and amygdala volumes in medication- advantages of high-resolution imaging in MDD and segmenta- naive females with major depressive disorder: A voxel-based tion of heterogeneous brain regions. morphometric magnetic resonance imaging study. Psychiatry Overall, we report significant changes in white matter Res 156:83–86. connection density in patients with MDD compared with con- 5. Bora E, Fornito A, Pantelis C, Yucel M (2012): Gray matter abnormalities in major depressive disorder: A meta-analysis of voxel based trol subjects that imply both increased and decreased struc- morphometry studies. J Affect Disord 138:9–18. tural connection strength. Increases in connection density in 6. Lindquist KA, Wager TD, Kober H, Bliss-Moreau E, Barrett LF (2012): the right hemisphere were driven specifically by the uncinate The brain basis of emotion: A meta-analytic review. Behav Brain Sci fasciculus and stria terminalis. Importantly, we show results 35:121–143. that differ between amygdala subnuclei, demonstrating the 7. Kamali A, Sair HI, Blitz AM, Riascos RF, Mirbagheri S, Keser Z, et al. importance and clinical significance of the separate consider- (2016): Revealing the ventral amygdalofugal pathway of the human limbic system using high spatial resolution diffusion tensor tractog- ation of amygdala substructures in MDD. raphy. Brain Struct Funct 221:3561–3569. 8. Nolte J (2002): The Human Brain. An Introduction to Its Functional Anatomy, 5th ed. Totowa, NJ: Humana Press. 9. Hutchins T, Herrod HC, Quigley E, Anerson J, Salzman K: Dissecting the ACKNOWLEDGMENTS AND DISCLOSURES white matter tracts: Interactive diffusion tensor imaging teaching atlas. This work was supported by the National Institutes of Health (Grant Nos. University of Utah Department of Neuroradiology. Available at: http:// R01 MH109544 [to PB] and R01 CA202911 [to PB]), Brain and Behavior www.asnr2.org/neurographics/7/1/26/White%20Matter%20Tract% Research Foundation National Alliance for Research on Schizophrenia and 20Anatomy/DTI%20tutorial%201.html. Accessed January 27, 2019. Depression Young Investigator Grant (to PB), Icahn School of Medicine 10. Bamiou DE, Sisodiya S, Musiek FE, Luxon LM (2007): The role of the Capital Campaign, Translational and Molecular Imaging Institute and interhemispheric pathway in hearing. Brain Res Rev 56:170–182.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI 191 Biological Psychiatry: CNNI Amygdala Subnuclei Structural Connectivity in MDD

11. Allen LS, Gorski RA (1991): Sexual dimorphism of the anterior 32. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, Haselgrove C, et al. commissure and massa intermedia of the human brain. J Comp Neurol (2002): Whole brain segmentation: Automated labeling of neuroana- 312:97–104. tomical structures in the human brain. Neuron 33:341–355. 12. Usunoff KG, Schmitt O, Itzev DE, Haas SJ, Lazarov NE, Rolfs A, et al. 33. Saygin ZM, Kliemann D, Iglesias JE, van der Kouwe AJW, Boyd E, (2009): Efferent projections of the anterior and posterodorsal regions Reuter M, et al. (2017): High-resolution magnetic resonance imaging of the medial nucleus of the amygdala in the mouse. Cells Tissues reveals nuclei of the human amygdala: Manual segmentation to Organs 190:256–285. automatic atlas. Neuroimage 155:370–382. 13. Tanaka Y, Miyazawa Y, Akaoka F, Yamada T (1997): Amnesia following 34. Veraart J, Novikov DS, Christiaens D, Ades-Aron B, Sijbers J, damage to the mammillary bodies. Neurology 48:160–165. Fieremans E (2016): Denoising of diffusion MRI using random matrix 14. Wood RI, Swann JM (2005): The bed nucleus of the stria terminalis in theory. Neuroimage 142:394–406. the Syrian hamster: subnuclei and connections of the posterior divi- 35. Veraart J, Fieremans E, Novikov DS (2016): Diffusion MRI noise mapping sion. Neuroscience 135:155–179. using random matrix theory. Magn Reson Med 76:1582–1593. 15. Amaral DG, Insausti R (1992): Retrograde transport of D-[3H]-aspar- 36. Tustison NJ, Avants BB, Cook PA, Zheng Y, Egan A, Yushkevich PA, tate injected into the monkey amygdaloid complex. Exp Brain Res et al. (2010): N4ITK: Improved N3 bias correction. IEEE Trans Med 88:375–388. Imaging 29:1310–1320. 16. Abivardi A, Bach DR (2017): Deconstructing white matter connectivity 37. Veraart J, Sijbers J, Sunaert S, Leemans A, Jeurissen B (2013): of human amygdala nuclei with thalamus and cortex subdivisions Weighted linear least squares estimation of diffusion MRI parameters: in vivo. Hum Brain Mapp 38:3927–3940. Strengths, limitations, and pitfalls. Neuroimage 81:335–346. 17. Rafal RD, Koller K, Bultitude JH, Mullins P, Ward R, Mitchell AS, et al. 38. Smith RE, Tournier JD, Calamante F, Connelly A (2012): Anatomically- (2015): Connectivity between the superior colliculus and the amygdala constrained tractography: Improved diffusion MRI streamlines trac- in humans and macaque monkeys: Virtual dissection with probabilistic tography through effective use of anatomical information. Neuroimage DTI tractography. J Neurophysiol 114:1947–1962. 62:1924–1938. 18. Catani M, Jones DK, Donato R, Ffytche DH (2003): Occipito-temporal 39. Dell’Acqua F, Tournier JD (2018): Modelling white matter with spherical connections in the human brain. Brain 126:2093–2107. deconvolution: How and why? NMR Biomed e3945. 19. Ramasubbu R, Konduru N, Cortese F, Bray S, Gaxiola-Valdez I, 40. Smith RE, Tournier JD, Calamante F, Connelly A (2015): SIFT2: Goodyear B (2014): Reduced intrinsic connectivity of amygdala in Enabling dense quantitative assessment of brain white matter con- adults with major depressive disorder. Front Psychiatry 5:17. nectivity using streamlines tractography. Neuroimage 119:338–351. 20. Cullen KR, Westlund MK, Klimes-Dougan B, Mueller BA, Houri A, 41. Jones DK, Knosche TR, Turner R (2013): White matter integrity, fiber Eberly LE, et al. (2014): Abnormal amygdala resting-state func- count, and other fallacies: The do’s and don’ts of diffusion MRI. tional connectivity in adolescent depression. JAMA Psychiatry Neuroimage 73:239–254. 71:1138–1147. 42. Kloke J, McKean JW (2012): Rfit: Rank-based estimation for linear 21. Fischer AS, Camacho MC, Ho TC, Whitfield-Gabrieli S, Gotlib IH models. The R Journal 4:57–64. (2018): Neural markers of resilience in adolescent females at fa- 43. R Core Team (2017): A Language and Environment for Statistical milial risk for major depressive disorder. JAMA Psychiatry Computing. R Foundation for Statistical Computing, Vienna, 75:493–502. Austria (2017). Available at: https://www.R-project.org/. Accessed 22. Seeberg I, Kjaerstad HL, Miskowiak KW (2018): Neural and behav- January 3, 2018. ioral predictors of treatment efficacy on mood symptoms and 44. Kang SG, Na KS, Choi JW, Kim JH, Son YD, Lee YJ (2017): Resting- cognition in mood disorders: A systematic review. Front Psychiatry state functional connectivity of the amygdala in suicide attempters 9:337. with major depressive disorder. Prog Neuropsychopharmacol Biol 23. Baur V, Hanggi J, Jancke L (2012): Volumetric associations be- Psychiatry 77:222–227. tween uncinate fasciculus, amygdala, and trait anxiety. BMC 45. Murphy ER, Barch DM, Pagliaccio D, Luby JL, Belden AC (2016): Neurosci 13:4. Functional connectivity of the amygdala and subgenual cingulate 24. Ding SL, Royall JJ, Sunkin SM, Ng L, Facer BA, Lesnar P, et al. (2016): during cognitive reappraisal of emotions in children with MDD Comprehensive cellular-resolution atlas of the adult human brain historyisassociatedwithrumination. Dev Cogn Neurosci 18:89– [published correction appears in J Comp Neurol 2017; 525:407]. 100. J Comp Neurol 524:3127–3481. 46. Ye X, Feng T, Tammineni P, Chang Q, Jeong YY, Margolis DJ, et al. 25. Janak PH, Tye KM (2015): From circuits to behaviour in the amygdala. (2017): Regulation of synaptic amyloid-beta generation through Nature 517:284–292. BACE1 retrograde transport in a mouse model of Alzheimer’s disease. 26. Kim H, Somerville LH, Johnstone T, Polis S, Alexander AL, Shin LM, J Neurosci 37:2639–2655. et al. (2004): Contextual modulation of amygdala responsivity to sur- 47. Connolly CG, Ho TC, Blom EH, LeWinn KZ, Sacchet MD, prised faces. J Cogn Neurosci 16:1730–1745. Tymofiyeva O, et al. (2017): Resting-state functional connectivity of the 27. Hrybouski S, Aghamohammadi-Sereshki A, Madan CR, Shafer AT, amygdala and longitudinal changes in depression severity in adoles- Baron CA, Seres P, et al. (2016): Amygdala subnuclei response cent depression. J Affect Disord 207:86–94. and connectivity during emotional processing. Neuroimage 48. Arnold JF, Zwiers MP, Fitzgerald DA, van Eijndhoven P, Becker ES, 133:98–110. Rinck M, et al. (2012): Fronto-limbic microstructure and structural 28. Balderston NL, Schultz DH, Hopkins L, Helmstetter FJ (2015): Func- connectivity in remission from major depression. Psychiatry Res tionally distinct amygdala subregions identified using DTI and high- 204:40–48. resolution fMRI. Soc Cogn Affect Neurosci 10:1615–1622. 49. Wu H, Sun H, Wang C, Yu L, Li Y, Peng H, et al. (2017): Abnormalities 29. First MD, Spitzer RL, Williams JBW, Gibbon M (1995): Structured in the structural covariance of emotion regulation networks in major Clinical Interview for DSM-IV Axis I Disorders–Patient Edition. New depressive disorder. J Psychiatr Res 84:237–242. York: New York Psychiatric Institute. 50. LeWinn KZ, Connolly CG, Wu J, Drahos M, Hoeft F, Ho TC, et al. 30. First MB, Williams JBW, Karg RS, Spitzer RL (2015): Structured Clin- (2014): White matter correlates of adolescent depression: structural ical Interview for DSM-5 Research Version (SCID-5-RV). Arlington, VA: evidence for frontolimbic disconnectivity. J Am Acad Child Adolesc American Psychiatric Association. Psychiatry 53:899–909, 909.e891–e897. 31. Marques JP, Gruetter R (2013): New developments and applications of 51. Huang H, Fan X, Williamson DE, Rao U (2011): White matter changes in the MP2RAGE sequence—focusing the contrast and high spatial healthy adolescents at familial risk for unipolar depression: A diffusion resolution R1 mapping. PLoS One 8:e69294. tensor imaging study. Neuropsychopharmacology 36:684–691.

192 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI Biological Psychiatry: Amygdala Subnuclei Structural Connectivity in MDD CNNI

52. Murphy ML, Frodl T (2011): Meta-analysis of diffusion tensor im- 56. Hollocks MJ, Lawrence AJ, Brookes RL, Barrick TR, Morris RG, aging studies shows altered fractional anisotropy occurring in Husain M, et al. (2015): Differential relationships between apathy and distinct brain areas in association with depression. Biol Mood depression with white matter microstructural changes and functional Anxiety Disord 1:3. outcomes. Brain 138:3803–3815. 53. Saygin ZM, Osher DE, Koldewyn K, Martin RE, Finn A, Saxe R, et al. 57. Kalin NH, Shelton SE, Davidson RJ (2004): The role of the central (2015): Structural connectivity of the developing human amygdala. nucleus of the amygdala in mediating fear and anxiety in the primate. PLoS One 10:e0125170. J Neurosci 24:5506–5515. 54. Iwabuchi SJ, Krishnadas R, Li C, Auer DP, Radua J, Palaniyappan L 58. Oler JA, Tromp DP, Fox AS, Kovner R, Davidson RJ, Alexander AL, (2015): Localized connectivity in depression: A meta-analysis of et al. (2017): Connectivity between the central nucleus of the amygdala resting state functional imaging studies. Neurosci Biobehav Rev and the bed nucleus of the stria terminalis in the non-human primate: 51:77–86. Neuronal tract tracing and developmental neuroimaging studies. Brain 55. Webster MJ, Ungerleider LG, Bachevalier J (1991): Connections of Struct Funct 222:21–39. inferior temporal areas TE and TEO with medial temporal-lobe struc- 59. LeDoux J (1998): Fear and the brain: Where have we been, and where tures in infant and adult monkeys. J Neurosci 11:1095–1116. are we going? Biol Psychiatry 44:1229–1238.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:184–193 www.sobp.org/BPCNNI 193 Biological Psychiatry: CNNI Archival Report

Elevated Amygdala Activity in Young Adults With Familial Risk for Depression: A Potential Marker of Low Resilience

Tracy Barbour, Avram J. Holmes, Amy H. Farabaugh, Stephanie N. DeCross, Garth Coombs, Emily A. Boeke, Rick P.F. Wolthusen, Maren Nyer, Paola Pedrelli, Maurizio Fava, and Daphne J. Holt

ABSTRACT BACKGROUND: Amygdala overactivity has been frequently observed in patients with depression, as well as in nondepressed relatives of patients with depression. A remaining unanswered question is whether elevated amygdala activity in those with familial risk for depression is related to the presence of subthreshold symptoms or to a trait-level vulnerability for illness. METHODS: To examine this question, functional magnetic resonance imaging data were collected in nondepressed young adults with (family history [FH1]) (n=27) or without (FH2)(n=45) a first-degree relative with a history of depression while they viewed images of “looming” or withdrawing stimuli (faces and cars) that varied in salience by virtue of their apparent proximity to the subject. Activation of the amygdala and 2 other regions known to exhibit responses to looming stimuli, the dorsal intraparietal sulcus (DIPS) and ventral premotor cortex (PMv), were measured, as well as levels of resilience, anxiety, and psychotic and depressive symptoms. RESULTS: Compared with the FH2 group, the FH1 group exhibited significantly greater responses of the amygdala, but not the dorsal intraparietal sulcus or ventral premotor cortex, to looming face stimuli. Moreover, amygdala responses in the FH1 group were negatively correlated with levels of resilience and unrelated to levels of subthreshold symptoms of psychopathology. CONCLUSIONS: These findings indicate that elevated amygdala activity in nondepressed young adults with a familial history of depression is more closely linked to poor resilience than to current symptom state. Keywords: Amygdala, Depression, Familial risk, fMRI, Resilience, Youth https://doi.org/10.1016/j.bpsc.2019.10.010

As the leading cause of disability worldwide (1), depression is a markers of risk and protective factors, we can more accurately major worldwide health concern. With the increasing prevalence identify individuals with the highest risk of developing depres- of depression (2,3), its economic and public health toll will likely sion who will benefit most from preventive interventions. escalate, as rising treatment needs strain already limited mental One general category of protective factors is referred to as health resources. Further, many people receiving treatment for emotional resilience (7,8). Resilience, as defined by the depression respond inadequately to available treatment options American Psychological Association, is a capacity to adapt and develop persistent or recurrent depression (4). Therefore, to well when facing adversity or significant sources of stress, or address the societal burden of depression, public health ap- the constellation of skills or traits that enable one to “bounce proaches such as early identification of at-risk individuals and back” from difficult experiences (9). It can also be defined prevention must be established. However, providing timely operationally as a positive outcome following stressors (10). preventative measures or early treatment strategies relies on the The building blocks of resilience are likely heterogeneous in identification of the most susceptible individuals (5). One terms of both the cognitive/affective processes and neurobi- vulnerable group is those individuals with a family history of ological mechanisms involved (7,11) and thus have been little depression (family history positive [FH1]), as they have a studied using neuroimaging methods to date. However, given threefold greater likelihood of developing depression in their that self-report measures of resilience, such as the Connor- lifetime than those without such a family history (family history Davidson Resilience Scale (CD-RISC) (12), can predict negative [FH2]) (6). However, further risk stratification within this positive outcomes following stressors (13,14), even cross- group is necessary to develop cost-effective, prevention- sectional investigations of the neurobiology of resilience may focused public health initiatives. Thus, by examining both begin to shed light on the interplay between biological and

194 ª 2020 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI ISSN: 2451-9022 Biological Psychiatry: Psychological Resilience Linked With Amygdala Response CNNI

environmentally determined aspects of resilience and psy- youths than in FH2 youths, as others have reported. Second, chopathology risk. we tested whether the FH1 group showed greater responses To investigate the neurobiology of depression risk and proof the dorsal attentional network than the FH2 group, tective factors such as resilience, the networks of the brain most consistent with prior findings in patients with depression. commonly implicated in the biology of depression must be Third, we tested whether the predicted higher responsivity of examined. Much research on depression has focused on the these regions in the familial high-risk group was related to trait- amygdala (15). In addition, some current models of depression like levels of emotional resilience or to subthreshold depressive have linked attentional deficits and abnormal cognitive control of or anxiety symptoms. emotion to core symptoms of the illness (16). Evidence for attentional deficits in patients with depression includes reports of abnormal attentional shifting (17), slowed reaction times (18), and METHODS AND MATERIALS mood-congruent attentional biases (19). Supporting this model, Overall Study Design both resting-state (20–22) and task-based (17,18,21)functional magnetic resonance imaging (fMRI) studies of depression have This study was conducted in a subset of subjects who participated in a larger study of mental health in college students (39), revealed involvement of circuitry supporting emotional processing and attention/cognitive control, such as the amygdala- in which on-campus screenings were conducted at several Boston-area universities. During these screenings, self-report centered and dorsal attention/frontoparietal networks, respectively. The functions of the amygdala-centered/limbic and questionnaires measuring a range of symptoms were administered (39). One of the goals of the study was to further attentional networks are closely linked; for example, the amygdala rapidly processes salient stimuli and then interacts with characterize young people with subthreshold symptoms of psychopathology, with mildly to moderately elevated scores on attention and cognitive control networks to influence behavioral responses (23–25). The dorsal attention network is involved in a measure of depression (the Beck Depression Inventory [BDI]) (40) and/or a measure of psychotic experiences (the Peters et al. orienting attention to spatial cues and in cognitive control, and these processes can be influenced by input from the amygdala Delusions Inventory [PDI]) (41). Students with elevated scores (BDI total score .5or.0 on BDI item 9 [measuring suicidal and other emotion-processing brain areas (26). Regions of the dorsal attention network, such as the dorsal intraparietal sulcus ideation], or PDI total score .7) and a small number of students with no depressive symptoms (total BDI scores of 0) were invited (DIPS) and ventral premotor cortex (PMv), support goal-directed attention and visuospatial processing (26), which are processes to participate in 1) a brief clinical assessment (administered by a Ph.D.-or M.D.-level clinician) in which the mood module of that are affected in depression (17). Neuroimaging studies of depression have identified abnor- the Structured Clinical Interview for DSM-IV Axis I Disorders was administered, 2) a baseline neuroimaging session, and 3) malities in both networks, repeatedly detecting abnormalities in amygdala responses to emotional and neutral faces in patients longitudinal follow-up assessments conducted at 6-month intervals (self-report scales completed online; analyses including with depression (27–29), as well as altered activation of frontoparietal regions of the dorsal attention network during visuo- the longitudinal data will be reported separately). The neuroimaging session included 1 T1 anatomical scan and 4 blood spatial processing (29–31). One meta-analysis of functional oxygen level–dependent (BOLD) scans (see Supplemental connectivity studies of depressed individuals showed that regions of the dorsal attention network are among the brain re- Methods for scan parameters), during which subjects viewed dynamic face and car stimuli and performed a low-level atten- gions most frequently affected in clinical depression (20). Similar findings within the amygdala and attentional regions have been tional task (24). Subjects with neurological disorders or serious medical illnesses, substance abuse or dependence, or contra- reported for clinically remitted patients and in never-depressed FH1 individuals (32–36), indicating that such abnormal activa- dictions to MRI scanning were excluded from participating in the scanning session. The presence or absence of a family tion patterns may not represent state-dependent effects of depression. Furthermore, a recent meta-analysis (37) showed history of depression in first-degree family members was assessed at baseline using 2 self-report questionnaires, which that cognitive (including attentional) deficits and deficits in visual scanning (38) are present in FH1 individuals when compared included questions about the psychiatric history of subjects’ family members. Resilience levels were measured using the with control subjects. However, it remains unclear whether such changes in the function of the amygdala and attentional net- CD-RISC (see Supplemental Methods for further details). works in unaffected relatives of individuals with depression reflect effects of subthreshold levels of symptoms, a marker of Participants resilience (in the face of genetic vulnerability), or a heightened A total of 131 subjects were scanned. For the current study, susceptibility for future illness. the data of 2 groups of subjects within this cohort were To investigate this question further, in the current study we examined: those with a first-degree relative with depression measured amygdala and frontoparietal cortical responses to (family history postive [FH1]) (n=29) and those without a first- dynamic social (faces) and nonsocial (cars) stimuli in young degree relative with depression (family history negative [FH2]) adults with or without a first-degree relative with a history of (n=47). Data of 4 subjects (2 FH1 and 2 FH2) were excluded depression. Specifically, we used a paradigm that activates following quality control procedures. See Supplemental areas of the dorsal attentional network (dorsal parietal cortex Methods for additional information about the participants and PMv) as well as the amygdala. who were screened and the exclusion criteria. Therefore, first, we determined whether, in our cohort, the All 72 subjects included in the analyses (27 FH1 and 45 amygdala showed greater responses in nondepressed FH1 FH2) did not meet DSM-IV criteria for depression at the time of

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI 195 Biological Psychiatry: CNNI Psychological Resilience Linked With Amygdala Response

scanning, based on the Structured Clinical Interview for DSM- by the task used here, showing significantly greater responses IV Axis I Disorders. FH2 subjects reported that their first- to approaching (i.e., appearing to “loom” toward the subject), degree relatives were without a history of any mental illness. compared with withdrawing, face stimuli in prior studies FH1 subjects reported the presence of a family history of (24,46). Hereafter, we refer to the approaching versus with- unipolar depression in at least one first-degree relative. drawing response or contrast as the looming response. The study protocol, including informed consent procedures, BOLD responses were extracted from each ROI for each was approved by the Partners HealthCare and Harvard Uni- condition. The average BOLD response to each face and car versity Institutional Review Boards, and written informed condition was compared with the BOLD response to the cross- consent was obtained from all subjects. hair baseline condition. The two contrasts of interest were 1) approach versus withdrawal (the looming response) for each Functional MRI Paradigm stimulus type (faces, cars) and 2) all faces versus all cars. These contrasts were chosen based on prior work demonstrating As described in Holt (24), during each functional scan, et al. that the DIPS and PMv regions respond preferentially to subjects viewed 2 types of stimuli: images of human faces looming stimuli, particularly faces (24), and that the amygdala (with neutral expressions) and images of cars. Each stimulus responds preferentially to face stimuli, compared with appeared to be moving toward or away from the subject at the nonface objects such as car stimuli (47,48). A repeated- speed of walking (112 cm/s) (Supplemental Figure S1). This measures analysis of variance, using a 3 region (DIPS, PMv, paradigm involves visuospatial processing of approaching amygdala) 3 2 hemisphere (left, right) 3 2 condition (approach, stimuli, which robustly activates frontoparietal regions of the withdrawal) 3 2 stimuli (faces, cars) 3 2 group (FH2,FH1) dorsal attentional network (23,24). Images of faces with neutral factorial design, was performed, to test our predictions that expressions were presented because some previous work has compared with the FH2 group, the FH1 group would show 1) identified significant differences 1) between control subjects greater amygdala responses to faces (both to looming faces and patients with depression (42) and 2) between individuals and to all faces vs. all cars) and 2) greater DIPS and PMv with and without familial risk for mood disorders (43)in response to looming faces. amygdala responses only to neutral (not to emotional) faces. Each of 4 conditions (Face Approach, Face Withdrawal, Car Approach, Car Withdrawal) was presented for 16 seconds. In Correlational Analyses. ROIs that showed significant each run, subjects viewed 2 blocks of each of the 4 stimuli between-group effects were then used to conduct correla- (8 blocks total), randomly presented. During each of the 4 tional and regression analyses to test whether changes in brain conditions, subjects performed a simple dot-detection task; to function associated with having a family history of depression distribute spatial attention evenly across the approach and were correlated with resilience levels [measured using the CD- withdrawal conditions, subjects were asked to press a button, RISC (12)] and/or with levels of subsyndromal depressive, while maintaining fixation, whenever a dot appeared at a psychotic-like, and anxiety symptoms [measured using the random location on the screen. The percentage of responses BDI, PDI (41), and Spielberger State-Trait Anxiety Inventory was calculated for each condition, in each subject. A run was (49), respectively]. excluded if the subject responded to ,40% of the dots during Subsequently, a sec- that run, as in Holt et al. (24). Secondary, Voxelwise Analysis. ondary, voxelwise analysis was conducted for the purpose of further localizing the findings of the anatomical ROI analysis MRI Data Analysis (see Results). This analysis was conducted using a Monte Data were analyzed using the FreeSurfer analysis stream Carlo simulation (10,000 iterations) whole-brain correction, (http://surfer.nmr.mgh.harvard.edu), with standard pre- using 2 cluster-forming height thresholds of p=.001 and processing methods and quality assessment procedures (see p = .05. In addition, percent signal change data extracted from Supplemental Methods for details). Images were spatially these maps (limited to regions showing between-group smoothed with a 6-mm Gaussian kernel (full width at half differences in the anatomical ROI analysis) were used to maximum) and a 3-dimensional spatial filter. confirm the findings of the regression and correlational analyses conducted using the anatomical ROIs. Anatomical Region-of-Interest Analysis. In the primary, hypothesis-testing analysis of this study, 3 a priori anatomical regions of interest (ROIs) (the DIPS, PMv, and amygdala) RESULTS (Supplemental Figure S2) were defined in each subject’sT1 anatomical scan, using an automated parcellation method that Participant Characteristics and Performance on the relies on well-known anatomical landmarks (44). We focused Dot-Detection Task on the amygdala because of the extensive prior literature There were no significant differences between the FH2 and demonstrating amygdala abnormalities in patients with FH1 groups in age, gender, ethnicity, childhood adversity depression (15,45) and their first-degree relatives (34,35). The levels, or baseline symptom or resilience levels (Table 1). DIPS and PMv were also examined because these two regions Analyses of subjects’ behavioral performance during scan- are central nodes of the dorsal attention network (which has ning showed no main effects of group in response rates during

shown abnormalities in patients with depression and in ﬁrst- the dot-detection task (F1,70 = 2.2, p=.14). A signiﬁcant degree relatives of patients with depression, as described group 3 condition effect (F1,70 = 4.3, p=.04) was driven by above) and because these two regions are robustly engaged greater accuracy of the FH1 group (vs. the FH2 group) during

196 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI Biological Psychiatry: Psychological Resilience Linked With Amygdala Response CNNI

Table 1. Demographic Information and Symptom Levels of the Participants FH2 (n = 45)a FH1 (n = 27)a Mean Min Max SD Mean Min Max SD Age, Years 19.49 18 22 1.31 19.59 18 24 1.60 Depressive Symptomsb 6.71 0 20 5.50 8.93 0 17 3.90 Psychotic Experiencesc 6.11 0 18 3.93 5.07 0 12 3.22 Resilienced 73.10 38 100 16.44 73.89 57 95 9.70 State Anxietye 28.58 0 56 16.72 32.85 0 62 18.27 Trait Anxietyf 31.31 0 59 17.70 34.41 0 57 17.35 Childhood Adversityg 30.11 0 48 8.93 31.96 0 77 17.47 Gender, Male/Female, n 12/33 9/18 Race, % Caucasian 51 44 African American 27 37 Asian 15 7 Other 8 12 Ethnicity, Hispanic, % 15 0 The FH2 and FH1 groups showed no signiﬁcant differences in age, education, gender, ethnicity, levels of symptoms of depression, psychotic experiences, suicidality, resilience, anxiety, and levels of childhood adversity, based on independent-sample t tests. There were no signiﬁcant 2 differences between groups in gender (Pearson’s c 1 = 0.3630, p=.547). FH1, family history of depression; FH2, no family history of depression. aResilience levels negatively correlated with depressive symptoms within the FH2 group (r=2.61, p , .01) but not within the FH1 group (r=2.10, p=.61). bMeasured using the Beck Depression Inventory. cMeasured using the Peters et al. Delusions Inventory. dMeasured using the Connor-Davidson Resilience Scale. eMeasured using the Spielberger State Anxiety Inventory. fMeasured using Spielberger Trait Anxiety Inventory. gMeasured using the Childhood Trauma Questionnaire.

the Car Withdrawal condition (t26 = 2.1, p=.04). Given that response to face compared with car stimuli in the left amygdala this result would not have any influence on our a priori hy- in both groups (FH2 group [t44 = 3.50, p=.001], FH1 group potheses, it was not explored further. [t26 = 2.50, p=.02]) and in the right amygdala in the FH2 group, with a similar trend in the FH1 group (right: FH2 group [ = 3.39, .002], FH1 group [ = 1.93, .06]). Functional MRI t44 p= t26 p= Hypothesis Testing: Anatomical ROI Analysis. As expected, the repeated-measures analysis of variance revealed Secondary, Voxelwise Analysis: Faces Approach Versus Withdrawal. To localize the between-group differ- a main effect of condition (F1,70 = 12.88, p=.001) due to the significant activation to looming (Approach . Withdrawal) ence observed in the amygdala in the anatomical ROI stimuli. Also, there was a significant 3-way interaction among analysis, a secondary voxelwise analysis of the Faces Approach versus Faces Withdrawal contrast was conducted. No signifi- group, condition, and stimulus type (F1,1 = 4.96, p = .029), which was due to a greater response of the FH1 group cant clusters were observed at the cluster-forming threshold compared with the FH2 group to looming faces in the bilateral of p = .001; however, at p=.05, a cluster within the left 23 amygdala was present, owing to a significantly greater amygdala (left [t70 = 22.89, p=5 3 10 ], right [t70 = 22.02, response of the FH1 group compared with the FH2 group p=.04]) (Figure 1 and Supplemental Table S1), with a trend toward a similar difference between the 2 groups in the bilat- (Montreal Neurological Institute coordinates [x, y, z] of the peak voxel: 226, 25, 223 [z = 4.5, p=6.6 3 1026]). A similar (slightly eral PMv (left [t70 = 21.92, p=.06], right [t70 = 21.83, p=.07]) weaker) pattern of findings was observed for the right amygdala but not in the DIPS (left [t70 = 20.67, p=.50], right [t70 = 20.76, (30, 28, 217 [ = 3.9, 1 3 1024]) (Figure 2). Follow-up p=.45]). There was no group 3 stimulus or group 3 z p= stimulus 3 region interaction, indicating that there was no 1-sample t tests confirmed that the FH1 group, but not the between-group difference in the responses of the amygdala or FH2 group, showed significant responses of the left amygdala 25 of the other 2 regions to faces compared with cars. Also, for all to looming faces (226, 25, 223 [z = 4.4, p=1 3 10 ]). There 3 ROIs, there were no differences between the 2 groups in was no significant activation at this threshold in the right responses to looming cars. amygdala in either group. Follow-up repeated-measures analysis of variances conducted within each group revealed a significant interaction Regression Analysis between region and stimulus for both the FH2 and FH1 A regression analysis using a multivariate general linear model groups. This interaction was due to a significantly greater was then conducted to test whether changes in brain function

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI 197 Biological Psychiatry: CNNI Psychological Resilience Linked With Amygdala Response

Figure 1. Results of the region-of-interest analysis. Bar plots of mean percent signal change (approaching vs. withdrawing [looming] faces) of the amygdala, ventral premotor cortex (PMv), and dorsal intraparietal sulcus (DIPS) regions of interest for the groups with (FH1) and without (FH2) a family history of depression are displayed. The FH1 group showed a signiﬁcantly greater response to looming faces in the left and right amygdala compared with the FH2 23 group (left [t70 = 22.89, p=5 3 10 ], right web 4C/FPO [t70 = 22.02, p=.04]). *p , .05.

associated with having a family history of depression were maps that were generated using 3 thresholds (p=.05, p=.01, associated with resilience or symptom levels. The regression and p=.001). These analyses revealed significant negative analysis was performed with amygdala BOLD response, correlations between resilience levels and amygdala re- extracted from the left and right amygdala anatomical ROIs, as sponses, regardless of the ROI used (Table 2). the dependent variable, group (FH1/FH2) as a between- subjects factor, and resilience/symptom scores as cova- DISCUSSION riates. Significant interactions between group and resilience or symptom levels on amygdala response were followed up with Summary of Main Findings Pearson’s correlations (2-tailed). There was a significant Consistent with prior work, this study demonstrated that interaction between group and resilience levels for the re- nondepressed young adults with a first-degree relative who sponses of the left, but not of the right, amygdala (left amyg- had a history of depression display greater amygdala dala [ = 8.94, .004], right amygdala [ = 1.69, F1,55 p= F1,55 p= responsivity than nondepressed young adults without such a .20]). There were no significant interactions between group and family history (33–35). Further analyses revealed that this levels of depressive (BDI), psychotic-like (PDI), or anxiety pattern of responses was not related to the presence of (Spielberger State-Trait Anxiety Inventory) symptoms for either subthreshold symptoms of psychopathology but rather to the left or right amygdala responses. Follow-up correlations low resilience levels. However, a parietofrontal cortical showed that levels of resilience were negatively correlated with network involved in attention did not show the same pattern responses of the left amygdala to looming faces in the FH1 of significantly elevated responsivity as seen in the amyg- group ( 2.41, .03) (Figure 3). This correlation remained r= p= dala in the relatives, suggesting that attentional systems significant after controlling for levels of subthreshold symp- may be disrupted in depressive illness but not in an at-risk toms of depression and anxiety. In contrast, the FH2 group cohort. did not show any significant correlations between amygdala responses to looming faces and resilience levels (left: .09; p= Amygdala Dysregulation as a Marker of Low right: p=.71) (Figure 3). There were no correlations between resilience levels and amygdala responses to faces Resilience and Risk for Illness overall (compared with cars) in either the FH1 or FH2 group Potentially related to these results, one recent study found that (all p . .3). lower resting-state connectivity of the amygdala with the Because having a history of childhood adversity is linked to orbitofrontal cortex in adolescents at risk for depression pre- an increased risk for developing depression (50), we repeated dicted the later development of depression (52); another prior this analysis controlling for levels of childhood adversity, study found that higher basal levels of amygdala activity was measured using the Childhood Trauma Questionnaire (51). The associated with low, self-reported resilience in older adults correlation between resilience levels and amygdala responses (53). Taken together with the current results, these findings to looming faces in the FH1 group remained significant after suggest that specific patterns of disrupted amygdala function (i.e., increases in basal or stimulus-elicited activity and reduced controlling for childhood adversity (r=2.42, p=.03). Also, there were no correlations between amygdala responses to functional coupling with prefrontal cortical regions) may looming faces and baseline levels of subsyndromal depres- be linked to poor resilience and vulnerability to depression. sion, anxiety, psychotic experiences, or levels of childhood However, exactly how poor resilience and depression risk are manifested and linked biologically remains to be understood. adversity in the FH1 group (all ps . .20) (Supplemental Table S2). Last, to confirm the above findings by repeating the ana- Resilience lyses using a slightly different approach, we conducted these While the definition of resilience and its components continues correlational analyses using 3 functionally defined amygdala to be debated, it is likely the result of a combination of a range ROIs, which were derived from between-group comparison of biological and environmental factors influencing one’s

198 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI Biological Psychiatry: Psychological Resilience Linked With Amygdala Response CNNI

depression, as amygdala responses to faces overall (in comparison with responses to cars) was not elevated in this group, nor were amygdala responses to faces (compared with cars) negatively correlated with resilience levels in either group. This may be due to the robustness of face-elicited responses of the amygdala (as compared with amygdala responses to nonface stimuli) in most individuals; there may not be sufﬁcient variance in these responses to reveal a dimensional relationship with resilience levels, in contrast to the variance observed in amygdala responses to salient [e.g., approaching and on a collision course with the body, or emotional (24,57)] versus nonsalient or safe-appearing (e.g., withdrawing or emotionally neutral) face images.

Task Performance There were no differences between the 2 groups in performance of the dot-detection task that was performed while the approaching and withdrawing face stimuli were presented, suggesting that differences between the 2 groups in levels of attention to the stimuli were not present and thus cannot account for the between-group differences in amygdala responses. The FH1 group did perform signiﬁcantly better on

FPO the task than the FH2 group during the Withdrawing = Cars condition, perhaps reflecting a generally greater level of vigilance that was only detectable during a rather nonsalient web 4C condition. Figure 2. Results of the voxelwise analyses. These whole brain–corrected maps ( , .05) display the amygdala responses to looming faces in the p (A) Dorsal Attention Network group without a family history of depression (FH2) and (B) the group with a family history of depression (FH1) and (C) the comparison between the We found no significant between-group differences in the responses of the two groups. Only the FH1 group (not the FH2 group) responses of the DIPS, and only a trend toward an elevated showed a significant response to looming faces, in the left amygdala. A, response of the PMv, in the FH1 compared with the FH2 Approach condition; R, right; W, Withdrawal condition. individuals. Thus, these results do not provide support for the hypothesis that primary abnormalities in the dorsal attentional system are responsible for attention-related abnor- response to adversity (54). In previous studies, resilience as malities in individuals at risk for depression. One possible measured by the self-report scale used here, the CD-RISC, has interpretation of this pattern of findings is that attentional been shown to moderate the relationship between adverse abnormalities in depression and depression risk are primarily events and severity of subsequent symptoms of anxiety and due to consequences of alterations in amygdala function (or depression (13,14), as well as predict treatment response in in an amygdala-centered network of regions), rather than due patients with depression (55) and posttraumatic stress disorder to fundamental changes in the neural circuitry that controls (56), irrespective of symptom severity. Together, these findings attention. However, it is also important to note that the task suggest that this self-report scale measures an aspect of used here did not engage all circuitry involved in attentional resilience that predicts both risk for the development of psy- control; for example, the ventral “bottom-up” attention chopathology and the capacity for rapid recovery from it. network was not examined. Thus, additional studies will be In our sample, average resilience levels did not differ be- needed to fully evaluate the neural systems mediating the tween the FH1 and FH2 groups. However, only the FH1 various forms of attention in individuals with risk factors for subjects with lower resilience levels showed elevated amyg- developing depression. dala responses to looming faces. Thus, in future studies, amygdala responses to salient stimuli and/or self-reported Elevated Responses of the Amygdala as a resilience levels could be tested as potential components of Transdiagnostic Risk Factor a tool for identifying individuals, among those with genetic risk for depression (conferring a threefold increase in risk), who Current models of depression (8,58) propose that a vulnera- have a particularly elevated risk for depression due to other, bility to depression may remain dormant for years, without perhaps nonfamilial, factors. leading to a depressive episode. Thus, a combination of factors, including environmental events and hormonal changes, as well as protective biological and psychosocial factors, may Specific Effect on Amygdala Responses to Looming determine whether clinically impairing depressive symptoms Faces emerge in vulnerable individuals. Notably, recent evidence Interestingly, in the current study, we did not find evidence for suggests that this model of depression may in fact correspond overall amygdala dysfunction in those with familial risk for to a more general model of psychopathology; some neural

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI 199 Biological Psychiatry: CNNI Psychological Resilience Linked With Amygdala Response

Figure 3. Correlations between amygdala responses and resilience levels. For these correlational analyses, responses of the left and right amygdala for each subject were extracted using anatomically defined regions of interest generated by FreeSurfer from the T1 scan of each subject. The scatter plot in (A) illustrates the significant correlation between resilience level (as measured by the Connor-Davidson Resilience Scale [CD- RISC]) and the response of the left amygdala to looming faces in the group with a family history of web 4C/FPO depression (FH1)(r=2.41, p=.03, n=27), whereas there was no significant correlation between responses of the left amygdala and resilience levels in the group without a family history of depression (FH2)(r=.26, p=.09). Moreover, the correlations between responses of the left amygdala and resilience levels in the FH1 and FH2 groups were significantly different from each other (z = 22.48, p=.02). (B) Responses of the right amygdala to looming faces were not significantly correlated with resilience levels in the FH1 (r=2.28, p=.16) or FH2 (r=.05, p=.71) groups. “vulnerability markers,” such as heightened amygdala re- been linked to resilience, such as physiologic and epigenetic sponses to salient social stimuli, may not be specificto markers, and test whether those and the putative marker of depression risk (59–63). For example, two studies have shown resilience described here, low or “normal” amygdala reactivity that the response of the amygdala to threatening stimuli prior to salient social stimuli, can be used to prospectively predict to a traumatic event predicts the later development of post- outcomes (i.e., resilient or not) following stressful events. traumatic stress disorder symptoms (59,64), and a third Follow-up work can also determine whether specific resilience- study showed that amygdala responses to negatively valenced boosting interventions can influence amygdala function, psy- faces predicted the later development of internalizing symp- chological responses to environmental stressors, and toms (65). The current data are generally in line with this and psychiatric outcomes. Future studies can also expand on the other evidence for amygdala dysfunction as a transdiagnostic experimental design of the functional MRI paradigm used here, marker of psychopathology risk. Specifically, taken together including increasing the representation of different ethnicities with prior work, these results suggest that reactivity of the in the social (face) stimuli, as well as the types of social and amygdala in response to salient social stimuli may be a marker nonsocial stimuli. of vulnerability for the context-dependent emergence of psy- Lastly, it is also important to emphasize that it is not possible to chopathology related to a reduced capacity to tolerate stress distinguish here, given the design of the current study, between (i.e., low resilience). an effect of greater loading for genetic variants associated with depression (6) and the influence of growing up with a depressed relative and the associated environmental stressors that may Limitations and Future Directions accompany that experience. Follow-up studies can investigate Resilience is a complex construct with no agreed-upon oper- whether aversive childhood experiences associated with the ational definition. Thus, one limitation of this study is that stress of familial depression are closely linked to elevated resilience was measured using only one self-report question- amygdala responses in at-risk individuals, or whether there is naire. Future studies can include additional measures that have instead (or in addition) a strongly genetic basis for this phenotype.

Table 2. Correlations Between Amygdala Responses to Looming Faces and Levels of Resilience Region of Interest Left Amygdala Right Amygdala Left 1 Right Amygdala Correlation p Correlation p Correlation p Region of Interest Definition Coefficient (r) Value Coefficient (r) Value Coefficient (r) Value Anatomical (FreeSurfer Segmentation) 2.41 .03 2.28 .16 2.36 .06 Functional (Voxel Threshold .05) 2.52 .01 2.42 .03 2.50 .01 Functional (Voxel Threshold .01) 2.52 .01 2.36 .06 2.46 .01 Functional (Voxel Threshold .001) 2.49 .01 2.35 .07 2.45 .02 Amygdala responses for each subject were extracted using regions of interest that were either defined anatomically (in the primary analysis) or defined using the between-group comparison (family history of depression vs. no family history of depression) voxelwise maps generated using 3 different thresholds (p=.05, p = .01, and p = .001). Significant correlations with resilience levels (measured using the Connor-Davidson Resilience Scale) in the family history–positive group [n=27]) are observed regardless of the type of amygdala region of interest used to extract the functional magnetic resonance imaging data.

200 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI Biological Psychiatry: Psychological Resilience Linked With Amygdala Response CNNI

Conclusions REFERENCES These findings show that a specific pattern of amygdala func- 1. World Health Organization (2017): Depression and Other Common tion (higher responses to salient social information) in those Mental Disorders: Global Health Estimates. Geneva, Switzerland: with familial risk for depression is linked to lower resilience World Health Organization. 2. Weinberger AH, Gbedemah M, Martinez AM, Nash D, Galea S, within this group. In the future these results could potentially Goodwin RD (2018): Trends in depression prevalence in the USA from provide a component of a quantitative tool that could be used 2005 to 2015: Widening disparities in vulnerable groups. Psychol Med to stratify individuals with a family history of depression into 48:1308–1315. high and low risk groups. Such a tool is a prerequisite for 3. Akushevich I, Kravchenko J, Yashkin AP, Yashin AI (2018): Time trends the development of objective screening strategies, individual- in the prevalence of cancer and non-cancer diseases among older U. ized risk profiles and targeted preventive interventions. S. adults: Medicare-based analysis. Exp Gerontol 110:267–276. 4. Ghaemi SN (2008): Why antidepressants are not antidepressants: STEP-BD, STAR*D, and the return of neurotic depression. Bipolar ACKNOWLEDGMENTS AND DISCLOSURES Disord 10:957–968. 5. Garber J, Clarke GN, Weersing VR, Beardslee WR, Brent DA, This work was supported by a Dupont Warren fellowship (to TB), a Stuart T. Gladstone TRG, et al. (2009): Prevention of depression in at-risk ad- Hauser Research Training Program in Biological and Social Psychiatry T32 olescents: A randomized controlled trial. JAMA 301:2215–2224. award (to TB), internal support from the Massachusetts General Hospital 6. Sullivan PF, Neale MC, Kendler KS (2000): Genetic epidemiology of Department of Psychiatry for neuroimaging research in psychiatry, and major depression: Review and meta-analysis. Am J Psychiatry Grant No. RO1MH109562 (Principal investigator, DJH). 157:1552–1562. We are very grateful to Tammy Moran, Marisa Hollinshead, and other 7. Charney DS (2004): Psychobiological mechanisms of resilience and staff of the Center for Brain Science of Harvard University for their invaluable vulnerability: Implications for successful adaptation to extreme stress. assistance with data acquisition. Am J Psychiatry 161:195–216. MF has received research support from following: Abbott Laboratories; 8. Southwick SM, Vythilingam M, Charney DS (2005): The psychobiology Acadia Pharmaceuticals; Alkermes, Inc.; American Cyanamid; Aspect Medi- of depression and resilience to stress: Implications for prevention and cal Systems; AstraZeneca; Avanir Pharmaceuticals; AXSOME Therapeutics; treatment. Annu Rev Clin Psychol 1:255–291. Biohaven; BioResearch; BrainCells Inc.; Bristol-Myers Squibb; CeNeRx Bio- 9. American Psychological Association. The road to resilience. Available at: Pharma; Cephalon; Cerecor; Clarus Funds; Clintara, LLC; Covance; Covidien; http://www.apa.org/helpcenter/road-resilience.aspx. Accessed June 29, Eli Lilly and Company; EnVivo Pharmaceuticals, Inc.; Euthymics Bioscience, 2018. Inc.; Forest Pharmaceuticals, Inc.; FORUM Pharmaceuticals; Ganeden 10. Kalisch R, Baker DG, Basten U, Boks MP, Bonanno GA, Biotech, Inc.; GlaxoSmithKline; Harvard Clinical Research Institute; Hoffman- Brummelman E, et al. (2017): The resilience framework as a strategy to LaRoche; Icon Clinical Research; i3 Innovus/Ingenix; Janssen R&D, LLC; Jed combat stress-related disorders. Nat Hum Behav 1:784–790. Foundation; Johnson & Johnson Pharmaceutical Research & Development; 11. Hermans EJ, Fernández G (2015): Heterogeneity of cognitive- Lichtwer Pharma GmbH; Lorex Pharmaceuticals; Lundbeck Inc.; Marinus neurobiological determinants of resilience. Behav Brain Sci 38:e103. Pharmaceuticals; MedAvante; Methylation Sciences Inc; National Alliance for 12. Connor KM, Davidson JRT (2003): Development of a new resilience Research on Schizophrenia and Depression; National Center for Comple- scale: The Connor-Davidson Resilience Scale (CD-RISC). Depress mentary and Alternative Medicine; National Coordinating Center for Inte- Anxiety 18:76–82. grated Medicine; National Institute of Drug Abuse; National Institute of Mental 13. Peng L, Zhang J, Li M, Li P, Zhang Y, Zuo X, et al. (2012): Negative life Health; Neuralstem, Inc.; NeuroRx; Novartis AG; Organon Pharmaceuticals; events and mental health of Chinese medical students: The effect of Otsuka Pharmaceutical Development, Inc.; PamLab, LLC.; Pfizer Inc.; resilience, personality and social support. Psychiatry Res 196:138– Pharmacia-Upjohn; Pharmaceutical Research Associates., Inc.; Pharmavite 141. LLC; PharmoRx Therapeutics; Photothera; Reckitt Benckiser; Roche Phar- 14. Campbell-Sills L, Cohan SL, Stein MB (2006): Relationship of resil- maceuticals; RCT Logic, LLC (formerly Clinical Trials Solutions, LLC); Sanofi- ience to personality, coping, and psychiatric symptoms in young Aventis US LLC; Shire; Solvay Pharmaceuticals, Inc.; Stanley Medical adults. Behav Res Ther 44:585–599. Research Institute; Synthelabo; Taisho Pharmaceuticals; Takeda Pharma- 15. Whalen PJ, Shin LM, Somerville LH, McLean AA, Kim H (2002): ceuticals; Tal Medical; VistaGen; and Wyeth-Ayerst Laboratories. DJH has Functional neuroimaging studies of the amygdala in depression. received research support from Forum Pharmaceuticals, Inc., and Janssen Semin Clin Neuropsychiatry 7:234–242. Scientific Affairs. All other authors report no biomedical financial interests or 16. Disner SG, Beevers CG, Haigh EAP, Beck AT (2011): Neural mech- potential conflicts of interest. anisms of the cognitive model of depression. Nat Rev Neurosci 12:467–477. 17. Schock L, Schwenzer M, Sturm W, Mathiak K (2011): Alertness and ARTICLE INFORMATION visuospatial attention in clinical depression. BMC Psychiatry 11:78. From the Department of Psychiatry (TB, AHF, SND, RPFW, MN, PP, MF, 18. Pardo JV, Pardo PJ, Humes SW, I Posner M (2006): Neurocognitive DJH), Massachusetts General Hospital; and Department of Psychiatry (TB, dysfunction in antidepressant-free, non-elderly patients with unipolar AHF, MN, PP, MF, DJH), Harvard Medical School, Boston; Department of depression: Alerting and covert orienting of visuospatial attention. Psychology (GC), Harvard University, Cambridge; and the Athinoula A. J Affect Disord 92:71–78. Martinos Center for Biomedical Imaging (DJH), Charlestown, Massachu- 19. de Fockert JW, Cooper A (2013): Higher levels of depression are setts; Department of Psychology (AJH), Yale University, New Haven, Con- associated with reduced global bias in visual processing. Cogn Emot necticut; Department of Psychology (EAB), New York University, New York, 28:541–549. New York; and the Translational Developmental Neuroscience Section 20. Kaiser RH, Andrews-Hanna JR, Wager TD, Pizzagalli DA (2015): Large- (RPFW), Department of Child and Adolescent Psychiatry, Faculty of Medi- scale network dysfunction in major depressive disorder: Meta-analysis cine Carl Gustav Carus, Technische Universität Dresden, Dresden, of resting-state functional connectivity. JAMA Psychiatry 72:603–611. Germany. 21. Sambataro F, Visintin E, Doerig N, Brakowski J, Holtforth MG, Address correspondence to Tracy Barbour, M.D., Massachusetts Gen- Seifritz E, Spinelli S (2017): Altered dynamics of brain connectivity in eral Hospital, Psychiatry, 149 13th Street, Room 2628, Boston, MA 02129; major depressive disorder at-rest and during task performance. Psy- E-mail: [email protected]. chiatry Res Neuroimaging 259:1–9. Received Mar 25, 2019; revised and accepted Oct 25, 2019. 22. Geng X, Xu J, Liu B, Shi Y (2018): Multivariate classification of major Supplementary material cited in this article is available online at http:// depressive disorder using the effective connectivity and functional doi.org/10.1016/j.bpsc.2019.10.010. connectivity. Front Neurosci 12:38.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI 201 Biological Psychiatry: CNNI Psychological Resilience Linked With Amygdala Response

23. Cléry J, Hamed SB (2018): Frontier of self and impact prediction. Front neuroanatomical structures in the human brain. Neuron 33:341– Psychol 9:1073. 355. 24. Holt DJ, Cassidy BS, Yue X, Rauch SL, Boeke EA, Nasr S, et al. (2014): 45. Abler B, Erk S, Herwig U, Walter H (2007): Anticipation of aversive Neural correlates of personal space intrusion. J Neurosci 34:4123–4134. stimuli activates extended amygdala in unipolar depression. 25. Graziano MSA, Cooke DF (2006): Parieto-frontal interactions, personal J Psychiatr Res 41:511–522. space, and defensive behavior. Neuropsychologia 44:2621–2635. 46. Holt DJ, Boeke EA, Coombs G, DeCross SN, Cassidy BS, 26. Corbetta M, Shulman GL (2002): Control of goal-directed and Stufflebeam S, et al. (2015): Abnormalities in personal space and stimulus-driven attention in the brain. Nat Rev Neurosci 3:201–215. parietal-frontal function in schizophrenia. NeuroImage Clin 9:233–243. 27. Jaworska N, Yang X-R, Knott V, MacQueen G (2015): A review of fMRI 47. Rolls ET (1984): Neurons in the cortex of the temporal lobe and in the studies during visual emotive processing in major depressive disorder. amygdala of the monkey with responses selective for faces. Hum World J Biol Psychiatry 16:448–471. Neurobiol 3:209–222. 28. Singh MK, Gotlib IH (2014): The neuroscience of depression: Impli- 48. Leonard CM, Rolls ET, Wilson FA, Baylis GC (1985): Neurons in the cations for assessment and intervention. Behav Res Ther 62:60–73. amygdala of the monkey with responses selective for faces. Behav 29. Desseilles M, Schwartz S, Dang-Vu TT, Sterpenich V, Ansseau M, Brain Res 15:159–176. Maquet P, Phillips C (2011): Depression alters “top-down” visual 49. Spielberger CD, Gorsuch RL (1970): STAI Manual for the State-trait attention: A dynamic causal modeling comparison between depressed Anxiety Inventory (“Self-evaluation Questionnaire”). Washington, DC: and healthy subjects. NeuroImage 54:1662–1668. Consulting Psychologists Press. 30. Beevers CG, Clasen P, Stice E, Schnyer D (2010): Depression symp- 50. Heim C, Nemeroff CB (2001): The role of childhood trauma in the toms and cognitive control of emotion cues: A functional magnetic neurobiology of mood and anxiety disorders: Preclinical and clinical resonance imaging study. Neuroscience 167:97–103. studies. Biol Psychiatry 49:1023–1039. 31. Rive MM, Koeter MWJ, Veltman DJ, Schene AH, Ruhé HG (2016): 51. Pennebaker JW, Susman JR (1988): Disclosure of traumas and psy- Visuospatial planning in unmedicated major depressive disorder and chosomatic processes. Soc Sci Med 26:327–332. bipolar disorder: Distinct and common neural correlates. Psychol Med 52. Fischer AS, Camacho MC, Ho TC, Whitfield-Gabrieli S, Gotlib IH 46:2313–2328. (2018): Neural markers of resilience in adolescent females at familial 32. Miskowiak KW, Glerup L, Vestbo C, Harmer CJ, Reinecke A, risk for major depressive disorder. JAMA Psychiatry 75:493–502. Macoveanu J, et al. (2015): Different neural and cognitive response to 53. Leaver AM, Yang H, Siddarth P, Vlasova RM, Krause B, St Cyr N, et al. emotional faces in healthy monozygotic twins at risk of depression. (2018): Resilience and amygdala function in older healthy and Psychol Med 45:1447–1458. depressed adults. J Affect Disord 237:27–34. 33. Chai XJ, Hirshfeld-Becker D, Biederman J, Uchida M, Doehrmann O, 54. Southwick SM, Bonanno GA, Masten AS, Panter-Brick C, Yehuda R Leonard JA, et al. (2015): Functional and structural brain correlates of (2014): Resilience definitions, theory, and challenges: Interdisciplinary risk for major depression in children with familial depression. Neuro- perspectives. Eur J Psychotraumatology 5:25338. Image Clin 8:398–407. 55. Min J-A, Lee N-B, Lee C-U, Lee C, Chae J-H (2012): Low trait anxiety, 34. Monk CS, Klein RG, Telzer EH, Schroth EA, Mannuzza S, Moulton JL, high resilience, and their interaction as possible predictors for treat- et al. (2008): Amygdala and nucleus accumbens activation to ment response in patients with depression. J Affect Disord 137:61–69. emotional facial expressions in children and adolescents at risk for 56. Davidson J, Stein DJ, Rothbaum BO, Pedersen R, Szumski A, major depression. Am J Psychiatry 165:90–98. Baldwin DS (2012): Resilience as a predictor of treatment response in 35. Swartz JR, Williamson DE, Hariri AR (2015): Developmental change in patients with posttraumatic stress disorder treated with venlafaxine amygdala reactivity during adolescence: Effects of family history of extended release or placebo. J Psychopharmacol 26:778–783. depression and stressful life events. Am J Psychiatry 172:276–283. 57. Delvecchio G, Fossati P, Boyer P, Brambilla P, Falkai P, Gruber O, 36. Samara Z, Evers EAT, Peeters F, Uylings HBM, Rajkowska G, et al. (2012): Common and distinct neural correlates of emotional Ramaekers JG, Stiers P (2018): Orbital and medial prefrontal cortex processing in bipolar disorder and major depressive disorder: A voxel- functional connectivity of major depression vulnerability and disease. based meta-analysis of functional magnetic resonance imaging Biol Psychiatry Cogn Neurosci Neuroimaging 3:348–357. studies. Eur Neuropsychopharmacol J 22:100–113. 37. MacKenzie LE, Uher R, Pavlova B (2019): Cognitive performance in 58. Krishnan V, Nestler EJ (2008): The molecular neurobiology of first-degree relatives of individuals with vs without major depressive depression. Nature 455:894–902. disorder: A meta-analysis. JAMA Psychiatry 76:297–305. 59. McLaughlin KA, Busso DS, Duys A, Green JG, Alves S, Way M, 38. Singh MK, Leslie SM, Bhattacharjee K, Gross M, Weisman EF, Sheridan MA (2014): Amygdala response to negative stimuli predicts Soudi LM, et al. (2018): Vulnerabilities in sequencing and task PTSD symptom onset following a terrorist attack. Depress Anxiety switching in healthy youth offspring of parents with mood disorders. 31:834–842. J Clin Exp Neuropsychol 40:606–618. 60. Hamilton JP (2015): Amygdala reactivity as mental health risk 39. Farabaugh A, Bitran S, Nyer M, Holt DJ, Pedrelli P, Shyu I, et al. (2012): endophenotype: A tale of many trajectories. Am J Psychiatry Depression and suicidal ideation in college students. Psychopathol- 172:214–215. ogy 45:228–234. 61. Pinkham AE, Loughead J, Ruparel K, Overton E, Gur RE, Gur RC 40. Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J (1961): An in- (2011): Abnormal modulation of amygdala activity in schizophrenia in ventory for measuring depression. Arch Gen Psychiatry 4:561–571. response to direct- and averted-gaze threat-related facial expressions. 41. Peters ER, Joseph SA, Garety PA (1999): Measurement of delusional Am J Psychiatry 168:293–301. ideation in the normal population: Introducing the PDI (Peters et al. 62. Holt DJ, Coombs G, Zeidan MA, Goff DC, Milad MR (2012): Failure of Delusions Inventory). Schizophr Bull 25:553–576. neural responses to safety cues in schizophrenia. Arch Gen Psychiatry 42. Oliveira L, Ladouceur CD, Phillips ML, Brammer M, Mourao-Miranda J 69:893–903. (2013): What does brain response to neutral faces tell us about major 63. Carlisi CO, Robinson OJ (2018): The role of prefrontal-subcortical depression? Evidence from machine learning and fMRI. PLoS One 8: circuitry in negative bias in anxiety: Translational, developmental and e60121. treatment perspectives. Brain Neurosci Adv 2:2398212818774223. 43. Mourão-Miranda J, Oliveira L, Ladouceur CD, Marquand A, 64. Admon R, Lubin G, Stern O, Rosenberg K, Sela L, Ben-Ami H, Brammer M, Birmaher B, et al. (2012): Pattern recognition and func- Hendler T (2009): Human vulnerability to stress depends on amyg- tional neuroimaging help to discriminate healthy adolescents at risk for dala’s predisposition and hippocampal plasticity. Proc Natl Acad Sci mood disorders from low risk adolescents. PLoS One 7:e29482. U S A 106:14120–14125. 44. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, Haselgrove C, 65. Swartz JR, Knodt AR, Radtke SR, Hariri AR (2015): A neural biomarker of et al. (2002): Whole brain segmentation: Automated labeling of psychological vulnerability to future life stress. Neuron 85:505–511.

202 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:194–202 www.sobp.org/BPCNNI Biological Psychiatry: Archival Report CNNI

Opponent Effects of Hyperarousal and Re-experiencing on Affective Habituation in Posttraumatic Stress Disorder

Katherine L. McCurry, B. Christopher Frueh, Pearl H. Chiu, and Brooks King-Casas

ABSTRACT BACKGROUND: Aberrant emotion processing is a hallmark of posttraumatic stress disorder (PTSD), with neurobiological models suggesting both heightened neural reactivity and diminished habituation to aversive stimuli. However, empirical work suggests that these response patterns may be speciﬁc to subsets of those with PTSD. This study investigates the unique contributions of PTSD symptom clusters (re-experiencing, avoidance and numbing, and hyperarousal) to neural reactivity and habituation to negative stimuli in combat-exposed veterans. METHODS: Ninety-ﬁve combat-exposed veterans (46 with PTSD) and 53 community volunteers underwent functional magnetic resonance imaging while viewing emotional images. This study examined the relationship between symptom cluster severity and hemodynamic responses to negative compared with neutral images (NEG.NEU). RESULTS: Veterans exhibited comparable mean and habituation-related responses for NEG.NEU, relative to civilians. However, among veterans, habituation, but not mean response, was differentially related to PTSD symptom severity. Hyperarousal symptoms were related to decreased habituation for NEG.NEU in a network of regions, including superior and inferior frontal gyri, ventromedial prefrontal cortex, superior and middle temporal gyri, and anterior insula. In contrast, re-experiencing symptoms were associated with increased habituation in a similar network. Furthermore, re-experiencing severity was positively related to amygdalar functional connectivity with the left inferior frontal gyrus and dorsal anterior cingulate cortex for NEG.NEU. CONCLUSIONS: These results indicate that hyperarousal symptoms in combat-related PTSD are associated with decreased neural habituation to aversive stimuli. These impairments are partially mitigated in the presence of re- experiencing symptoms, such that during exposure to negative stimuli, re-experiencing symptoms are positively associated with amygdalar connectivity to prefrontal regions implicated in affective suppression. Keywords: Affective neuroscience, Emotion, fMRI, Habituation, Heterogeneity, PTSD https://doi.org/10.1016/j.bpsc.2019.09.006

Following traumatic events, individuals often experience acute responsiveness to negative stimuli (11) and difficulties down- stress responses, which are adaptive and usually dissipate regulating emotions (12). In contrast, findings indicate that re- over a period of days or weeks. However, in posttraumatic experiencing symptoms may be associated with effortful stress disorder (PTSD), trauma-related disturbances persist or suppression of intrusive emotions and thoughts (13,14). While intensify over time and include co-occurring symptoms of the use of inhibitory control strategies may be associated with trauma re-experiencing, avoidance and numbing, and hyper- lower sympathetic arousal in the short term (15), over time, use arousal (1). Considerable work suggests that PTSD confers has been linked to increased PTSD symptoms (2,15). Addi- abnormalities in affective processing ranging from exagger- tionally, avoidance and numbing symptoms may manifest as a ated emotional responses to trauma cues and emotion regu- general disengagement from emotional processing with lation deficits (2), to alexithymia (3). Furthermore, both reduced neural responsiveness across stimuli (16–18). Here we heightened physiological reactivity to aversive stimuli (4–6) and seek to directly examine the relationships among symptom diminished habituation of these reactions (4,7) have been clusters and neural correlates of reactivity to negative stimuli implicated in PTSD-related difficulties. However, each of these and habituation over time. response patterns characterizes only subsets of individuals Neurobiological models of emotional difficulties in PTSD with PTSD (8–10), suggesting that heterogeneity of symptom implicate exaggerated amygdalar reactivity to aversive stimuli profiles may be associated with distinct aspects of emotion (19) coupled with inadequate modulation by ventromedial difficulties. In particular, evidence suggests that hyperarousal prefrontal cortical and hippocampal regions (20). Diminished symptoms may be associated with increased neural habituation to negative stimuli has also been identified using

SEE COMMENTARY ON PAGE 135

Published by Elsevier Inc on behalf of Society of Biological Psychiatry. This is an open access article under the 203 CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI Biological Psychiatry: CNNI Opponent Symptom Cluster Effects on Habituation in PTSD

both skin conductance and hemodynamic measures (4,12), Freedom, and Operation New Dawn) were recruited from the providing preliminary evidence for PTSD-related habituation community as well as from Veterans Affairs medical centers. deficits in the amygdala (7,21) and subgenual cingulate cortex Data from 3 veterans were excluded for excessive head motion (21,22), in contrast to healthy individuals who exhibit habitua- during functional magnetic resonance imaging (MRI), resulting tion of neural responses in the amygdalae (23,24), as well as in a final sample of 95 veterans (current PTSD: n = 46; no prefrontal (23,25,26) and parietal cortices (25). current PTSD: n = 49). Given our interest in dimensional effects Despite evidence linking PTSD to altered affective proof PTSD symptoms, we included veterans who did not meet cessing, inconsistencies in the presence and pattern of the full diagnostic criteria for PTSD but exhibited symptoms indi- underlying neural correlates exist (5,6,10,27). For example, cating subthreshold PTSD [as defined by Blanchard et al. (38); meta-analytic evidence of PTSD-related hyperactive amygda- n = 16]. Of note, all veterans, regardless of PTSD status, re- lar responses to negative stimuli has been mixed, with one ported experiencing one or more traumatic event(s) during meta-analysis finding both hyperactivity and hypoactivity deployment that met criterion A1 of the DSM-IV diagnostic within the amygdala (5), one finding right amygdala hyperac- criteria for PTSD (39). Additionally, 53 community volunteers tivity (19), and two others finding PTSD-related hyperactive were recruited to provide a civilian comparison group; our amygdalar responses under some but not all conditions primary purpose for including this group was to illustrate (6,27). Discrepant findings may be explained, in part, by normative brain responses to affective stimuli in a represen- heterogeneity of symptom presentation. That is, although tative sample of the population (i.e., not restricted to age and many individuals with PTSD exhibit heightened emotional gender demographic characteristics of our Operation Enduring responses to aversive stimuli and increased physiological Freedom, Operation Iraqi Freedom, and Operation New Dawn responses, some experience a detached or dissociative veterans). emotional reaction coupled with stable or decreased physi- Exclusion criteria included MR contraindications; claustro- ological responses (8,9). Similar symptom-specific relation- phobia; pregnancy; substance use disorders, other than ships have been identified in a variety of behavioral and nicotine dependence, during the past month; and head injury neural correlates of PTSD, including measures of interper- with loss of consciousness .30 minutes. For the veteran sonal functioning (28), treatment response (29), regional brain group, additional psychiatric exclusion criteria were significant volumes (30), resting-state functional connectivity (31), and current suicidal or homicidal ideation, or history of schizo- functional brain networks (32). However, research on the re- phrenia, schizoaffective disorder, delusional disorder, or lationships between symptom clusters and biomarkers of organic psychosis. For the civilian comparison group, use of affective responding in PTSD, a cardinal feature of the dis- psychotropic medication was an additional exclusion criterion. order, has thus far been limited [see (16,33,35)]. To test the possibility of systematic associations between Assessment of Psychiatric Disorders heterogeneity of affective responding and PTSD symptom clus- In the veteran sample, the Clinician-Administered PTSD ters, we examined the respective relationships of re-experiencing, Scale for DSM-IV (CAPS) (40) and the Structured Clinical avoidance and numbing, and hyperarousal symptom severity with Interview for DSM-IV-TR Axis I Disorders, Non-Patient neural responses to aversive (compared with neutral) images in a Edition (41) were used to assess PTSD diagnosis and large cohort of combat-exposed veterans. Given prior work severity of symptoms (42), and other Axis I disorders, suggesting that neural habituation is a more reliable metric of respectively (Supplemental Methods and Materials). The past affective responding than average neural activation (24), we pri- month’s symptom cluster severity scores for re-experiencing, marily focused on the relationships between symptoms and avoidance and numbing, and hyperarousal as measured by neural habituation. Specifically, we hypothesized that greater the CAPS were used as our primary covariates of interest. severity of hyperarousal symptoms would be associated with diminished neural habituation to negative versus neutral images in Emotion Paradigm limbic and salience network regions (12,36). Additionally, based on work showing greater re-experiencing severity with increased During functional MRI, participants viewed images from the suppression of affective and physiological responses to aversive International Affective Picture System (43)(Figure 1A). Each stimuli (13,15,37), we hypothesized that greater severity of re- image was presented for 4 seconds, and 8 to 10 images of the experiencing symptoms would be associated with enhanced same valence (negative [NEG], positive, or neutral [NEU]) were neural habituation to negative versus neutral images. presented in each block. Twenty-four blocks (8 of each type) were pseudo-randomly presented, separated by jittered fixation blocks of 4 to 12 seconds for a total task duration of METHODS AND MATERIALS approximately 18 minutes (Supplemental Methods and All procedures were carried out in accordance with the Institu- Materials). tional Review Boards of Baylor College of Medicine and the Salem Veterans Affairs Medical Center. After receiving a description of Image Acquisition and Preprocessing the study’s procedures and being given the opportunity to ask Magnetic resonance images were collected using 3T questions, all participants provided written informed consent. Siemens Trio MR scanners (Siemens, Erlangen, Germany). Whole-brain functional images were continuously acquired Participants during a single run and a high-resolution T1-weighted Ninety-eight veterans who were deployed during post-9/11 structural scan was acquired. MR images were analyzed conflicts (Operation Enduring Freedom, Operation Iraqi using Statistical Parametric Mapping 12 (SPM12; Wellcome

204 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI Biological Psychiatry: Opponent Symptom Cluster Effects on Habituation in PTSD CNNI

Analysis of effects related to positive image blocks is beyond the scope of this study. Here, we speciﬁcally focus on effects associated with NEG and NEU image blocks. Accordingly, b maps for contrasts of interest (i.e., NEG_- MEAN.NEU_MEAN, NEG_LIN.NEU_LIN) were brought to the second-level for group analyses. Whole-brain multivariate regressions were conducted on NEG_MEAN.NEU_MEAN and NEG_LIN.NEU_LIN contrasts, with simultaneous inclusion of 3 covariates of interest corresponding to current re-experiencing, avoidance and numbing, and hyperarousal symptom cluster severity. Depression severity, presence of probable mild traumatic brain injury (mTBI), combat exposure severity, and use of psychotropic medication were included as controlling covariates, as prior research indicates that these variables may be associated with distinct neural alterations (44–47)(Supplemental Methods and FPO

= Materials). Given the expected correlation among PTSD symptom cluster severity scores (and to a lesser degree, controlling covariates), variance inflation factors were calcu- web 4C lated for the covariates included in regression analyses; all Figure 1. Experimental paradigm and neural response to negative and covariates had variance inflation factors ,5, indicating that neutral images. (A) Blocks of 8 to 10 images of the same valence (negative, positive, or neutral) were presented, followed by a fixation screen of 4 to 12 multicollinearity was not a significant issue (48). seconds. (B) Block order was pseudo-randomized with #2 blocks of the To facilitate comparison to prior work that focused on the same valence occurring consecutively. Eight blocks of each valence were overall impact of PTSD, we conducted secondary analyses to displayed (24 blocks total) for a task duration of 17 minutes and 44 seconds. assess diagnostic and dimensional effects of PTSD. Specifically, Regressors of interest for the imaging analyses modeled the overall effect of for the contrasts of interest, 2-sample between-group Student’s each valence (negative or neutral) and a parametric modulator used to capture the effect of habituation across blocks of each valence. For the t tests were used to assess the effect of PTSD diagnosis within contrast of negative.neutral, areas of significance for the overall mean (C) the veteran cohort (excluding veterans with subthreshold PTSD, and for habituation (D) in the veteran group (n = 95) are shown (p , .05, n = 16). To examine dimensional effects of PTSD severity within familywise error–cluster-corrected, coronal slices displayed at y = 21in the veteran cohort, whole-brain multivariate regressions were Montreal Neurological Institute standard space). Red and orange indicate conducted on the contrasts of interest, with overall PTSD significant positive t values at cluster-defining primary thresholds of p , symptom severity included as a covariate of interest and .001 and , .005, respectively. Similarly, blue and light blue indicate sig- p depression, mTBI, combat exposure severity, and psychotropic nificant negative t values at cluster-defining primary thresholds of p , .001 and p , .005, respectively. Comparable results for the civilian group (n = 53) medication usage included as controlling covariates. can be found in Supplemental Figure S1. Owing to copyright restrictions Unless otherwise noted, all imaging results were assessed associated with the International Affective Picture System, images in (A) are for significance using a cluster-level familywise error–corrected taken from the Open Affective Standardized Image Set (92). threshold of p , .05 with a cluster-defining primary threshold of p , .001. This thresholding approach has been shown to appropriately control for familywise error in cluster-level ana- Trust Centre for Neuroimaging, London, UK) and MATLAB lyses employing similar methodology (49). R2010b (The MathWorks, Inc., Natick, MA). Standard preprocessing was performed. Additional details regarding scanning parameters and preprocessing methods are pro- Exploratory Psychophysiological Interaction vided in the Supplemental Methods and Materials. Analysis The results of our primary analyses suggested a mitigating role of re-experiencing symptoms on hyperarousal-related impair- Whole-Brain Analyses ments in affective habituation to aversive stimuli, which we First-level general linear models included 2 regressors of in- hypothesized may be related to the use of suppression tech- terest (mean and habituation) for each valence (NEG, positive, niques. To examine this possibility, we assessed the impact of and NEU) (Figure 1B). Specifically, valence-specific mean re- PTSD symptom clusters on valence-specific (NEG.NEU) sponses (hereafter, MEAN) were modeled as boxcar functions functional connectivity of the amygdalae on cortical regions corresponding to each block’s duration, and valence-specific associated with suppression of thoughts and emotions. Spe- patterns of habituation and/or sensitization of hemodynamic cifically, we performed an exploratory generalized psycho- responses (hereafter, LIN) were modeled as a linear parametric physiological interaction analysis (PPI) using the gPPI toolbox modulation of each block corresponding to block order over (50). An amygdalar seed region was defined as all significant time. For habituation analyses, positive b values reflect habit- voxels within a bilateral anatomical amygdalar mask for the uation and negative b values reflect sensitization. Regressors contrast, NEG_MEAN.NEU_MEAN, with a primary threshold were convolved with a canonical hemodynamic response of p , .005, for the veteran cohort. Additional gPPI analytic function. Six motion parameters were included as regressors, details are included in the Supplemental Methods and and a 180-second high-pass temporal filter was applied. Materials. At the group level, the relationships between PTSD

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI 205 Biological Psychiatry: CNNI Opponent Symptom Cluster Effects on Habituation in PTSD

symptom clusters and valence-specific, whole-brain amygda- expected, veterans with PTSD reported more severe symptoms lar connectivity were analyzed, for the contrast of interest, of depression, current PTSD, and lifetime PTSD; veterans with NEG_PPI.NEU_PPI, using a multivariate regression with the PTSD also had a significantly higher incidence of a current mood same set of covariates as the main analyses. disorder diagnosis compared with veterans without PTSD. Vet- erans with PTSD also reported greater combat exposure, inci- RESULTS dence of mTBI, and use of psychotropic medication, relative to veterans without PTSD. The Supplement provides additional Demographic and Clinical Characteristics of information about psychotropic medication use, characteristics Participants of the veteran cohort when separated into 3 diagnostic groups Demographic and clinical characteristics of participants are (current, subthreshold, no PTSD), and characteristics of the shown in Table 1 (40,41,51–54). The veteran sample included civilian cohort (Supplemental Methods and Materials; 82 male and 13 female participants, with a mean age of 32.2 Supplemental Results;andSupplemental Tables S1–S3). years (range: 21–57, SD: 8.2). The civilian cohort included 26 male and 27 female participants, with a mean age of Neural Responses to Negative Images Compared 27.2 years (range: 18–48, SD: 7.6). With Neutral Images Veterans with and without PTSD did not differ on age, gender, Mean Effects. Veterans exhibited significant positive race, education, and household income; the diagnostic groups hemodynamic response for the contrast NEG_MEAN. also did not differ on incidence of current anxiety disorders (other NEU_MEAN in regions previously implicated in emotion pro- than PTSD) or incidence of past substance use disorders. As cessing including the amygdalae and thalamus as well as the

Table 1. Demographic and Clinical Characteristics of Civilians and Combat-Exposed Veterans With and Without Current PTSD Civilians Combat-Exposed Veterans Veterans With PTSD vs. No PTSD Characteristic All (n = 53) All (n = 95) PTSDa (n = 46) No PTSDa (n = 49) p Valueb Age, Years 27.2 (7.6) 32.2 (8.2) 32.4 (8.1) 32.0 (8.3) .82 Education, Years 15.5 (2.0) 14.6 (1.5) 14.3 (1.3) 14.9 (1.7) .08 Household Income, 1000s of Dollarsc 45.1 (30.8) 41.3 (24.9) 48.6 (35.2) .27 Combat Exposured 19.1 (9.5) 21.2 (8.5) 17.0 (10.0) .03 Depression Symptomse 4.9 (5.9) 17.8 (12.5) 25.1 (11.7) 10.9 (8.9) ,.001 Current PTSD Symptomsa 44.7 (30.4) 69.1 (18.3) 21.8 (19.7) ,.001 Re-experiencing 10.2 (8.8) 16.7 (7.2) 4.1 (5.0) ,.001 Avoidance and numbing 18.0 (13.6) 28.9 (8.7) 7.7 (8.1) ,.001 Hyperarousal 16.5 (10.5) 23.5 (6.2) 10.0 (9.5) ,.001 Lifetime PTSD Symptomsa 70.9 (39.6) 100.5 (14.5) 43.1 (35.3) ,.001 Re-experiencing 19.5 (12.3) 28.8 (6.1) 10.7 (10.1) ,.001 Avoidance and numbing 27.5 (17.1) 40.0 (7.3) 15.7 (15.1) ,.001 Hyperarousal 23.9 (12.2) 31.7 (4.0) 16.7 (12.7) ,.001 Gender, Female, n (%) 27 (51) 13 (14) 7 (15) 6 (12) .90 Race, Nonwhitef, n (%) 23 (43) 37 (39) 20 (43) 17 (35) .50 Current Mood Disorderg, n (%) 32 (34) 28 (61) 4 (8) ,.001 Current Anxiety Disorderg,h, n (%) 12 (13) 9 (20) 3 (6) .10 Past Substance Abuseg, n (%) 55 (58) 29 (63) 26 (53) .44 Positive mTBI Screeni, n (%) 29 (31) 22 (48) 7 (14) ,.001 Psychotropic Medicationj, n (%) 37 (39) 27 (59) 10 (20) ,.001 Unless otherwise indicated, data are reported as mean (SD) of group. See Supplemental Table S2 for demographic and clinical characteristics of the veteran cohort when separated into 3 groups (PTSD, subthreshold PTSD, and no PTSD). mTBI, mild traumatic brain injury; PTSD, posttraumatic stress disorder. aPast month’s diagnosis and severity based on DSM-IV criteria as assessed by the Clinician-Administered PTSD Scale (40). bThe p values are based on 2-sample Student’s t tests for continuous variables and c2 tests for dichotomous variables. cData not reported for 6 veterans (4 with current PTSD and 2 without current PTSD). dTotal raw score on Combat Exposure Scale (51). Data missing for 3 veterans (1 with current PTSD and 2 without current PTSD). eTotal score on Beck Depression Inventory-II (52). fBased on participant’s self-report. gDiagnosis as assessed by the Structured Clinical Interview for DSM-IV (41); “current” defined as meeting criteria during the past month; “past” defined as meeting lifetime criteria prior to the past month. hAnxiety disorders other than PTSD. iPositive screen on the Brief Traumatic Brain Injury Screen (53). jPsychotropic medication use was defined as self-reported current use of one or more medication(s) listed in the National Institute of Mental Health’s publication on Mental Health Medications (54).

206 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI Biological Psychiatry: Opponent Symptom Cluster Effects on Habituation in PTSD CNNI

precentral gyrus, fusiform gyrus, and inferior frontal gyrus (IFG) experiencing and hyperarousal symptoms were related, in (Figure 1C; Supplemental Table S4). Additionally, veterans opposite directions, to widespread neural habituation to exhibited significantly negative hemodynamic response bilat- negative versus neutral images (i.e., for NEG_LIN.NEU_LIN, erally in the superior and inferior temporal gyri, angular gyri, with increasing habituation related to re-experiencing severity inferior parietal lobules, posterior insular cortex, precuneus, and increasing sensitization with hyperarousal severity) and orbitofrontal cortex, as well as in the left anterior cingulate (Figure 2B). Avoidance and numbing severity was not related cortex (ACC), right posterior cingulate cortex, and right cere- to neural habituation in this network of regions. bellum. Similar response patterns were seen in the civilian group (Supplemental Figure S1A, Supplemental Table S5). A Exploratory Analysis of Opponent Effects of Re- 2-sample Student’s t test comparing veterans to civilians experiencing and Hyperarousal. Furthermore, in this found no significant differences. same network of regions, we conducted an exploratory analysis of the additive effects of the re-experiencing and hyper- Habituation and Sensitization Effects. Consistent with arousal symptom clusters on habituation. Given the observed prior findings of affective habituation, as a group, veterans opponent effects of hyperarousal and re-experiencing symp- exhibited greater habituation across negative blocks than across tom severity on neural habituation, we calculated each in- neutral blocks (NEG_LIN.NEU_LIN) bilaterally in middle temporal dividual’s severity discrepancy score (i.e., CAPS hyperarousal gyri, fusiform gyri, parahippocampal gyri, supplementary motor score 2 CAPS re-experiencing score) and entered the area, frontal eye fields, and precuneus (Figure 1D; Supplemental scores as independent variables in a linear regression analysis Table S6). For the contrast of interest, no regions showed signif- to predict neural habituation in the region of interest for icantly less habituation. Similar response patterns were seen in NEG_LIN.NEU_LIN. As depicted in Figure 3, we found a the civilian group (Supplemental Figure S1B, Supplemental significant negative relationship (adjusted r2 = .255, p = 1.1 3 Table S7). A 2-sample Student’s t test comparing habitua- 1027), such that as the severity discrepancy between hyper- tion patterns of veterans to civilians found no significant arousal and re-experiencing symptoms increased, veterans differences. exhibited diminishing neural habituation in the shared habituation network. These data suggest that combat-exposed vet- Effects of PTSD Symptom Cluster Severity erans with prominent hyperarousal symptoms in the absence of significant re-experiencing symptoms are most likely to Mean Effects. Severity of re-experiencing, avoidance and exhibit disrupted habituation to aversive stimuli. The numbing, and hyperarousal symptoms was not significantly Supplement provides additional details regarding the related to the contrast of NEG_MEAN.NEU_MEAN; further- relationship between re-experiencing and hyperarousal symptom more, none of the controlling covariates were significantly severity in this sample (Supplemental Figure S3)aswellas related to the NEG_MEAN.NEU_MEAN contrast. analysis of the relationship of these symptoms with habituation and/or sensitization to neutral blocks only Habituation and Sensitization Effects. When account- (Supplemental Results). ing for effects of controlling covariates and other symptom To confirm that these results were robust to individual dif- clusters, severity of re-experiencing symptoms was posi- ferences in initial reactivity, secondary analyses were con- tively correlated with habituation to negative versus neutral ducted in which absolute habituation, according to Montagu images (i.e., greater habituation) in a widespread network (55), was calculated to obtain habituation metrics that were including the superior and middle temporal gyri, superior and independent of individuals’ initial responses (Supplemental medial frontal gyri, ACC, supplementary motor area, pre- Methods and Materials). Similar opponent effects of re- cuneus, and IFG (Figure 2A,left;Supplemental Table S8). No experiencing and hyperarousal were seen when comparing significant relationship was found between severity of absolute habituation across negative blocks to absolute avoidance and numbing symptoms and neural response to habituation across neutral blocks (Supplemental Figure S4). NEG_LIN.NEU_LIN (Figure 2A, middle). Severity of hyperarousal symptoms was negatively related to neural re- Exploratory PPI Analysis. For the contrast of interest, no sponses to NEG_LIN.NEU_LIN (i.e., less habituation) in significant effects were seen for avoidance and numbing portions of the superior and middle temporal gyri, ACC, IFG, symptom severity or hyperarousal symptom severity; however, ventromedial prefrontal cortex, precuneus, and anterior greater severity of re-experiencing symptoms was significantly insula (Figure 2A, right; Supplemental Table S8). associated with increased amygdalar connectivity with the left For NEG_LIN.NEU_LIN, no significant relationship was IFG and dorsal ACC for the negative condition relative to the seen between hemodynamic responses and combat exposure neutral condition (Figure 4). severity or use of psychotropic medication; depression symptom severity was related to increased habituation in the right middle occipital gyrus, and presence of probable mTBI was related to increased habituation bilaterally in the pre- Effects of Overall PTSD Severity cuneus as well as right midcingulate cortex (Supplemental Overall PTSD symptom severity was not significantly related Figure S2, Supplemental Table S9). to hemodynamic responses for either the contrast of NEG_MEAN.NEU_MEAN or the contrast of NEG_LIN. Opponent Effects of Re-experiencing and Hyper- NEU_LIN (see Supplemental Results for test of PTSD diag- arousal. In an overlapping network of regions, re- nostic group differences).

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI 207 Biological Psychiatry: CNNI Opponent Symptom Cluster Effects on Habituation in PTSD

Figure 2. Habituation and sensitization of neural response across blocks of negative and neutral images related to posttraumatic stress disorder (PTSD) symptom clusters. (A) The images show the relations between neural habituation for the contrast of negative condition compared to neutral condition (NEG.NEU) and severity of each PTSD symptom cluster (p , .05, familywise error–cluster-corrected with cluster deﬁning threshold of p , .001, axial and coronal slices shown at Montreal Neurological Insti- tute coordinates z = 211 and y = 21, respectively). Increasing severity of re-experiencing symptoms is positively associated with neural habituation in NEG.NEU (left), while increasing severity of hyperarousal symptoms is positively associated with neural sensitization in NEG.NEU (right). No signiﬁ- cant relation is found between avoidance and numbing symptoms and habituation or sensitization to NEG.NEU (middle). To visualize the separable

web 4C/FPO effects of each symptom cluster on neural habituation, we deﬁned a region of interest as the conjunction of re-experiencing and hyperarousal effects, seen in (A), and within this region of interest, extracted the ﬁrst eigenvariate from the habituation- related NEG.NEU contrast, as well as from habituation-related NEG and NEU contrasts separately. (B) The extracted values were used to calculate b estimates of the effect of each symptom cluster controlling for remaining PTSD clusters, depression, mild traumatic brain injury, combat exposure, and use of psychotropic medication. Bar plots illustrate mean b estimates by PTSD symptom cluster severity. Errors bars depict the standard error of the mean for each bin. Differences in habituation and sensitization for NEG.NEU contrast are depicted (top), as well as the separate effects across NEG (middle) and NEU (bottom) conditions. For the habituation-related NEG.NEU contrast, greater severity of re-experiencing is related to greater neural habituation and greater severity of hyperarousal is related to less neural habituation (for effects of controlling covariates, see Supplemental Figure S2). CAPS, Clinician-Administered PTSD Scale for DSM-IV.

DISCUSSION attentional biases (61). Taken together, these findings are in Here, we sought to determine the associations between line with a conceptualization of hyperarousal symptoms symptom clusters of PTSD and neural responsivity to negative resulting from a failure to habituate, oversensitization, or a emotional stimuli. We found that neural habituation to aversive combination of both processes (12). stimuli relative to neutral stimuli is diminished among veterans PTSD-related attentional biases result not only from with greater hyperarousal symptoms and enhanced among increased orientation toward threatening stimuli (62), but also those with greater re-experiencing symptoms. Of note, the from difficulty disengaging from threatening stimuli (63). Such opponent associations of hyperarousal and re-experiencing attentional difficulties are associated with greater use of mal- symptoms with neural habituation were evident across a set adaptive coping strategies, such as thought suppression, of regions previously implicated in attention, cognitive control, which has been found to mediate the relationship between and affective processing (56), suggesting that both symptom attentional interference and re-experiencing symptoms (13). clusters are related to widespread modulation of habituation. Similarly, decreases in physiological arousal during attempts to Additionally, a significant negative relationship was found be- reduce negative emotion have been associated with later tween the severity discrepancy of hyperarousal and re- increases in intrusive memories (37). Here, we found greater re- experiencing symptoms and neural habituation in this set of experiencing severity to be associated with stronger amygdalar regions, such that veterans who had more severe hyperarousal connectivity to left IFG and dorsal ACC when viewing negative than re-experiencing symptoms showed the greatest impair- relative to neutral images, regions consistently implicated in ment in habituation to negative versus neutral cues. suppression of thoughts and emotions (64–67). One possible Hyperarousal’s negative effect on neural habituation was interpretation of this finding is that when confronted with seen in regions previously implicated in affective habituation aversive cues, individuals with prominent re-experiencing in healthy individuals including the right spatial attention symptoms may engage in coping behaviors such as thought network, ventrolateral prefrontal cortex, dorsomedial pre- suppression through attentional control, which in the short term frontal cortex (57), and anterior insula (58). Diminished may decrease physiological responses but in the long term may habituation in the spatial attention network and other re- also lead to a rebound of intrusive thoughts. Thus, the finding gions involved in attentional control, such as the right IFG that greater re-experiencing severity is associated with greater (59), is consistent with hypervigilance, a symptom of hy- neural habituation for the contrast NEG_LIN.NEU_LIN may perarousal that has been associated with increased visual reflect a process by which individuals with severe re- scanning and autonomic arousal (60). Threat-related atten- experiencing symptoms engage in suppression in response tional biases have been repeatedly implicated in PTSD, to negative stimuli, leading to greater decreases in neural re- lending support to the suggestion that hyperarousal symp- sponses across the negative blocks, while potentially also toms may result, in part, from enhancement of these resulting in subsequent rebound of intrusive thoughts.

208 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI Biological Psychiatry: Opponent Symptom Cluster Effects on Habituation in PTSD CNNI FPO = web 4C Figure 4. Relation of re-experiencing severity with context-modulated amygdalar connectivity for negative versus neutral conditions. Analysis of the effects of posttraumatic stress disorder symptom clusters on a psy-

Neural Habituation to Negative > Neutral Negative Habituation to Neural chophysiological interaction of amygdalar connectivity and valence (negative.neutral) found a positive relationship between re-experiencing severity and context-modulated amygdalar connectivity with the left inferior FPO = frontal gyrus (Tpeak = 4.56; cluster size: 91 voxels) and dorsal anterior cingulate cortex (Tpeak = 3.21; cluster size = 70 voxels) (p , .05, familywise Hyperarousal Severity - Re-experiencing Severity error–cluster-corrected with cluster defining threshold of p , .001). Cluster web 4C peaks are located at Montreal Neurological Institute coordinates (251, 11, Figure 3. Relation between hyperarousal and re-experiencing severity 10) and (0, 21, 43) and shown above in coronal slices at y = 11 and y = 21. discrepancy and neural habituation in veterans with and without posttraumatic stress disorder. Using each individual’s severity discrepancy score (hyperarousal severity 2 re-experiencing severity) as a predictor confounds including combat exposure, comorbid depressive variable, a linear regression analysis was conducted to predict neural habituation to negative.neutral for the network of regions in which habit- symptoms, probable exposure to mild traumatic brain injury, uation showed opponent relations with hyperarousal and re-experiencing and use of psychotropic medications. severity (shown in Figure 2A). The scatter plot depicts the significant Another reason for the discrepant results between this negative relation (adjusted r2 = .255, p , .001) with data from veterans with study and previous studies may be related to statistical power. and without posttraumatic stress disorder represented by filled (C) and Notably, many previous neuroimaging studies of affective open ( ) points, respectively; the solid line represents the linear association processing in PTSD have had substantially smaller sample and dashed lines show 695% confidence intervals. sizes than that of the current study [e.g., (76–82)]. While sample size concerns are often focused on increased false-negative No significant differences in neural activity or habituation to rates, low statistical power (as is seen in studies with small aversive images (compared with neutral images) were seen sample sizes) also makes false-positive rates more likely when veterans with PTSD were compared with veterans (83,84). Additionally, some past neuroimaging studies in this without PTSD; similarly, overall PTSD severity was not signif- domain have employed statistical thresholds and/or correc- icantly associated with differences in neural activity or habit- tions for multiple comparisons that have subsequently been uation to negative relative to neutral images. Given that prior associated with inflated false-positive rates (49). work using similar affective paradigms have found significant diagnostic and dimensional effects of PTSD [e.g., (6,18,27,68–72)], these null results were somewhat surprising. Limitations While the reasons for this inconsistency are unclear, differ- The current study focuses on combat-exposed veterans of ences in study design may be a potential factor. Prior studies post-9/11 conflicts, and almost all of our veteran participants of emotional reactivity in PTSD have sometimes selected for (n = 90; 95%) reported experiencing one or more combat- individuals based on heightened psychological or physiolog- related trauma(s). In addition, a relatively small number of ical responses to emotional stimuli (73,74). Thus, between- women were included in our sample (n = 13; 14%). Thus, while group comparisons may have be driven primarily by our data have the advantage of sample homogeneity, gener- hyperarousal symptoms, associated here with diminished alizability across gender as well as across other types of habituation. Significant differences in neural responses seen in trauma is unknown at this time. Additionally, we did not previous studies may also be attributable to the use of trauma- conduct toxicology screens on the day of scanning, so specific and/or threat-specific stimuli, rather than to the more although individuals in our veteran sample were screened via general negatively-valenced stimuli used in the present study clinical interview to assess criteria for current substance use (17,75). Finally, some previous studies have not accounted for disorders, we cannot exclude the possibility that one or more potential confounding factors such as severity of trauma individual(s) used illicit substances. Also, as our study exam- exposure [e.g., (69)] and comorbid disorders [e.g., (68,70)]; ined habituation to negative, relative to neutrally valenced thus, between-group differences may be attributable to one or images in general, we cannot draw conclusions about the more confounding factor(s) rather than to PTSD diagnosis impact of PTSD symptoms on neural habituation to trauma- alone. A strength of this study was the accounting for potential specific images. Future research using trauma-specific

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI 209 Biological Psychiatry: CNNI Opponent Symptom Cluster Effects on Habituation in PTSD

stimuli may further elucidate the effect of PTSD symptoms on Portions of this work were presented, in poster form, at the 2016 As- neural habituation. sociation for Behavioral and Cognitive Therapies Annual Convention in New York City, New York. The authors report no biomedical financial interests or potential conflicts Conclusions of interest. To our knowledge, the present data are among the first to ARTICLE INFORMATION show that re-experiencing and hyperarousal symptoms are differentially related to negative affective neural habituation. From the Salem Veterans Affairs Medical Center (KLMcC, PHC, BK-C), Salem; Fralin Biomedical Research Institute at Virginia Tech Carilion Moreover, negative affective neural habituation, compared (KLMcC, BCF, PHC, BK-C), Virginia Tech, and Department of Psychiatry and with neutral habituation, is most adversely impacted not by the Behavioral Medicine (PHC, BK-C), Virginia Tech Carilion School of Medicine, magnitude of hyperarousal symptoms alone, but rather by the Roanoke; Department of Psychology (KLMcC, PHC, BK-C), Virginia Tech, relative severity of hyperarousal symptoms and re- and School of Biomedical Engineering and Sciences (BK-C), Virginia Tech– experiencing symptoms. Re-experiencing and hyperarousal Wake Forest University, Blacksburg, Virginia; Department of Psychology symptoms are generally highly correlated (85), and previous (BCF), University of Hawaii at Hilo, Hilo, Hawaii; and Trauma and Resilience Center (BCF), Department of Psychiatry, University of Texas Health research on neural heterogeneity in PTSD has more often Sciences Center, Houston, Texas. focused on differences between symptom clusters thought to Address correspondence to Pearl H. Chiu, Ph.D., and Brooks King- reflect heightened responses (i.e., re-experiencing and hyper- Casas, Ph.D., Fralin Biomedical Research Institute at Virginia Tech Car- arousal) and those believed to represent diminished responses ilion, 2 Riverside Circle, Roanoke, VA 24016; E-mail: [email protected] or (i.e., avoidance and numbing) (8,34). However, the common- [email protected]. ality of re-experiencing and hyperarousal clusters is cautioned Received Dec 21, 2018; revised Sep 6, 2019; accepted Sep 9, 2019. Supplementary material cited in this article is available online at https:// by animal models of PTSD suggesting that hyperarousal and doi.org/10.1016/j.bpsc.2019.09.006. context-specific responses similar to re-experiencing may develop independently and may be differentially changed by treatment (86–88). REFERENCES The positive association of re-experiencing symptoms 1. Yehuda R, LeDoux J (2007): Response variation following trauma: A with negative affective habituation suggests the intriguing translational neuroscience approach to understanding PTSD. Neuron possibility of a protective compensatory mechanism that 56:19–32. offsets adverse effects of heightened arousal. Although re- 2. Seligowski AV, Lee DJ, Bardeen JR, Orcutt HK (2014): Emotion regulation and posttraumatic stress symptoms: A meta-analysis. Cogn experiencing is maladaptive in the long term (e.g., Behav Ther 44:87–102. repeated intrusive memories result in increased distress), 3. Frewen PA, Dozois DJA, Neufeld RWJ, Lanius RA (2008): Meta-anal- these symptoms have also been associated with coping ysis of alexithymia in posttraumatic stress disorder. J Traum Stress behaviors [e.g., thought suppression (13,89)] that in the 21:243–246. short term may prevent arousal overload (15). Additional 4. Pole N (2007): The psychophysiology of posttraumatic stress disorder: research is needed to replicate this relationship and gain a A meta-analysis. Psychol Bull 133:725–746. 5. Etkin A, Wager TD (2007): Functional neuroimaging of anxiety: A meta- clearer understanding of the impact of re-experiencing on analysis of emotional processing in PTSD, social anxiety disorder, and negative affective habituation. specific phobia. Am J Psychiatry 164:1476–1488. These results also suggest ways in which an individual’s 6. Hayes JP, Hayes SM, Mikedis AM (2012): Quantitative meta-analysis symptom presentation may be informative for treatment de- of neural activity in posttraumatic stress disorder. Biol Mood Anxiety cisions. For instance, prior work has shown that individuals Disord 2:9–9. with severe hyperarousal symptoms are more likely to be 7. Protopopescu X, Pan H, Tuescher O, Cloitre M, Goldstein M, Engelien W, et al. (2005): Differential time courses and specificity of treatment nonresponders (29). Moreover, individuals for whom amygdala activity in posttraumatic stress disorder subjects and normal hyperarousal is the most prominent initial symptom cluster control subjects. Biol Psychiatry 57:464–473. show less symptom improvement over time (90). For in- 8. Lanius RA, Bluhm RL, Pain C (2006): A review of neuroimaging studies dividuals with prominent hyperarousal symptoms accompa- in PTSD: Heterogeneity of response to symptom provocation. nied by minimal re-experiencing symptoms, treatments that J Psychiatr Res 40:709–729. facilitate the development of general habituation capacity to 9. McTeague LM, Lang PJ, Laplante M-C, Cuthbert BN, Shumen JR, Bradley MM (2010): Aversive imagery in posttraumatic stress disorder: aversive relative to neutral stimuli may improve treatment Trauma recurrence, comorbidity, and physiological reactivity. Biol outcomes (91). Psychiatry 67:346–356. 10. Grupe DW, Heller AS (2016): Brain imaging alterations in posttraumatic stress disorder. Psychiatr Ann 46:519–526. ACKNOWLEDGMENTS AND DISCLOSURES 11. Stevens JS, Jovanovic T, Fani N, Ely TD, Glover EM, Bradley B, This work was supported in part by the Department of Veterans Affairs, Ressler KJ (2013): Disrupted amygdala-prefrontal functional connec- Office of Research and Development, Rehabilitation Research and Devel- tivity in civilian women with posttraumatic stress disorder. J Psychiatr opment Grant Nos. D2354R and D7030R (to BK-C), and National Institutes Res 47:1469–1478. of Health Grant Nos. MH074468 (to BCF), MH115221 (to BK-C), and 12. Lissek S, van Meurs B (2015): Learning models of PTSD: Theoretical MH087692 and MH106756 (to PHC). accounts and psychobiological evidence. Int J Psychophysiol 98:594– We thank Wright Williams and Matt Estey (who were supported by the 605. Department of Veterans Affairs, Office of Research and Development, 13. Wisco BE, Pineles SL, Shipherd JC, Marx BP (2013): Attentional Rehabilitation Research and Development Grant No. B7760P [to WW]), and interference by threat and post-traumatic stress disorder: The role of Jessica Eiseman, Kat Gardner, David Graham, LaRaun Lindsey, Robert thought control strategies. Cogn Emot 27:1314–1325. McNamara, and April Sanders, for their research support. We also gratefully 14. Mairean C, Ceobanu CM (2016): The relationship between sup- acknowledge discussions with Vanessa Brown and Nina Lauharatanahirun. pression and subsequent intrusions: The mediating role of

210 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI Biological Psychiatry: Opponent Symptom Cluster Effects on Habituation in PTSD CNNI

peritraumatic dissociation and anxiety. Anxiety Stress Coping profiles during threat anticipation in combat veterans. Psychol Med 30:304–316. 46:1885–1895. 15. Bardeen JR, Daniel TA (2017): A longitudinal examination of the role of 34. Hopper JW, Frewen PA, van der Kolk BA, Lanius RA (2007): attentional control in the relationship between posttraumatic stress Neural correlates of reexperiencing, avoidance, and dissociation and threat-related attentional bias: An eye-tracking study. Behav Res in PTSD: Symptom dimensions and emotion dysregulation in Ther 99:67–77. responses to script-driven trauma imagery. J Traum Stress 16. Felmingham KL, Falconer EM, Williams L, Kemp AH, Allen A, 20:713–725. Peduto A, Bryant RA (2014): Reduced amygdala and ventral striatal 35. Simmons AN, Matthews SC, Strigo IA, Baker DG, Donovan HK, activity to happy faces in PTSD is associated with emotional numbing. Motezadi A, et al. (2011): Altered amygdala activation during face PLoS One 9:e103653. processing in Iraqi and Afghanistani war veterans. Biol Mood Anxiety 17. Forster GL, Simons RM, Baugh LA (2017): Revisiting the role of the Disord 1:6. amygdala in posttraumatic stress disorder. In: Ferry B, editor. The 36. Liberzon I, Abelson JL (2016): Context processing and the neurobi- Amygdala—Where Emotions Shape Perception, Learning and Mem- ology of post-traumatic stress disorder. Neuron 92:14–30. ories. London, UK: InTechOpen, 113–135. 37. Shepherd L, Wild J (2014): Emotion regulation, physiological arousal 18. Phan KL, Britton JC, Taylor SF, Fig LM, Liberzon I (2006): Corticolimbic and PTSD symptoms in trauma-exposed individuals. J Behav Ther blood flow during nontraumatic emotional processing in posttraumatic Exp Psychiatry 45:360–367. stress disorder. Arch Gen Psychiatry 63:184. 38. Blanchard EB, Hickling EJ, Taylor AE, Loos WR, Gerardi RJ (1994): 19. Schulze L, Schulze A, Renneberg B, Schmahl C, Niedtfeld I (2018): Psychological morbidity associated with motor vehicle accidents. Neural correlates of affective disturbances. A comparative meta- Behav Res Ther 32:283–290. analysis of negative affect processing in borderline personality 39. American Psychiatric Association (2000): Diagnostic and Statistical disorder, major depression, and posttraumatic stress disorder. Biol Manual of Mental Disorders: DSM-IV-TR. Washington, DC: American Psychiatry Cogn Neurosci Neuroimaging 4:1–89. Psychiatric Association. 20. Rauch SL, Shin L, Phelps EA (2006): Neurocircuitry models 40. Blake DD, Weathers FW, Nagy LM, Kaloupek DG, Gusman FD, of posttraumatic stress disorder and extinction: Human neuro- Charney DS, Keane TM (1995): The development of a clinician- imaging research—past,present,andfuture.BiolPsychiatry administered PTSD scale. J Traum Stress 8:75–90. 60:376–382. 41. First MB, Spitzer RL, Gibbon M, Williams JBW (2002): Structured 21. Stevens JS, Kim YJ, Galatzer-Levy IR, Reddy R, Ely TD, Nemeroff CB, Clinical Interview for DSM-IV-TR Axis I Disorders, Research Version, et al. (2017): Amygdala reactivity and anterior cingulate habituation Non-Patient Edition (SCID-I/NP). New York, NY: Biometrics Research predict PTSD symptom maintenance after acute civilian trauma. Biol Department. Psychiatry 15:1023–1029. 42. Weathers FW, Keane TM, Davidson JR (2001): Clinician-administered 22. Tuescher O, Protopopescu X, Pan H, Cloitre M, Butler T, Goldstein M, PTSD scale: A review of the first ten years of research. Depress et al. (2011): Differential activity of subgenual cingulate and brainstem Anxiety 13:132–156. in panic disorder and PTSD. J Anxiety Disord 25:251–257. 43. Lang PJ, Bradley MM, Cuthbert BN (2008): International Affective 23. Wright CI, Fischer H, Whalen PJ, McInerney SC, Shin LM, Rauch SL Picture System (IAPS): Affective Ratings of Pictures and Instruction (2001): Differential prefrontal cortex and amygdala habituation to Manual. Gainesville: University of Florida, National Institute of Mental repeatedly presented emotional stimuli. Neuroreport 12:379–383. Health Center. 24. Plichta MM, Grimm O, Morgen K, Mier D, Sauer C, Haddad L, et al. 44. Lanius RA, Frewen PA, Girotti M, Neufeld RWJ, Stevens TK, (2014): Amygdala habituation: A reliable fMRI phenotype. Neuroimage Densmore M (2007): Neural correlates of trauma script-imagery in 103:383–390. posttraumatic stress disorder with and without comorbid major 25. Feinstein JS, Goldin PR, Stein MB, Brown GG, Paulus MP (2002): depression: A functional MRI investigation. Psychiatry Res 155:45–56. Habituation of attentional networks during emotion processing. Neu- 45. Simmons AN, Matthews SC (2012): Neural circuitry of PTSD with or roreport 13:1255–1258. without mild traumatic brain injury. Neuropharmacology 62:598–606. 26. Phan KL, Liberzon I, Welsh RC, Britton JC, Taylor SF (2003): Habitu- 46. van Wingen GA, Geuze E, Caan MWA, Kozicz T, Olabarriaga SD, ation of rostral anterior cingulate cortex to repeated emotionally salient Denys D, et al. (2012): Persistent and reversible consequences of pictures. Neuropsychopharmacology 28:1344–1350. combat stress on the mesofrontal circuit and cognition. Proc Natl 27. Patel R, Spreng RN, Shin LM, Girard TA (2012): Neurocircuitry models Acad Sci U S A 109:15508–15513. of posttraumatic stress disorder and beyond: A meta-analysis of 47. McCabe C, Mishor Z, Cowen PJ, Harmer CJ (2010): Diminished neural functional neuroimaging studies. Neurosci Biobehav Rev 36:2130– processing of aversive and rewarding stimuli during selective seroto- 2142. nin reuptake inhibitor treatment. Biol Psychiatry 67:439–445. 28. Shea MT, Vujanovic AA, Mansfield AK, Sevin E, Liu F (2010): Post- 48. Tabachnick BG, Fidell LS (2013): Using Multivariate Statistics: Pearson traumatic stress disorder symptoms and functional impairment among New International Edition. Upper Saddle River, NJ: Pearson Higher OEF and OIF National Guard and Reserve veterans. J Traum Stress Education. 23:100–107. 49. Eklund A, Nichols TE, Knutsson H (2016): Cluster failure: Why fMRI 29. Stein NR, Dickstein BD, Schuster J, Litz BT, Resick PA (2012): Tra- inferences for spatial extent have inflated false-positive rates. Proc jectories of response to treatment for posttraumatic stress disorder. Natl Acad Sci U S A 113:7900–7905. Behav Ther 43:790–800. 50. McLaren DG, Ries ML, Xu G, Johnson SC (2012): A generalized form of 30. Pietrzak RH, Averill LA, Abdallah CG, Neumeister A, Krystal JH, Levy I, context-dependent psychophysiological interactions (gPPI): A com- Harpaz-Rotem I (2015): Amygdala-hippocampal volume and the parison to standard approaches. Neuroimage 61:1277–1286. phenotypic heterogeneity of posttraumatic stress disorder: A cross- 51. Keane TM, Fairbank JA, Caddell JM, Zimering RT, Taylor KL, Mora CA sectional study. JAMA Psychiatry 72:396. (1989): Clinical evaluation of a measure to assess combat exposure. 31. Tursich M, Ros T, Frewen PA, Kluetsch RC, Calhoun VD, Lanius RA Psychol Assess 1:53–55. (2015): Distinct intrinsic network connectivity patterns of post- 52. Beck AT, Steer RA, Ball R, Ranieri W (1996): Comparison of Beck traumatic stress disorder symptom clusters. Acta Psychiatr Scand Depression Inventories–Ia and –Ii in psychiatric outpatients. J Pers 132:29–38. Assess 67:588–597. 32. Spielberg JM, McGlinchey RE, Milberg WP, Salat DH (2015): Brain 53. Schwab KA, Ivins B, Cramer G, Johnson W, Sluss-Tiller M, Kiley K, network disturbance related to posttraumatic stress and traumatic et al. (2007): Screening for traumatic brain injury in troops returning brain injury in veterans. Biol Psychiatry 78:210–216. from deployment in Afghanistan and Iraq: initial investigation of the 33. Grupe DW, Wielgosz J, Davidson RJ, Nitschke JB (2016): Neurobio- usefulness of a short screening tool for traumatic brain injury. J Head logical correlates of distinct post-traumatic stress disorder symptom Trauma Rehabil 22:377–389.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI 211 Biological Psychiatry: CNNI Opponent Symptom Cluster Effects on Habituation in PTSD

54. U.S. Department of Health and Human Services, National Institute of and sound in Vietnam combat veterans with and without posttraumatic Mental Health (2008): Mental Health Medications. Bethesda, MD: U.S. stress disorder: A positron emission tomography study. Biol Psychi- Government Printing Ofﬁce. atry 45:806–816. 55. Montagu JD (1963): Habituation of the psycho-galvanic reﬂex during 74. Pissiota A, Frans O, Fernandez M, von Knorring L, Fischer H, serial tests. J Psychosom Res 7:199–214. Fischer H, Fredrikson M (2002): Neurofunctional correlates of post- 56. Kober HH, Barrett LF, Joseph J, Bliss-Moreau E, Lindquist K, traumatic stress disorder: a PET symptom provocation study. Eur Arch Wager TD (2008): Functional grouping and cortical–subcortical in- Psychiatry Clin Neurosci 252:68–75. teractions in emotion: A meta-analysis of neuroimaging studies. 75. Liberzon I, Martis B (2006): Neuroimaging studies of emotional re- Neuroimage 42:998–1031. sponses in PTSD. Ann N Y Acad Sci 1071:87–109. 57. Denny BT, Fan J, Liu X, Guerreri S, Mayson SJ, Rimsky L, et al. (2014): 76. Cohen JE, Shalev H, Hefetz S, Gasho CJ, Shachar LJ, Shelef I, Insula-amygdala functional connectivity is correlated with Friedman A (2013): Emotional brain rhythms and their impairment in habituation to repeated negative images. Soc Cogn Affect Neurosci post-traumatic patients. Hum Brain Mapp 34:1344–1356. 9:1660–1667. 77. Fani N, Jovanovic T, Ely TD, Bradley B, Gutman D, Tone EB, 58. Ishai A, Pessoa L, Bikle PC, Ungerleider LG (2004): Repetition sup- Ressler KJ (2012): Neural correlates of attention bias to threat in post- pression of faces is modulated by emotion. Proc Natl Acad Sci U S A traumatic stress disorder. Biol Psychol 90:134–142. 101:9827–9832. 78. Herzog JI, Niedtfeld I, Rausch S, Thome J, Mueller-Engelmann M, 59. Hampshire A, Chamberlain SR, Monti MM, Duncan J, Owen AM Steil R, et al. (2017): Increased recruitment of cognitive control in the (2010): The role of the right inferior frontal gyrus: Inhibition and presence of traumatic stimuli in complex PTSD. Eur Arch Psychiatry attentional control. Neuroimage 50:1313–1319. Clin Neurosci 71:1–13. 60. Kimble M, Boxwala M, Bean W, Maletsky K, Halper J, Spollen K, 79. Hou C, Liu J, Wang K, Li L, Liang M, He Z, et al. (2007): Brain re- Fleming K (2014): The impact of hypervigilance: Evidence for a forward sponses to symptom provocation and trauma-related short-term feedback loop. J Anxiety Disord 28:241–245. memory recall in coal mining accident survivors with acute severe 61. Sadeh N, Spielberg JM, Warren SL, Miller GA, Heller W (2014): Aber- PTSD. Brain Res 1144:165–174. rant neural connectivity during emotional processing associated with 80. Kim MJ, Chey J, Chung A, Bae S, Khang H, Ham B, et al. (2008): posttraumatic stress. Clin Psychol Sci 2:748–755. Diminished rostral anterior cingulate activity in response to threat- 62. Vythilingam M, Blair KS, McCaffrey D, Scaramozza M, Jones M, related events in posttraumatic stress disorder. J Psychiatr Res Nakic M, et al. (2007): Biased emotional attention in post-traumatic 42:268–277. stress disorder: A help as well as a hindrance? Psychol Med 81. Landré L, Destrieux C, Andersson F, Barantin L, Quidé Y, Tapia G, 37:1445–1455. et al. (2012): Working memory processing of traumatic material in 63. Aupperle RL, Melrose AJ, Stein MB, Paulus MP (2012): Executive women with posttraumatic stress disorder. J Psychiatry Neurosci function and PTSD: Disengaging from trauma. Neuropharmacology 37:87–94. 62:686–694. 82. Moser DA, Aue T, Suardi F, Kutlikova H, Cordero MI, Rossignol AS, 64. Phan KL, Fitzgerald DA, Nathan PJ, Moore GJ, Uhde TW, Tancer ME et al. (2015): Violence-related PTSD and neural activation when seeing (2005): Neural substrates for voluntary suppression of negative affect: emotionally charged male-female interactions. Soc Cogn Affect Neu- A functional magnetic resonance imaging study. Biol Psychiatry rosci 10:645–653. 57:210–219. 83. Christley RM (2010): Power and error: Increased risk of false positive 65. Frank DW, Dewitt M, Hudgens-Haney M, Schaeffer DJ, Ball BH, results in underpowered studies. Open Epidemiol J 3:16–19. Schwarz NF, et al. (2014): Emotion regulation: Quantitative meta- 84. Button KS, Munafò MR (2013): Power failure: Why small sample size analysis of functional activation and deactivation. Neurosci Biobehav undermines the reliability of neuroscience. Nat Rev Neurosci 14:365– Rev 45:202–211. 376. 66. Murakami H, Katsunuma R, Oba K, Terasawa Y, Motomura Y, 85. Solomon Z, Horesh D, Ein-Dor T (2009): The longitudinal course of Mishima K, Moriguchi Y (2015): Neural networks for mindfulness and posttraumatic stress disorder symptom clusters among war veterans. emotion suppression. PLoS One 10:e0128005–e0128018. J Clin Psychiatry 70:837–843. 67. Morawetz C, Bode S, Derntl B, Heekeren HR (2017): The effect of 86. Siegmund A, Wotjak CT (2007): Hyperarousal does not depend on strategies, goals and stimulus material on the neural mechanisms of trauma-related contextual memory in an animal model of post- emotion regulation: A meta-analysis of fMRI studies. Neurosci Bio- traumatic stress disorder. Physiol Behav 90:103–107. behav Rev 72:111–128. 87. Golub Y, Mauch CP, Dahlhoff M, Wotjak CT (2009): Consequences of 68. Brunetti M, Sepede G, Mingoia G, Catani C, Ferretti A, Merla A, et al. extinction training on associative and non-associative fear in a mouse (2010): Elevated response of human amygdala to neutral stimuli in mild model of posttraumatic stress disorder (PTSD). Behav Brain Res post traumatic stress disorder: Neural correlates of generalized 205:544–549. emotional response. Neuroscience 168:670–679. 88. Costanzi M, Cannas S, Saraulli D, Rossi-Arnaud C, Cestari V (2011): 69. Bryant RA, Kemp AH, Felmingham KL, Liddell B, Olivieri G, Peduto A, Extinction after retrieval: Effects on the associative and nonassociative et al. (2008): Enhanced amygdala and medial prefrontal activation components of remote contextual fear memory. Learn Mem 18:508– during nonconscious processing of fear in posttraumatic stress dis- 518. order: An fMRI study. Hum Brain Mapp 29:517–523. 89. Shipherd JC, Beck JG (2005): The role of thought suppression in 70. Felmingham K, Williams LM, Kemp AH, Liddell B, Falconer E, posttraumatic stress disorder. Behav Ther 36:277–287. Peduto A, Bryant RA (2010): Neural responses to masked fear faces: 90. Schell TL, Marshall GN, Jaycox LH (2004): All symptoms are not Sex differences and trauma exposure in posttraumatic stress disorder. created equal: The prominent role of hyperarousal in the natural course J Abnorm Psychol 119:241–247. of posttraumatic psychological distress. J Abnorm Psychol 113:189– 71. van Rooij SJH, Rademaker AR, Kennis M, Vink M, Kahn RS, Geuze E 197. (2015): Neural correlates of trauma-unrelated emotional processing in 91. Seppälä EM, Nitschke JB, Tudorascu DL, Hayes A, Goldstein MR, war veterans with PTSD. Psychol Med 45:575–587. Nguyen DTH, et al. (2014): Breathing-based meditation decreases 72. Williams LM, Kemp AH, Felmingham K, Barton M, Olivieri G, Peduto A, posttraumatic stress disorder symptoms in U.S. military veterans. et al. (2006): Trauma modulates amygdala and medial prefrontal re- J Traum Stress 27:397–405. sponses to consciously attended fear. Neuroimage 29:347–357. 92. Kurdi B, Lozano S, Banaji MR (2016): Introducing the Open Af- 73. Bremner JD, Staib LH, Kaloupek D, Southwick SM, Soufer R, fective Standardized Image Set (OASIS). Behav Res Meth Charney DS (1999): Neural correlates of exposure to traumatic pictures 49:457–470.

212 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:203–212 www.sobp.org/BPCNNI Biological Psychiatry: Archival Report CNNI

Cognitive Control Network Homogeneity and Executive Functions in Late-Life Depression

Matteo Respino, Matthew J. Hoptman, Lindsay W. Victoria, George S. Alexopoulos, Nili Solomonov, Aliza T. Stein, Maria Coluccio, Sarah Shizuko Morimoto, Chloe J. Blau, Lila Abreu, Katherine E. Burdick, Conor Liston, and Faith M. Gunning

ABSTRACT BACKGROUND: Late-life depression is characterized by network abnormalities, especially within the cognitive control network. We used alternative functional connectivity approaches, regional homogeneity (ReHo) and network homogeneity, to investigate late-life depression functional homogeneity. We examined the association between cognitive control network homogeneity and executive functions. METHODS: Resting-state functional magnetic resonance imaging data were analyzed for 33 older adults with depression and 43 healthy control subjects. ReHo was performed as the correlation between each voxel and the 27 neighbor voxels. Network homogeneity was calculated as global brain connectivity restricted to 7 networks. T-maps were generated for group comparisons. We measured cognitive performance and executive functions with the Dementia Rating Scale, Trail-Making Test (A and B), Stroop Color Word Test, and Digit Span Test. RESULTS: Older adults with depression showed increased ReHo in the bilateral dorsal anterior cingulate cortex (dACC) and the right middle temporal gyrus, with no signiﬁcant ﬁndings for network homogeneity. Hierarchical linear regression models showed that higher ReHo in the dACC predicted better performance on Trail-Making Test B (p , .001; R2 = .49), Digit Span Backward (p , .05; R2 = .23), and Digit Span Total (p , .05; R2 = .23). Used as a seed, the dACC cluster of higher ReHo showed lower functional connectivity with bilateral precuneus. CONCLUSIONS: Higher ReHo within the dACC and right middle temporal gyrus distinguish older adults with depression from control subjects. The correlations with executive function performance support increased ReHo in the dACC as a meaningful measure of the organization of the cognitive control network and a potential compensatory mechanism. Lower functional connectivity between the dACC and the precuneus in late-life depression suggests that clusters of increased ReHo may be functionally segregated. Keywords: Aging, Cognitive control, Depression, Executive functions, Functional connectivity, MRI https://doi.org/10.1016/j.bpsc.2019.10.013

A network dysfunction model may help explain the neurobi- Although region-of-interest, seed-based rsFC studies have ological underpinnings of late-life depression (LLD) (1). provided valuable insights into network abnormalities in LLD, Resting-state functional connectivity (rsFC) measures the there are several limitations to this approach. Seed-based temporal coherence between single voxels or groups of approaches require a priori decisions for seed placement. voxels at rest (2). Region-of-interest, seed-based rsFC studies These a priori decisions are, by definition, hypothesis driven often show a lower rsFC in the cognitive control network and introduce a potential bias to network analyses. In addition, (CCN) among patients with LLD compared with healthy con- the placement of seeds of varying sizes and coordinates to trol subjects (3,4). interrogate the same network can result in different rsFC pat- Executive dysfunction is a clinical expression of disrupted terns. These different patterns contribute to ambiguity when CCN functions. Patients with LLD and executive dysfunction, interpreting results both within and across laboratories. or depression-executive dysfunction syndrome, have difficulty Furthermore, seed selection can overlook the complexity of a inhibiting irrelevant stimuli and engaging in goal-directed be- network because within-network activity often depends on haviors (5). Depression-executive dysfunction syndrome is multiple functionally heterogeneous subregions (10). Although characterized by disability and poor response to antidepres- the model-free independent component analysis (11) ad- sants (6–8). Impairments in executive functions often persist dresses some of these seed-based limitations, independent even after remission of depression (9). Thus, understanding component analysis often relies on relatively arbitrary judg- specific patterns of abnormal rsFC in LLD may inform novel ments to select meaningful patterns among the automatically treatment approaches. generated components (12).

SEE COMMENTARY ON PAGE 138

ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. 213 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI Biological Psychiatry: CNNI Functional Homogeneity and Executive Functions in LLD

Alternative rsFC analytic techniques can address some of clusters of CCN abnormal functional homogeneity (ReHo, the limitations of both seed-based and independent compo- NeHo) as data-driven seeds for exploratory whole-brain FC nent analysis approaches by focusing on “homogeneity.” analysis. Homogeneity, as measured by resting-state functional magnetic resonance imaging (fMRI), is conceptualized as the METHODS AND MATERIALS synchronization of all voxels in a well-defined area of the brain. The term homogeneity has been applied to two main resting- Participants state fMRI approaches: regional homogeneity (ReHo) (13) This study included 33 participants with LLD and 43 elderly and network homogeneity (NeHo) (14). nonpsychiatric comparison participants (age range 60–85 ReHo (13) is a measure of the temporal synchronization of years). The group with depression consisted of participants blood oxygen level–dependent time series between every who met DSM-IV criteria for MDD without psychotic features, voxel and its nearest neighbors. ReHo provides a measure of with a score of 18 or greater on the 24-item Hamilton local connectivity and has the advantage of probing a well- Depression Rating Scale (HDRS) (28) and a score of 26 or defined anatomical area by assessing the synchronization of greater on the Mini-Mental State Examination (29). Participants spatially adjacent voxels. ReHo allows the definition of were recruited through print and radio advertisements along boundaries between functionally heterogeneous within- with referrals from outpatient mental health providers. Healthy network subregions and the detection of within-network hubs control subjects had no history or presence of any psychiatric of functional abnormality (15). A meta-analysis of ReHo in disorder. All participants signed informed consent as approved adults with major depressive disorder (MDD) suggests wide- by local institutional review boards. spread abnormalities involving primarily areas of the default Participants were excluded for the presence of any of the mode network (DMN) (e.g., increased ReHo in the medial following: 1) high suicide risk; 2) any Axis I psychiatric disorder prefrontal cortex) (16). Reports of ReHo in LLD show abnor- other than MDD or generalized anxiety disorder; 3) history of malities involving the CCN and the DMN, but demonstrate psychiatric disorders other than MDD or generalized anxiety inconsistent directionality across samples and do not show a disorder; 4) mild cognitive impairment or dementia; 5) acute or relationship of ReHo to executive functions among older adults severe medical illness; 6) any neurological brain disease; with active depression (17–20). ReHo can locate highly ho- 7) history of electroconvulsive therapy; 8) ongoing treatment mogeneous clusters within the brain and serve as a data- with drugs associated with depression (i.e., steroids, alpha- driven method to identify seeds for rsFC studies (21). methyl-dopa, clonidine, reserpine, tamoxifen, or cimetidine); Furthermore, ReHo correlates with measures of functional or 9) metal implants or other contraindications to MRI. segregation such as clustering coefficient and local efficiency (22) and is thought to have a substantially similar biological Assessment meaning, where higher ReHo (within-region highly synchronous activity) could indicate decreased communication with In participants with LLD, a diagnosis of MDD was made by remote brain regions. Structured Clinical Interview for DSM Disorders criteria. NeHo (14) is a voxelwise measure of the correlation of a Furthermore, depression severity was assessed with the HDRS voxel with all other voxels within a network, thus allowing an prior to study entry. Two clinician investigators specializing in unbiased examination of within-network connectivity. NeHo geriatric psychiatry reached consensus on the diagnosis of can be investigated with a graph-theory–based approach MDD and ruled out the possibility of mild cognitive impairment called restricted global brain connectivity (rGBC) (23). rGBC based on review of neuropsychological tests and overall provides a measure of the connectivity between each voxel function. If mild cognitive impairment was suspected, a and every other voxel in a predefined restricted space (i.e., a comprehensive neuropsychological battery was administered. network mask). As such, rGBC allows the evaluation of the Participants with depression and control subjects were averaged within-network connectivity of an entire network. In administered a baseline neuropsychological battery. Overall adults with MDD, NeHo analyses suggest homogeneity ab- cognitive function was assessed with the Mini-Mental State normalities in prefrontal regions of the CCN and in the DMN Examination (29) and the Dementia Rating Scale (DRS) (30). (24,25). Processing speed and executive function were assessed with Employing connectivity measures less reliant on a priori the Trail Making Test, Parts A and B (TMT-A and TMT-B) (31); assumptions about network organization is particularly the Stroop Color Word Test (32); and the Digit Span Test important in understanding network abnormalities in the aging Forward, Backward, and Total score (33). Vascular risk was brain, because brain long-range connectivity is particularly extracted from the Charlson Comorbidity Index (namely, hy- vulnerable to aging (26) and so are intranetwork and internet- pertension, smoking status, and diabetes items) (34). works connectivity (27). The primary objectives of this article are to examine 1) group MRI Data Acquisition differences in rsFC homogeneity measures in LLD and MRI scans were acquired on a 3T Siemens TiM Trio (Siemens, nonpsychiatric comparison participants and 2) the association Erlangen, Germany) equipped with a 32-channel head coil at between functional homogeneity and executive performance in the Center for Biomedical Imaging and Neuromodulation of the LLD. We hypothesized that 1) the group differences in ReHo Nathan Kline Institute for Psychiatric Research. For patients and NeHo will involve the CCN and 2) CCN homogeneity will who were under antidepressant treatment, 1-week washout be associated with executive performance. Secondarily, we was done before MRI data acquisition. Anatomical imaging investigate broader network organizational levels by using included a dual-echo T1-weighted magnetization prepared

214 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI Biological Psychiatry: Functional Homogeneity and Executive Functions in LLD CNNI

rapid acquisition gradient-echo for coregistration with func- voxelwise manner. Concordance was computed on 27 voxels tional data (repetition time = 2500 ms, echo time = 3.5 ms, slice (the node voxel plus the 26 neighboring voxels), which is thickness = 1 mm, inversion time = 1200 ms, 256 axial slices, suggested as the more appropriate cluster size to cover all matrix = 256 3 256, 1-mm isovoxel, field of view = 256 mm). directions in 3-dimensional (3D) space (15). ReHo maps were Resting-state images were acquired using a single-shot, T2- then standardized using the Fisher r-to-z transformation (zFC). weighted echo planar blood oxygen level–dependent contrast image, which allowed whole brain coverage (repeti- NeHo: rGBC Data Analysis tion time = 2500 ms, echo time = 30 ms, flip angle = 80 , slice DPARSF was used to calculate the time series for each sub- thickness = 3 mm, 38 axial slices, matrix = 72 3 72, 3-mm ject. To calculate the correlation between each voxel time isovoxel, field of view = 216 mm, integrated parallel acquisi- series and all the other voxels within brain networks, the tion techniques factor = 2). Acquisition time was 6 minutes, 15 resulting preprocessed 4D NIfTI files were then used as inputs seconds (150 volumes). Patients were instructed to stay awake in AFNI software (41) using 3dTcorrMap. We used the masking with eyes closed. Wakefulness during the scan was verified at option to restrict the analysis to 7 networks’ liberal masks (42): the end of the resting-state sequence by the MR technician. visual network, somatomotor network, dorsal attention network, ventral attention network, limbic network, frontopar- Image Preprocessing ietal network, and DMN. To examine regional effects that might The preprocessing of resting-state data was performed using reflect local hubs of abnormal within-network connectivity, a Data Processing Assistant for Resting-State fMRI (DPARSF) group-comparison map was created for each of the 7 4.3 Advanced Edition (35), a software plug-in within networks. DPABI_V3.0_171210 (http://rfmri.org/dpabi), which is based on SPM (http://www.fil.ion.ucl.ac.uk/spm/software/spm8). Secondary Analysis: Homogeneity-Seeded The first 5 volumes of the blood oxygen level–dependent Functional Connectivity sequence were discarded to reduce relaxation effects. The Clusters of significant CCN abnormal functional homogeneity remaining images were then corrected for slice timing and (ReHo or NeHo) from previous steps were first saved as binary head motion. T1 images were skull-stripped, coregistered to masks and then entered as seeds in seed-based functional functional images, and segmented into gray matter, white connectivity analysis. FC maps were calculated from pre- matter, and cerebrospinal fluid based on SPM priors. Global processed data in a voxelwise manner and zFCs were calcu- signal regression was not applied because it can bias effects lated. For the seed-to-voxel functional connectivity analysis, on local and long-range correlations (36). In lieu of global signal we used a voxel p-value threshold of p , .01 and cluster regression, CompCor, a component-based noise correction threshold of p , .05. method, was used to reduce the effect of physiological noise (37). Nuisance regression was applied using white matter, Statistical Analysis cerebrospinal fluid, and Friston 24 motion parameters as SPSS Statistics 25 (IBM Corp., Armonk, NY) was used to covariates. The images were segmented and spatially perform 2-sample Student’s tests and chi-squared tests on normalized using DARTEL (38), resampled to 3-mm isovoxels, t demographic and clinical variables. DPARSF Statistical Analysis and smoothed with a Gaussian kernel of 4-mm full width half tool was used to perform a 2-sample Student’s test on the maximum. In the ReHo data analysis (13) only, smoothing was t groups’ ReHo maps, the rGBC maps and zFC maps with sex, not performed during preprocessing but just after ReHo education, and mean framewise displacement as covariates. calculation. Lastly, the resulting fMRI data were filtered (0.01 , Sex was regressed based on the evidence of its influence on , 0.1 Hz) to reduce low-frequency drift and high-frequency f ReHo values at rest (43). To further control for the effect of physiological noise. motion artifacts, mean framewise displacement values were To minimize physiologic sources of head displacement, we also used as a covariate. Finally, education was regressed adopted the following exclusion strategies: 1) Participants with owing to the significant difference between the 2 groups. Age Van Dijk framewise displacement (FD) (mean relative root- was not included as a covariate because the 2 samples are well mean-square) (39) greater than 2 SD above the mean were matched on that variable (Table 1). Multiple comparison excluded (40), as well as those with absolute mean FD . 0.2 correction was carried out using Gaussian random field with mm. 2) Participants with single-frame maximum head motion smoothness estimated on statistical image directly. Given the of .2.5 mm of displacement in any direction (x, y, or z) or 2.5 recent debate on spatial clustering (44,45), we adopted the of angular motion were also excluded. While this is a some- following threshold for both the ReHo and NeHo analyses: voxel what lenient motion correction approach, these parameters value , .005, cluster value , .05, 1-tailed. We used 1-tailed were selected because a more conservative approach may not p p Student’s tests in all comparisons, because we had a priori have been appropriate for our older sample with difficulties in t hypotheses that ReHo would be higher in the CCN and DMN, executive function. as per the review by Iwabuchi et al. (16). Similarly, we hypothesized that for ReHo-seeded functional connectivity, higher ReHo Data Analysis ReHo would be associated with lower functional connectivity. ReHo was performed with DPARSF (35). Individual ReHo maps To examine the association between ReHo or NeHo abnor- were generated based on the Kendall’s coefficient of concor- malities and cognitive performance variables, zFC connectivity dance, which is computed as the correlation between the time values were extracted at the subject level from each signifi- series of each voxel and those of its nearest neighbors (13)ina cantly different cluster and input into SPSS for further analysis.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI 215 Biological Psychiatry: CNNI Functional Homogeneity and Executive Functions in LLD

Table 1. Comparison Between Older Adults With Depression and Healthy Control Subjects on Demographic, Clinical, and Cognitive Variables Variable LLD Group (n = 33) HC Group (n = 43) t Test p Value Age, Years 72.2 (6.6) 73.4 (6.5) 0.75 ns Female, n (%) 21 (63.6%) 25 (58.1%) 0.24a ns Education, Years 14.4 (2.7) 17.0 (2.0) 4.65 ,.001 HDRS Total Score 22.7 (4.3) 1.1 (1.1) 227.9 ,.001 DRS Total (n = 75) 138.8 (4.2) 140.8 (3.1) 2.3 ,.05 DRS Attention 36.0 (1.2) 36.3 (1.4) 1.0 ns DRS I/P (n = 75) 36.2 (1.5) 36.7 (0.7) 1.6 ns DRS Construction 5.8 (0.6) 5.9 (0.4) 1.0 ns DRS Conceptualization 37.3 (1.7) 37.5 (2.9) 0.2 ns DRS Memory 23.3 (2.4) 24.5 (0.7) 2.8 ,.01 TMT-A (n = 74) 62.5 (51.0) 38.6 (14.2) 22.6 ,.05 TMT-B (n = 74) 137.4 (74.7) 86.3 (33.4) 23.6 ,.001 Stroop Word 85.4 (16.6) 98.5 (16.7) 3.4 ,.01 Stroop Color 55.2 (11.8) 64.3 (11.6) 3.4 ,.01 Stroop Color Word 31.0 (8.3) 35.0 (9.2) 2.0 ,.05 Digit Span Forward (n = 75) 7.1 (2.3) 8.2 (2.4) 2.0 .052 Digit Span Backward (n = 74) 5.9 (2.2) 6.9 (1.8) 2.2 ,.05 Digit Span Total (n = 74) 13.0 (3.8) 15.1 (3.5) 2.6 ,.05 Vascular Risk (n = 73) 1.0 (1.1) 0.6 (0.8) 21.5 ns Mean FD 0.071 (0.03) 0.068 (0.03) 20.43 ns Values are mean (SD) unless otherwise indicated. DRS, Dementia Rating Scale; FD, framewise displacement, measured as van Dijk's mean relative root-mean-square; HC, healthy control; HDRS, Hamilton Depression Rating Scale; I/P, Initiation/Perseveration; LLD, late-life depression; ns, not signiﬁcant; TMT, Trail-Making Test (Parts A and B). ac2 test.

Correlations between significant clusters’ connectivity values subjects (n = 43), the group with depression (n = 33) had fewer and those cognitive variables that showed to be significantly years of education; greater HDRS total score; and poorer different between groups were calculated, as well as with age, performance on DRS Total, DRS Memory, TMT-A, TMT-B, education, depression severity, and vascular risk. Cognitive Stroop Word, Stroop Color, Stroop Color Word, Digit Span variables that were found to be significantly correlated with Backward, and Digit Span Total (all p , .05). We also report ReHo or NeHo were included in listwise hierarchical linear that vascular risk and framewise displacement did not signifi- regression models as outcome variables. Each regression cantly differ between groups. model included 2 steps, the first accounting for the effect of covariates and the second for the independent effect of ReHo ReHo in Patients With Depression Versus Control or NeHo. Subjects A significant group difference between older adults with Exploratory Analysis. We used SPSS Statistics 25 to depression and control subjects in ReHo was detected in the conduct 2 additional exploratory analyses. First, we examined dorsal anterior cingulate cortex (dACC) bilaterally, a region on sex as a moderator in the association of ReHo and cognitive the edge of paracingulate and medial frontal gyri, and the right performance. The hierarchical regression models described middle temporal gyrus (rMTG), with patients with depression above were run with an additional third step that entered a having greater ReHo relative to healthy control subjects. Re- ReHo 3 sex interaction term to identify the moderating effect sults are summarized in Table 2. Figure 1 shows the depressed of sex differences in accounting for the relationship between versus control groups map comparison. ReHo and cognitive variables. We also examined the association between ReHo and cognitive variables that did not differ Association Between ReHo and Executive Function between groups at baseline. ReHo values were extracted at the subject level and correlated with cognitive scores using To investigate the relationship between increased ReHo in Spearman’s rank-order correlation approach. the 2 clusters (ACC and rMTG) and executive functions, Spearman correlations were performed within the two groups. Within older adults with depression, ACC ReHo RESULTS significantly correlated with TMT-B (r = 2.552, p = .001), Digit Span Backward (r =.459,p , .01), Digit Span Total Participants score (r =.468,p , .01), and education (r = .509, p , .01) Demographic, clinical, and cognitive characteristics of the (all false discovery rate [FDR] q , 0.05); the only correlation sample are summarized in Table 1. Compared with control with rMTG ReHo that survived FDR correction was with

216 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI Biological Psychiatry: Functional Homogeneity and Executive Functions in LLD CNNI

Table 2. Clusters of Abnormal ReHo and ReHo-Seeded FC (Older Adults With Depression Versus Healthy Control Subjects) Clusters R/L Cluster Size Peak MNI Coordinates (x, y, z) Peak Effect Size Increased ReHo ACC Bilateral 80 26, 39, 36 4.14a Middle temporal gyrus Right 96 57, 221, 26 4.76a ACC ReHo-Seeded FC Precuneus Bilateral 208 212, 251, 69 23.56b ACC, anterior cingulate cortex; FC, functional connectivity; GRF, Gaussian random ﬁeld; MNI, Montreal Neurological Institute; ReHo, regional homogeneity; R/L, right/left. aVoxel p , .005, GRF-corrected cluster p , .05. bVoxel p , .01, GRF-corrected cluster p , .05.

education (r =.464,p , .01). Within healthy control subjects, executive functions independent of depression severity. In a no correlation survived FDR correction. Scatterplots of the second step, we added ACC ReHo to the model to examine relation between ACC ReHo and executive measures are its contribution to the explained variance above and beyond displayed in Figure 2. covariates. In models 2 and 3 (Digit Span Backward and Digit Based on these correlations, 3 regression models were Span Total as outcomes, respectively), we first entered only utilized. We respectively entered TMT-B, Digit Span Back- HDRS as covariate because age and education did not ward, and Digit Span Total scores as outcome variables in significantly correlate with Digit Span Backward or Digit Span each model (Table 3). In model 1 (TMT-B as outcome), we first Total performance. Among patients, in model 1, ReHo entered age, education, and HDRS total score as covariates. significantly predicted TMT-B scores (DR2 =.08,p =.047) Age and education because they correlated with TMT-B above and beyond covariates; this final model predicted (respectively, age: r =.365,p , .05; education: r = 2.433, almost half of the variance in TMT-B scores (R2 = .49; F = p , .05) and HDRS total score to detect the effect of ReHo on 6.61, p , .001). In model 2 (outcome Digit Span Backward) and model 3 (outcome Digit Span Total score) ACC ReHo significantly predicted scores on cognitive outcomes (model 2: R2 =.23,F =4.35,p = .022; model 3: R2 = .23, F =4.40,p = .021) above and beyond HDRS (model 2: DR2 =.20,p =.011; model 3: DR2 = .18, p = .014). None of these models were significant in the healthy comparisons group.

ReHo Exploratory Analyses. We performed exploratory analyses to examine the potential moderator effects of sex differences on the association between ReHo and measures of executive function and to examine the relationship between ReHo and performance on tasks that did not show group differences at baseline between participants with depression and healthy control subjects. We performed exploratory regression models with sex as a moderator. These models included the variables and covariates reported above and in Table 3. In model 1 (TMT-B as outcome), age, education, and HDRS total were entered as covariates in the first step, ACC ReHo was entered as a predictor in the next step, and an ACC ReHo 3 sex interaction term was entered in the next step to explore the moderating effect. In model 1, the addition of the ACC ReHo 3 sex moderator was not significant (DR2 = 2.11, p = .12). In models 2 and 3 (Digit Span Backward and Digit Span Total as outcomes, respectively), HDRS total was entered as a covariate in the first step, ACC ReHo was entered as a predictor in the next step, and an ACC ReHo 3 sex interaction term was entered in

FPO the next step to explore the moderating effect. In model 2, the = addition of the ACC ReHo 3 sex moderator was also not significant (DR2 = 2.004, p = .70). In model 3, the addition of web 4C the ACC ReHo 3 sex moderator was also not significant Figure 1. Clusters of significant regional homogeneity (ReHo) increase in (D 2 = 2.03, = .34). These exploratory results suggest that older adults with depression compared with healthy control subjects. R p sex differences did not account for a significant amount of the Clusters Gaussian random field–corrected voxel p , .005, cluster p , .05. ACC, anterior cingulate cortex; L, left; R, right; R-MTG, right middle temporal variance in explaining the observed relationship between ACC gyrus; T, t score. ReHo and executive function.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI 217 Biological Psychiatry: CNNI Functional Homogeneity and Executive Functions in LLD FPO = web 4C Figure 2. Scatterplots of the relations between dorsal anterior cingulate (ACC) regional homogeneity (ReHo) values and executive performance in older adults with depression and healthy control subjects. (Left) A 3-dimensional view of dorsal ACC cluster of increased ReHo in patients compared with control subjects (BrainNet Viewer) (62). (Center to right) Scatterplots of the relation between ACC ReHo r-to-z values to executive performance. TMT-B, Trail Making Test B.

We also examined the relationship of ACC and MTG ReHo connectivity involving the bilateral precuneus. Given the to the measures of the cognitive battery that did not show exploratory nature of this secondary analysis, here we applied signiﬁcant group differences at baseline (i.e., the following a less conservative threshold of voxel p value , .01, cluster p subscales of the DRS: Attention, Initiation/Perseveration, value , .05 (Figure 3). Within older adults with depression, zFC Construction, and Conceptualization). There were no signiﬁ- values extracted from this area correlated with DRS total score cant correlations between ACC or MTG ReHo and DRS sub- (r = 2.367, p , .05) and education (r = 2.353, p , .05), scale scores in the LLD group or in the healthy control although neither survived FDR correction. subjects.

NeHo in Patients With Depression Versus Control DISCUSSION Subjects The principal finding of this study is that relative to healthy NeHo, as measured by rGBC, did not significantly differ be- elderly control subjects, elderly participants with depression tween participants with depression and control subjects in any display increased functional ReHo in the CCN. Specifically, local of the 7 networks analyzed. Among the network group com- homogeneity of the dACC, as measured by ReHo, was higher in parison maps, no regional effect survived Gaussian random participants with depression relative to control subjects. field correction in any of the 7 networks of rGBC. Therefore, Furthermore, increased ReHo in the ACC predicted better NeHo values were not extracted and correlations with cogni- cognitive control performance as measured by brief clinical tive performance were not performed. measures of working memory and task switching. Another non- CCN region, the rMTG, displayed increased functional ReHo but Secondary Analysis: ReHo-Seeded Functional did not predict performance on objective clinical measures of Connectivity cognitive control. Our results expand on previous reports of We used the ACC cluster of increased ReHo as a data-driven aberrant spontaneous activation at rest in the CCN of individuals seed to explore seed-based zFC maps in both groups. Group with LLD (3,4,46). However, the application of ReHo in this study comparisons were performed through independent 2-sample improves on some of the limitations inherent to previously used Student’s t test and revealed a region of decreased seed-based approaches.

Table 3. Final Regression Models R2 DR2 B SE b tpValue Model 1: TMT-B as Outcome Variable Age –– 3.15 1.61 .28 1.96 .061 Education ––29.27 4.36 2.34 22.12 .043 HDRS Total Score ––22.64 2.40 2.15 21.10 .282 ACC ReHo .49 .08 272.45 34.82 2.34 22.08 .047 Overall Model: F = 6.61, p , .001 Model 2: Digit Span Backward as Outcome Variable HDRS Total Score ––20.07 0.08 2.14 20.88 .383 ACC ReHo .23 .20 2.72 1.00 .44 2.71 .011 Overall Model: F = 4.35, p = .022 Model 3: Digit Span Total Score as Outcome Variable HDRS Total Score ––20.16 0.14 2.19 21.16 .255 ACC ReHo .23 .18 4.57 1.75 .43 2.60 .014 Overall Model: F = 4.40, p = .021 ACC, anterior cingulate cortex; HDRS, Hamilton Depression Rating Scale; ReHo, regional homogeneity; TMT-B, Trail-Making Test Part B.

218 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI Biological Psychiatry: Functional Homogeneity and Executive Functions in LLD CNNI

ReHo correlates with measures of “functional segregation” such as local efficiency and clustering coefficients (22). Therefore, increased ReHo may have similar biological significance reflecting a restriction of information transfer to spatially close areas, at the expense of broader connections with more distant brain regions. Regions of increased ReHo may be functionally segregated from distant hubs and respond to lesser information transfer by synchronizing their activity. Within the aging brain, long-range connections with distant hubs, rather than short-range connections, are preferentially impaired (26,27). Of note, we did not observe significant relationships between ReHo and executive functions in the elderly adults without depression. It is possible that the functional segregation that higher ReHo may represent in older

FPO adults with depression, which is reflected as weaker executive = performance, may be below a “threshold” in healthy older adults that does not contribute to detectable difficulties on web 4C measures of executive functions. Figure 3. Clusters of significantly decreased regional homogeneity– Converging evidence shows reduced functional connectiv- seeded functional connectivity in older adults with depression compared with healthy control subjects. A 3-dimensional view (BrainNet Viewer) (62)of ity between ACC and spatially distal brain regions in LLD precuneus clusters Gaussian random field–corrected voxel p , .01, cluster (3,46). We found that ReHo-seeded ACC has decreased con- p , .05. L, left; R, right; T, t score. nectivity with posterior regions of the bilateral precuneus. The specific region of the precuneus to which we observed Within older adults with depression, higher ReHo of the dACC decreased functional connectivity is a posterior component of was primarily associated with better cognitive flexibility and the frontoparietal network. The precuneus is believed to act in working memory, as revealed by the significant contribution of concert with more anterior aspects of the frontoparietal dACC ReHo to explain variance in performance on TMT-B, Digit network to support cognitive control processes, and this Span Backward, and Digit Span Total. Increased ReHo in the dACC/precuneus disruption may reflect aging-related disrup- rMTG did not relate to measures of cognitive control. The relation of long-range connections of the frontoparietal network tionship between dACC ReHo and stronger cognitive control (53–55). Furthermore, this disruption in connectivity to the performance in older adults with depression indicates that ReHo precuneus converges with the putative role of the precuneus in may be a meaningful neurobiological measure of the organization higher-order cognitive functions relevant to depression, such of the CCN at rest. Because LLD patients show increased ReHo as self-referential thinking and first-person perspective taking values in the dACC, which in turns relates to better executive (56), as well as sustained attention, cognitive flexibility, and functions, we propose that a higher ReHo in the ACC may be a task switching (57). ReHo may play a compensatory role compensatory mechanism necessary to support select execu- through the segregation of information processing at the local tive functions in LLD. level, in accordance with prior reports indicating that the Although there have been some reports of abnormal ReHo strength of ACC-seeded internetwork connectivity relates to regions within the CCN in older adults with depression the expression of depression (58). (17–19,47), these findings have not been consistently linked to In contrast with the examination of potential widespread, the clinical expression of poor cognitive control. A preliminary across-network ReHo abnormalities, our approach to NeHo study (20), conducted in a small sample of patients in remis- (rGBC) allowed us to examine group differences in the average sion, revealed a negative relation between ReHo of other CCN connectivity of 7 independent networks. Contrary to our hy- regions (superior frontal gyrus) and TMT-B performance. potheses, the comparison of rGBC in older adults with Our findings support the association between executive depression versus in control subjects did not show signifi- functions in depression and CCN connectivity (3,46). Within cantly different clusters of across-voxel connectivity in any of the CCN, we observed increased ReHo in the dACC. This the 7 networks analyzed. This result suggests that averaging subregion acts in concert with other components of the the correlation between voxels of vast brain regions may mask CCN, and in healthy individuals it is associated with subtle effects detectable at a more local level. Although con- increased engagement of other cognitive control regions ducting a functional homogeneity analysis at the network level (e.g., posterior parietal cortices, lateral prefrontal cortex) avoids the bias of seed selection and placement process, in (10,48,49). Across a number of psychiatric conditions, gray this case the application of a priori network templates may matter volume reduction of the dorsal ACC relates to exec- have contributed to the lack of significant group differences. utive dysfunction (48). Dorsal ACC activation during cognitive The use of data-driven network masks, in place of a priori control tasks may predict subsequent cognitive decline in network templates, could facilitate further studies in identifying LLD (50) and is associated with conflict monitoring during foci of abnormal NeHo in the aging brain. Stroop tasks (51). Furthermore, dorsal ACC volume predicts Our findings should be interpreted within the context of its treatment response in LLD (52), consistent with the idea that limitations. First, the cross-sectional nature of the study does the dACC may be an integral node of the CCN that plays a not allow us to infer causality. We applied a somewhat lenient role in illness course. motion correction threshold; this was done to avoid possible

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI 219 Biological Psychiatry: CNNI Functional Homogeneity and Executive Functions in LLD

sample bias, because more cognitively impaired subjects tend Data presented in this manuscript was presented at the 2017 Annual to move more in the scanner (59); to mitigate motion effect, we Meeting of the Society of Biological Psychiatry. accounted for motion in both the preprocessing and the GSA serves as a member of the Speakers Bureau for Allergan, Plc, Otsuka Pharmaceutical Co., Ltd., Sunovion Pharmaceuticals Inc., and group-level analyses. An additional methodological limitation is Takeda Pharmaceutical Company Ltd. All other authors report no biomed- the length of the resting-state scan (6 minutes, 15 seconds), ical financial interests or potential conflicts of interest. which is of a relatively short duration by current standards. ClinicalTrials.gov: Emotional and Cognitive Control in Late-Onset Work by Zuo et al. (60) supports the stationarity of ReHo over Depression; https://clinicaltrials.gov/ct2/show/NCT01728194? time, showing high intrascan reliability across multiple scan term=NCT01728194&draw=2&rank=1; NCT01728194. sessions with resting-state scans of a shorter duration (5 minutes). However, data quality may be improved with longer scan sequences and improved multiband acquisition param- ARTICLE INFORMATION eters (61). In future studies, it will be important to replicate From the Department of Psychiatry (MR, LWV, GSA, NS, ATS, MC, SSM, LA, these current results in scans of longer length to address CL, FMG), Joan & Sanford I. Weill Medical College of Cornell University, New York; Nathan S. Kline Institute for Psychiatric Research (MJH, CJB), concerns of stationarity. The analyses were limited to the use Orangeburg, New York; and Department of Psychiatry (KEB), Brigham & of a 3D ReHo technique instead of 2D ReHo, which in- Women’s Hospital, Harvard Medical School, Boston, Massachusetts. vestigates the brain surface (rather than brain volume) and may Address correspondence to Faith M. Gunning, Ph.D., 21 Bloomingdale be more specific to the organization of the cortical mantle; Road, White Plains, NY 10605; E-mail: [email protected]. however, we adopted 3D ReHo at 27 voxels (instead of 9 Received Aug 16, 2019; revised Oct 9, 2019; accepted Oct 26, 2019. voxels) to cover all directions in 3D space (15). Additionally, TMT-B is not a “pure” measure of set shifting and may include a processing speed component. Finally, our subjects were REFERENCES instructed to stay awake with eyes closed, a condition un- 1. Tadayonnejad R, Ajilore O (2014): Brain network dysfunction in late-life controlled for arousal level; however, wakefulness status was depression: A literature review. J Geriatr Psychiatry Neurol 27:5–12. verified for each participant at the end of the session. 2. Biswal B, Zerrin Yetkin F, Haughton VM, Hyde JS (1995): Functional connectivity in the motor cortex of resting human brain using echo- Conclusions and Future Directions planar MRI. Magn Reson Med 34:537–541. 3. Alexopoulos GS, Hoptman MJ, Kanellopoulos D, Murphy CF, Lim KO, We observed that local functional homogeneity abnormalities Gunning FM (2012): Functional connectivity in the cognitive control in the ACC, within the CCN, and in the MTG distinguished network and the default mode network in late-life depression. J Affect between older adults with depression and healthy control Disord 139:56–65. subjects. In contrast, averaging the functional connectivity at 4. Andreescu C, Tudorascu DL, Butters MA, Tamburo E, Patel M, Price J, et al. (2013): Resting state functional connectivity and treatment the entire network level did not distinguish between the pa- response in late-life depression. Psychiatry Res 214:313–321. tients with depression and healthy control subjects on any of 5. Alexopoulos GS, Kiosses DN, Klimstra S, Kalayam B, Bruce ML (2002): the 7 networks analyzed. Increased ReHo within the CCN Clinical presentation of the “depression-executive dysfunction syn- predicted better cognitive flexibility and working memory in drome” of late life. Am J Geriatr Psychiatry 10:98–106. LLD, and the ReHo-seeded ACC cluster showed decreased 6. Alexopoulos GS, Kiosses DN, Heo M, Murphy CF, Shanmugham B, connectivity to posterior regions of the DMN. Increased ReHo Gunning-Dixon F (2005): Executive dysfunction and the course of geriatric depression. Biol Psychiatry 58:204–210. within the CCN of older adults with depression may represent a 7. Manning KJ, Alexopoulos GS, Banerjee S, Morimoto SS, Seirup JK, segregation mechanism that attempts to preserve executive Klimstra SA, et al. (2015): Executive functioning complaints and esci- performance. These findings provide a preliminary character- talopram treatment response in late-life depression. Am J Geriatr ization of the brain’s functional topography in LLD and expand Psychiatry 23:440–445. on the existing model of CCN abnormalities in LLD. 8. Potter GG, Kittinger JD, Wagner HR, Steffens DC, Krishnan KRR (2004): These findings can be used to inform novel targets for in- Prefrontal neuropsychological predictors of treatment remission in late-life depression. Neuropsychopharmacology 29:2266–2271. terventions in individuals with depression with executive 9. Rock PL, Roiser JP, Riedel WJ, Blackwell AD (2014): Cognitive dysfunction who do not respond to typical antidepressant impairment in depression: A systematic review and meta-analysis. treatments. Such findings can be used to inform neuro- Psychol Med 44:2029–2040. stimulation approaches, including transcranial magnetic 10. Cole MW, Schneider W (2007): The cognitive control network: Integrated simulation and cognitive interventions that attempt to rescue cortical regions with dissociable functions. Neuroimage 37:343–360. dysfunctional circuitry. Furthermore, future investigation of 11. Fox MD, Raichle ME (2007): Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging. Nat Rev LLD should focus on interactions between local and remote Neurosci 8:700–711. levels of connectivity. 12. Greicius M (2008): Resting-state functional connectivity in neuropsychiatric disorders. Curr Opin Neurol 21:424–430. 13. Zang Y, Jiang T, Lu Y, He Y, Tian L (2004): Regional homogeneity ACKNOWLEDGMENTS AND DISCLOSURES approach to fMRI data analysis. Neuroimage 22:394–400. This work was supported by the National Institute of Mental Health, Grant 14. Uddin LQ, Kelly AMC, Biswal BB, Margulies DS, Shehzad Z, Shaw D, Nos. R01 MH097735 (to FMG, principal investigator) and T32 MH019132 (to et al. (2008): Network homogeneity reveals decreased integrity of GSA, principal investigator). default-mode network in ADHD. J Neurosci Methods 169:249–254. We thank the staff of Weill Cornell’s Institute of Geriatric Psychiatry for 15. Jiang L, Zuo XN (2016): Regional homogeneity: A multimodal, multi- assistance with recruitment of participants and collection of cognitive and scale neuroimaging marker of the human connectome. Neuroscientist mood data. We thank the staff of the Center for Biomedical Neuroimaging 22:486–505. and Brain Modulation of the Nathan Kline Institute for Psychiatric Research 16. Iwabuchi SJ, Krishnadas R, Li C, Auer DP, Radua J, Palaniyappan L for assistance with conduct of the MRI aspects of the described work. (2015): Localized connectivity in depression: A meta-analysis of

220 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI Biological Psychiatry: Functional Homogeneity and Executive Functions in LLD CNNI

resting state functional imaging studies. Neurosci Biobehav Rev 51:77– individual variation in 1000 functional connectomes. Neuroimage 86. 80:246–262. 17. Chen JD, Liu F, Xun GL, Chen HF, Hu MR, Guo XF, et al. (2012): Early 41. Cox RW (1996): AFNI: Software for analysis and visualization of and late onset, first-episode, treatment-naive depression: Same clin- functional magnetic resonance neuroimages. Comput Biomed Res ical symptoms, different regional neural activities. J Affect Disord 29:162–173. 143:56–63. 42. Yeo BTT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, 18. Liu F, Hu M, Wang S, Guo W, Zhao J, Li J, et al. (2012): Abnormal Hollinshead M, et al. (2011): The organization of the human cerebral regional spontaneous neural activity in first-episode, treatment-naive cortex estimated by intrinsic functional connectivity. J Neurophysiol patients with late-life depression: A resting-state fMRI study. Prog 106:1125–1165. Neuropsychopharmacol Biol Psychiatry 39:326–331. 43. Xu C, Li C, Wu H, Wu Y, Hu S, Zhu Y, et al. (2015): Gender differences 19. Yue Y, Yuan Y, Hou Z, Jiang W, Bai F, Zhang Z (2013): Abnormal in cerebral regional homogeneity of adult healthy volunteers: A resting- functional connectivity of amygdala in late-onset depression was state fMRI study. Biomed Res Int 2015:183074. associated with cognitive deficits. PLoS One 8:e75058. 44. Eklund A, Nichols TE, Knutsson H (2016): Cluster failure: Why fMRI 20. Yuan Y, Zhang Z, Bai F, Yu H, Shi Y, Qian Y, et al. (2008): Abnormal inferences for spatial extent have inflated false-positive rates. Proc neural activity in the patients with remitted geriatric depression: A Natl Acad Sci U S A 113:7900–7905. resting-state functional magnetic resonance imaging study. J Affect 45. Slotnick SD (2017): Cluster success: fMRI inferences for spatial extent Disord 111:145–152. have acceptable false-positive rates. Cogn Neurosci 8:150–155. 21. Jiang L, Xu Y, Zhu XT, Yang Z, Li HJ, Zuo XN (2015): Local-to-remote 46. Gandelman JA, Albert K, Boyd BD, Park JW, Riddle M, Woodward ND, cortical connectivity in early-and adulthood-onset schizophrenia. et al. (2019): Intrinsic functional network connectivity is associated with Transl Psychiatry 5:e566. clinical symptoms and cognition in late-life depression. Biol Psychiatry 22. Lee TW, Xue SW (2017): Linking graph features of anatomical archi- Cogn Neurosci Neuroimaging 4:160–170. tecture to regional brain activity: A multi-modal MRI study. Neurosci 47. Ma Z, Li R, Yu J, He Y, Li J (2013): Alterations in regional homogeneity Lett 651:123–127. of spontaneous brain activity in late-life subthreshold depression. 23. Anticevic A, Brumbaugh MS, Winkler AM, Lombardo LE, Barrett J, PLoS One 8:e53148. Corlett PR, et al. (2013): Global prefrontal and fronto-amygdala dys- 48. McTeague LM, Goodkind MS, Etkin A (2016): Transdiagnostic connectivity in bipolar I disorder with psychosis history. Biol Psychiatry impairment of cognitive control in mental illness. J Psychiatr Res 73:565–573. 83:37–46. 24. Abdallah CG, Averill LA, Collins KA, Geha P, Schwartz J, Averill C, 49. Botvinick MM, Cohen JD, Carter CS (2004): Conflict monitoring and et al. (2017): Ketamine treatment and global brain connectivity in major anterior cingulate cortex: An update. Trends Cogn Sci 8:539–546. depression. Neuropsychopharmacology 42:1210–1219. 50. Wang L, Potter GG, Krishnan RKR, Dolcos F, Smith GS, Steffens DC 25. Guo W, Liu F, Zhang J, Zhang Z, Yu L, Liu J, et al. (2014): Abnormal (2012): Neural correlates associated with cognitive decline in late-life default-mode network homogeneity in first-episode, drug-naive major depression. Am J Geriatr Psychiatry 20:653–663. depressive disorder. PLoS One 9:e91102. 51. Kerns JG, Cohen JD, MacDonald AW, Cho RY, Stenger VA, Carter CS 26. Tomasi D, Volkow ND (2012): Aging and functional brain networks. Mol (2004): Anterior cingulate conflict monitoring and adjustments in Psychiatry 17:471, 549–558. control. Science 303:1023–1026. 27. Archer JA, Lee A, Qiu A, Chen S-HA (2016): A comprehensive analysis of 52. Gunning FM, Cheng J, Murphy CF, Kanellopoulos D, Acuna J, connectivity and aging over the adult life span. Brain Connect 6:169–185. Hoptman MJ, et al. (2009): Anterior cingulate cortical volumes and 28. Hamilton MC (1960): A rating scale for depression. J Neurol Neurosurg treatment remission of geriatric depression. Int J Geriatr Psychiatry Psychiatry 23:56–62. 24:829–836. 29. Folstein MF, Folstein SE, McHugh PR (1975): “Mini-mental state”:A 53. La Corte V, Sperduti M, Malherbe C, Vialatte F, Lion S, Gallarda T, et al. practical method for grading the cognitive state of patients for the (2016): Cognitive decline and reorganization of functional connectivity clinician. J Psychiatr Res 12:189. in healthy aging: The pivotal role of the salience network in the predic- 30. Mattis S (1988): Dementia Rating Scale Professional Manual. Lutz, FL: tion of age and cognitive performances. Front Aging Neurosci 8:204. Psychological Assessment Resources. 54. Li R, Yin S, Zhu X, Ren W, Yu J, Wang P, et al. (2017): Linking inter- 31. Reitan RM (1958): Validity of the Trail Making Test as an indicator or individual variability in functional brain connectivity to cognitive abil- organic brain damage. Percept Motor Skills 8:271–276. ity in elderly individuals. Front Aging Neurosci 9:385. 32. Stroop JR (1935): Studies of interference in serial verbal reactions. 55. Davis SW, Dennis NA, Buchler NG, White LE, Madden DJ, Cabeza R J Exp Psychol 18:643–662. (2009): Assessing the effects of age on long white matter tracts using 33. Wechsler D (2014): Weschler Adult Intelligence Scale - Fourth Edition diffusion tensor tractography. Neuroimage 46:530–541. (WAIS-IV). San Antonio, TX: NCS, Pearson, 22, 498. 56. Cavanna AE, Trimble MR (2006): The precuneus: A review of its 34. Charlson ME, Pompei P, Ales KL, MacKenzie CR (1987): A new functional anatomy and behavioural correlates. Brain 129:564–583. method of classifying prognostic comorbidity in longitudinal studies: 57. Piguet C, Cojan Y, Sterpenich V, Desseilles M, Bertschy G, Vuilleumier P Development and validation. J Chronic Dis 40:373–383. (2016): Alterations in neural systems mediating cognitive flexibility and 35. Yan CG, Zang YF (2010): DPARSF: A MATLAB toolbox for “pipeline” inhibition in mood disorders. Hum Brain Mapp 37:1335–1348. data analysis of resting-state fMRI. Front Syst Neurosci 4:13. 58. Sheline YI, Price JL, Yan Z, Mintun MA (2010): Resting-state functional 36. Saad ZS, Gotts SJ, Murphy K, Chen G, Jo HJ, Martin A, et al. (2012): MRI in depression unmasks increased connectivity between networks Trouble at rest: How correlation patterns and group differences become via the dorsal nexus. Proc Natl Acad Sci U S A 107:11020–11025. distorted after global signal regression. Brain Connect 2:25–32. 59. Wylie GR, Genova H, Deluca J, Chiaravalloti N, Sumowski JF (2014): 37. Behzadi Y, Restom K, Liau J, Liu TT (2007): A component based noise Functional magnetic resonance imaging movers and shakers: Does correction method (CompCor) for BOLD and perfusion based fMRI. subject-movement cause sampling bias? Hum Brain Mapp 35:1–13. Neuroimage 37:90–101. 60. Zuo XN, Xu T, Jiang L, Yang Z, Cao XY, He Y, et al. (2013): Toward 38. Klein A, Andersson J, Ardekani BA, Ashburner J, Avants B, Chiang MC, reliable characterization of functional homogeneity in the human brain: et al. (2009): Evaluation of 14 nonlinear deformation algorithms applied Preprocessing, scan duration, imaging resolution and computational to human brain MRI registration. Neuroimage 46:786–802. space. Neuroimage 65:374–386. 39. van Dijk KRA, Sabuncu MR, Buckner RL (2012): The influence of 61. Zuo XN, Xing XX (2014): Test-retest reliabilities of resting-state FMRI head motion on intrinsic functional connectivity MRI. Neuroimage measurements in human brain functional connectomics: A systems 59:431–438. neuroscience perspective. Neurosci Biobehav Rev 45:100–118. 40. Yan CG, Craddock RC, Zuo XN, Zang YF, Milham MP (2013): Stan- 62. Xia M, Wang J, He Y (2013): BrainNet Viewer: A network visualization dardizing the intrinsic brain: Towards robust measurement of inter- tool for human brain connectomics. PLoS One 8:e68910.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:213–221 www.sobp.org/BPCNNI 221 Biological Psychiatry: CNNI Archival Report

Inhibition-Related Cortical Hypoconnectivity as a Candidate Vulnerability Marker for Obsessive-Compulsive Disorder

Adam Hampshire, Ana Zadel, Stefano Sandrone, Eyal Soreq, Naomi Fineberg, Edward T. Bullmore, Trevor W. Robbins, Barbara J. Sahakian, and Samuel R. Chamberlain

ABSTRACT BACKGROUND: Obsessive-compulsive disorder (OCD) is a prevalent neuropsychiatric condition, with biological models implicating disruption of cortically mediated inhibitory control pathways, ordinarily serving to regulate our environmental responses and habits. The aim of this study was to evaluate inhibition-related cortical dysconnectivity as a novel candidate vulnerability marker of OCD. METHODS: In total, 20 patients with OCD, 18 clinically asymptomatic first-degree relatives of patients with OCD, and 20 control participants took part in a neuroimaging study comprising a functional magnetic resonance imaging stop signal task. Brain activations during the contrasts of interest were cluster thresholded, and a three-dimensional watershed algorithm was used to decompose activation maps into discrete clusters. Functional connections between these key neural nodes were examined using a generalized psychophysiological interaction model. RESULTS: The three groups did not differ in terms of age, education level, gender, IQ, or behavioral task parameters. Patients with OCD exhibited hyperactivation of the bilateral occipital cortex during the task versus the other groups. Compared with control participants, patients with OCD and their relatives exhibited significantly reduced connectivity between neural nodes, including frontal cortical, middle occipital cortical, and cerebellar regions, during the stop signal task. CONCLUSIONS: These findings indicate that hypoconnectivity between anterior and posterior cortical regions during inhibitory control represents a candidate vulnerability marker for OCD. Such vulnerability markers, if found to generalize, may be valuable to shed light on etiological processes contributing not only to OCD but also obsessive- compulsive–related disorders more widely. Keywords: Compulsivity, Disinhibition, Inhibition, OCD, Phenotype, Phenotyping https://doi.org/10.1016/j.bpsc.2019.09.010

Obsessive-compulsive disorder (OCD) constitutes a global mental disorders, new insights may be gleaned into underlying public health concern (1–3) and has been estimated to affect causal mechanisms, including genetic ones, cutting across 2% to 3% of the population worldwide (4,5). Family and twin conventionally discrete obsessive-compulsive and related studies have provided strong evidence of a heritable contri- disorders (13,14). bution to the disorder (6), yet attempts to identify specific In prior work, it was suggested that objective measures of genetic loci have met with only partial success. For example, loss of inhibitory control constitute candidate latent pheno- particular single nucleotide polymorphisms regulating cortical types for OCD (15). Deficits on neuropsychological tasks of (especially serotonergic and dopaminergic) neurotransmission motor inhibition, including the stop signal task (SST) (16,17), have been implicated in OCD, but inconsistently and typically have been observed in patients with OCD versus control par- with individually small effect sizes (7). It has been proposed ticipants, as now also shown by a meta-analysis (18). These that such limitations may be overcome in the future by using deficits have also been found in clinically asymptomatic first- intermediate biomarkers such as those combining imaging and degree relatives of patients with OCD in several studies cognition (8–11). Fundamentally, OCD can be considered as (16,19), highlighting their potential value as intermediate the mechanistic end point of underlying psychological and phenotypic markers of vulnerability. Cortico-subcortical cir- brain processes (12). Intermediate-level, biologically grounded cuits have been centrally implicated in OCD symptomatology vulnerability markers for OCD are lacking. By identifying latent (20). While initial OCD models focused on the prefrontal cortex, vulnerability markers (phenotypes) linked with underpinning recent data implicate other cortical regions and the cerebellum psychological processes contributing to a range of related in their pathophysiology (11,21–24). In a recent meta-analysis

222 ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI ISSN: 2451-9022 Biological Psychiatry: Inhibition-Related Dysconnectivity in OCD CNNI

of the available functional imaging literature, OCD was asso- Adult Reading Test estimates IQ. For patients with OCD, ciated with hypoactivation during inhibitory control tasks in the symptom severity was assessed via interview using the Yale- anterior cingulate cortex, anterior insula/frontal operculum, Brown Obsessive-Compulsive Scale (37). supramarginal gyrus, orbitofrontal cortex, and thalamus/ Inclusion criteria across all groups were being of adult age, caudate (25). The SST is contingent on frontal lobe integrity being right-handed according to the Edinburgh Handedness and activates a distributed neural network, including frontal but Inventory (38), and being willing to provide written informed also posterior brain regions (26–28). This task has been found consent. Exclusion criteria across all groups were the inability to exhibit abnormal activation in patients with OCD and their to tolerate scanning procedures (e.g., owing to history of clinically unaffected first-degree relatives (29) and so may be claustrophobia), contraindication to scanning (e.g., metallic valuable for addressing connectivity vulnerability markers of implant, pregnancy), current depression (defined as those in- the disorder. dividuals meeting DSM criteria on the MINI and/or those with While functional imaging has been widely used to explore an MADRS score .15), current mental health disorder on the case-control differences in brain activation in OCD MINI (except OCD in the OCD group), history of neurologic (16,21,25,30,31), subsequent research has also elicited differ- disorders (e.g., Tourette’s syndrome, tics, major head trauma), ences in the functional connectivity between different cortical history of psychosis, and history of bipolar disorder. In the regions. In a meta-analysis of seed-based resting-state func- OCD group, participants were required to meet DSM criteria for tional imaging studies, OCD was associated with hypo- the disorder based on clinical interview and the MINI, to have connectivity between frontoparietal (executive), salience, and primarily washing/checking symptoms, and to have a Yale- default mode networks (22). Using the Multi-Source Interfer- Brown Obsessive-Compulsive Scale total score .16. Our ence task, which examines aspects of cognitive control, a prior rationale for including patients with mainly washing/checking study found altered regional connectivities in patients with OCD symptoms was that washing symptoms in particular are compared with control participants, including in paralimbic, extremely common in OCD (5), and we wished to include the sensorimotor, and default mode networks (32). In a functional same symptom-related criteria as in our previous case- imaging study using an SST, (33), abnormal negative coupling relative-control behavioral study (19). Patients with OCD with was found in patients with OCD versus control participants clinically significant hoarding were excluded because hoarding between the inferior frontal gyrus and amygdala. Similar results differs from mainstream OCD and is now listed separately from were evident, but to a lesser degree, in first-degree relatives of OCD in diagnostic nosological systems (39). In the OCD rela- patients versus control participants (33). tives group and control group, participants were required to be The aim of this study therefore was to examine brain dys- free from history of OCD (including no clinically significant connectivity during response inhibition as a candidate latent symptoms based on extended clinical assessment such as the vulnerability marker for OCD. We hypothesized that patients MINI), to be free from other mainstream mental disorders (e.g., with OCD and their clinically asymptomatic first-degree rela- mood disorder, anxiety disorder), and to not be receiving tives would exhibit reduced connectivity between frontal and psychotropic medication(s). posterior brain regions within the inhibitory control network.

METHODS AND MATERIALS Stop Signal Task Participants completed pretraining on the SST (40) prior to Participants functional magnetic resonance imaging (fMRI), with a view to Patients with OCD were recruited from a National Health minimizing between-group differences in behavioral measures Service treatment center in the United Kingdom. Each patient that can confound interpretation of imaging connectivity data. entering into the study gave permission for the study team to Participants then completed the task during fMRI. We used a contact a ﬁrst-degree relative (by preference this was a same- version of the task optimized for fMRI as described elsewhere gendered, similarly aged sibling when possible). Healthy con- (41). In brief, individuals viewed a series of left- and right- trol participants were recruited using media advertisements. pointing arrows (the go signals) and were instructed to Participants provided written informed consent after having the respond as quickly as possible by clicking the button with their opportunity to read the information sheets and ask questions right hand, depending on which direction the arrow was of the study team. The study was approved by the Cambridge pointing. Intermittently, a down-pointing arrow (the stop signal) Research Ethics Committee. would appear on the screen for a variable time interval (initially All study participants participated in an extended clinical 200 ms) after a go signal, and participants were instructed to interview supplemented by the Mini International Neuropsy- stop their initiated response when it appeared. By modulating chiatric Interview (MINI; DSM-IV/ICD-10 version) (34), the the go–stop gap as previously described, the task was Montgomery–Åsberg Depression Rating Scale (MADRS) (35), designed for a 50% successful inhibition outcome and was and the National Adult Reading Test (36). The MINI version performed by each participant for approximately 8 minutes. used identiﬁes the following mental disorders: major depres- The stop signal reaction time was calculated using the simple/ sive disorder, dysthymia, suicidality, manic episodes, panic standard way for such designs, that is, by subtracting the disorder, agoraphobia, social phobia, posttraumatic stress mean go–stop interval from the mean reaction time. Scanner disorder, alcohol dependence/abuse, substance dependence/ behavioral data recorded for each participant are presented in abuse, psychotic disorders, anorexia nervosa, bulimia nervosa, the Supplement, with analyses indicating that the task design generalized anxiety disorder, and antisocial personality disor- functioned correctly [no behavioral differences between der. The MADRS rates depressive symptoms, and the National groups and p(inhibit) close to 50% in each group as expected].

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI 223 Biological Psychiatry: CNNI Inhibition-Related Dysconnectivity in OCD

Functional Imaging Acquisition thresholded at p , .05 voxelwise, and false discovery rate Imaging data were acquired at the Wolfson Brain Imaging cluster correction was then applied at p , .05. Significant ef- Centre at the University of Cambridge. Participants were fects of group were further interpreted by fitting 5-mm-radius scanned with a 3T Siemens TIM Trio scanner (Siemens Corp., spheres at the peak coordinates of a given significant F test Erlangen, Germany). While the participants were undertaking map and conducting post hoc permutation tests for each the SST, blood oxygen level–dependent sensitive three- groupwise comparison (10,000 permutations per test). dimensional volume images were acquired every 2 seconds. The first 10 images were discarded to account for equilibrium Regions of Interest effects of T1. Each image volume consisted of 32 slices of 4 Regions of interest (ROIs) were generated by our in-house mm thickness, with in-plane resolution of 3 3 3 mm and three-dimensional watershed transform algorithm (42,43). The orientated parallel with the anterior commissure–posterior method was used because it can accurately and efficiently commissure line. A standard echo-planar imaging sequence decompose thresholded statistical activation maps into was used with 78 flip angle, 30 ms echo time, and temporal discrete clusters even when the clusters are contiguous. It was resolution of 1.1 seconds in a continuous descending conducted at the group level based on the thresholded sta- sequence. The field of view of images was 192 3 192 mm, a tistical maps to enable connectivity across the activated 64 3 64 matrix, 0.51 ms echo spacing, and 2232 Hz/pixel network to be examined. When generating the ROIs, the bandwidth. In addition, a 1-mm resolution magnetization pre- within-subject contrasts (1 and 2) were also thresholded vox- pared rapid acquisition gradient-echo structural scan was elwise at p , .01 to focus on the most active brain regions. The collected for each individual with a 256 3 240 3 192 matrix, ROIs formed the basis of the connectivity analyses. 900 ms inversion time, 2.99 ms echo time, and 9 flip angle. Scan preprocessing was conducted using the standard pro- Connectivity Analysis cedure in SPM12. Data for each participant were motion cor- Measures of task-evoked network connectivity were estimated rected, registered to the structural magnetization prepared using psychophysiological interaction (PPI) models, which rapid acquisition gradient-echo, spatially warped onto the quantify how the correlation in activity between pairs of brain standard Montreal Neurological Institute template using DAR- regions differs across task conditions. Notably, the classic PPI TEL toolbox, upsampled to 2-mm cubed voxels, and spatially method focuses on a single task contrast across task condi- smoothed using a Gaussian filter (8 mm full width at half tions. More recently, a generalized form of PPI (gPPI) was maximum Gaussian kernel). developed that simultaneously assesses the impact on connectivity of multiple task conditions. We used a custom General Linear Modeling Analysis MATLAB (The MathWorks, Inc., Natick, MA) implementation of fMRI data were analyzed to determine blood oxygen level– the following gPPI model: dependent signal changes in response to participants per- T ¼ b 1 ½ S; ð Þ; b 1 ½ S ð Þb 1 ; forming the SST. General linear model analysis was applied at Y 0 Y H X E G Y H X j e the individual participant level in SPM12. The data were high- pass filtered (cutoff period = 180 seconds) to remove low- where X was the matrix containing psychological time courses frequency drifts in the MRI signal. Regressor functions for (i.e., time courses for encode, maintain, and probe events) and each condition were created by convolving timing functions HðXÞ was the hemodynamic response function convolution of indicating the onset of each of six event types, with a basis that matrix; Y T was the target time series and Y S was the function representing the canonical hemodynamic response. source time series; E was the nuisance regressor matrix defined previously in the preprocessing stage; b included The event types were successfully versus unsuccessfully G weights of no interest and b included the weights for the PPI inhibited left or right responses and the left or right responses j in go trials. Six regressors were included representing rotations predictors, which were the target of further analysis; b0 was the and translations for the x-, y-, and z-axes. intercept and ewas the residual error. This model was estimated for all pairs of connections to form a connectivity matrix, Group-Level Analysis and upper and lower triangles were averaged to form an undirected weighted connectivity matrix for each condition in the Whole-brain maps depicting beta weights for the experimental design matrix. gPPI models included successful inhibition, predictor functions from the first-level models were collated for failed inhibition, and go trials for each participant group. Two group-level analyses using a full-factorial 2 3 2 3 3 design, contrasts were generated: all stop signals minus all go trials where outcome of the stop trials (successful inhibition or un- and successful minus failed stop signal trials. Mixed analyses successful inhibition) and the direction with which the of variance were applied to test for significant differences response was made were the within-subject factors and group among the three groups in the pattern of PPI estimates across (OCD, relative, or control) was the between-subject factor. The ROIs. Pairwise tests were then applied at an uncorrected following four a priori voxelwise contrasts were estimated: 1) t , .01 threshold to characterize the basis of any significant the positive effect of condition ( contrast of the mean of all p t interactions. stop trials vs. 0), which captures regions of the brain that are significantly active during stop trials; 2) successful minus failed stop trials; 3) the main effect of group; and 4) the group 3 RESULTS condition interaction. To correct for multiple comparisons In total, 20 patients with OCD, 18 of their nonsymptomatic across the whole-brain mass, contrast images were first-degree relatives, and 20 control participants completed

224 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI Biological Psychiatry: Inhibition-Related Dysconnectivity in OCD CNNI

Table 1. Demographic and Clinical Characteristics of Patients With Obsessive-Compulsive Disorder, Their Unaffected First- Degree Relatives, and Healthy Control Participants Patients Relatives Control Participants (n = 20) (n = 18) (n = 20) Statistic p Demographic Measures

Age, years 37.6 6 14.6 40.7 6 10.8 36.3 6 8.3 F2 = 0.7115 .4954 2 Gender, n (male:female) 20 (17:3) 18 (13:5) 20 (15:5) c 2 = 1.007 .6044 NART IQ 115.4 6 5.2 114.6 6 7.2 115.9 6 6.2 Kruskal–Wallis statistic = 0.23 .8914 Clinical Measures MADRS 7.5 6 7.5 2.33 6 3.3 1.30 6 3.4 Kruskal–Wallis statistic = 16.05 .0003 Y-BOCS obsessions 11.15 6 2.58 –– – – Y-BOCS compulsions 11.85 6 2.62 –– – – Y-BOCS total 22.50 6 5.30 –– – – Values are presented as mean 6 standard deviation unless otherwise stated. One-way analysis of variance or Kruskal–Wallis statistical tests were used depending on the normality of data. MADRS, Montgomery–Åsberg Depression Rating Scale; NART, National Adult Reading Test; Y-BOCS, Yale-Brown Obsessive Compulsive Scale.

the study. The demographic and clinical features of the sample discovery rate ps , .05). There was a main effect of group are presented in Table 1, where it can be seen that the groups (Figure 1A), yielding group differences mainly in the occipital were well matched on age, gender, and IQ. As expected, pa- lobes, specifically in the left and right occipital cortex (Brod- tients with OCD scored significantly higher on MADRS total mann areas 18 and 19), the temporal occipital fusiform cortex scores than the other groups, but mean scores were well (Brodmann area 37), and the cerebellum. Post hoc permutation below the threshold for clinically significant depression, in tests indicated that the group effect was due to hyper- keeping with the exclusion criteria used. Task-related behav- activation in patients with OCD versus both other groups ioral measures did not differ significantly among the groups maximal in the bilateral lateral occipital complex (both ps , (see Supplement). The following numbers of patients were .001) (Supplemental Figure S1). Brain regions significantly taking psychotropic medication: eight selective serotonin re- activated during stop signal trials, across all participants, are uptake inhibition monotherapy and two selective serotonin shown in Figure 1B. It can be seen that the SST activated the reuptake inhibitor plus low-dose antipsychotic medication. distributed inhibitory control network, including the bilateral One patient was also taking occasional lorazepam but had not inferior frontal gyrus, insula, and anterior cingulate cortex. For taken this within 48 hours of study participation. the contrast of successful minus failed stops across all participants, relative hypoactivation was observed (Figure 1C)in Activation Results for the SST regions associated with motor responses (including Brodmann Activation differences for the SST contrasts of interest, along areas 4 and 6). This is consistent with failed stops activating with the extracted ROIs, are summarized in Figure 1 (all false relevant motor areas owing to action as compared with there FPO = web 4C Figure 1. Significant activation maps for the contrasts of interest during the stop signal task. (A) Brain regions showing a main effect of group (false discovery rate p , .05). (B) Brain regions activated during stop signal trials (false discovery rate p , .05). (C) Brain regions underactivated for successful minus failed stops (false discovery rate p , .05). (D) Regions of interest for subsequent connectivity analyses on a brain map and also labeled, generated from the above contrasts and color coded in keeping with (A) to (C). Ant, anterior; Inf, inferior; L, left; Mid, middle; Oper, operculum; Post, posterior; R, right; Sup, superior.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI 225 Biological Psychiatry: CNNI Inhibition-Related Dysconnectivity in OCD

being no motor response for successful stops. The interaction with OCD and their relatives had higher connectivity for failed of group 3 successful minus failed inhibition did not yield stops and/or lower connectivity for successful stops compared significant regions. with control participants. Ultimately, determining what this means on a process level requires further investigation Group Differences in Connectivity for the SST examining causal dynamics. However, the implicated neural The 29 functional ROIs from the above activation maps regions are likely to operate via mutual bidirectional connec- (Figure 1D) were used for the subsequent connectivity anal- tions to facilitate response inhibition (48). It is interesting to ysis. For the SST contrast (stop signal minus go trials), there note that certain frontal brain regions found to be abnormally was no significant main effect of group on gPPI connectivity connected here during response inhibition (i.e., inferior frontal cortex/insula) were previously found to exhibit reduced striatal- (F = 0.69, p = .50), there was a significant effect of connection related connectivity in OCD in association with cognitive (F = 1.78, p = .011), and there was no significant interaction rigidity (49). (F = 1.17, p = .19) (all Greenhouse–Geisser corrected). When applied to the success minus fail contrast, there was a sig- The most commonly dysconnected nodes common to patients and their asymptomatic first-degree relatives included nificant main effect of group (F = 3.67, p = .032) and connec- frontal cortical, occipital, and cerebellar regions (Figure 2). tion (F = 1.71, p = .016) and a significant interaction (F = 1.38, Conventional neurobiological models of OCD have focused on p = .041) (all Greenhouse–Geisser corrected). These results indicated that the task conditions affected network connec- the frontal lobes, whereas the current findings implicate tivity in different ways across the three groups. To characterize abnormal connections involving not only frontal brain regions the basis of the effects at the node level, the coefficients were but also these other brain regions. This is in keeping with contrasted pairwise for patients and their relatives versus several tiers of OCD research more broadly (21,24,50), including connectivity studies. For example, resting-state control participants, thresholded at p , .01 two tailed (Figure 2). A widespread pattern of reduced connectivity was connectivity changes in OCD were maximal in the cerebellar evident in patients with OCD and their relatives. Summing the crus 1 region (51), and machine learning algorithms designed number of supra-threshold connections for each node high- to discriminate patients with OCD from control participants lighted a high degree of abnormality affecting cerebellum area based on resting-state connectivity indicated important con- crus 1 connectivity bilaterally, middle occipital gyrus bilaterally, tributions from not only frontal regions but also occipital and superior frontal gyrus and superior medial frontal cortex, left cerebellar regions (52). To our knowledge, only one previous middle temporal, and left postcentral gyri. study has examined task-related functional dysconnectivity as a candidate vulnerability marker for OCD (53). This study found reduced functional connectivity between the right dorsolateral DISCUSSION prefrontal cortex and the basal ganglia (putamen) during ex- This study evaluated functional brain dysconnectivity during ecutive planning (53). Resting-state connectivity changes have response inhibition as a candidate vulnerability marker for also been described in the literature, in patients with OCD and OCD. Consistent with our hypothesis, the key finding was that their relatives, involving distributed brain regions (54,55). patients with OCD and their first-degree relatives had abnor- Collectively, the emerging evidence thus suggests important mally reduced functional connectivity during the SST between dysconnectivity not only between cortical and subcortical re- frontal and posterior brain regions, including the frontal cortex, gions but also between anatomically distant cortical regions in occipital cortex, and cerebellum. These novel findings accord OCD, findings that are likely to be contingent on the nature of well with the notion that functional connectomics constitutes a the cognitive probe used to explore such neural circuitry. candidate vulnerability marker for OCD, supporting neurobio- In terms of group differences in SST-related brain activation logical models of the disorder implicating loss of cortically (as opposed to functional connectivity), we found differences mediated inhibitory control, not only constrained to the frontal in posterior brain regions, maximal in the bilateral lateral oc- lobes but also involving distant posterior brain regions (15,44). cipital complex. This result was attributable to hyperactivation Conventional analysis confirmed that the fMRI SST acti- in patients versus both other groups, whereas activation in vated neural circuitry, including the bilateral inferior frontal relatives did not differ from control participants in this region. cortex and anterior cingulate cortex as well as more posterior There was no group 3 successful minus failed inhibition parts of the brain playing a role in visual attention streams interaction, indicating that this abnormality was common to (Figure 1). This is in keeping with prior lines of research inhibition trials on the task whether or not inhibition was suc- implicating such regions in cortically mediated motor inhibition cessful. The lateral occipital complex plays an important role in processes (26,45–47). We generated a set of ROIs using an visual attentional processing, including representation and innovative watershed algorithm to examine connectivity dif- perception of objects (56) and faces (57). One interpretation of ferences between groups using a gPPI model. This identified the current finding is that hyperactivation of this visual pro- widespread patterns of hypoconnectivity, common to patients cessing region may be related to hypervigilance in OCD or an with OCD and their relatives, versus control participants in expectation of an environmental threat. Owing to the unpre- frontal and posterior brain regions (Figure 2). Overall group dicted nature of this result, replication is required before firm differences in connectivity during the SST were specifically conclusions can be made. Nonetheless, this result suggests detected during the success–fail contrast, with connectivity that tasks designed to probe visual attentional streams may be being lower in patients and relatives versus control partici- valuable in OCD research. pants. In the absence of significant overall stop–go differences Although this is the first study to address inhibitory control– in connectivity among the groups, this suggests that patients related functional connectivity as a candidate vulnerability

226 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI Biological Psychiatry: Inhibition-Related Dysconnectivity in OCD CNNI web 4C/FPO

Figure 2. Results from connectivity analyses. (A) Schemaball showing abnormally hypoconnected regions in patients with obsessive-compulsive disorder and relatives versus control participants. Each region of interest (ROI) is indicated by a peripheral label. Curved lines within the circle indicate ROI–ROI connections that were signiﬁcantly hypoconnected in patients and relatives versus control participants. Thicker curved lines indicate greater abnormality (mean psychophysiological interaction coefﬁcient). The outer circumference of the circle is color coded to indicate the contrast of interest as per Figure 1, and the size of nodes on the peripheral circle represents the total number of suprathreshold abnormal connections (i.e., nodal degree). (B) Glass brain representation of abnormal connections from (A) to show anatomical extents. (C) List of all ROIs and the number of suprathreshold connections with other regions for each ROI. Color codings refer to the task contrasts of interest. AA, Automated Anatomical Labeling; Inf, inferior; L, left; Mid, middle; MNI, Montreal Neurological Institute; Post, posterior; R, right; Sup, superior.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI 227 Biological Psychiatry: CNNI Inhibition-Related Dysconnectivity in OCD

marker for OCD, several limitations should be considered. We OCD. Future studies could use such cognitive probe connec- recruited patients with primarily washing/checking OCD tivity approaches to help delineate etiological factors involved symptoms who did not have comorbidities. As such, it remains in OCD and extend research into other obsessive-compulsive– to be demonstrated whether the findings generalize to patients related disorders (59). with other primary symptoms or to those who have comorbidities. Owing to the sample size, power may be limited. Our approach could be viewed as conservative because nodes of ACKNOWLEDGMENTS AND DISCLOSURES interest were generated using false discovery rate , .05; SRC’s role in this study was funded by a Wellcome Trust Clinical Fellowship p (Grant No. 110049/Z/15/Z). AH’s team was supported by the Dementia hence, and in view of the sample size, some neural nodes Research Institute and the National Institute for Health Research (NIHR) implicated in OCD, but with a smaller effect size, may have Imperial Biomedical Research Centre. TWR’s role in this study was funded been overlooked. Presupplementary motor activation abnor- by the Wellcome Trust (Grant No. 104631/Z/14/Z). malities were previously found in patients with OCD and their We thank all members of the study team, including radiologists at the relatives (29), but we could not replicate this finding in the Wolfson Brain Imaging Centre, and the study participants. relevant ROIs (see Supplemental Table S2). Likely because AZ completed analyses for this study in fulfillment of a postgraduate research master’s program in experimental neuroscience at Imperial participants were pretrained, they did not differ on stop signal College. behavioral measures; this is an advantage because it simplifies SRC consults for Promentis and Ieso Digital Health; he receives a sti- imaging interpretation, but the corollary is that our study did pend for his work as associate editor at Neuroscience and Biobehavioral not measure neural changes related to impaired inhibition but Reviews and at Comprehensive Psychiatry. AH is founder and director of rather measured neural changes related to inhibition per se. Future Cognition Ltd. and H2 Cognitive Designs. BJS consults for Cam- Owing to the nature of the gPPI analysis, it could not be bridge Cognition, Greenfield BioVentures, and Cassava Sciences. TWR undertakes consulting work for Cambridge Cognition, Unilever, and established whether there was heightened connectivity during Greenfield BioVentures; he has received research grants from Shionogi, go trials or decreased connectivity during stop trials in the GlaxoSmithKline, and Small Pharma; he receives royalties from Cambridge patients and relatives. Our connectivity difference was in the Cognition; and he receives editorial honoraria from Springer Nature and contrast of successful–failed stop trials. Control participants Elsevier. NF has recently held research or networking grants from the Eu- showed heightened connectivity when stopping was suc- ropean College of Neuropsychopharmacology (ECNP), UK NIHR, and EU cessful relative to unsuccessful. In OCD, this effect was Horizon 2020, accepted travel and hospitality expenses from the British Association for Psychopharmacology, ECNP, Royal College of Psychia- reduced. This is an interesting pattern of connectivity differ- trists, and International College of Neuropsychopharmacology, and received ence. Patients with OCD may be engaging the network more honoraria from Taylor & Francis and Elsevier for editorial duties; she leads a during unsuccessful stop trials, in line with abnormal post-error National Health Service treatment service for OCD and holds board mem- processing. Or, it may be that they engage the network less berships for various registered charities linked to OCD. ETB is employed half during the successful stop trials. The fact that we see this time by the University of Cambridge and half time by GlaxoSmithKline; he difference but no cross- group difference for stop–go suggests holds stock in GlaxoSmithKline. The other authors report no biomedical financial interests or potential conflicts of interest. that it is both. This aspect could be assessed in future studies by including rest blocks, allowing activity and connectivity during routine responding to be estimated separate from the ARTICLE INFORMATION resting baseline. While some patients with OCD were receiving psychotropic medications, functional dysconnectivity was also From the Computational, Cognitive and Clinical Neuroimaging Laboratory (AH, AZ, SS, ES), Division of Brain Sciences, Imperial College London, found in these patients’ relatives who were not receiving any London; Department of Psychiatry (NF, ETB, BJS, SRC), University of psychotropic medications. Hence, while we cannot address Cambridge, Addenbrooke’s Hospital, Department of Experimental Psy- effects of such pharmacotherapies on connectivity owing to chology (TWR), and Behavioural and Clinical Neurosciences Institute (TWR), the sample size, our key findings were not due to such effects. University of Cambridge, Cambridge, United Kingdom. Prior work found treatment-related changes in activation AH and AZ are joint first authors. during a Stroop task, which examines attentional inhibition Address correspondence to Samuel R. Chamberlain, MB B.Chir, Ph.D., MRCPsych, Department of Psychiatry, University of Cambridge, Adden- processes, in patients with OCD (58). Future work should brooke’s Hospital, Cambridge CB2 0QQ, United Kingdom; E-mail: src33@ examine effects of treatment on functional connectivity during cam.ac.uk. inhibition tasks in OCD. We did not observe robust differences Received Jul 10, 2019; revised Sep 19, 2019; accepted Sep 27, 2019. between the OCD and first-degree relative groups in functional Supplementary material cited in this article is available online at https:// connectivity. Identification of differences between these two doi.org/10.1016/j.bpsc.2019.09.010. types of group using larger samples in future work may be valuable to identify mechanisms associated with chronicity/ REFERENCES instantiation of OCD as opposed to vulnerability toward OCD. 1. Stein DJ (2019): Obsessive-compulsive disorder and global mental Lastly, the current study focused on cortical functional con- health. Indian J Psychiatry 61(suppl 1):S4–S8. nectivity; however, given the prominent role of the basal 2. Hollander E, Doernberg E, Shavitt R, Waterman RJ, Soreni N, ganglia in OCD models, future work should also look at Veltman DJ, et al. (2016): The cost and impact of compulsivity: A cortico-subcortical connectivity on the SST, with there already research perspective. Eur Neuropsychopharmacol 26:800–809. being evidence of abnormalities in OCD using an executive 3. Hollander E, Kwon JH, Stein DJ, Broatch J, Rowland CT, Himelein CA planning task (53). (1996): Obsessive-compulsive and spectrum disorders: Overview and quality of life issues. J Clin Psychiatry 57(suppl 8):3–6. In conclusion, we found that hypoconnectivity during 4. Fontenelle LF, Mendlowicz MV, Versiani M (2006): The descriptive response inhibition, involving frontal and posterior brain re- epidemiology of obsessive-compulsive disorder. Prog Neuro- gions, may constitute a candidate vulnerability marker for psychopharmacol Biol Psychiatry 30:327–337.

228 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI Biological Psychiatry: Inhibition-Related Dysconnectivity in OCD CNNI

5. Ruscio AM, Stein DJ, Chiu WT, Kessler RC (2010): The epidemiology 26. Aron AR, Fletcher PC, Bullmore ET, Sahakian BJ, Robbins TW (2003): of obsessive-compulsive disorder in the National Comorbidity Survey Stop-signal inhibition disrupted by damage to right inferior frontal Replication. Mol Psychiatry 15:53–63. gyrus in humans. Nat Neurosci 6:115–116. 6. Hettema JM, Neale MC, Kendler KS (2001): A review and meta- 27. Aron AR, Behrens TE, Smith S, Frank MJ, Poldrack RA (2007): Trian- analysis of the genetic epidemiology of anxiety disorders. Am J Psy- gulating a cognitive control network using diffusion-weighted mag- chiatry 158:1568–1578. netic resonance imaging (MRI) and functional MRI. J Neurosci 7. Sampaio AS, Hounie AG, Petribu K, Cappi C, Morais I, Vallada H, et al. 27:3743–3752. (2015): COMT and MAO-A polymorphisms and obsessive-compulsive 28. Bari A, Robbins TW (2013): Inhibition and impulsivity: Behavioral and disorder: A family-based association study. PLoS One 10:e119592. neural basis of response control. Prog Neurobiol 108:44–79. 8. Grunblatt E, Marinova Z, Roth A, Gardini E, Ball J, Geissler J, et al. (2018): 29. de Wit SJ, de Vries FE, van der Werf YD, Cath DC, Heslenfeld DJ, Combining genetic and epigenetic parameters of the serotonin transporter Veltman EM, et al. (2012): Presupplementary motor area hyperactivity gene in obsessive-compulsive disorder. J Psychiatr Res 96:209–217. during response inhibition: A candidate endophenotype of obsessive- 9. Noh HJ, Tang R, Flannick J, O’Dushlaine C, Swofford R, Howrigan D, compulsive disorder. Am J Psychiatry 169:1100–1108. et al. (2017): Integrating evolutionary and regulatory information with a 30. Rauch SL, Whalen PJ, Curran T, Shin LM, Coffey BJ, Savage CR, et al. multispecies approach implicates genes and pathways in obsessive- (2001): Probing striato-thalamic function in obsessive-compulsive compulsive disorder. Nat Commun 8:774. disorder and Tourette syndrome using neuroimaging methods. Adv 10. Grunblatt E, Hauser TU, Walitza S (2014): Imaging genetics in Neurol 85:207–224. obsessive-compulsive disorder: Linking genetic variations to alter- 31. Breiter HC, Rauch SL (1996): Functional MRI and the study of OCD: ations in neuroimaging. Prog Neurobiol 121:114–124. From symptom provocation to cognitive-behavioral probes of cortico- 11. Burguiere E, Monteiro P, Mallet L, Feng G, Graybiel AM (2015): Striatal striatal systems and the amygdala. Neuroimage 4:S127–S138. circuits, habits, and implications for obsessive-compulsive disorder. 32. Cocchi L, Harrison BJ, Pujol J, Harding IH, Fornito A, Pantelis C, et al. Curr Opin Neurobiol 30:59–65. (2012): Functional alterations of large-scale brain networks related to 12. Pallanti S, Hollander E (2008): Obsessive-compulsive disorder spec- cognitive control in obsessive-compulsive disorder. Hum Brain Mapp trum as a scientific “metaphor.” CNS Spectr 13:6–15. 33:1089–1106. 13. Cuthbert BN, Insel TR (2013): Toward the future of psychiatric diag- 33. van Velzen LS, de Wit SJ, Curcic-Blake B, Cath DC, de Vries FE, nosis: The seven pillars of RDoC. BMC Med 11:126. Veltman DJ, et al. (2015): Altered inhibition-related frontolimbic con- 14. Chamberlain SR, Stochl J, Redden SA, Grant JE (2018): Latent traits of nectivity in obsessive-compulsive disorder. Hum Brain Mapp 36:4064– impulsivity and compulsivity: Toward dimensional psychiatry. Psychol 4075. Med 48:810–821. 34. Sheehan DV, Lecrubier Y, Sheehan KH, Amorim P, Janavs J, Weiller E, 15. Chamberlain SR, Blackwell AD, Fineberg NA, Robbins TW, et al. (1998): The Mini-International Neuropsychiatric Interview Sahakian BJ (2005): The neuropsychology of obsessive compulsive (M.I.N.I.): The development and validation of a structured diagnostic disorder: The importance of failures in cognitive and behavioural in- psychiatric interview for DSM-IV and ICD-10. J Clin Psychiatry hibition as candidate endophenotypic markers. Neurosci Biobehav 59(suppl 20):22–33; quiz 34–57. Rev 29:399–419. 35. Montgomery SA, Asberg M (1979): A new depression scale designed 16. Menzies L, Achard S, Chamberlain SR, Fineberg N, Chen CH, del to be sensitive to change. Br J Psychiatry 134:382–389. Campo N, et al. (2007): Neurocognitive endophenotypes of obsessive- 36. Nelson HE (1982): National Adult Reading Test (NART) Manual. compulsive disorder. Brain 130:3223–3236. Windsor, UK: NFER–Nelson. 17. Chamberlain SR, Fineberg NA, Blackwell AD, Robbins TW, 37. Goodman WK, Price LH, Rasmussen SA, Mazure C, Fleischmann RL, Sahakian BJ (2006): Motor inhibition and cognitive flexibility in Hill CL, et al. (1989): The Yale-Brown Obsessive Compulsive Scale: I. obsessive-compulsive disorder and trichotillomania. Am J Psychiatry Development, use, and reliability. Arch Gen Psychiatry 46:1006–1011. 163:1282–1284. 38. Oldfield RC (1971): The assessment and analysis of handedness: The 18. Lipszyc J, Schachar R (2010): Inhibitory control and psychopathology: Edinburgh Inventory. Neuropsychologia 9:97–113. A meta-analysis of studies using the stop signal task. J Int Neuro- 39. Mataix-Cols D, Fernandez de la Cruz L (2018): Hoarding disorder has psychol Soc 16:1064–1076. finally arrived, but many challenges lie ahead. World Psychiatry 19. Chamberlain SR, Fineberg NA, Menzies LA, Blackwell AD, 17:224–225. Bullmore ET, Robbins TW, et al. (2007): Impaired cognitive flexibility 40. Logan GD, Cowan WB, Davis KA (1984): On the ability to inhibit simple and motor inhibition in unaffected first-degree relatives of patients with and choice reaction time responses: A model and a method. J Exp obsessive-compulsive disorder. Am J Psychiatry 164:335–338. Psychol Hum Percept Perform 10:276–291. 20. Graybiel AM, Rauch SL (2000): Toward a neurobiology of obsessive- 41. Chamberlain SR, Hampshire A, Mueller U, Rubia K, del Campo N, compulsive disorder. Neuron 28:343–347. Craig K, et al. (2009): Atomoxetine modulates right inferior frontal 21. Menzies L, Chamberlain SR, Laird AR, Thelen SM, Sahakian BJ, activation during inhibitory control: A pharmacological functional Bullmore ET (2008): Integrating evidence from neuroimaging and neuro- magnetic resonance imaging study. Biol Psychiatry 65:550–555. psychological studies of obsessive-compulsive disorder: The 42. Meyer F (1994): Topographic distance and watershed lines. Signal orbitofronto-striatal model revisited. Neurosci Biobehav Rev 32:525–549. Process 38:113–125. 22. Gursel DA, Avram M, Sorg C, Brandl F, Koch K (2018): Frontoparietal 43. Soreq E, Leech R, Hampshire A (2019): Dynamic network coding of areas link impairments of large-scale intrinsic brain networks with working-memory domains and working-memory processes. Nat aberrant fronto-striatal interactions in OCD: A meta-analysis of resting- Commun 10:936. state functional connectivity. Neurosci Biobehav Rev 87:151–160. 44. van Velzen LS, Vriend C, de Wit SJ, van den Heuvel OA (2014): 23. Hu X, Du M, Chen L, Li L, Zhou M, Zhang L, et al. (2017): Meta-analytic Response inhibition and interference control in obsessive-compulsive investigations of common and distinct grey matter alterations in spectrum disorders. Front Hum Neurosci 8:419. youths and adults with obsessive-compulsive disorder. Neurosci 45. White CN, Congdon E, Mumford JA, Karlsgodt KH, Sabb FW, Biobehav Rev 78:91–103. Freimer NB, et al. (2014): Decomposing decision components in the 24. Nakao T, Okada K, Kanba S (2014): Neurobiological model of obsessive- stop-signal task: A model-based approach to individual differences in compulsive disorder: Evidence from recent neuropsychological and inhibitory control. J Cogn Neurosci 26:1601–1614. neuroimaging findings. Psychiatry Clin Neurosci 68:587–605. 46. Aron AR, Robbins TW, Poldrack RA (2014): Inhibition and the right 25. Norman LJ, Taylor SF, Liu Y, Radua J, Chye Y, De Wit SJ, et al. (2019): inferior frontal cortex: One decade on. Trends Cogn Sci 18:177–185. Error processing and inhibitory control in obsessive-compulsive dis- 47. Swick D, Ashley V, Turken U (2011): Are the neural correlates of order: A meta-analysis using statistical parametric maps. Biol Psy- stopping and not going identical? Quantitative meta-analysis of two chiatry 85:713–725. response inhibition tasks. Neuroimage 56:1655–1665.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI 229 Biological Psychiatry: CNNI Inhibition-Related Dysconnectivity in OCD

48. Hampshire A, Sharp DJ (2015): Contrasting network and modular 54. Hou JM, Zhao M, Zhang W, Song LH, Wu WJ, Wang J, et al. (2014): perspectives on inhibitory control. Trends Cogn Sci 19:445–452. Resting-state functional connectivity abnormalities in patients with 49. Vaghi MM, Vertes PE, Kitzbichler MG, Apergis-Schoute AM, van der obsessive-compulsive disorder and their healthy first-degree relatives. Flier FE, Fineberg NA, et al. (2017): Specific frontostriatal circuits for J Psychiatry Neurosci 39:304–311. impaired cognitive flexibility and goal-directed planning in obsessive- 55. de Vries FE, de Wit SJ, van den Heuvel OA, Veltman DJ, Cath DC, van compulsive disorder: Evidence from resting-state functional connec- Balkom A, et al. (2019): Cognitive control networks in OCD: A resting- tivity. Biol Psychiatry 81:708–717. state connectivity study in unmedicated patients with obsessive- 50. Piras F, Piras F, Caltagirone C, Spalletta G (2013): Brain circuitries of compulsive disorder and their unaffected relatives. World J Biol obsessive compulsive disorder: A systematic review and meta- Psychiatry 20:230–242. analysis of diffusion tensor imaging studies. Neurosci Biobehav Rev 56. Grill-Spector K, Kourtzi Z, Kanwisher N (2001): The lateral 37:2856–2877. occipital complex and its role in object recognition. Vision Res 51. Xu T, Zhao Q, Wang P, Fan Q, Chen J, Zhang H, et al. (2019): Altered 41:1409–1422. resting-state cerebellar-cerebral functional connectivity in obsessive- 57. Nagy K, Greenlee MW, Kovacs G (2012): The lateral occipital cortex in compulsive disorder. Psychol Med 49:1156–1165. the face perception network: An effective connectivity study. Front 52. Takagi Y, Sakai Y, Lisi G, Yahata N, Abe Y, Nishida S, et al. (2017): Psychol 3:141. A neural marker of obsessive-compulsive disorder from whole-brain 58. Nabeyama M, Nakagawa A, Yoshiura T, Nakao T, Nakatani E, functional connectivity. Sci Rep 7:7538. Togao O, et al. (2008): Functional MRI study of brain activation alter- 53. Vaghi MM, Hampshire A, Fineberg NA, Kaser M, Bruhl AB, ations in patients with obsessive-compulsive disorder after symptom Sahakian BJ, et al. (2017): Hypoactivation and dysconnectivity of a improvement. Psychiatry Res 163:236–247. frontostriatal circuit during goal-directed planning as an endopheno- 59. Hollander E, Kim S, Braun A, Simeon D, Zohar J (2009): Cross-cutting type for obsessive-compulsive disorder. Biol Psychiatry Cogn Neu- issues and future directions for the OCD spectrum. Psychiatry Res rosci Neuroimaging 2:655–663. 170:3–6.

230 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:222–230 www.sobp.org/BPCNNI Biological Psychiatry: Archival Report CNNI

Duration of Untreated Psychosis Correlates With Brain Connectivity and Morphology in Medication-Naïve Patients With First-Episode Psychosis

Jose O. Maximo, Eric A. Nelson, William P. Armstrong, Nina V. Kraguljac, and Adrienne C. Lahti

ABSTRACT BACKGROUND: In the United States, the average duration of untreated psychosis (DUP) is 21 months, and it remains unknown how longer DUP may affect brain functioning in antipsychotic-naïve patients with first-episode psychosis. The objective was to determine the effects of DUP on functional connectivity and brain morphology measured with resting-state functional and structural magnetic resonance imaging. METHODS: Medication-naïve patients with first-episode psychosis were referred from various clinical settings. After accounting for exclusion criteria, attrition, and data quality, final analyses included 55 patients (35 male and 20 female; mean age, 24.18 years). Patients with first-episode psychosis were subjected to a 16-week trial of risperidone, a commonly used antipsychotic drug. Treatment response was calculated as change in the psychosis subscale of the Brief Psychiatric Rating Scale between baseline and 16 weeks. Resting-state functional connectivity magnetic resonance imaging and brain morphology (surface area and cortical thickness) were assessed. RESULTS: Longer DUP was associated with worse treatment response and reduced functional connectivity—more specifically in the default, salience, and executive networks. Moreover, longer DUP was associated with reduced surface area in the salience and executive networks and with increased cortical thickness in the default mode and salience networks. When the functional connectivity of the default mode network was added as a mediator, the relationship between DUP and treatment response was no longer significant. CONCLUSIONS: These data suggest that several neurobiological alterations in the form of reduced functional connectivity and surface area and increased cortical thickness underpin the effect of prolonged DUP. Keywords: Brain networks, Duration of untreated psychosis, Functional connectivity, Morphology, Resting state, Treatment response https://doi.org/10.1016/j.bpsc.2019.10.014

Meta-analyses (1–3) have consistently identified an associa- imaging (20), functional magnetic resonance imaging (MRI) tion between the duration of untreated psychosis (DUP), which (21,22), and MR spectroscopy (23). While not all studies have is the duration between the onset of positive symptoms and detected a direct relationship between DUP and brain structure treatment, and clinical outcomes (1,4,5). This relationship is or function [see review by (24)], methodological limitations found across various lengths of follow-up periods, suggesting such as prior exposure to antipsychotic drugs (APDs) and that DUP influences the long-term course of the illness (6). variable illness durations at the time of assessment may have Given that in the United States the average DUP is 74 weeks disguised existing associations (25–28). To address these (7), the National Institute of Mental Health has deemed it critical limitations, we designed a prospective multimodal imaging to identify strategies to reduce the DUP (8) in an effort to study with the goal to investigate the impact of the DUP on alleviate the overall disease burden in patients on the psy- brain structure and function in medication-naïve patients with chosis spectrum. first-episode psychosis (FEP). Importantly, active psychosis might adversely affect the Connectome analyses have provided abundant data sug- brain (9), possibly via NMDA receptor hypofunctioning (10–12), gesting that 3 large-scale brain networks deemed critical for increased dopaminergic activity (13), or persistent catechol- higher-order cognitive processes—the default mode network aminergic and hypothalamic-pituitary-adrenal axis activity (14). (DMN), salience network (SN), and central executive network Several studies have attempted to characterize the effects of (CEN) (29)—are affected in the illness. Importantly, deficits are DUP on the brain with noninvasive brain imaging techniques observed in both functional connectivity (FC) and cortical including T1-weighted imaging (15–19), diffusion-weighted morphology (30,31), suggesting a possible relationship

ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. 231 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:231–238 www.sobp.org/BPCNNI Biological Psychiatry: CNNI DUP, Brain Connectivity, and Morphology in FEP Patients

between functional and structural alterations in these brain Neuropsychological Status were used to assess symptom networks. The goal of this longitudinal multimodal imaging severity and cognition (33,34). Treatment response (TR) was study was twofold: ﬁrst, to examine the relationship among the deﬁned as the percentage of change on the BPRS psychosis DUP and the 3-network FC and cortical morphology in medi- (BPRS1) subscale from baseline to 16 weeks of risperidone, cation-naïve patients with FEP; and second, to evaluate their where a greater percentage indicates a greater reduction in

contributions to eventual treatment response following APD positive symptoms: [(BPRS1Baseline – BPRS1Week 16)/ treatment. Because it can take up to 16 weeks for patients with (BPRS1Baseline)] 3 (2100). TR data were available for 41 FEP to respond (32), we measured treatment response over medication-naïve patients with FEP. this period. We hypothesized that longer DUP would be associated with reduced FC and altered cortical morphology, Data Acquisition as well as with poor treatment response. We further hypothe- All imaging was performed on a 3T whole-body Siemens sized that these functional and morphological alterations MAGNETOM Prisma MRI scanner (Siemens AG, Erlangen, would mediate the link between DUP and treatment response. Germany) equipped with a 20-channel head coil. A high- To our knowledge, this is the first study to examine DUP in resolution T1-weighted structural scan was acquired for medication-naïve patients with FEP from a multimodal imaging anatomical reference and morphological analyses (magneti- approach. zation prepared rapid acquisition gradient-echo: repetition time = 2400 ms; echo time = 2.22 ms; inversion time = 1000 METHODS AND MATERIALS ms; flip angle = 8; generalized autocalibrating partially parallel acquisitions (GRAPPA) factor = 2; voxel size = 0.8 mm3). A T2- Participants weighted image was also obtained (repetition time = 3200 ms; We screened 169 medication-naïve patients with FEP, echo time = 563.0 ms; flip angle = 8; GRAPPA factor = 2, 208 recruited from outpatient clinics, inpatient units, and the slices, voxel size = 0.8 mm3). Resting-state functional MRI emergency department at the University of Alabama at Bir- data were acquired in opposing phase encoding directions mingham. Studies were approved by the University of Alabama (anterior . posterior and posterior . anterior; repetition time = at Birmingham Institutional Review Board, and written 1550 ms; echo time = 37.80 ms; flip angle = 71, field of view = informed consent was obtained before enrollment (patients 104 mm2; multiband acceleration factor = 4; voxel size= 2 with FEP had to be deemed competent to provide consent). mm3; 225 volumes, and 72 axial slices). Subjects were Exclusion criteria were major neurological or medical condi- instructed to keep their eyes open and let their mind wander. tions, history of significant head trauma, substance use disorders (excluding nicotine and cannabis) within 1 month of Data Preprocessing imaging, $5 days of lifetime antipsychotic exposure, preg- Morphological Data. Structural images were processed nancy or breastfeeding, and MRI contraindications. There were and analyzed using FreeSurfer 6.0 (http://surfer.nmr.mgh. 103 medication-naïve patients with FEP who did not meet in- harvard.edu/)(35,36) using standard preprocessing steps. clusion criteria and 7 who dropped before start of antipsy- Briefly, T1- and T2-weighted images were skull stripped (37), chotic trial, which left us with a sample size of 59. This number segmented, and parceled into units based on gyral and sulcal dropped to 55 after quality control (see Data Preprocessing). structure in FSL, resulting in estimates for volume, cortical Scans were obtained prior to treatment. Patients underwent thickness, and surface area (38). Data quality was assessed a 16-week treatment with risperidone using a flexible dosing with the Qoala-T Tool (Lara Wierenga, Ph.D., and Eduard regimen. Risperidone was started at 1 to 2 mg and titrated in 1- Klapwijk, Ph.D.; https://github.com/Qoala-T/Qoala-T), which to 2-mg increments; dosing was based on therapeutic effect uses a supervised-learning model to reduce rater bias and and side effects. Use of concomitant medications was permitted misclassification (39). Scans with a Qoala-T score ,70% were as clinically indicated, and these medications were prescribed visually inspected for accuracy in pial and white matter surface after the MRI session. During the trial, the following medications registration. Two structural scans were removed from final were prescribed for 24 medication-naïve patients with FEP: analysis owing to irreparable artifacts, resulting in 53 medica- amphetamine (n = 1), benztropine (n = 16), sertraline (n = 7), tion-naïve patients with FEP for morphometric analyses. escitalopram (n = 1), lithium (n = 1), trazodone (n = 2), lorazepam ( = 1), valproic acid ( = 1), and bupropion ( = 1). Seven pa- n n n Resting-State Functional MRI Data. Data were analyzed tients were taking more than one medication. Compliance was using the CONN toolbox version 18a (https://web.conn-toolbox. monitored with pill counts at each visit. The interval between org/)(40). After discarding the first 10 volumes of each scan scanning and antipsychotic initiation was ,24 hours. allowing for signal equilibration, field inhomogeneities were corrected in FSL’s topup and merged, resulting in a single 4- Clinical Assessment dimensional image of 430 total volumes (41). Functional im- Consensus diagnoses were made according to DSM-5 criteria ages were then slice-timing and motion corrected using rigid- by 2 board-certified psychiatrists from all historical and direct body realignment, coregistered to the structural image, assessment information available (ACL and NVK). Determina- normalized to Montreal Neurological Institute space, bandpass tion of DUP was based on information provided by the patient filtered (0.008 , f , 0.08 Hz), and spatially smoothed with a 4- and caregivers during screening as well as any time during the mm full width at half maximum Gaussian kernel. 16-week trial period. The Brief Psychiatric Rating Scale (BPRS) Framewise displacement and percentage of censored data and Repeatable Battery for the Assessment of were then calculated (42). Motion outliers as detected by the

232 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:231–238 www.sobp.org/BPCNNI web 4C=FPO eal oentok D,fledsoeyrt;L et H ethmshr;LP,lf otro aitlcre;R ih;R,rgthemisphere; right RH, right; R, cortex; parietal posterior left LPPC, hemisphere; executiv left central network. CEN, LH, salience analyses). left; (morphological SN, L, volume cortex; rate; intracranial parietal discovery and sex, posterior false age, FDR, for network; controlling mode from default residuals (std.) standardized are data All connectivity functional For networks. executive FreeSurfer on rendered U,BanCnetvt,adMrhlg nFPPatients FEP in Morphology and Connectivity, Brain DUP, iue1. Figure lseso signi of Clusters ilgclPyhar:CgiieNuocec n eriaigFbur 00 5:231 2020; February Neuroimaging and Neuroscience Cognitive Psychiatry: Biological ’ svrg ri sdfrbanmrhlg nlssadprilcreainposfor plots correlation partial and analyses morphology brain for used brain fsaverage s ﬁ atprilcreain ewe uaino nrae scoi DP ihfntoa onciiy( connectivity functional with (DUP) psychosis untreated of duration between correlations partial cant – U orltos inlfo l lseswr vrgdadpotdwt U o aavsaiainpurposes. visualization data for DUP with plotted and averaged were clusters all from signal correlations, DUP (A) eal mode, default – 238 www.sobp.org/BPCNNI (B) aine and salience, p FDR , ewr;DMN, network; e 0)admasks and .05) PC right RPPC, (C) central 233 CNNI Psychiatry: Biological Biological Psychiatry: CNNI DUP, Brain Connectivity, and Morphology in FEP Patients

Artifact Detection Tools toolbox (NeuroImaging Tools and Re- Table 1. Demographics, Clinical Measures, and Data sources Collaboratory, https://www.nitrc.org/projects/artifact_ Quality (n = 55) detect/) were censored (composite volume-to-volume motion Demographic Variables .0.9 mm and intensity .5 SDs), and the 6 motion parameters Age, years 24.18 6 6.27 (14–40) derived from rigid-body realignment and their derivatives, as well Sex, male, % 64 as the first 5 component time series derived from cerebrospinal Parental occupationa 5.91 6 4.83 (1–18) fluid and white matter using aCompCor within the CONN toolbox Cannabis users, % 31 and corresponding derivatives, were regressed out from the signal. Smokers, % 42 After excluding patients who dropped out from the study, Smoking, packs per day 0.22 6 0.34 (0–1) did not have a baseline MRI scan (n = 1), had irreparable ar- Clinical Variables tifacts (n = 1), or had excessive head motion (n = 2), our final sample included 55 patients. Diagnosis, n Schizophrenia 27 Statistical Analyses Schizoaffective disorder 11 Bipolar disorder with psychosis 3 For morphometric analyses (volume, cortical thickness, and surface), labels of the DMN, SN, and CEN defined from the Schizophreniform disorder 2 CorticalParcellation_Yeo2011 (FreeSurfer) (43) were projected Psychosis NOS 10 onto each subject. Measures were extracted from right and left Brief psychotic disorder 1 hemisphere labels and entered into partial correlations with Major depressive disorder with psychosis 1 DUP (log transformed to account for nonnormal distribution of DUP, months 19.65 6 39.42 (0.5–180) data) while controlling for age and estimated total intracranial Treatment response, %b 60.43 6 21.06 (9–82) volume. Because sex and estimated total intracranial volume BPRS baselinec were highly correlated, we only controlled for the latter. Total 51.4 6 11.99 (32–84) For FC analyses, 4 CONN toolbox-based regions of interest Positive 11.74 6 3.55 (3–20) were used: DMN (posterior cingulate cortex), SN (right anterior Negative 6.23 6 3.38 (3–16) insula), and CEN (left and right posterior parietal cortexes). We BPRS week 16d then created individual whole-brain -transformed correlation z Total 29.56 6 5.82 (20–45) maps for each region of interest to each voxel of the brain. Positive 4.44 6 1.91 (3–10) Second-level analyses were performed for each correlation map using separate general linear models with DUP with age, Negative 5.53 6 2.49 (3–12) sex, and framewise displacement as covariates. All analyses Risperidone dose at week 16, mg 4.76 6 2.42 (1–8) RBANSe were corrected using voxel (puncorrected , .01) and cluster-level Total index 75.67 6 15.04 (50–117) and false discovery rate (FDR) correction (pFDR , .05). Finally, a mediation analysis was performed to examine Immediate memory 83.02 6 18.37 (44–123) whether brain morphology and connectivity mediated the rela- Visuospatial/constructional 76.40 6 17.20 (50–121) tionship between DUP and TR. A factor analysis via principal Language 83.73 6 16.21 (40–112) component analysis was performed on the combined FC and Attention 81.56 6 16.38 (43–109) morphological results (Figure 1) to reduce the data and was Delayed memory 79.69 6 13.26 (40–102) examined using a varimax rotation on SPSS, v25 (IBM Corp., Scan Quality Data Armonk, NY). In accordance with Judd and Kenny (44), we tested Volumes after scrubbing, % 93.77 6 6.97 (70–100) whether 1) the effect of DUP on TR is significant, 2) a significant Framewise displacement, mm 0.32 6 0.19 (0.11–0.89) relationship exists between DUP and brain data, 3) the relationship between brain data and TR is significant, and 4) the direct Values are mean 6 SD (range), unless otherwise indicated. BPRS, Brief Psychiatric Rating Scale; DUP, duration of untreated effect of DUP on TR adjusted for brain data is not significant. psychosis; NOS, not otherwise specified; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status. aRanks determined from Diagnostic Interview for Genetic Studies RESULTS (1–18 scale); higher rank (lower numerical value) corresponds to higher socioeconomic status. Demographics and Clinical Data bData available for 48 patients. Demographic data are summarized in Table 1. The average cData available for 53 patients. DUP was 19.65 months (SD = 39.42 months), and the median dData available for 43 patients. was 6 months. Age and sex were not correlated with DUP. eData available for 41 patients. Longer DUP prior to treatment was associated with worse TR after controlling for age and sex (b = 20.376, p = .019). FC in bilateral medial and pre- and postcentral regions (pFDR , Repeatable Battery for the Assessment of Neuropsychological .05) (Table 2). On the other hand, cortical thickness of the right Status scores were not correlated with DUP. hemisphere was positively correlated with DUP (r =.25,p = .04) (Figure 1A, Table 3). For the SN, significant clusters showing Brain Functional Connectivity and Morphology negative correlations between DUP and baseline right anterior Results for DMN revealed significant clusters of negative cor- insula FC were found in bilateral temporal regions, left middle relations between DUP and baseline posterior cingulate cortex frontal gyrus, right medial prefrontal gyrus, and right precentral

234 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:231–238 www.sobp.org/BPCNNI Biological Psychiatry: DUP, Brain Connectivity, and Morphology in FEP Patients CNNI

Table 2. Regions Exhibiting Signiﬁcant Correlations Between FC and DUP, Separately for DMN, SN, and CEN MNI Coordinates Seed (Network) Location, Hemisphere xyz Cluster Size, Voxels R2 PCC (DMN) Frontal medial cortex 212 42 212 196 .35 Precentral gyrus, L 22 232 68 132 .25 Postcentral gyrus, R 36 228 64 124 .25 RAI (SN) Planum temporale, R 64 232 8 530 .39 Middle temporal gyrus, L 260 246 24 313 .41 Middle frontal gyrus, L 238 32 22 197 .41 Supramarginal gyrus, L 242 246 40 173 .28 Precentral gyrus, L 52 2 44 129 .31 Lateral occipital cortex, L 230 264 56 118 .29 Planum temporale, L 260 226 10 111 .28 Paracingulate gyrus, L 4 28 32 97 .28 RPPC (CEN) Supramarginal gyrus, R 54 240 36 165 .31 Inferior temporal gyrus, L 248 252 212 141 .28 Temporal occipital, R 32 256 210 136 .38 Precentral gyrus, L 258 4 34 115 .33 Middle frontal gyrus, R 48 16 34 112 .31 LPPC (CEN) Supramarginal gyrus, R 66 238 28 372 .33 CEN, central executive network; DMN, default mode network; DUP, duration of untreated psychosis; FC, functional connectivity; L, left; LPPC, left posterior parietal cortex; MNI, Montreal Neurological Institute; PCC, posterior cingulate cortex; R, right; RAI, right anterior insula; RPPC, right posterior parietal cortex; SN, salience network.

gyrus (pFDR , .05) (Table 2). Similarly, negative and positive with the highest explained variance (34%) included loadings correlations with DUP were found between left hemisphere from FC of DMN and SN after a varimax rotation solution; surface area (r = 2.27, p = .03) and right hemisphere cortical however, it did not meet the requirements for being a potential thickness (r =.28,p = .03), respectively (Figure 1B, Table 3). mediator (item 3 [see Statistical Analyses]). As an exploratory Lastly, for CEN, significant clusters showing a negative corre- approach, we tested separately the DMN and SN FC data for lation between DUP and baseline right posterior parietal cortex potential mediation, given that the factor loadings came mostly FC were found in left precentral, right angular, and ventral from these networks. We found that FC of the DMN that temporal regions. In contrast, a cluster showing a positive correlated with DUP (clusters from Figure 1A) met all re- correlation was found in right middle gyrus, and when using the quirements for being a potential mediator, but FC of the SN did left posterior parietal cortex seed, a cluster showing a negative not (44). When FC of the DMN was included as a mediator in correlation between DUP and baseline connectivity was found the model, the direct path from DUP to TR (b = 20.376, p = in the right angular gyrus (pFDR , .05) (Table 2). Similarly, .019) was no longer significant (b = 20.396, p = .072), which negative correlations between left and right hemisphere surface indicated that DUP exerted its effect on TR through FC of the were found with DUP (r = 2.28, p = .02 and r = 2.25, p = .04, DMN (Figure 2). This analysis showed a significant mediated respectively) (Figure 1C, Table 3). No significant correlations effect using a Sobel test (z = 21.94, p = .03). emerged between DUP and cortical volume.

Mediation Analysis DISCUSSION A total of 6 factors (of 22 total) had eigenvalues .1.0, cumu- To our knowledge, we are the ﬁrst to simultaneously investigate latively accounting for 77% of the total variance. The factor brain function and structure in 3 major neural networks in antipsychotic-naïve patients with FEP to understand the neural Table 3. Correlations Between DUP and Brain Morphology, substrates of the relationship between DUP and clinical out- Separately for DMN, SN, and CEN (n = 53) comes. We demonstrated that longer DUP was associated with Cortical reduced FC in all 3 networks and a single cluster of increased Surface Area Thickness Volume FC in the CEN network. Furthermore, we found that greater DUP Network Left Right Left Right Left Right was associated with reduced surface area in the SN and CEN, DMN 2.21a 2.21a .17 .25b 2.11 2.05 and we reported positive correlations between cortical thickness SN 2.27b 2.23a .15 .28b 2.18 2.08 and DUP in the SN and DMN. Importantly, our data empirically support that DMN connectivity mediates the relationship be- CEN 2.25b 2.28b .20a .22a 2.14 2.12 tween DUP and treatment response, implicating brain network All correlations are controlled for age and intracranial volume. connectivity as a neurobiological underpinning of the relation- CEN, central executive network; DMN, default mode network; DUP, duration of untreated psychosis; SN, salience network. ship between longer DUP and poorer clinical outcomes. ap , .1. Our ﬁndings are consistent with the only two studies that bp , .05. examined FC in relation to DUP (21,22). Sarpal et al. (22)

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:231–238 www.sobp.org/BPCNNI 235 Biological Psychiatry: CNNI DUP, Brain Connectivity, and Morphology in FEP Patients

may cause these alterations is currently unknown, although different mechanisms have been proposed (10,15). Our principal component analysis of all signiﬁcant brain morphology and functional connectivity matrix revealed a factor explaining 34% of the variance that is composed of FC only, indicating its greater contribution to the association with the DUP. Shorter DUP associated with better outcome implicates a process by which APDs attenuate the pathophysiological process underlying the DUP. Our results indicate that DMN connectivity positively predicted TR and mediated the relationship between DUP and TR. Links between the DMN and dopamine activity exist. Modulation of the DMN network has been achieved with various dopaminergic compounds (57–60) and is associated with striatal dopamine transporter activity (61) and genetic variation of the COMT (catechol-O- FPO

= methyltransferase) enzyme (62). We previously reported that the FC between the ventral tegmental area, which is the origin of mesocorticolimbic dopamine projections, and the DMN web 4C predicted TR (27,28,63). In addition, a recent study that eval- Figure 2. Duration of untreated psychosis (DUP) significantly predicts treatment response (TR) (indicated by a). When the signal from significant uated both structural and functional predictors of TR identified clusters from Figure 1A (default mode network functional connectivity [FC DMN connectivity as the single most important predictor of DMN]) is added to the model (indicated by b), DUP no longer predicts TR response (64). Together these results indicate an important (Sobel test [z = 21.94, p = .03]). std., standardized. role of the DMN FC in treatment response, as well as being a neurobiological underpinning of the relationship between DUP reported correlations between corticostriatal connectivity and and treatment response. DUP in early phase schizophrenia with ,2 years of antipsy- Like Bora et al. (65) and Goldberg et al. (4), we did not find chotic exposure. Similar to us, they found that corticostriatal an association between DUP and cognitive deficits measured connectivity mediated the relationship between DUP and with the Repeatable Battery for the Assessment of Neuro- treatment response. More recently, Manivannan et al. (21) re- psychological Status. This may be because cognitive deficits ported correlations between task-based functional MRI using a are already established at the onset of psychosis and do not visuospatial working memory task with DUP in patients with worsen in the early stage of illness, or that more fine-grained FEP (22 medication naïve and 12 with prior APD exposure). measures of cognition are needed. Additionally, the hetero- We observed negative correlations between surface area of geneity within patients with FEP who show different levels of SN and CEN, and positive correlations between cortical cognitive impairment, could potentially mask this association thickness of DMN and SN with DUP. Previous studies reported (66). a set of mixed results, where some have found a direct link There are several strengths and limitations to our study. We between brain morphology and DUP (16–18) and others have enrolled a sample of medication-naïve patients with FEP at first failed to do so (19,24). However, the majority of these studies treatment contact in our medical center; the vast majority of only examined volume. Because cortical thickness and surface them had never been exposed to antipsychotic medications area have distinct developmental trajectories and genetic in- before the first scan, allowing us to study psychosis with fluences, they index different biological entities (45,46). Sur- negligible confound of medications on measurements, although face area is a loose representation of the number of medication before scanning could affect connectivity. We minicolumns running perpendicular to the surface of the brain, chose not to exclude patients with cannabis use; approximately whereas cortical thickness reflects the size, density, and 30% of our sample tested positive for cannabis at study entry. arrangement of neurons, neuroglia, and nerve fibers (47). Though exposure to cannabis may affect brain structure and Reduction in neuropil and GABAergic (gamma-aminobutyric function, it is among the major risk factors for developing a acidergic) interneurons (which are found in cortical mini- psychotic illness and therefore highly clinically relevant. columns) previously found in schizophrenia in postmortem Including only patients without a history of cannabis use would studies may help explain reduction in surface area in frontal have inadvertently biased our sample and limited the general- and temporal areas (48–51). Similarly, increased cortical izability of our data. We also obtained DUP from patients and thickness in medial prefrontal and insular regions may be their next of kin in a retrospective manner. A quantitative review associated with brain inflammation, which has been found in comparing methods of gauging DUP concluded that clinical the early stages of schizophrenia and may be the result of interviews are no less reliable than standardized assessment global and focal reduction, hypodensities, and deficits in tools, but that the definition of treatment onset as first-ever myelin and activated microglia (52–56). Overall, reduction of antipsychotic medication prescription or first hospitalization surface area and increased cortical thickness as a function of may have greater validity than other methods such as first DUP strongly point to alterations in the functional organization adequate response to treatment (67). of the neocortex with longer DUP. To summarize, we simultaneously investigated brain function While these findings suggest that one or several patho- and structure in 3 major neural networks in medication-naïve physiologic processes underlie the DUP, the mechanism that patients with FEP and found that FC of the DMN mediated the

236 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:231–238 www.sobp.org/BPCNNI Biological Psychiatry: DUP, Brain Connectivity, and Morphology in FEP Patients CNNI

relationship between DUP and treatment response, suggesting deficits and brain morphology in first-episode schizophrenia. Am J a neurobiological mechanism underlying this phenomenon. Psychiatry 160:142–148. Overall, our findings add knowledge to the emerging theme of 6. Perkins DO (2006): Review: Longer duration of untreated psychosis is associated with worse outcome in people with first episode psychosis. the National Institute of Mental Health of further understanding Evid Based Ment Health 9:36. the neural correlates of DUP in psychopathology. Finally, these 7. Kane JM, Robinson DG, Schooler NR, Mueser KT, Penn DL, findings highlight the importance of reducing DUP and initiation Rosenheck RA, et al. (2016): Comprehensive versus usual community of treatment as soon as possible to mitigate the detrimental care for first-episode psychosis: 2-Year outcomes from the NIMH effects of psychosis on long-term clinical outcomes. RAISE early treatment program. Am J Psychiatry 173:362–372. 8. Marshall M, Lewis S, Lockwood A, Drake R, Jones P, Croudace T (2005): Association between duration of untreated psychosis and ACKNOWLEDGMENTS AND DISCLOSURES outcome in cohorts of first-episode patients: A systematic review. Arch Gen Psychiatry 62:975–983. This work was supported by grants from the National Institutes of Health 9. Wyatt RJ (1991): Neuroleptics and the natural course of schizophrenia. (Grant Nos. R01MH102951 and R01MH113800 [to ACL]). Schizophr Bull 17:325–351. The funding agency had no role in the design and conduct of the study; 10. Lisman JE, Coyle JT, Green RW, Javitt DC, Benes FM, Heckers S, collection, management, analysis, and interpretation of the data; prepara- Grace AA (2008): Circuit-based framework for understanding neuro- tion, review, or approval of the manuscript; and decision to submit the transmitter and risk gene interactions in schizophrenia. Trends Neu- manuscript for publication. rosci 31:234–242. ACL had full access to all of the data in the study and takes responsibility 11. Olney JW, Farber NB (1995): Glutamate receptor dysfunction and for the integrity of the data and the accuracy of the data analysis. ACL schizophrenia. Arch Gen Psychiatry 52:998–1007. conceived and designed the study. EAN, WPA, and NVK acquired the data. 12. Olney JW, Wozniak DF, Jevtovic-Todorovic V, Farber NB, Bittigau P, All authors analyzed and interpreted the data. JOM, NVK, and ACL drafted Ikonomidou C (2002): Drug-induced apoptotic neurodegeneration in the manuscript. JOM performed the statistical analyses. ACL obtained the the developing brain. Brain Pathol 12:488–498. funding. ACL obtained administrative, technical, or material support. ACL 13. Simantov R, Blinder E, Ratovitski T, Tauber M, Gabbay M, Porat S supervised the study. (1996): Dopamine-induced apoptosis in human neuronal cells: Inhibi- We give special thanks to the patients and their families. tion by nucleic acids antisense to the dopamine transporter. Neuro- This study was presented in part at the Schizophrenia International science 74:39–50. Research Society (SIRS) Conference, April 13, 2019, Orlando, Florida. 14. Keshavan MS, Haas GL, Kahn CE, Aguilar E, Dick EL, Schooler NR, The authors report no biomedical financial interests or potential conflicts et al. (1998): Superior temporal gyrus and the course of early schizophrenia: of interest. Progressive, static, or reversible? J Psychiatr Res 32:161–167. ClinicalTrials.gov: Trajectories of Treatment Response as Window into 15. Anderson KK, Voineskos A, Mulsant BH, George TP, McKenzie KJ the Heterogeneity of Psychosis: A Longitudinal Multimodal Imaging Study; (2014): The role of untreated psychosis in neurodegeneration: A review https://clinicaltrials.gov/ct2/show/NCT03442101; NCT03442101. Gluta- of hypothesized mechanisms of neurotoxicity in first-episode psy- mate, Brain Connectivity, and Duration of Untreated Psychosis (DUP); chosis. Can J Psychiatry 59:513–517. https://clinicaltrials.gov/ct2/show/NCT02034253; NCT02034253. 16. Goff DC, Zeng B, Ardekani BA, Diminich ED, Tang Y, Fan X, et al. (2018): Association of hippocampal atrophy with duration of untreated psychosis and molecular biomarkers during initial antipsychotic ARTICLE INFORMATION treatment of first-episode psychosis. JAMA Psychiatry 75:370–378. 17. Guo X, Li J, Wei Q, Fan X, Kennedy DN, Shen Y, et al. (2013): Duration From the Department of Psychiatry and Behavioral Neurobiology (JOM, of untreated psychosis is associated with temporal and occipito- WPA, NVK, ACL) and Department of Psychology (EAN), University of Ala- temporal gray matter volume decrease in treatment naive schizo- bama at Birmingham, Birmingham, Alabama. phrenia. PLoS One 8:e83679. Address correspondence to Adrienne C. Lahti, M.D., University of 18. Malla AK, Bodnar M, Joober R, Lepage M (2011): Duration of untreated Alabama at Birmingham, Department of Psychiatry and Behavioral psychosis is associated with orbital-frontal grey matter volume re- Neurobiology, SC 501, 1530 3rd Avenue South, Birmingham, AL 35294- ductions in first episode psychosis. Schizophr Res 125:13–20. 0017; E-mail: [email protected]. 19. Rapp C, Canela C, Studerus E, Walter A, Aston J, Borgwardt S, Received Sep 24, 2019; revised and accepted Oct 29, 2019. Riecher-Rossler A (2017): Duration of untreated psychosis/illness and brain volume changes in early psychosis. Psychiatry Res 255:332–337. 20. Lee SW, Lee A, Choi TK, Kim B, Lee KS, Bang M, Lee SH (2018): White matter abnormalities of the tapetum and their associations with REFERENCES duration of untreated psychosis and symptom severity in first-episode 1. Perkins DO, Gu H, Boteva K, Lieberman JA (2005): Relationship be- psychosis. Schizophr Res 201:437–438. tween duration of untreated psychosis and outcome in first-episode 21. Manivannan A, Foran W, Jalbrzikowski M, Murty VP, Haas GL, schizophrenia: A critical review and meta-analysis. Am J Psychiatry Tarcijonas G, et al. (2019): Association between duration of untreated 162:1785–1804. psychosis and frontostriatal connectivity during maintenance of vi- 2. Boonstra N, Klaassen R, Sytema S, Marshall M, De Haan L, suospatial working memory. Biol Psychiatry Cogn Neurosci Neuro- Wunderink L, Wiersma D (2012): Duration of untreated psychosis and imaging 4:454–461. negative symptoms—A systematic review and meta-analysis of indi- 22. Sarpal DK, Robinson DG, Fales C, Lencz T, Argyelan M, Karlsgodt KH, vidual patient data. Schizophr Res 142:12–19. et al. (2017): Relationship between duration of untreated psychosis 3. Farooq S, Large M, Nielssen O, Waheed W (2009): The relationship and intrinsic corticostriatal connectivity in patients with early phase between the duration of untreated psychosis and outcome in low-and- schizophrenia. Neuropsychopharmacology 42:2214–2221. middle income countries: A systematic review and meta-analysis. 23. Galinska B, Szulc A, Tarasow E, Kubas B, Dzienis W, Czernikiewicz A, Schizophr Res 109:15–23. Walecki J (2009): Duration of untreated psychosis and proton mag- 4. Goldberg TE, Burdick KE, McCormack J, Napolitano B, Patel RC, netic resonance spectroscopy (1H-MRS) findings in first-episode Sevy SM, et al. (2009): Lack of an inverse relationship between dura- schizophrenia. Med Sci Monit 15:CR82–CR88. tion of untreated psychosis and cognitive function in first episode 24. Anderson KK, Rodrigues M, Mann K, Voineskos A, Mulsant BH, schizophrenia. Schizophr Res 107:262–266. George TP, McKenzie KJ (2015): Minimal evidence that untreated 5. Ho BC, Alicata D, Ward J, Moser DJ, O’Leary DS, Arndt S, psychosis damages brain structures: A systematic review. Schizophr Andreasen NC (2003): Untreated initial psychosis: Relation to cognitive Res 162:222–233.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:231–238 www.sobp.org/BPCNNI 237 Biological Psychiatry: CNNI DUP, Brain Connectivity, and Morphology in FEP Patients

25. Chakos MH, Lieberman JA, Bilder RM, Borenstein M, Lerner G, 46. Wierenga LM, Langen M, Oranje B, Durston S (2014): Unique devel- Bogerts B, et al. (1994): Increase in caudate nuclei volumes of first- opmental trajectories of cortical thickness and surface area. Neuro- episode schizophrenic patients taking antipsychotic drugs. Am J image 87:120–126. Psychiatry 151:1430–1436. 47. Rakic P (1988): Defects of neuronal migration and the pathogenesis of 26. Navari S, Dazzan P (2009): Do antipsychotic drugs affect brain cortical malformations. Prog Brain Res 73:15–37. structure? A systematic and critical review of MRI findings. Psychol 48. Casanova MF (2009): Schizophrenia seen as a deficit in the modulation Med 39:1763–1777. of cortical minicolumns by monoaminergic systems. Int Rev Psychiatry 27. Hadley JA, Kraguljac NV, White DM, Ver Hoef L, Tabora J, Lahti AC 4:361–372. (2016): Change in brain network topology as a function of treatment 49. Glausier JR, Lewis DA (2013): Dendritic spine pathology in schizo- response in schizophrenia: A longitudinal resting-state fMRI study phrenia. Neuroscience 251:90–107. using graph theory. NPJ Schizophr 2:16014. 50. Raghanti MA, Spocter MA, Butti C, Hof PR, Sherwood CC (2010): 28. Kraguljac NV, White DM, Hadley JA, Visscher K, Knight D, ver Hoef L, A comparative perspective on minicolumns and inhibitory GABAergic et al. (2016): Abnormalities in large scale functional networks in un- interneurons in the neocortex. Front Neuroanat 4:3. medicated patients with schizophrenia and effects of risperidone. 51. Sweet RA, Bergen SE, Sun Z, Sampson AR, Pierri JN, Lewis DA Neuroimage Clin 10:146–158. (2004): Pyramidal cell size reduction in schizophrenia: Evidence for 29. Menon V (2011): Large-scale brain networks and psychopathology: A involvement of auditory feedforward circuits. Biol Psychiatry unifying triple network model. Trends Cogn Sci 15:483–506. 55:1128–1137. 30. Brandl F, Avram M, Weise B, Shang J, Simoes B, Bertram T, et al. (2019): 52. Bernstein HG, Steiner J, Guest PC, Dobrowolny H, Bogerts B (2015): Specific substantial dysconnectivity in schizophrenia: A transdiagnostic Glial cells as key players in schizophrenia pathology: Recent insights multimodal meta-analysis of resting-state functional and and concepts of therapy. Schizophr Res 161:4–18. structural magnetic resonance imaging studies. Biol Psychiatry 85:573– 53. Korschenhausen DA, Hampel HJ, Ackenheil M, Penning R, Muller N 583. (1996): Fibrin degradation products in post mortem brain tissue of 31. Glahn DC, Ragland JD, Abramoff A, Barrett J, Laird AR, Bearden CE, schizophrenics: A possible marker for underlying inflammatory pro- Velligan DI (2005): Beyond hypofrontality: A quantitative meta-analysis cesses. Schizophr Res 19:103–109. of functional neuroimaging studies of working memory in schizo- 54. Muller N, Schwarz M (2006): Schizophrenia as an inflammation- phrenia. Hum Brain Mapp 25:60–69. mediated dysbalance of glutamatergic neurotransmission. Neurotox 32. Gallego JA, Robinson DG, Sevy SM, Napolitano B, McCormack J, Res 10:131–148. Lesser ML, Kane JM (2011): Time to treatment response in first- 55. Muller N, Weidinger E, Leitner B, Schwarz MJ (2015): The role of episode schizophrenia: Should acute treatment trials last several inflammation in schizophrenia. Front Neurosci 9:372. months? J Clin Psychiatry 72:1691–1696. 56. Pasternak O, Westin CF, Bouix S, Seidman LJ, Goldstein JM, Woo TU, 33. Overall JE, Gorham DR (1962): The brief psychiatric rating scale. et al. (2012): Excessive extracellular volume reveals a neurodegener- Psychol Rep 10:799–812. ative pattern in schizophrenia onset. J Neurosci 32:17365–17372. 34. Randolph C, Tierney MC, Mohr E, Chase TN (1998): The Repeatable 57. Kelly C, de Zubicaray G, Di Martino A, Copland DA, Reiss PT, Klein DF, Battery for the Assessment of Neuropsychological Status (RBANS): et al. (2009): L-dopa modulates functional connectivity in striatal Preliminary clinical validity. J Clin Exp Neuropsychol 20:310–319. cognitive and motor networks: A double-blind placebo-controlled 35. Fischl B, Sereno MI, Dale AM (1999): Cortical surface-based analysis. study. J Neurosci 29:7364–7378. II: Inflation, flattening, and a surface-based coordinate system. Neu- 58. Nagano-Saito A, Liu J, Doyon J, Dagher A (2009): Dopamine modu- roimage 9:195–207. lates default mode network deactivation in elderly individuals during 36. Dale AM, Fischl B, Sereno MI (1999): Cortical surface-based analysis. the Tower of London task. Neurosci Lett 458:1–5. I. Segmentation and surface reconstruction. Neuroimage 9:179–194. 59. Argyelan M, Carbon M, Ghilardi MF, Feigin A, Mattis P, Tang C, et al. 37. Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, (2008): Dopaminergic suppression of brain deactivation responses Johansen-Berg H, et al. (2004): Advances in functional and structural during sequence learning. J Neurosci 28:10687–10695. MR image analysis and implementation as FSL. Neuroimage 60. Tomasi D, Volkow ND, Wang GJ, Wang R, Telang F, Caparelli EC, et al. 23(suppl 1):S208–S219. (2011): Methylphenidate enhances brain activation and deactivation 38. Desikan RS, Segonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, responses to visual attention and working memory tasks in healthy et al. (2006): An automated labeling system for subdividing the human controls. Neuroimage 54:3101–3110. cerebral cortex on MRI scans into gyral based regions of interest. 61. Tomasi D, Volkow ND, Wang R, Telang F, Wang GJ, Chang L, et al. Neuroimage 31:968–980. (2009): Dopamine transporters in striatum correlate with deactivation 39. Klapwijk ET, van de Kamp F, van der Meulen M, Peters S, in the default mode network during visuospatial attention. PLoS One 4: Wierenga LM (2019): Qoala-T: A supervised-learning tool for e6102. quality control of FreeSurfer segmented MRI data. Neuroimage 62. Liu B, Song M, Li J, Liu Y, Li K, Yu C, Jiang T (2010): Prefrontal-related 189:116–129. functional connectivities within the default network are modulated by 40. Whitfield-Gabrieli S, Nieto-Castanon A (2012): CONN: A functional COMT val158met in healthy young adults. J Neurosci 30:64–69. connectivity toolbox for correlated and anticorrelated brain networks. 63. Hadley JA, Nenert R, Kraguljac NV, Bolding MS, White DM, Brain Connect 2:125–141. Skidmore FM, et al. (2014): Ventral tegmental area/midbrain functional 41. Glasser MF, Sotiropoulos SN, Wilson JA, Coalson TS, Fischl B, connectivity and response to antipsychotic medication in schizo- Andersson JL, et al. (2013): for the WU-Minn HCP Consortium. The phrenia. Neuropsychopharmacology 39:1020–1030. minimal preprocessing pipelines for the Human Connectome Project. 64. Doucet GE, Moser DA, Luber MJ, Leibu E, Frangou S (2018): Baseline Neuroimage 80:105–124. brain structural and functional predictors of clinical outcome in the 42. Power JD, Mitra A, Laumann TO, Snyder AZ, Schlaggar BL, early course of schizophrenia [published online ahead of print Oct 3]. Petersen SE (2014): Methods to detect, characterize, and remove Mol Psychiatry. motion artifact in resting state fMRI. Neuroimage 84:320–341. 65. Bora E, Yalincetin B, Akdede BB, Alptekin K (2018): Duration of un- 43. Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, treated psychosis and neurocognition in first-episode psychosis: A et al. (2011): The organization of the human cerebral cortex estimated by meta-analysis. Schizophr Res 193:3–10. intrinsic functional connectivity. J Neurophysiol 106:1125–1165. 66. Bora E (2016): Differences in cognitive impairment between schizo- 44. Judd CM, Kenny DA (1981): Process analysis estimating mediation in phrenia and bipolar disorder: Considering the role of heterogeneity. treatment evaluations. Eval Rev 5:602–608. Psychiatry Clin Neurosci 70:424–433. 45. Panizzon MS, Fennema-Notestine C, Eyler LT, Jernigan TL, Prom- 67. Register-Brown K, Hong LE (2014): Reliability and validity of methods Wormley E, Neale M, et al. (2009): Distinct genetic influences on cortical for measuring the duration of untreated psychosis: A quantitative re- surface area and cortical thickness. Cereb Cortex 19:2728–2735. view and meta-analysis. Schizophr Res 160:20–26.

238 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:231–238 www.sobp.org/BPCNNI Biological Psychiatry: Archival Report CNNI

Regulation of Craving and Negative Emotion in Alcohol Use Disorder

Shosuke Suzuki, Maggie Mae Mell, Stephanie S. O’Malley, John H. Krystal, Alan Anticevic, and Hedy Kober

ABSTRACT BACKGROUND: Alcohol use disorder (AUD) is a chronic, relapsing condition with poor treatment outcomes. Both alcohol craving and negative affect increase alcohol drinking, and—in healthy adults—can be attenuated using cognitive strategies, which rely on the prefrontal cortex (PFC). However, AUD is associated with cognitive impairments and PFC disruptions. Thus, we tested whether individuals with AUD can successfully recruit the PFC to effectively regulate craving and negative emotions, whether neural mechanisms are shared between the two types of regulation, and whether individual differences inﬂuence regulation success. METHODS: During functional magnetic resonance imaging, participants with AUD completed the regulation of craving task (n = 17) that compares a cue-induced craving condition with an instructed regulation condition. They also completed the emotion regulation task (n = 15) that compares a negative affect condition with an instructed regulation condition. Regulation strategies were drawn from cognitive behavioral therapy treatments for AUD. Self-reported craving and negative affect were collected on each trial. RESULTS: Individuals with AUD effectively regulated their craving and negative affect when instructed to do so using cognitive behavioral therapy–based strategies. Regulation was associated with recruitment of both common and distinct PFC regions across tasks, as well as with reduced activity in regions associated with craving and negative affect (e.g., ventral striatum, amygdala). Effective regulation of craving was associated with negative alcohol expectancies. CONCLUSIONS: Both common and distinct regulatory systems underlie regulation of craving and negative emotions in AUD, with notable individual differences. This has important implications for AUD treatment. Keywords: Alcohol use disorder, Craving, Emotion regulation, fMRI, Negative affect, Regulation of craving https://doi.org/10.1016/j.bpsc.2019.10.005

Today, alcohol remains the most commonly used psychoac- Importantly, craving can be triggered by stimuli that have tive substance (other than caffeine) (1,2), despite mounting previously been associated with the substance (26). Such cue- evidence of its harmful effects (3). Importantly, 9% to 17% of induced craving has been directly linked to drug and alcohol drinkers meet criteria for alcohol use disorder (AUD), a chronic, use (27–31). Additionally, intensity of cue-induced craving relapsing condition (2,4,5) with enormous societal costs (6). predicts relapse to drugs (20,32) and alcohol (33–36), and in- Many individuals with AUD never receive or delay treatment creases in cue-induced responses are reported months into (7–9), and the modal treatment outcome is relapse (10,11). abstinence (37). As such, cue-induced craving contributes to Therefore, it is important to understand the factors that high rates of relapse, and thereby to the chronicity of SUDs contribute to substance use disorders (SUDs), including AUD. and difﬁculty of treatment. Given the role of cue-induced craving in relapse and SUD maintenance, regulation of craving (ROC) has become an Craving and Its Regulation in AUD important clinical target. In fact, cognitive behavioral therapy One factor that contributes to drug and alcohol use is craving, (CBT) for SUDs focuses on situations in which individuals may deﬁned as “a strong desire or urge” (12). Craving has been be exposed to drug cues or crave or seek the substance (38). linked to drug and alcohol use across retrospective reports To cope with such “high-risk situations,” CBT teaches cogni- (13–15), prospective clinical studies (16–21), and ecological tive strategies, such as orienting individuals to focus on the momentary assessments (22–25). Such studies have found negative consequences of drug use [vs. pleasurable conse- that craving is cross-sectionally and prospectively associated quences (38)]. Importantly, recent experiments have demon- with increased drug use. Thus, craving was added as a diag- strated that cognitive strategies can reduce cue-induced nostic criterion for SUDs in the DSM-5 (12). craving for cigarettes (39–41), stimulants (42,43), and food

ª 2019 Society of Biological Psychiatry. Published by Elsevier Inc. All rights reserved. 239 ISSN: 2451-9022 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI Biological Psychiatry: CNNI Regulation of Craving and Negative Emotion in AUD

(39,40,44–46). Clinically, implementation of such strategies has Given the relationship between negative affect and drug been associated with reduced smoking (47), and acquisition of use, regulation of negative affect has important implications for such skills mediates CBT treatment success for drug and/or SUD treatment (along with ROC) (56). A rich literature has alcohol use (48). These data suggest that cognitive strategies demonstrated that healthy adults can use cognitive strategies are an active component of CBT, at least in part owing to their to regulate affective responses to aversive stimuli (73). A recent effects on reducing craving. meta-analysis demonstrated that regulation is associated with However, only one study, to our knowledge, has experi- recruitment of the dlPFC, vlPFC, and dmPFC (similar to ROC) mentally investigated ROC in AUD (49). In this previous study, and with reduced activation in the amygdala (74). However, we taught participants to focus on the negative consequences cognitive emotion regulation (ER) has been poorly investigated of drinking during cue-induced alcohol craving (38–40). We in SUDs, with only 2 published studies to date (to our knowl- observed that—compared with social drinkers—individuals edge). In these studies, cognitive strategies reduced negative with AUD were less able to regulate craving using this CBT- emotional responses toward negative images in nicotine- based cognitive strategy (49). This is consistent with abun- dependent cigarette smokers (75) and patients with cocaine dant evidence linking alcohol drinking to deficits in cognitive use disorder (76). control [for a review, see Day et al. (50)] and studies that further link deficits to subsequent alcohol use [e.g., (51)]. Shared Neural Systems Supporting ROC and Despite the clinical importance of the ability to regulate Negative Emotion craving, its underlying neural mechanisms have not been An important remaining question is whether shared neural investigated in AUD. Using functional magnetic resonance mechanisms underlie the implementation of cognitive strate- imaging (fMRI), we previously showed that ROC in cigarette gies to regulate craving and negative affect. Prior studies smokers depends on recruitment of prefrontal brain regions, suggest that neural activation during different types of self- such as the dorsolateral prefrontal cortex (dlPFC), ventrolateral control may converge in the PFC (77,78). One study in PFC (vlPFC), and dorsomedial PFC (dmPFC) (40). This was cigarette smokers found that the dlPFC, vlPFC, and dmPFC/ accompanied by relative deactivations in reward-related re- presupplementary motor area were activated across self- gions, including the ventral striatum (VS) and ventromedial PFC control of emotion, craving, and motor impulses (78), sug- (vmPFC) (40). Similar findings have been reported in cocaine gesting a common prefrontal pathway supporting various use disorder (43) and with regulation of food craving (52–54). forms of cognitive regulation. However, this has yet to be replicated in any population and has not been investigated in Alcohol Expectancies AUD. Importantly, this question has implications for whether Regulation strategies require individuals to focus on the regulation success in one domain may generalize to other negative effects of alcohol; thus, individual differences in ex- domains of regulation. pectancies about alcohol may affect ROC. For example, in- Notably, AUD has consistently been linked to disruptions in dividuals who drink the most report higher expectations for the PFC function and structure [for a review, see Goldstein and positive effect of alcohol (55) and may disregard the negative Volkow (79)]. Human imaging studies have documented consequences of excessive drinking. However, it is unknown reduced prefrontal gray matter volume in AUD (80–82), which whether individuals who have higher negative expectations are was further associated with executive function impairments better at regulating craving. Importantly, identification of indi- (80). In addition, alcohol drinkers exhibit altered PFC activity, vidual differences that modulate regulation processes may including hyperactivity in the vmPFC and anterior cingulate help optimize and individualize treatments. cortex (ACC) during relaxing imagery (71), hypoactivity in the medial PFC (mPFC) and ACC during threat processing (83), Negative Affect and Its Regulation in AUD and disruptions in the PFC during negative emotional processing (84–86). Investigating whether individuals with AUD Negative affect is also thought to contribute to drug and can effectively recruit PFC mechanisms during regulation will alcohol use (56–58). Indeed, many theories of SUDs and AUD inform treatment approaches that utilize cognitive strategies focus on reduction of negative affect as motives for con- (48). sumption (57,59–61). In turn, the reduction of negative affect is considered negatively reinforcing, increasing the likelihood of continued use (62). Consistently, several lines of work link The Current Study negative affect, drug use, and SUDs. First, SUDs (including To address these questions, we used fMRI to scan individuals AUD) frequently co-occur with mood and anxiety disorders with AUD while they completed the ROC and ER tasks. These (6,56,63–65), and major depression predicts AUD (66). Second, tasks are specifically designed to examine the neural mecha- levels of self-reported negative affect have been linked to drug nisms underlying cognitive ROC and negative affect—invoked use. For example, negative affect predicts relapse in cigarette using alcohol-related and negatively valenced pictures, smokers (16) and correlates with drinking problems in adult respectively—given the importance of these processes to drinkers (67). Third, negative affective states trigger craving AUD. In addition, we tested whether there are shared prefrontal across drug types. Specifically, acute induction of negative mechanisms across these forms of cognitive regulation. affect has been shown to increase craving for cigarettes (68), Further, we used self-report to assess regulation success cocaine (69), opiates (70), and alcohol (33,71,72). Fourth, across tasks, and whether regulation success for alcohol negative affect and stress have been directly linked to drug use craving is modulated by individual differences in negative and relapse after treatment (72). alcohol expectancies.

240 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI Biological Psychiatry: Regulation of Craving and Negative Emotion in AUD CNNI

Table 1. Participant Demographics for Each Task runs). All but the first 2 participants were scanned in the af- ROC Task ER Task ternoon (w4 PM) to minimize any diurnal variations in alcohol craving; therefore, the 2 participants scanned in the morning Demographic (n=17) (n=15) Characteristics Mean 6 SD % Mean 6 SD % were excluded from ROC analysis. In addition, ROC data from 1 participant and ER data from 3 participants were excluded Age, Years 33.35 6 10.97 32.33 6 9.98 for excessive motion or technical errors. Thus, the final sample Education, Years 14.29 6 1.21 14.07 6 1.22 included 17 participants for the ROC task and 15 participants Female 41.2 40 for the ER task. All procedures were approved by Yale Uni- Hispanic 17.6 20 versity’s Institutional Review Board. Race White 47.1 40.0 Procedures African American 47.1 46.7 Participants were instructed to abstain from alcohol on scan- More than 1 race 5.9 6.7 ning day and from eating for 2 hours prior to their visit. Alcohol NA 0.0 6.7 abstinence was confirmed using a breathalyzer (BACtrack, San Diagnosis Francisco, CA). Psychoactive substance use was tested via Dependence 76.5 73.3 urinalysis (Redwood Toxicology Laboratory, Santa Rosa, CA). Abuse 23.5 26.7 Time Since Last Drink, 18.60 6 14.35 19.79 6 15.42 Negative Alcohol Expectancy Questionnaire Hours The Negative Alcohol Expectancy Questionnaire (87) assessed Average Drinks/Day During 8.22 6 5.02 7.63 6 4.76 participants’ expectations about the negative consequences of Past Month alcohol (3 subscales: same day, next day, long term). The long- ER, emotion regulation; NA, not available; ROC, regulation of term subscale is known to predict abstinence 3 months later (88). craving. ROC Task METHODS AND MATERIALS The ROC task assesses craving and ROC (39,40,42,45,49,89) using an event-related design. On each trial, participants were Participants presented with a unique picture of alcohol or food shown to Twenty-six individuals with AUD (Table 1) participated in this induce craving in prior studies (Figure 1A)(39,40,49). Prior to the 2-visit outpatient study. Recruitment was conducted via adver- picture, an instruction cue appeared in the center of the screen, tisements in the New Haven community. Participants were orienting participants to view the picture in one of two ways: eligible if they 1) met diagnostic criteria for alcohol abuse or 1) craving condition: “think about the immediate effects of dependence (assessed via the Structured Clinical Interview for consuming the item,” indicated by the word NOW; or 2) regu- DSM-IV Axis I Disorders); 2) did not meet criteria for any other lation condition (based on CBT): “think about the long-term ef- Axis I disorder, except nicotine dependence; 3) consumed $25 fects of regularly consuming the item,” indicated by the word drinks/week for men and $20 drinks/week for women; LATER. Importantly, to minimize demand characteristics, par- 4) consumed alcohol on $4 days/week; and 5) could under- ticipants were not explicitly told to reduce craving. Then, stand and consent to study procedures. Participants were following an exponentially jittered interstimulus interval (w2.73 excluded if they 1) had contraindications for MRI (e.g., pregnancy, seconds) (90), participants rated the intensity of their craving on metallic implants); 2) reported currently taking centrally active a scale from 1 (not at all) to 5 (very much). Trials were separated medications; 3) tested positive for psychoactive drugs (cocaine, by exponentially jittered intertrial intervals (w3.43 seconds) (90). opiates, methamphetamine, amphetamine, PCP, or benzodiaz- Participants completed 4 runs (20 trials/run). epines) as indicated by urinalysis; or 4) provided false information during screening. All participants provided informed consent ER Task during the initial screening visit and returned for fMRI scanning at The ER task assesses the regulation of negative affect Yale University’s Magnetic Resonance Research Center. (73,74,91,92) in response to negative images using an event- A priori sample size and data collection stopping targets related design. On each trial, participants were shown an in- were based on sample and effect sizes reported at the time in struction cue orienting them to view a subsequent picture in the extant ER literature [i.e., commonly around 16–20 partici- one of two ways: 1) look condition: “simply look at the picture,” pants; for a meta-analysis, see Buhle et al. (74)], on our prior indicated by the word LOOK, for neutral and negative pictures; study with the ROC task (n=21) (40), and on availability of or 2) regulation condition: “think about the image in a less funding. Four participants could not complete the MRI session negative way,” indicated by the word REAPPRAISE, for owing to unanticipated claustrophobia. One participant pro- negative pictures only (Figure 1B). Unique negative and neutral vided false information during screening, which was discov- pictures were drawn from the International Affective Picture ered following completion of the study. This participant’s data System (93). Following exponentially jittered interstimulus in- were removed prior to analysis (specifically, we discovered that tervals (w2.8 seconds) (90), participants rated the intensity of the participant was previously enrolled in another study in our their negative affect on a scale from 1 (not at all) to 5 (very lab and had provided conflicting information regarding his or negative). Trials were separated by exponentially jittered her substance use). Data from another participant were intertrial intervals (w3.5 seconds) (90). Participants completed excluded due to noncompliance (e.g., no responses in some 4 runs (15 trials/run).

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI 241 Biological Psychiatry: CNNI Regulation of Craving and Negative Emotion in AUD

Figure 1. Schematic representation of a single A trial of the (A) regulation of craving (ROC) and (B) emotion regulation (ER) tasks. Trials in the two tasks followed a parallel structure: each trial started with an instruction presented in the center of the screen ALCOHOL FOOD (NOW or LATER in the ROC task; LOOK or REAP- PRAISE in the ER task) for 2100 ms. The instructions oriented participants to focus on the subsequently presented picture in particular ways. Then, a picture NOW How much do ROC you want this was presented for 6300 ms (alcohol or food in the or Task + item? + ROC task; negative or neutral image in the ER task). LATER 1—2—3—4—5 A jittered interstimulus interval (1400–8400 ms) followed the image presentation. Participants were then asked to make a rating of their craving (ROC) or 2100ms 6300ms 1400-8400ms 2800ms 2100-9100ms negative affect (ER) on a 5-point scale (2800 ms). The intertrial interval was jittered at 2100–9100 ms. IAPS, International Affective Picture System. LOOK ER How negave do or [IAPS Image] you feel? web 4C/FPO + + Task REAPPRAISE 1—2—3—4—5

NEGATIVE NEUTRAL

[Negave [Neutral IAPS IAPS B Image] Image]

Behavioral Data Acquisition and Analysis fMRI Data Acquisition and Analysis The tasks were programmed in E-Prime version 2.0 (Psy- Participants were scanned in a 3T Siemens TIM Trio scanner chology Software Tools, Sharpsburg, PA) and presented (Siemens AG, Munich, Germany). One participant was scanned using a back-projection mirror. Participants provided ratings in a 3T Siemens Prisma (Siemens AG) due to scanner upgrade using an MRI-compatible 5-button box. Order of tasks was during the study period (between-scanner differences are counterbalanced. Analyses of behavioral data were con- minimal and do not compare with physiological noise) (97). ducted in SPSS, version 22 (IBM Corp., Armonk, NY). For the Importantly, excluding this participant did not change the main ROC task, we conducted a 2 instruction (NOW/LATER) 3 2 findings; thus, we included these data in analyses. Scan pa- image type (alcohol/food) repeated-measures analysis of rameters followed Human Connectome Project recommenda- variance. For the ER task, we conducted a 1-way analysis tions (98). Functional images were acquired with T2*-weighted of variance with condition as a within-subjects factor with echo-planar pulse sequences with a multiband acceleration 3 levels (REAPPRAISE-Negative/LOOK-Negative/LOOK- factor of 6 (repetition time/echo time = 700 ms/31 ms; flip Neutral). Significant effects were followed up using post hoc angle = 55; field of view = 210 3 210 mm; 54 3 2.50 mm pairwise comparisons. Finally, following prior studies using the slices). High-resolution structural images were acquired using ROC and ER tasks [e.g., (40,49,89,92,94)], we calculated a a single-shot, magnetization prepared rapid acquisition “regulation success” score for each participant, for each task. gradient-echo sequence (repetition time/echo time = 2400 ms/ Specifically, for the ROC task, the mean self-reported craving 2.01 ms; flip angle = 8; field of view = 256 3 256 mm; 224 3 during the regulation condition was subtracted from that 0.80 mm slices). during the craving condition, and for the ER task, the mean Functional images were preprocessed using SPM8 (Well- self-reported negative affect during the REAPPRAISE- come Trust Centre for Neuroimaging, London, United Negative condition was subtracted from that during the Kingdom) following our prior work [e.g., (99)]. Specifically, LOOK-Negative condition. Thus, for each task, positive functional images were co-registered to the structural image, regulation success scores indicated successful reduction in motion-corrected, warped to the Montreal Neurological Insti- craving or negative affect, whereas negative scores indicated tute template, and smoothed using a Gaussian filter (5-mm full an increase in craving or negative affect during the regulation width at half maximum). Both before and after preprocessing, condition. We then assessed whether negative alcohol ex- data were subjected to multiple tests for quality assurance and pectancies correlated with regulation success in the ROC inspected for signal spiking and motion. Volumes were dis- task, and whether ROC regulation success correlated with carded if the root mean square of motion parameters excee- regulation success in the ER task. A missing value on the ded half of a single voxel size (100). First-level modeling of trial Negative Alcohol Expectancy Questionnaire was imputed for 1 events was conducted using robust regression to reduce the participant (95,96). An alpha level of p , .05 was used across influence of strong outliers (99,101,102). Following prior work analyses. [e.g., (40,92)], for both tasks, the instruction cue and picture

242 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI Biological Psychiatry: Regulation of Craving and Negative Emotion in AUD CNNI

Behavioral ratings (y-axis) from NOW Figure 2. A B the ( ) regulation of craving and ( ) emotion 5 LATER 5 A B * regulation tasks, separated by condition *** **** *** (x-axis). Signiﬁcance is noted: *p , .05, **p , .01, ***p , .005, and ****p , .001. 4 ** 4

3 3 Craving 2 2 Negative Affect Negative

1 1 Alcohol Food Neutral- Negative- Negative- LOOK LOOK REAPPRAISE Image Type Condition Instruction: F =13.04, p=.002 (1,16) Condition: F =57.70, p<.001 F p (1,14) Image Type: (1,16)=5.50, =.03

presentation periods were modeled together as boxcar re- affect as parametric regressor (k = 68), and ROC task and ER gressors convolved with the hemodynamic response function, task conjunction map (k = 2). The VS/subgenual ACC (sgACC) using the general linear model. The rating period, motion pa- and vmPFC/orbitofrontal cortex (OFC) in the ROC task and the rameters, their squares, and high-pass filter parameters were amygdala in the ER task were selected as a priori regions of included as additional regressors of no interest. Second-level interest (ROIs), given the evidence implicating these areas in contrasts identified regions that were differentially activated craving and affective processing, respectively. To do this, the between conditions. Specifically, we calculated the amygdala ROI was defined using an anatomical mask, while LATER.NOW contrast for the ROC task and the REAP- the VS/sgACC and vmPFC/OFC ROIs were defined PRAISE-Negative.LOOK-Negative contrast for the ER task. using spherical ROIs (radius = 6 mm) centered around peaks Additionally, parametric analyses were performed to identify previously observed during ROC in cigarette smokers (VS/ regions that covaried trial by trial with self-reported levels of sgACC: 23, 11, 22; vmPFC/OFC: 0, 56, 22) (40). These craving in the ROC task and negative affect in the ER task. regions were considered significant at a familywise error rate of To identify regions that were mutually activated during p , .05 using a voxelwise threshold of p , .005 and small regulation across tasks, we performed a formal conjunction volume correction. analysis between the primary group contrasts from the two tasks using the minimum statistic approach, as previously described (103,104) and as we implemented previously RESULTS (99,101). The conjunction map was defined by setting the statistical value at each voxel to the smaller of the two con- Self-reported Craving in the ROC Task trasts (LATER.NOW for the ROC task and REAPPRAISE- We found a main effect of instruction (F1,16 = 13.04, p=.002), Negative.LOOK-Negative for the ER task). This allowed us to such that participants reported lower craving in LATER than in formally represent voxels that were statistically significant NOW, as expected (t16 = 23.58, p=.003) (Figure 2A, across both contrasts. Finally, to test whether the regions Supplemental Figure S1). We also found a main effect of image identified in the conjunction overlapped with those previously type (F1,16 = 5.50, p=.03), such that alcohol images induced reported as commonly activated across self-control types in significantly greater craving than food (t16 = 2.48, p=.02). We smokers, we masked the conjunction results with 6-mm did not find a significant interaction (p=.24). Regulation suc- spheres centered around these peaks (78): vlPFC (248, cess did not significantly differ between image types (p=.24). 20, 26), dlPFC (244, 24, 30), and dmPFC/presupplementary motor area (24, 16, 52). Results were familywise error corrected at a combined Self-reported Negative Affect in the ER Task voxelwise and cluster threshold (k)ofp , .05, with a voxelwise We found a main effect of condition (F1,14 = 57.70, p , .001), threshold of p , .001 (105) (see Supplement for analyses such that participants reported greater negative affect when corrected with p , .005 voxelwise threshold). The estimated looking at negative images than at neutral images (t14 = 10.59, spatial smoothness of the residual was used to determine the k p , .001), and reported lower negative affect when instructed threshold for each second-level map: ROC task to reappraise negative images compared with looking at them

LATER.NOW (k = 27), ROC task effect of craving as para- (t14 = 23.40, p=.004), consistent with prior work (Figure 2B, metric regressor (k = 56), ER task REAPPRAISE-NEG- Supplemental Figure S1). Regulation success in the ER and ATIVE.LOOK-Negative (k = 27), ER task effect of negative ROC tasks was not signiﬁcantly correlated (p=.41).

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI 243 Biological Psychiatry: CNNI Regulation of Craving and Negative Emotion in AUD

Figure 3. Regions modulated by cognitive regu- A ROC task lation in the (A) regulation of craving (ROC) (LAT- ER.NOW) and (B) emotion regulation (ER) (REAPPRAISE-Negative.LOOK-Negative) tasks. Maps are familywise error corrected for multiple comparisons at p , .05 with a voxelwise threshold dlPFC of p , .001. Subcortical regions of interest (ventral striatum [VS] and ventromedial prefrontal cortex [vmPFC] for ROC, amygdala for ER) are shown at vlPFC p , .005 (uncorrected). dlPFC, dorsolateral vmPFC prefrontal cortex; dmPFC, dorsomedial prefrontal cortex; preSMA, presupplementary motor area; vlPFC, ventrolateral prefrontal cortex.

VS z=-12

web 4C/FPO B ER task preSMA/ dmPFC dlPFC

vlPFC

amygdala

Correlation With Negative Alcohol Expectancy Parametric Modulation Regulation success during alcohol trials of the ROC task was We observed a signiﬁcant positive association between trial- signiﬁcantly and positively correlated with long-term negative by-trial brain activity and craving in the vmPFC/OFC and VS/

alcohol expectancies (r15 = .54, p=.03), such that participants sgACC (Figure 4, Table 2). We did not observe any parametric with greater expectancies about the long-term negative effects associations between brain activity and negative affect during of alcohol were better able to regulate their craving. Regulation the ER task. success did not correlate significantly with the same-day or next-day subscales (ps . .12). Conjunction Between Regulation of Craving and Emotion We observed significant coactivation in the left vlPFC between Neural Activity During ROC ROC (LATER.NOW) and negative affect (ER; REAPPRAISE- We observed greater recruitment of the dlPFC and vlPFC Negative.LOOK-Negative) (Figure 5, Table 2) (see during cognitive ROC (LATER.NOW) (Figure 3A, Table 2). We Supplement for dlPFC in analysis using a p , .005 threshold). also observed relative deactivations in the VS/sgACC and As a result, this analysis revealed that subregions of the right vmPFC/OFC (a priori ROIs). We did not find significant mod- dlPFC and left vlPFC were uniquely active during ROC but not ulation of activity in the amygdala. during ER (Table 2). When the conjunction analysis was masked with regions previously observed in smokers during various forms of cognitive control, we observed significant Neural Activity During ER overlap in the left vlPFC. We observed increased activations in the dlPFC, vlPFC, and dmPFC during ER (REAPPRAISE-Negative.LOOK-Negative) DISCUSSION (Figure 3B, Table 2). We also observed a relative deactivation The results indicate that individuals with AUD can regulate in the amygdala (a priori ROI). We did not find significant craving as well as negative affect using cognitive strategies. modulation of activity in the VS/sgACC or vmPFC/OFC. Further, those who reported greater long-term negative alcohol

244 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI Biological Psychiatry: Regulation of Craving and Negative Emotion in AUD CNNI

Table 2. Peak Coordinates of Regions Identiﬁed in Whole-Brain Analyses Peak MNI Coordinates Region of Activation Hemisphere x y z k Maximum Mean ROC (LATER.NOW) dlPFC/middle frontal gyrus Right 45 12 51 39 5.47 4.37 vlPFC/inferior frontal gyrus Left 242 21 23 63 4.87 4.33 VS and subgenual anterior cingulatea Left 0 15 212 6 23.63 23.30 vmPFC/medial frontal gyrusa Left 23572645 23.18 23.03 ER (REAPPRAISE-Negative.LOOK-Negative) Middle temporal gyrus Left 248 218 227 97 7.08 4.87 dlPFC/middle frontal gyrus Left 245 6 54 38 6.47 4.85 dmPFC/superior frontal gyrus Left 215 9 72 111 6.47 4.82 vlPFC/inferior frontal gyrus Left 239 21 26 61 5.94 4.63 Amygdalaa Left 230 3 221 5 24.66 23.88 ROC Parametric Analysis vmPFC/dmPFC/medial frontal gyrusa Left 3 51 3 40 5.30 3.97 VSa Left 26623 15 3.98 3.50 Conjunction: ROC (LATER.NOW) and ER (REAPPRAISE-Negative.LOOK-Negative) Inferior frontal gyrusb Left 242 21 26 20 5.01 4.50 Inferior frontal gyrus Left 248 21 12 3 4.42 4.27 Results are p , .05 (familywise error corrected), with applied voxelwise threshold of p , .001. k indicates number of 2.5-mm isometric voxels. Maximum indicates the maximum t statistic in region; mean indicates the average t statistic. dlPFC, dorsolateral prefrontal cortex; dmPFC, dorsomedial prefrontal cortex; ER, emotion regulation; MNI, Montreal Neurological Institute; ROC, regulation of craving; vlPFC, ventrolateral prefrontal cortex; vmPFC, ventromedial prefrontal cortex; VS, ventral striatum. aRegion of interest small volume familywise error corrected at p , .05 with a voxelwise threshold of p , .005. bRegions in the conjunction analysis that were also observed after masking with previously identiﬁed regulation region from Tabibnia et al.(78).

expectancies were more successful at regulating their craving. craving and the amygdala for negative affect. Furthermore, the Importantly, regulation was accompanied by increased activity VS/sgACC and mPFC/medial OFC activity parametrically var- in the dlPFC, vlPFC, and dmPFC, and relative deactivations in ied with moment-to-moment craving. Finally, formal conjunc- regulation targets, namely the VS/sgACC and vmPFC/OFC for tion analysis showed that activations during both types of regulation spatially converged in vlPFC (and dlPFC at a more relaxed threshold; see Supplement). These findings have important implications for our understanding of AUD, CBT, and y=6 SUD treatment more broadly. Our results demonstrate that individuals with AUD can use cognitive strategies to reduce cue-induced craving for both alcohol and food, consistent with our prior behavioral findings (49). This is an important replication given the wealth of research linking alcohol use and AUD to cognitive impairments (106). In addition, we provide novel evidence suggesting that VS individuals with AUD can successfully regulate negative af- vmPFC fective responses toward negative stimuli. As such, these findings provide experimental validation of cognitive strategies that are used in psychological AUD treatments. Interestingly, although results suggest that individuals with AUD are able to use cognitive strategies to regulate craving and negative emotion in a laboratory setting, they may not FPO

= spontaneously do so in everyday life. Indeed, AUD is clinically characterized by impaired control of emotion (12,56). This z=0 seeming contradiction—whereby individuals with AUD can use web 4C regulatory strategies in the lab despite apparent deficits in Figure 4. Regions that parametrically covaried with levels of craving in the regulation of craving task trial by trial. The ventral striatum (VS) and everyday life—parallels findings in depression. Specifically, ventromedial prefrontal cortex (vmPFC) were regions of interest (shown studies have demonstrated that individuals with depression uncorrected; p , .005). can use cognitive reappraisal to reduce negative affect in

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI 245 Biological Psychiatry: CNNI Regulation of Craving and Negative Emotion in AUD

somewhat surprising given the consistent reports of disruptions in prefrontal function and structure in AUD [e.g., (80–82)]. Importantly, along with PFC recruitment, we observed relative deactivations in the VS/sgACC and vmPFC/OFC during ROC. Parametric modeling demonstrated that VS/sgACC and mPFC/OFC activity tracked with moment-to-moment changes in craving across image types and instructions. These results are consistent with meta-analyses showing that craving for drugs and food is associated with VS/sgACC and mPFC/OFC activations (115–118) and with the broader role proposed for these regions in value computation (119–122). vlPFC Taken together, these results provide further evidence for the role of a prefrontal-striatal pathway in the ROC (40). Across both tasks, conjunction revealed coactivation of the FPO = vlPFC during regulation (and dlPFC at a more relaxed threshold; see Supplement). Interestingly, this region over-

web 4C lapped with a region previously identified across 3 domains of Figure 5. Regions coactivated during regulation of craving (LATER.NOW) cognitive control in cigarette smokers (78). We also observed and emotion regulation (REAPPRAISE-Negative.LOOK-Negative). Maps are nonoverlapping activations in PFC subregions during ROC and familywise error corrected for multiple comparisons at p , .05 with a negative emotion. These spatial differences may be partially voxelwise threshold of , .001. vlPFC, ventrolateral prefrontal cortex. p attributable to differences in strategy implementation between the ROC and ER tasks: the strategy used for alcohol cues laboratory studies [e.g., (107,108)]; nevertheless, they are also involved focusing on the negative consequences of con- less likely to spontaneously do so (109,110). This may also be sumption, whereas the strategy used for negative stimuli true for individuals with AUD who are able to regulate craving involved generating positive reinterpretations. Thus, differential but may not do so in everyday life. Thus, the underlying engagement of prefrontal regions associated with autobio- problem may not be in capability, but rather in a more down- graphical memory [e.g., (123)], prospective thinking [e.g., stream process that negatively affects the tendency to execute (124)], self-referential processing [e.g., (125)], and affective regulation strategies. Alternatively, individuals with AUD may processing [e.g., (126)] is likely. not regulate their craving in everyday life simply because it is In addition, given that domain-specific subcortical targets more difficult to do so, relative to the laboratory setting. In were modulated during ROC and negative emotion, different either case, the findings suggest that targeted training in ROC cortical mechanisms may be engaged during each type of and negative emotion might be helpful for this population, to regulation, forming distinct regulatory pathways. That is, enhance individuals’ ability to regulate even in difficult mo- although PFC activation during both types of regulation largely ments and to make strategies more accessible. This could overlap, different cortical circuits are likely engaged in the increase the likelihood of regulation in everyday life modulation of distinct subcortical targets. Importantly, these (89,111,112). neural differences may account for the absence of correlation Notably, successful regulation of alcohol craving was between regulation success for craving and regulation success associated with long-term negative expectancies about for negative emotion. In other words, behavioral differences alcohol. This finding suggests that the current strategy— between the two tasks may be due to the domain-specificity of namely, thinking about the long-term negative effects of pathways that underlie these two types of regulation. alcohol—may work best in individuals who already have Importantly, the sample size was relatively small. Although negative expectancies. This is consistent with prior reports that this is often the case with fMRI studies in AUD (127–130) and in negative alcohol expectancies are associated with abstinence clinical populations more generally [for meta-analysis, see motivation prior to alcohol treatment (113) and predict sub- Picó-Pérez et al. (131)], this is a limitation that reduces the sequent abstinence (88). Thus, one intriguing possibility is that likelihood of detecting small effects and may limit generaliz- negative expectancies influence treatment outcome via their ability of the findings. We hope that future studies will expand effect on ROC. Alternatively, negative expectancies may on this work with larger samples. Such studies could also represent individuals’ motivation for change (7–9). Both ac- include healthy control participants and social drinkers to counts point to enhancing negative expectancies as an examine whether regulation success relates to drinking important target of treatment. severity. Such studies could also test whether AUD affects Neuroimaging analyses revealed that during cognitive ROC prefrontal-subcortical pathways within and across domains of and negative emotion, individuals with AUD recruit prefrontal regulation (e.g., selecting appropriate regulatory outputs). regions, including the dlPFC, vlPFC, and dmPFC. These re- Additionally, longitudinal designs could test the relationships sults echo findings from prior imaging work on ER in healthy between negative alcohol expectancy, ROC, and clinical out- adults (74,114) and ROC in cigarette smokers (26) and cocaine comes. Finally, given that individuals with AUD are capable of users (43). However, to our knowledge, this is the first study to regulating craving when instructed to do so, the next important examine the neural mechanisms underlying these cognitive step may be to identify ways to train them to increase their regulation strategies in AUD. Moreover, the current findings are tendency to use such strategies in their daily lives.

246 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI Biological Psychiatry: Regulation of Craving and Negative Emotion in AUD CNNI

ACKNOWLEDGMENTS AND DISCLOSURES 7. Cohen E, Feinn R, Arias A, Kranzler HR (2007): Alcohol treatment utilization: Findings from the National Epidemiologic Survey on This work was supported by Center for the Translational Neuroscience Alcohol and Related Conditions. Drug Alcohol Depend 86:214– of Alcoholism Grant No. P50 AA0123870 (Principal investigator, JHK; to AA 221. and HK) and National Institute on Drug Abuse Grant No. K12-DA00167 8. Chapman C, Slade T, Hunt C, Teesson M (2015): Delay to first (to HK). treatment contact for alcohol use disorder. Drug Alcohol Depend We thank Nicholas Franco, Michael Kleinberg, and Matthew G. Schafer 147:116–121. for their assistance with data collection. 9. Tuithof M, Ten Have M, Van Den Brink W, Vollebergh W, De Graaf R SSO reports being a consultant or an advisory board member of (2016): Treatment seeking for alcohol use disorders: treatment gap or Alkermes, Amygdala, Indivior, Mitsubishi Tanabe, and Opiant; a member of adequate self-selection? Eur Addict Res 22:277–285. the American Society of Clinical Pharmacology Alcohol Clinical Trials 10. Moos RH, Moos BS (2006): Rates and predictors of relapse after Initiative supported by Amygdala, Ethypharm, Lilly, Lundbeck, Otsuka, natural and treated remission from alcohol use disorders. Addiction Pfizer, Arbor Pharmaceuticals, and Indivior; and a member of the National 101:212–222. Institute on Drug Abuse Clinical Trials Network Data Safety and Monitoring 11. Witkiewitz K, Marlatt GA (2007): Modeling the complexity of post- Board with honorarium from the Emmes Corporation and donated study treatment drinking: It’s a rocky road to relapse. Clin Psychol Rev medications from AstraZeneca, Pfizer, and Novartis. JHK has received 27:724–738. grants and contracts from the Department of Veterans Affairs, National 12. American Psychiatric Association (2013): Diagnostic and Statistical Institute on Alcohol Abuse and Alcoholism, and National Institute of Mental Manual of Mental Disorders, 5th ed. Washington, DC: American Health; has received consultancy fees from Biogen, Idec, Biomedisyn Cor- Psychiatric Association. poration, Janssen Research & Development, Otsuka America, Spring Care, 13. Connors GJ, Maisto SA, Zywiak WH (1998): Male and female alco- Sunovion Pharmaceuticals, Takeda Industry, and Taisho Pharmaceutical; holics’ attributions regarding the onset and termination of relapses owns stock or options in ArRETT Neuroscience, Biohaven Pharmaceuticals and the maintenance of abstinence. J Subst Abuse 10:27–42. Medical Sciences, BlackThorn Therapeutics, Luc Therapeutics, and Spring 14. Heather N, Stallard A, Tebbutt J (1991): Importance of substance Care; and has a share of a patent licensed by Janssen Research related to cues in relapse among heroin users: Comparison of two methods of the intranasal administration of ketamine for mood disorders and its use to investigation. Addict Behav 16:41–49. treat suicide risk. AA consults for and is a scientific advisory board member 15. Norregaard J, Tonnesen P, Petersen L (1993): Predictors and reasons for BlackThorn Therapeutics. HK participated in a scientific advisory board for relapse in smoking cessation with nicotine and placebo patches. and is a consultant for Indivior. SS and MMM report no biomedical financial Prev Med 22:261–271. interests or potential conflicts of interest. 16. Abrantes AM, Strong DR, Lejuez CW, Kahler CW, Carpenter LL, Price LH, et al. (2008): The role of negative affect in risk for early lapse among low distress tolerance smokers. Addict Behav ARTICLE INFORMATION 33:1394–1401. 17. Allen AM, Allen SS, Lunos S, Pomerleau CS (2010): Severity of From the Department of Psychiatry (SS, SSO, JHK, AA, HK), Yale University withdrawal symptomatology in follicular versus luteal quitters: The School of Medicine, New Haven, Connecticut; and the Department of combined effects of menstrual phase and withdrawal on smoking Neuroscience (MMM), Medical University of South Carolina, Charleston, cessation outcome. Addict Behav 35:549–552. South Carolina. 18. Bottlender M, Soyka M (2004): Impact of craving on alcohol relapse Address correspondence to Hedy Kober, Ph.D., Yale University, Clinical during, and 12 months following, outpatient treatment. Alcohol & Affective Neuroscience Lab, One Church Street, Suite 701, New Haven, Alcohol 39:357–361. CT 06510; E-mail: [email protected]. 19. Brady KT, Back SE, Waldrop AE, McRae AL, Anton RF, Received Jun 9, 2019; revised Oct 1, 2019; accepted Oct 2, 2019. Upadhyaya HP, et al. (2006): Cold pressor task reactivity: Predictors Supplementary material cited in this article is available online at https:// of alcohol use among alcohol-dependent individuals with and without doi.org/10.1016/j.bpsc.2019.10.005. comorbid posttraumatic stress disorder. Alcohol Clin Exp Res 30:938–946. 20. Fatseas M, Denis C, Massida Z, Verger M, Franques-Rénéric P, REFERENCES Auriacombi M (2011): Cue-induced reactivity, cortisol response and 1. Gowing LR, Ali RL, Allsop S, Marsden J, Turf EE, West R, et al. (2015): substance use outcome in treated heroin dependent individuals. Biol Global statistics on addictive behaviours: 2014 Status report. Psychiatry 70:720–727. Addiction 110:904–919. 21. Flannery BA, Poole SA, Gallop RJ, Volpicelli JR (2003): Alcohol 2. Substance Abuse and Mental Health Services Administration (2018): craving predicts drinking during treatment: An analysis of three Results From the 2017 National Survey on Drug Use and Health: assessment instruments. J Stud Alcohol 64:120–126. Detailed Tables. Rockville, MD: Substance Abuse and Mental Health 22. Buckner JD, Crosby RD, Silgado J, Wonderlich SA, Schmidt NB Services Administration. (2012): Immediate antecedents of marijuana use: An analysis from 3. Institution of Health Metrics and Evaluation (2017): GBD Results Tool. ecological momentary assessment. J Behav Ther Exp Psychiatry Seattle, WA: Institution of Health Metrics and Evaluation, University 43:647–655. of Washington. 23. Cooney NL, Litt MD, Cooney JL, Pilkey DT, Steinburg HR, 4. McLellan AT, Lewis DC, O’Brien CP, Kleber HD (2000): Drug Oncken CA (2007): Alcohol and tobacco cessation in alcohol- dependence, a chronic medical illness: Implications for treatment, dependent smokers: Analysis of real-time reports. Psychol Addict insurance, and outcomes evaluation. JAMA 284:1689–1695. Behav 21:277–286. 5. Grant BF, Chou SP, Saha TD, Pickering RP, Kerridge BT, 24. Epstein DH, Willner-Reid J, Vahabzadeh M, Mezghanni M, Lin JL, Ruan WJ, et al. (2017): Prevalence of 12-month alcohol use, high- Preston KL (2009): Real-time electronic diary reports of cue exposure risk drinking, and DSM-IV alcohol use disorder in the United and mood in the hours before cocaine and heroin craving and use. States, 2001-2002 to 2012-2013: Results from the National Arch Gen Psychiatry 66:88–94. Epidemiologic Survey on Alcohol and Related Conditions. JAMA 25. Litt MD, Cooney NL, Morse P (2000): Reactivity to alcohol-related Psychiatry 74:911–923. stimuli in the laboratory and in the field: Predictors of craving in 6. Suzuki S, Kober H (2018): Substance-related and addictive disorders. treated alcoholics. Addiction 95:889–900. In: Butcher JN, Hooley J, Kendall PC, editors. APA Handbook of 26. Kober H, Mell MM (2015): Neural mechanisms underlying craving and Psychopathology: Psychopathology: Understanding, Assessing, and the regulation of craving. In: Wilson SJ, editor. Handbook on the Treating Adult Mental Disorders, Vol. 1. Washington, DC: American Cognitive Neuroscience of Addiction. Oxford, United Kingdom: Psychological Association, 481–506. Wiley-Blackwell, 195–218.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI 247 Biological Psychiatry: CNNI Regulation of Craving and Negative Emotion in AUD

27. Carpenter MJ, Saladin ME, DeSantis S, Gray KM, LaRowe SD, 47. O’Connell KA, Fears BA, Cook MR, Gerkovich MM, Zechmann A Upadhyaya HP (2009): Laboratory-based, cue-elicited craving and (1991): Overcoming the urge to smoke: The strategies of long-term cue reactivity as predictors of naturally occurring smoking behavior. abstainers and late relapsers. Psychol Addict Behav 5:1–8. Addict Behav 34:536–541. 48. Kiluk BD, Nich C, Babuscio T, Carroll KM (2010): Quality versus 28. Conklin CA, Vella EJ, Joyce CJ, Salkeld RP, Perkins KA, quantity: Acquisition of coping skills following computerized cogni- Parzynski CS (2015): Examining the relationship between cue- tive behavioral therapy for substance use disorders. Addiction induced craving and actual smoking. Exp Clin Psychopharmacol 105:2120–2127. 23:90–96. 49. Naqvi NH, Ochsner KN, Kober H, Kuerbis A, Feng T, Wall M, et al. 29. McHugh RK, Fitzmaurice GM, Carroll KM, Griffin ML, Hill KP, (2015): Cognitive regulation of craving in alcohol-dependent and Wasan AD, et al. (2014): Assessing craving and its relationship to social drinkers. Alcohol Clin Exp Res 39:343–349. subsequent prescription opioid use among treatment-seeking 50. Day A, Kahler C, Ahern D, Clark U (2015): Executive functioning in prescription opioid dependent patients. Drug Alcohol Depend alcohol use studies: A brief review of findings and challenges in 145:121–126. assessment. Curr Drug Abuse Rev 8:26–40. 30. Palfai TP (2006): Activating action tendencies: The influence of action 51. Berking M, Margraf M, Ebert D, Wupperman P, Hofmann SG, priming on alcohol consumption among male hazardous drinkers. Junghanns K (2011): Deficits in emotion-regulation skills predict J Stud Alcohol 67:926–933. alcohol use during and after cognitive-behavioral therapy for alcohol 31. Petrakis IL, Buonopane A, O’Malley S, Cermik O, Trevisan L, dependence. J Consult Clin Psychol 79:307–318. Boutros NN, et al. (2002): The effect of tryptophan depletion on 52. Giuliani NR, Mann T, Tomiyama AJ, Berkman ET (2014): Neural alcohol self-administration in non-treatment-seeking alcoholic in- systems underlying the reappraisal of personally craved foods. dividuals. Alcohol Clin Exp Res 26:969–975. J Cogn Neurosci 26:1390–1402. 32. Waters AJ, Shiffman S, Sayette MA, Paty JA, Gwaltney CJ, 53. Hollmann M, Hellrung L, Pleger B, Schlogl H, Kabisch S, Stumvoll M, Balabanis MH (2004): Cue-provoked craving and nicotine replace- et al. (2012): Neural correlates of the volitional regulation of the desire ment therapy in smoking cessation. J Consult Clin Psychol 72:1136– for food. Int J Obes (Lond) 36:648–655. 1143. 54. Yokum S, Stice E (2013): Cognitive regulation of food craving: Effects 33. Cooney NL, Litt MD, Morse PA, Bauer LO, Gaupp L (1997): Alcohol of three cognitive reappraisal strategies on neural response to cue reactivity, negative-mood reactivity, and relapse in treated palatable foods. Int J Obes (Lond) 37:1565–1570. alcoholic men. J Abnorm Psychol 106:243–250. 55. Carey KB (1995): Alcohol-related expectancies predict quantity and 34. Papachristou H, Nederkoorn C, Giesen JC, Jansen A (2014): Cue frequency of heavy drinking among college students. Psychol Addict reactivity during treatment, and not impulsivity, predicts an initial Behav 9:236–241. lapse after treatment in alcohol use disorders. Addict Behav 56. Kober H (2014): Emotion regulation in substance use disorders. In: 39:737–739. Gross JJ, editor. Handbook of Emotion Regulation, 2nd ed. New 35. Sinha R, Fox HC, Hong KA, Hansen J, Tuit K, Kreek MJ (2011): Ef- York, NY: Guilford Press, 428–446. fects of adrenal sensitivity, stress-and cue-induced craving, and 57. Khantzian EJ (1985): The self-medication hypothesis of addictive anxiety on subsequent alcohol relapse and treatment outcomes. disorders - Focus on heroin and cocaine dependence. Am J Psy- Arch Gen Psychiatry 68:942–952. chiatry 142:1259–1264. 36. Wrase J, Makris N, Braus D, Mann K, Smolka M, Kennedy D, et al. 58. Sinha R, Shaham Y, Heilig M (2011): Translational and reverse (2008): Amygdala volume associated with alcohol abuse relapse and translational research on the role of stress in drug craving and craving. Am J Psychiatry 165:1179–1184. relapse. Psychopharmacology 218:69–82. 37. Bedi G, Preston KL, Epstein DH, Heishman SJ, Marrone GF, 59. Baker TB, Piper ME, McCarthy DE, Majeskie MR, Fiore MC (2004): Shaham Y, et al. (2011): Incubation of cue-induced cigarette craving Addiction motivation reformulated: An affective processing model of during abstinence in human smokers. Biol Psychiatry 69:708–711. negative reinforcement. Psychol Rev 111:33–51. 38. Kadden R, Carroll KM, Donovan D, Cooney N, Monti P, Abrams D, 60. Cox WM, Klinger E (1988): A motivational model of alcohol use. et al. (1995): Cognitive-Behavioral Coping Skills Therapy Manual: A J Abnorm Psychol 97:168–180. Clinical Research Guide for Therapists Treating Individuals With 61. Cooper ML, Frone MR, Russell M, Mudar P (1995): Drinking to Alcohol Abuse and Dependence. Bethesda, MD: National Institute on regulate positive and negative emotions: A motivational model of Alcohol Abuse and Alcoholism. alcohol use. J Pers Soc Psychol 69:990–1005. 39. Kober H, Kross EF, Mischel W, Hart CL, Ochsner KN (2010): Regu- 62. Koob GF, Le Moal M (2008): Neurobiological mechanisms for lation of craving by cognitive strategies in cigarette smokers. Drug opponent motivational processes in addiction. Philos Trans R Soc Alcohol Depend 106:52–55. Lond B Biol Sci 363:3113–3123. 40. Kober H, Mende-Siedlecki P, Kross EF, Weber J, Mischel W, Hart CL, 63. Grant BF, Stinson FS, Dawson DA, Chou SP, Dufour MC, et al. (2010): Prefrontal-striatal pathway underlies cognitive regulation Compton W, et al. (2004): Prevalence and co-occurrence of sub- of craving. Proc Natl Acad Sci U S A 107:14811–14816. stance use disorders and independent mood and anxiety disorders: 41. Littel M, Franken IHA (2011): Intentional modulation of the late pos- Results from the National Epidemiologic Survey on Alcohol and itive potential in response to smoking cues by cognitive strategies in Related Conditions. Arch Gen Psychiatry 61:807–816. smokers. PLoS One 6:e27519. 64. Lai HMX, Cleary M, Sitharthan T, Hunt GE (2015): Prevalence of 42. Lopez R, Onyemekwu C, Hart CL, Ochsner KN, Kober H (2015): comorbid substance use, anxiety and mood disorders in epidemio- Boundary conditions of methamphetamine craving. Exp Clin Psy- logical surveys, 1990–2014: A systematic review and meta-analysis. chopharmacol 23:436–444. Drug Alcohol Depend 154:1–13. 43. Volkow ND, Fowler JS, Wang GJ, Telang F, Logan J, Jayne M, et al. 65. Grant BF, Goldstein RB, Saha TD, Chou SP, Jung J, Zhang H, et al. (2010): Cognitive control of drug craving inhibits brain reward regions (2015): Epidemiology of DSM-5 alcohol use disorder: Results from in cocaine abusers. Neuroimage 49:2536–2543. the National Epidemiologic Survey on Alcohol and Related Condi- 44. Giuliani NR, Calcott RD, Berkman ET (2013): Piece of cake. Cognitive tions III. JAMA Psychiatry 72:757–766. reappraisal of food craving. Appetite 64:56–61. 66. Brière FN, Rohde P, Seeley JR, Klein D, Lewinsohn PM (2014): Co- 45. Meule A, Kubler A, Blechert J (2013): Time course of electrocortical morbidity between major depression and alcohol use disorder from food-cue responses during cognitive regulation of craving. Front adolescence to adulthood. Compr Psychiatry 55:526–533. Psychol 4:669. 67. Johnson PB, Gurin G (1994): Negative affect, alcohol expectancies 46. Stice E, Yokum S, Burger K, Rohde P, Shaw H, Gau JM (2015): A pilot and alcohol-related problems. Addiction 89:581–586. randomized trial of a cognitive reappraisal obesity prevention pro- 68. Saladin ME, Gray KM, Carpenter MJ, LaRowe SD, DeSantis SM, gram. Physiol Behav 138:124–132. Upadhyaya HP (2012): Gender differences in craving and cue

248 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI Biological Psychiatry: Regulation of Craving and Negative Emotion in AUD CNNI

reactivity to smoking and negative affect/stress cues. Am J Addict residential treatment program: A one-month and three-month follow- 21:210–220. up study in men. J Stud Alcohol 55:543–548. 69. Sinha R, Catapano D, O’Malley S (1999): Stress-induced craving and 89. Boswell RG, Sun W, Suzuki S, Kober H (2018): Training in cognitive stress response in cocaine dependent individuals. Psychopharma- strategies reduces eating and improves food choice. Proc Natl Acad cology 142:343–351. Sci U S A 115:E11238–E11247. 70. Childress AR, Ehrman R, McLellan AT, MacRae J, Natale M, 90. Ollinger JM, Corbetta M, Shulman GL (2001): Separating processes O’Brien CP (1994): Can induced moods trigger drug-related within a trial in event-related functional MRI: II. Analysis. Neuroimage responses in opiate abuse patients? J Subst Abuse Treat 13:218–229. 11:17–23. 91. Gross JJ (1998): Antecedent- and response-focused emotion regu- 71. Seo D, Lacadie CM, Tuit K, Hong KI, Constable RT, Sinha R (2013): lation; divergent consequences for experience, expression, and Disrupted ventromedial prefrontal function, alcohol craving, and physiology. J Pers Soc Psychol 74:224–237. subsequent relapse risk. JAMA Psychiatry 70:727–739. 92. Ochsner KN, Bunge SA, Gross JJ, Gabrieli JD (2002): Rethinking 72. Sinha R, Li CSR (2007): Imaging stress- and cue-induced drug and feelings: An FMRI study of the cognitive regulation of emotion. alcohol craving: Association with relapse and clinical implications. J Cogn Neurosci 14:1215–1229. Drug Alcohol Rev 26:25–31. 93. Lang PJ, Bradley MM, Cuthbert BN (1997): International Affective 73. Gross JJ (2014): Emotion regulation: Conceptual and empirical Picture System (IAPS): Technical Manual and Affective Ratings. foundations. In: Gross JJ, editor. Handbook of Emotion Regulation, Gainesville, FL: NIMH Center for the Study of Emotion and 2nd ed. New York, NY: Guilford Press, 3–20. Attention. 74. Buhle JT, Silvers JA, Wager TD, Lopez R, Onyemekwu C, Kober H, 94. Denny BT, Ochsner KN, Weber J, Wager TD (2013): Anticipatory brain et al. (2014): Cognitive reappraisal of emotion: A meta-analysis of activity predicts the success or failure of subsequent emotion regu- human neuroimaging studies. Cereb Cortex 24:2981–2990. lation. Soc Cogn Affect Neurosci 9:403–411. 75. Wu L, Winkler MH, Wieser MJ, Andreatta M, Li Y, Pauli P (2015): 95. Peyre H, Leplège A, Coste J (2011): Missing data methods for dealing Emotion regulation in heavy smokers: Experiential, expressive and with missing items in quality of life questionnaires. A comparison by physiological consequences of cognitive reappraisal. Front Psychol simulation of personal mean score, full information maximum likeli- 6:1555. hood, multiple imputation, and hot deck techniques applied to the 76. Albein-Urios N, Verdejo-Román J, Asensio S, Soriano-Mas C, Mar- SF-36 in the French 2003 decennial health survey. Qual Life Res tínez-González JM, Verdejo-García A (2014): Re-appraisal of negative 20:287–300. emotions in cocaine dependence: Dysfunctional corticolimbic acti- 96. Raaijmakers QA (1999): Effectiveness of different missing data vation and connectivity. Addict Biol 19:415–426. treatments in surveys with Likert-type data: Introducing the relative 77. Tabibnia G, Monterosso JR, Baicy K, Aron AR, Poldrack RA, mean substitution approach. Educ Psychol Meas 59:725–748. Chakrapani S, et al. (2011): Different forms of self-control share a 97. Noble S, Scheinost D, Finn ES, Shen X, Papademetris X, neurocognitive substrate. J Neurosci 31:4805–4810. McEwen SC, et al. (2017): Multisite reliability of MR-based functional 78. Tabibnia G, Creswell JD, Kraynak T, Westbrook C, Julson E, connectivity. Neuroimage 146:959–970. Tindle HA (2014): Common prefrontal regions activate during self- 98. Van Essen DC, Barch DM (2015): The human connectome in health control of craving, emotion, and motor impulses in smokers. Clin and psychopathology. World Psychiatry 14:154–157. Psychol Sci 2:611–619. 99. Kober H, Brewer JA, Tuit K, Sinha R (2017): Neural stress reactivity 79. Goldstein RZ, Volkow ND (2011): Dysfunction of the prefrontal cortex relates to smoking outcomes and differentiates between mindfulness in addiction: Neuroimaging findings and clinical implications. Nat Rev and cognitive-behavioral treatments. Neuroimage 151:4–13. Neurosci 12:652–669. 100. Power JD, Barnes KA, Snyder AZ, Schlaggar BL, Petersen SE (2012): 80. Chanraud S, Martelli C, Delain F, Kostogianni N, Douaud G, Spurious but systematic correlations in functional connectivity MRI Aubin HJ, et al. (2007): Brain morphometry and cognitive perfor- networks arise from subject motion. Neuroimage 59:2142–2154. mance in detoxified alcohol-dependents with preserved psychoso- 101. Kober H, DeVito EE, DeLeone CM, Carroll KM, Potenza MN (2014): cial functioning. Neuropsychopharmacology 32:429–438. Cannabis abstinence during treatment and one-year follow-up: 81. Fein G, Di Sclafani V, Cardenas V, Goldmann H, Tolou-Shams M, Relationship to neural activity in men. Neuropsychopharmacology Meyerhoff DJ (2002): Cortical gray matter loss in treatment-naive 39:2288–2298. alcohol dependent individuals. Alcohol Clin Exp Res 26:558–564. 102. Wager TD, Keller MC, Lacey SC, Jonides J (2005): Increased sensi- 82. Makris N, Oscar-Berman M, Jaffin SK, Hodge SM, Kennedy DN, tivity in neuroimaging analyses using robust regression. Neuroimage Caviness VS, et al. (2008): Decreased volume of the brain reward 26:99–113. system in alcoholism. Biol Psychiatry 64:192–202. 103. Nichols T, Brett M, Andersson J, Wager T, Poline JB (2005): Valid 83. Yang H, Devous MD, Briggs RW, Spence JS, Xiao H, Kreyling N, et al. conjunction inference with the minimum statistic. Neuroimage (2013): Altered neural processing of threat in alcohol-dependent men. 25:653–660. Alcohol Clin Exp Res 37:2029–2038. 104. Friston KJ, Holmes AP, Price CJ, Büchel C, Worsley KJ (1999): 84. Gilman JM, Hommer DW (2007): Modulation of brain response to Multisubject fMRI studies and conjunction analyses. Neuroimage emotional images by alcohol cues in alcohol-dependent patients. 10:385–396. Addict Biol 13:423–434. 105. Eklund A, Nichols TE, Knutsson H (2016): Cluster failure: Why fMRI 85. O’Daly OG, Trick L, Scaife J, Marshall J, Ball D, Phillips ML, et al. inferences for spatial extent have inflated false-positive rates. Proc (2012): Withdrawal-associated increases and decreases in functional Natl Acad Sci U S A 113:7900–7905. neural connectivity associated with altered emotional regulation in 106. Schweizer TA, Vogel-Sprott M (2008): Alcohol-impaired speed and alcoholism. Neuropsychopharmacology 37:2267–2276. accuracy of cognitive functions: A review of acute tolerance and 86. Salloum JB, Ramchandani VA, Bodurka J, Rawlings R, Momenan R, recovery of cognitive performance. Exp Clin Psychopharmacol George D, et al. (2007): Blunted rostral anterior cingulate response 16:240–250. during a simplified decoding task of negative emotional facial ex- 107. Ellis AJ, Vanderlind WM, Beevers CG (2013): Enhanced anger reac- pressions in alcoholic patients. Alcohol Clin Exp Res 31:1490–1504. tivity and reduced distress tolerance in major depressive disorder. 87. McMahon J, Jones BT (1994): Social drinkers’ negative alcohol ex- Cogn Ther Res 37:498–509. pectancy relates to their satisfaction with current consumption: 108. Liu DY, Thompson RJ (2017): Selection and implementation of Measuring motivation for change with the NAEQ. Alcohol Alcohol emotion regulation strategies in major depressive disorder: An inte- 29:687–690. grative review. Clin Psychol Rev 57:183–194. 88. Jones BT, McMahon J (1994): Negative and positive alcohol ex- 109. Joormann J, Gotlib IH (2010): Emotion regulation in depression: pectancies as predictors of abstinence after discharge from a Relation to cognitive inhibition. Cogn Emotion 24:281–298.

Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI 249 Biological Psychiatry: CNNI Regulation of Craving and Negative Emotion in AUD

110. Joormann J, Stanton CH (2016): Examining emotion regulation in 122. Wilson RP, Colizzi M, Bossong MG, Allen P, Kempton M, Abe N, et al. depression: A review and future directions. Behav Res Ther 86:35–49. (2018): The neural substrate of reward anticipation in health: A meta- 111. Cohen N, Ochsner KN (2018): From surviving to thriving in the face of analysis of fMRI ﬁndings in the monetary incentive delay task. Neu- threats: The emerging science of emotion regulation training. Curr ropsychol Rev 28:496–506. Opin Behav Sci 24:143–155. 123. Svoboda E, McKinnon MC, Levine B (2006): The functional neuro- 112. Denny BT, Ochsner KN (2014): Behavioral effects of longitudinal anatomy of autobiographical memory: A meta-analysis. Neuro- training in cognitive reappraisal. Emotion 14:425–433. psychologia 44:2189–2208. 113. Heather N, Adamson SJ, Raistrick D, Slegg GP (2010): Initial pref- 124. Addis DR, Wong AT, Schacter DL (2007): Remembering the past and erence for drinking goal in the treatment of alcohol problems: I. imagining the future: Common and distinct neural substrates during Baseline differences between abstinence and non-abstinence event construction and elaboration. Neuropsychologia 45:1363– groups. Alcohol Alcohol 45:128–135. 1377. 114. Wager TD, Davidson ML, Hughes BL, Lindquist MA, Ochsner KN 125. Northoff G, Heinzel A, De Greck M, Bermpohl F, Dobrowolny H, (2008): Prefrontal-subcortical pathways mediating successful Panksepp J (2006): Self-referential processing in our brain—a meta- emotion regulation. Neuron 59:1037–1050. analysis of imaging studies on the self. Neuroimage 31:440–457. 115. Engelmann JM, Versace F, Robinson JD, Minnix JA, Lam CY, Cui Y, 126. Wager TD, Phan KL, Liberzon I, Taylor SF (2003): Valence, et al. (2012): Neural substrates of smoking cue reactivity: A meta- gender, and lateralization of functional brain anatomy in emotion: analysis of fMRI studies. Neuroimage 60:252–262. A meta-analysis of ﬁndings from neuroimaging. Neuroimage 116. Chase HW, Eickhoff SB, Laird AR, Hogarth L (2011): The neural basis 19:513–531. of drug stimulus processing and craving: An activation likelihood 127. Alba-Ferrara L, Müller-Oehring E, Sullivan E, Pfefferbaum A, estimation meta-analysis. Biol Psychiatry 70:785–793. Schulte T (2016): Brain responses to emotional salience and reward 117. Kuhn S, Gallinat J (2011): Common biology of craving across legal in alcohol use disorder. Brain Imaging Behav 10:136–146. and illegal drugs - a quantitative meta-analysis of cue-reactivity brain 128. Elkins RL, Richards TL, Nielsen R, Repass R, Stahlbrandt H, response. Eur J Neurosci 33:1318–1326. Hoffman HG (2017): The neurobiological mechanism of chemical 118. Tang DW, Fellows LK, Small DM, Dagher A (2012): Food and drug aversion (emetic) therapy for alcohol use disorder: An fMRI study. cues activate similar brain regions: A meta-analysis of functional MRI Front Behav Neurosci 11:182. studies. Physiol Behav 106:317–324. 129. Karch S, Keeser D, Hümmer S, Paolini M, Kirsch V, Karali T, et al. 119. Bartra O, McGuire JT, Kable JW (2013): The valuation system: A (2015): Modulation of craving related brain responses using real- coordinate-based meta-analysis of BOLD fMRI experiments exam- time fMRI in patients with alcohol use disorder. PLoS One 10: ining neural correlates of subjective value. Neuroimage 76:412–427. e0133034. 120. Hutcherson CA, Plassmann H, Gross JJ, Rangel A (2012): Cognitive 130. Zehra A, Lindgren E, Wiers CE, Freeman C, Miller G, Ramirez V, et al. regulation during decision making shifts behavioral control between (2019): Neural correlates of visual attention in alcohol use disorder. ventromedial and dorsolateral prefrontal value systems. J Neurosci Drug Alcohol Depend 194:430–437. 32:13543–13554. 131. Picó-Pérez M, Radua J, Steward T, Menchón JM, Soriano-Mas C 121. Knutson B, Greer SM (2008): Anticipatory affect: Neural correlates (2017): Emotion regulation in mood and anxiety disorders: A meta- and consequences for choice. Philos Trans R Soc Lond B Biol Sci analysis of fMRI cognitive reappraisal studies. Prog Neuro- 363:3771–3786. psychopharmacol Biol Psychiatry 79:96–104.

250 Biological Psychiatry: Cognitive Neuroscience and Neuroimaging February 2020; 5:239–250 www.sobp.org/BPCNNI