A Two-Pronged Approach to Preeclampsia: Understanding Gene Expression and Targeting Sflt1 Using Rnai

University of Massachusetts Medical School eScholarship@UMMS

GSBS Dissertations and Theses Graduate School of Biomedical Sciences

2017-07-10

A Two-Pronged Approach to Preeclampsia: Understanding Gene Expression and Targeting sFlt1 using RNAi

Ami Ashar-Patel University of Massachusetts Medical School

Let us know how access to this document benefits ou.y Follow this and additional works at: https://escholarship.umassmed.edu/gsbs_diss

Part of the Biology Commons, Female Urogenital Diseases and Pregnancy Complications Commons, and the Genetics and Genomics Commons

Repository Citation Ashar-Patel A. (2017). A Two-Pronged Approach to Preeclampsia: Understanding Gene Expression and Targeting sFlt1 using RNAi. GSBS Dissertations and Theses. https://doi.org/10.13028/M2RH4N. Retrieved from https://escholarship.umassmed.edu/gsbs_diss/909

This material is brought to you by eScholarship@UMMS. It has been accepted for inclusion in GSBS Dissertations and Theses by an authorized administrator of eScholarship@UMMS. For more information, please contact [email protected]. i

A TWO-PRONGED APPROACH TO PREECLAMPSIA: UNDERSTANDING GENE EXPRESSION AND TARGETING sFlt1 using RNAi

A Dissertation Presented

Ami Ashar-Patel

Submitted to the Faculty of the University of Massachusetts Graduate School of Biomedical Sciences, Worcester in partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY

July 10, 2017

INTERDISCIPLINARY GRADUATE PROGRAM ii

A TWO-PRONGED APPROACH TO PREECLAMPSIA: UNDERSTANDING GENE EXPRESSION AND TARGETING sFlt1 using RNAi

A Dissertation Presented By AMI ASHAR-PATEL This work was undertaken in the Graduate School of Biomedical Sciences Interdisciplinary graduate program

The signature of the thesis advisor signifies validation of the Dissertation content

______Melissa J. Moore, Ph.D., Thesis Advisor

The signatures of the Dissertation Defense Committee signifies completion and approval as to style and content of the Dissertation

______Jane E. Freedman, M.D., Member of Committee

______Manuel Garber, Ph.D., Member of Committee

______Tiffany A. Moore Simas, M.D., Member of Committee

______Manoj Bhasin, Ph.D., External Member of Committee

The signature of the Chair of the Committee signifies that the written dissertation meets the requirements of the Dissertation Committee

______Jeremy Luban, M.D., Chair of Committee

The signature of the Dean of the Graduate School of Biomedical Sciences signifies that the student has met all graduation requirements of the School

Anthony Carruthers, Ph.D., Dean of the Graduate School of Biomedical Sciences

Interdisciplinary Graduate Program July 10, 2017 iii

DEDICATION

This thesis is dedicated to all the women, their babies and their families who suffered due to preeclampsia.

I also want to dedicate my thesis to my maternal grandfather and paternal grandmother, who are not amongst us today. They always believed that one day

I would come to the United States of America, and fulfill my academic dreams. I am forever grateful to them for their confidence in me, and my abilities. I would also like to dedicate this thesis to my maternal grandmother, who cannot wait to call me Dr. Ami Ashar-Patel.

This thesis would have been impossible without the love and never-ending support of my family. My parents, Jagdish and Priti Ashar, and my brother Mihir

Ashar; they have been my side through thick and thin. My in-laws Nayana and

Nalin Patel and my brother-in-law Biajun Patel have been a constant source of positivity, for which I am very grateful.

Finally, this thesis is dedicated to my husband, Hemal Patel, who has been my guiding star throughout graduate school. Thank you for being by my side every day, and through the ups and downs of my PhD. Thank you for loving me, for believing in me, cooking for me, and staying positive no matter what. iv

ACKNOWLEDGEMENTS

It takes a village to get a PhD, and I want to thank my village for not giving up on me through my PhD!

I want to thank my roommates, and now my dearest friends, Jillian Lindblad and

Ozge Yildiz, who held my hands in the valley of PhD qualifying darkness with their steadfast support and kind words. I am forever grateful for our friendship. I would also like to acknowledge the never-ending support of my best friends,

Ashka Patel and Dr. Harjeet Johal. I would not have survived my PhD without you two, and our crazy innumerable weekend get together.

I want to thank all the past and present Moore lab members who have provided friendship and a friendly shoulder to cry upon, and with whom I have shared elation, frustration, tears and companionship. Thank you Harleen Saini, our friendship and afternoon coffee and singing will not be forgotten after we graduate. Mihir Metkar thanks for your help with bioinformatics, pushing me to think beyond my comfort zone, and all the bad jokes! I have to give special thanks to Dr. Alicia Bicknell who over the years has tirelessly provided feedback on my scientific writing. Thank you Alicia for willing to provide a shoulder to cry on whenever I needed it. Thank you Dr. Erin Heyer for countless scientific discussions and improving my presentation skills. v

A special thanks to Dr. Guramrit Singh who got me started with my PhD and is now a faculty at OSU. I miss our science and philosophical discussions. I also want to thank Dr. Blandine Mercier for taking over charge of my PhD career after

Guramrit left. She was a tough and fantastic mentor, who always pushed me to advocate my scientific knowledge. Thank you Dr. Emiliano Ricci for patiently teaching me several scientific techniques, and always being positive despite the pressure of handling multiple projects. Thank you Dr. Akiko Noma for sharing the lab afternoons with cups of coffee and giggles. Thank you to Dr. Jennifer

Broderick for the long walks in the park and warning me about the perils of attempting PhD. She was always willing to listen to my frustrations without judgment. I also want to thank Laureen Murtha and Rina Paladino for keeping the

Moore lab running and handling logistics (meetings, QE, TRAC) over the years.

In 2010, I started graduate school with two amazing scientists - Dr. Meetu Seth and Dr. Vahbiz Jokhi. Meetu, your friendship has been a blessing and I could not have asked for a better friend to survive graduate school. Vahbiz, your kindness and strength never cease to amaze me. Thank you both for your friendship through the PhD years and beyond. Thank you Dr. Shreya Kulkarni for your friendship, weekends filled with board games and singing songs. Many thanks to

Shiuli Agrawal for the friendship, laughter, delicious food, and always having my back.

I would like to acknowledge the support and interest of my thesis research advisor committee members, Dr. Jeremy Luban, Dr. Tiffany Moore Simas, Dr.

Jane Freedman and Dr. Manuel Garber. They have given me helpful scientific advise over the years. I would also like to thank my external committee member

Dr. Manoj Bhasin, who has taken the time to come to Worcester for my defense.

I would like to give special thanks to our collaborators, Dr. Ananth Karumanchi and Dr. Augustine Rajakumar without whom my thesis project would have not been possible. I would also like to acknowledge the tremendous support of Dr.

Karumanchi over the last several years and also coming to Worcester to attend my defense seminar.

Finally, I want to thank Dr. Melissa J. Moore, in whose lab I have conducted research for the last nine years (two years as a Research Associate, and seven years as a PhD student). I am very grateful for her mentoring, scientific training, freedom and the encouragement she has given me to pursue my scientific and non-scientific interests. It was a privilege to work with her on the preeclampsia project, which is close to her heart, both professionally and personally. Melissa, I am a better scientist today than when I had first entered your lab, and it has been an honor to be your student.

vii

ABSTRACT Preeclampsia (PE) is a disorder affecting 2-10% of pregnancies worldwide.

Clinical signs include high blood pressure (HBP) and proteinuria in the mother after the 20th week of pregnancy. Currently, the only cure for PE is delivery of the fetus, which is often necessary preterm and thus dangerous for both mother and fetus. Maternal symptoms of PE are caused by excess anti-angiogenic proteins of placental origin called soluble Flt1s (sFlt1s). sFlt1 mRNA isoforms are produced by alternative polyadenylation (APA) of full-length Flt1 (fl-Flt1) pre- mRNA. While fl-Flt1 encodes a transmembrane protein, sFlt1s encode truncated proteins that are soluble. Multiple sFlt1 isoforms exist, and their respective contribution to the pathophysiology of PE is unclear. Furthermore, it is unknown whether there is a genome-wide role for APA in PE.

In my thesis research, I developed a polyadenylation site sequencing method, and used this method to simultaneously quantify transcriptome-wide polyadenylation site usage and gene expression levels in normal, early-onset PE, and late-onset PE human placentae. I observed distinct expression profiles amongst the three groups, with differential expression of genes in several functional categories, including angiogenesis. I found that three sFlt1 isoforms account for >94% of all placental FLT1 transcripts, and that increased transcription of the entire FLT1 locus drives upregulation of both fl-Flt1 and sFlt1 in PE. I found that APA does not contribute substantially to PE pathophysiology. I viii

also identified siRNAs that knock down sFlt1 mRNA efficiently in cell lines that pave the way for further development of novel RNAi based therapeutics to alleviate PE.

Table of Contents TITLE PAGE ...... i SIGNATURE PAGE ...... ii DEDICATION ...... iii ACKNOWLEDGEMENTS ...... iv ABSTRACT ...... vii TABLE OF CONTENTS……………………………………………………………………………………..ix LIST OF FIGURES ...... xiii LIST OF TABLES ...... xv LIST OF ABBREVIATIONS ...... xvi COPYRIGHT MATERIAL ...... xviii 1. CHAPTER I: INTRODUCTION ...... 1 1.1 Pregnancy and pregnancy complications ...... 1 1.2 Preeclampsia: a silent killer ...... 2 1.3 History of preeclampsia ...... 4 1.4 Pathophysiology of preeclampsia ...... 6 1.4.1 Angiogenic Factors and their receptors ...... 7 1.4.2 Oxidative stress ...... 10 1.4.3 Immune response ...... 13 1.4.4 Other factors that cause PE ...... 14 1.5 Risk Factors of preeclampsia ...... 17 BMI, High Blood Pressure and sFlt1 : PlGF ratio ...... 17 1.6 Long term effects of preeclampsia on mother and baby ...... 21 1.6.1 Long-term effects on mother ...... 21 1.6.2 Long-term effects on fetus ...... 21 1.7 FLT1 and PE ...... 22 1.7.1 Role of Flt1 in PE ...... 22 1.7.2 sFlt1: Anti-angiogenic effect on mFlt1 and its signaling pathways ...... 23 1.8 Current treatments for preeclampsia ...... 24 1.9 siRNA therapeutics ...... 26 SUMMARY ...... 28 2. CHAPTER II: FLT1 and transcriptome-wide polyadenylation site (PAS) analysis in preeclampsia ...... 33 2.1 Introduction ...... 34 2.2 Results ...... 39 2.2.1 Placental PAS-Seq datasets ...... 39 2.2.2 EO-PE and LO-PE placentas exhibit distinct gene expression signature ...... 54 2.2.3 Comparison to previous RNA-Seq and microarray studies ...... 60 x

2.2.4 Non-coding RNAs and intergenic regions ...... 64 2.2.5 All Flt1 isoforms are upregulated in PE ...... 72 2.2.6 Other genes with altered isoform abundances ...... 76 2.3 Methods ...... 82 2.3.1 Human Subjects ...... 82 2.3.2 Ethics ...... 82 2.3.3 Tissue collection, storage and RNA isolation ...... 82 2.3.4 PAS-Seq library preparation ...... 83 2.3.5 PAS-Seq bioinformatics analysis ...... 84 2.4 Discussion ...... 87 3. CHAPTER III: PILOT SCREENING of siRNAs AGAINST Homo sapiens and Mus musculus sFlt1 mRNAs ...... 93 3.1. Introduction ...... 94 3.2. Results ...... 99 3.2.1 Pilot testing of anti-human-sFlt1 siRNAs in 393-D1 cells ...... 99 3.2.2 Two highly efficient anti-sFlt1 siRNAs identified ...... 99 3.2.3 Pilot testing of anti-mice-sFlt1 siRNAs in 393-D1 cells ...... 100 3.3. Methods ...... 106 3.3.1. Plasmids ...... 106 3.3.2. Cell Culture and Transfection ...... 106 3.3.3. Luciferase Assays ...... 107 3.3.4. Data Analysis ...... 107 3.4. Discussion ...... 108 4. Chapter IV: DEVELOPING anti-sFlt1 siRNAs to alleviate preeclampsia .... 109 4.1. Introduction ...... 110 4.2. Results ...... 112 4.2.1. sFlt1-i13 and sFlt1-e15a isoforms of FLT1 are overexpressed in human placentas ...... 112 4.2.2. Screen to identify hsiRNAs selectively silencing sFlt1-i13 and i15 sFlt1 isoforms ...... 113 4.2.3. Efficient delivery of fully chemically stabilized hydrophobically modified siRNAs to placental labyrinth ...... 117 4.2.4. Systemic administration of siRNA in vivo resulted in efficient delivery to placental labyrinth ...... 120 4.2.5. Efficient in vivo modulation of sFlt1 in placenta have no impact on mouse progeny survival and ability to strive ...... 121 4.2.6. Combinatorial silencing of sFlt1-i13 and i15a isoforms in vitro and in vivo .. 125 4.3. Methods ...... 128 4.3.1 siRNA Design ...... 128 4.3.2 Oligonucleotide synthesis, deprotection and purification ...... 128 4.3.3 Cell Culture ...... 130 4.3.4 Direct Delivery (passive uptake) of oligonucleotides ...... 131 4.3.5 mRNA quantification in cells and tissue punches ...... 132 4.3.6 Fluorescence Imaging ...... 133 4.3.7 PNA Hybridization assay ...... 133 4.4. Discussion ...... 134 xi

5. CHAPTER V: DISCUSSION AND FUTURE PERSPECTIVES ...... 140 5.1. Discussion ...... 141 5.2. Future Directions ...... 147 5.3. Conclusions ...... 150 APPENDIX A: RNA SEQUENCING ...... 152 Introduction ...... 152 Results ...... 152 Placental RNA-Seq samples ...... 152 Bioinformatics analyses ...... 155 RNA-Seq library issues ...... 155 Methods ...... 157 Isolation of mRNA for deep sequencing ...... 157 Library Preparation ...... 157 RNA-Seq Bioinformatics analysis ...... 157 RNA-Seq Differential gene expression ...... 160 Discussion ...... 165 APPENDIX B: Staufen1 senses overall transcript secondary structure to regulate translation ...... 166 Abstract ...... 167 Introduction ...... 168 Results ...... 169 Transcriptome-wide mapping of Stau-binding sites ...... 169 Stau associates with translating ribosomes ...... 174 Stau binds paired Alu elements in 3′ UTRs ...... 175 Non-Alu 3′-UTR targets ...... 181 dsRBD-dependent binding of Stau to CDS regions ...... 185 Gene ontology analysis ...... 189 Functional consequences of varying Stau protein levels ...... 192 Methods ...... 201 Plasmids and cell lines ...... 201 Generation of Stau1-knockdown cell line ...... 201 Stau1 RIPiT ...... 202 RIPiT RNA extraction ...... 204 Preparation of samples for RNA-Seq ...... 204 Poly(A)-site sequencing (PAS-Seq) ...... 204 Quantitative RT-PCR ...... 205 Sucrose-gradient sedimentation of Flag–Stau1-WT cells...... 206 Oligo(dT) pulldown of poly(A) RNAs after UV exposure of living cells ...... 207 Stau1 RNA reassociation test ...... 207 Ribosome profiling ...... 208 Illumina cDNA library construction ...... 210 Mapping of high-throughput sequencing reads ...... 211 Transcript-level quantification and normalization for all high-throughput sequencing libraries ...... 211 Transcriptome-wide pairing of inverted and tandem Alu elements ...... 212 xii

Definition of distal 3′-UTR regions ...... 212 Counting of sequencing reads for Alu pairs separated by different distances ...... 213 Secondary-structure analysis of inverted Alu pairs ...... 213 Analysis of ribosome-profiling reads ...... 214 Accession codes ...... 214 Discussion ...... 215 APPENDIX C: Alterations in mRNA 3' UTR isoform abundance accompany gene expression changes in human Huntington's disease brains ...... 221 Introduction ...... 222 Results ...... 224 The abundance of HTT mRNA 3′UTR isoforms changes in Huntington’s disease ..... 224 HTT mRNA isoform changes extend to liver and muscle and arise from both alleles 227 PolyA site sequencing identified a novel conserved mid-3′UTR isoform of HTT mRNA ...... 231 The HTT mRNA 3′UTR isoforms have different localization and half-lives ...... 235 Many other genes exhibit changes in isoform abundance ...... 240 Most genes with isoform changes in HD are not differentially expressed ...... 249 Decreasing expression of the RNA binding protein CNOT6 leads some genes to change isoform abundance ...... 251 Methods ...... 256 Samples ...... 256 Public 3′Seq data analysis ...... 256 RNA extraction ...... 256 qRT-PCR ...... 256 Nuclear and cytoplasmic isoform abundance ...... 257 In vitro transcription of ethinyl-uridine (EU) pulse chase spike-in ...... 257 EU pulse chase and isoform-specific qRT-PCR ...... 257 PolyA tail length assay ...... 258 MicroRNA and RNA binding protein target site prediction ...... 258 MicroRNA transfection ...... 258 PAS-seq library preparation ...... 259 PAS-Seq analysis ...... 259 PAS-seq analysis: isoform weighted change ...... 260 PAS-seq analysis: HD versus control gene expression comparison ...... 260 Gene ontology analysis ...... 260 CNOT6 siRNA transfection ...... 261 Discussion ...... 261 REFERENCES ...... 265

xiii

LIST OF FIGURES

Figure 1.1 VEGF mediated FLT1 signaling pathway ...... 9 Figure 1.2 FLT1 RNA and its isoforms ...... 11 Figure 2.1 FLT1 expression and patient selection ...... 36 Figure 2.2 Overview of bioinformatics analyses ...... 52 Figure 2.3 Analysis of differentially expressed genes ...... 58 Figure 2.4 FLT1 alternative polyadenylation isoform analyses ...... 62 Figure 2.5 Frequency distribution of called peaks and called intergenic PAS peaks ...... 69 Figure 2.6 Differential PAS isoform analyses on FLT1, SDC1 and ADAM12 genes ...... 74 Figure 2.7 Top 20 abundant genes in normal, EO-PE and LO-PE and analyses 80 Figure 3.1 Position of tested siRNAs on sFlt1-i13 S ...... 98 Figure 3.2 siRNAs against human sFlt1-i13 isoforms ...... 103 Figure 3.3 siRNAs against mouse sFlt1-i13 isoform ...... 104 Figure 4.1 Development of hydrophobically modified chemically stabilized siRNA compounds against sFlt1 ...... 115 Figure 4.2 Modified siRNA enables selective delivery to placental labyrinth .... 118 Figure 4.3 Efficient silencing of sFlt1 mRNA and protein by sFlt1-i13_2283 siRNA in pregnant mice ...... 123 Figure 4.4 Efficient silencing of sFlt1 by siRNA mixture (sFlt1-i13_2283/sFlt1- e15a_2519) in vitro and in vivo ...... 127 Figure A.1 RNA-Seq method flowchart ...... 154 Figure A.2 BoxWhisker plot showing the quality score across all bases on every position in the fastq file ...... 156 Figure A.3 Scatter plot showing differential gene expression between CTRL and EO-PE samples...... 161 Figure A.4 Scatter plot showing differential gene expression between CTRL and LO-PE samples ...... 162 Figure B.1 Mapping of Stau1 RNA-binding sites reveal coding regions and 3'UTRs as major occupancy sites ...... 172 Figure B.2 Inverted Alu pairs are an important class of Stau1 binding sites ..... 178 Figure B.3 Characterization of the structural features of Stau1 Alu-binding sites ...... 183 Figure B.4 Examples of 3'-UTR non-Alu Staufen-binding sites ...... 188 Figure B.5 Stau1 binding to 3'-UTRs correlates with GC content and predicted secondary-structure free energy ...... 190 Figure B.6 Stau1 occupancy on the CDS strongly correlates with GC content and predicted secondary-structure free energy ...... 194 Figure B.7 Consequences of Stau1 binding on RNA levels and ribosome density ...... 198 xiv

Figure B.8 Model of Stau1 RNA binding and its functional role in translation ... 200 Figure C.1 HTT 3'UTR isoforms change their relative amounts in HD ...... 226 Figure C.2 HTT isoform changes extend to muscle and liver and both HTT alleles change 3'UTR isoform amounts in HD ...... 229 Figure C.3 PAS-Seq revealed a third HTT 3'UTR isoform whose relative abundance also changes in HD ...... 233 Figure C.4 HTT 3'UTR isoforms are metabolized differently ...... 238 Figure C.5 Transcriptome-wide PAS-Seq analysis identifies a large subset of genes with 3'UTR isoform changes in HD motor cortex ...... 244 Figure C.6 Gene expression analysis shows most genes with 3'UTR isoform changes do not exhibit expression changes in HD ...... 247 Figure C.7 Changes in the expression of RNA binding proteins may influence alterations in isoform abundances ...... 254

LIST OF TABLES Table 1.1 Symptoms of PE ...... 16 Table 1.2 Risk Factors for PE ...... 20 Table 2.1 Clinical characteristics of patients ...... 41 Table 2.2 PAS-Seq read mapping statistics ...... 43 Table 2.3 Top 20 most abundant genes in placenta ...... 51 Table 2.4 Differentially expressed intergenic PAS in EO-PE ...... 68 Table 2.5 Differentially expressed intergenic PAS in LO-PE ...... 71 Table 2.6 FLT1 isoforms ...... 79 Table 2.7 PAS-Seq RT primer sequences ...... 86 Table 3.1 Sequence information of tested human siRNAs ...... 101 Table 3.2 Sequence information of mice siRNAs ...... 101 Table 3.3 siRNA IC50s and efficiencies ...... 105 Table A.1 RNA-Seq RT primer sequences ...... 159 Table A.2 RNA-Seq mapping statistics ...... 163 Table A.3 RNA-Seq reads mapping to rRNA sub-units ...... 164

xvi

LIST OF ABBREVIATIONS 3’ UTR 3’untranslated region BMI Body mass index bp base pair BP Blood pressure CDS coding sequences CTB Cytotrophoblasts inner layer of the trophoblast cells, undifferentiated cells CTRL Control samples DBP Diastolic blood pressure dsRBDs dsRNA-binding domains dsRBPs dsRNA-binding proteins dsRNA double-stranded RNA ENG Endoglin EO-PE Early-Onset Preeclampsia FACS Fluorescence-activated cell sorting FLT1 Fms-Like Tyrosine Kinase-1 GDM Gestational Diabetes Mellitus GDW Gestational Delivery Week HD Huntington’s disease HELLP haemolysis, elevated liver enzymes and low platelet count hs Homo sapiens HTT Huntingtin IP immunoprecipitation IUGR Intra-uterine growth restriction mRNA messenger RNA mFlt1 Membrane-bound Fms-Like Tyrosine Kinase-1 miRNA microRNA mm Mus musculus MUT mutant lncRNA Long non-coding RNA LO-PE Late-Onset Preeclampsia nt Nucleotide NTC Non-target control qPCR Quantitative polymerase chain reaction RIPiT RNA immunoprecipitation in tandem RNA Ribose Nucleic Acid RNAi RNA interference xvii

RNA-Seq RNA Sequencing PAS-Seq Polyadenylation Site Sequencing PAS Polyadenylation Site PGF/PlGF Placenta Growth Factor PE Preeclampsia SGA Small for Gestational Age SBP Systolic blood pressure SC subcutaneous SD Standard deviation siRNA short interfering RNA shRNA short hairpin RNA sFlt1 soluble Fms-Like Tyrosine Kinase-1 SNP single nucleotide polymorphism snRNA small nuclear RNA snoRNA small nucleolar RNA Stau Staufen ST syncytiotrophoblasts fully differentiated cells, in direct contact with maternal blood tRNA transfer RNA Trophoblasts outer layer of blastocyst, provides nutrients to growing embryo and often termed as placental cells VEGF Vascular Endothelial Growth Factor WT Wild Type Yac Yeast Artificial Chromosome

xviii

Parts of this dissertation have been submitted for publication as:

Patel A.A., Kaymaz Y., Rajakumar A., Bailey J.A., Karumanchi S.A., Moore M.J. FLT1 and transcriptome-wide polyadenylation site (PAS) analysis in preeclampsia (Manuscript under review at Nature Scientific Reports)

Turanov A.A., Hassler M.R., Patel A.A., Makris A., Lo A., Alterman J.F., Coles A.H., Haraszti R.A., Godinho B. MDC, Echeverria D., Karumanchi S.A., Moore M.J., Hennessy A. and Khvorova A. Development of Therapeutic anti-sFlt1 siRNA for the Treatment of Preeclampsia (Manuscript in preparation)

Lindsay Romo, Ami Ashar-Patel, Edith Pfister, and Neil Aronin Alterations in mRNA 3′UTR isoform abundance accompany gene expression changes in human Huntington’s disease brains (Manuscript in preparation)

Parts of this dissertation have been published as:

Ricci, E. P., Kucukural, A., Cenik, C., Mercier, B. C., Singh, G., Heyer, E. E., Ashar-Patel A., Peng L. et al Staufen1 senses overall transcript secondary structure to regulate translation. Nature Structural & Molecular Biology, 21(1), 26–35. http://doi.org/10.1038/nsmb.2739 1

1. CHAPTER I: INTRODUCTION 1.1 Pregnancy and pregnancy complications

Pregnancy in humans begins with conception when the sperm fertilizes the ovum creating a zygote (this stage is called conception). The zygote then implants itself in the uterus of the woman. Pregnancy typically lasts for forty- two weeks and is divided into three trimesters as follows. Week’s 1-12 form the first trimester (conception and implantation), weeks 13 to 28 form the second trimester and weeks 29 to 40 make up the last trimester. Full term pregnancy refers to the pregnancy that lasts 37-42 weeks and preterm pregnancy is typically earlier than 37 week of pregnancy. In 2015, it was estimated that around 303,000 women (WHO, 2011) worldwide died due to complications during pregnancy. Ninety nine percent of these deaths occurred in developing countries.

A few of the common pregnancy complications include but are not limited to severe bleeding, infections, gestational diabetes mellitus (GDM) and. PIH affects around 10% of pregnancies worldwide (American College of

Obstetricians and GynecologistsTask Force on Hypertension in Pregnancy,

2013) (ACOG Committee on Practice Bulletins--Obstetrics, 2002; WHO,

2011). Hypertensive disorders of pregnancy encompass gestational hypertension, preeclampsia [PE] and HELLP (Hemolysis, Elevated Liver 2

enzymes and Low Platelet count) syndrome. Gestational hypertension in pregnant women is diagnosed as de novo blood pressure elevations above

140/90 after 20 weeks, with the absence of excessive protein in the urine. PE is diagnosed as gestational hypertension or pregnancy-onset hypertension with the presence of excessive protein in the urine (termed as proteinuria)(ACOG Committee on Practice Bulletins--Obstetrics, 2002) or evidence of end-organ damage. Symptoms that are associated with PE are headaches; right upper quadrant abdominal pain and visual disturbances. A severe form of PE is called HELLP syndrome (Buchbinder et al., 2002; Haram et al., 2009). It is estimated that 10-20% of women who are diagnosed with

PE, will develop HELLP syndrome during pregnancy and even after delivery of the baby (https://www.preeclampsia.org/health-information/hellp- syndrome)(Sibai, 2004). There is a 25% chance that a woman with a history of HELLP pregnancy will develop HELLP syndrome in future pregnancies. As both PE and the HELLP syndrome are very dangerous to the mother and the fetus, delivery of the placenta is the only treatment.

1.2 Preeclampsia: a silent killer

Amongst the hypertensive disorders associated with pregnancy, PE is the number one cause of severe morbidity and mortality worldwide in pregnant women(WHO, 2011). If not diagnosed correctly PE can be fatal for the mother and the developing baby. By definition, it can only be diagnosed after the 20th 3

week (fifth month) of pregnancy. PE that occurs within 48 hours post-partum

(after delivering of the baby) is termed postpartum PE. It is also known to occur up to six weeks after delivery. PE is a heterogeneous disorder and can be classified as mild or severe PE depending on factors like severity of hypertension, symptoms and various laboratory criteria indicating end organ damage. It can further be classified into early-onset PE (EO-PE) versus late- onset PE (LO-PE), depending on the gestational delivery week (GDW) of the baby. In cases of EO-PE, the premature fetus is usually delivered before 34 weeks (GDW<34) and after 34 weeks in case of LO-PE (GDW>34).

Furthermore, EO-PE is sometimes termed as “placental” PE wherein there is abnormal placentation leading to oxidative stress and hypoxia(Ness and

Roberts, 1996). The LO-PE is referred as “maternal” PE wherein there is normal placentation but there is an inflammatory response(Ness and Roberts,

1996) later in gestation.

PE is one of the major causes of maternal and neonatal morbidity and mortality(Dadelszen et al., 2011). PE leads to a wide spread organ damage in the mother. Depending on the severity of the symptoms PE is managed by fetal monitoring (amniotic fluid volume, ultrasound to determine fetal growth and non-stress testing of fetal well-being) and assessing mother’s health

(monitoring symptoms, blood pressure, proteinuria, complete blood count, liver enzyme levels and renal function). In case of deteriorating condition of 4

the mother or evidence of fetal growth restriction, the only treatment available currently is to deliver the placenta i.e. the premature fetus irrespective of the gestational age. Premature delivery, thus, is one of the main mediators of neonatal morbidity and mortality(Larroque, 2004). In cases of EO-PE, the fetus is subject to nutritional insufficiency and asphyxia and the administration of corticosteroids for fetal lung development may have to be administered.

1.3 History of preeclampsia

According to the current standard, PE is defined as a hypertensive disorder.

However, this modern day definition did not come into play until 1881. In

1881, the world’s first blood pressure measuring device called sphygmomanometer was invented by Samuel Siegfried Karl Ritter von Basch.

Prior to 1880s, PE was defined by observing the symptoms of the pregnant woman, though it was not classified as a pregnancy disorder. For example, symptoms like convulsions are common between PE and epilepsy and in fact, eclampsia in a pregnant woman was considered to be one of the four types of epilepsy in the 15th century(Bell, 2010; Chesley, 1985). It was Bossier de

Sauvages who in 1739 distinguished between the seizures of epilepsy and

PE(Bell, 2010; Chesley, 1985). It was only in the 16th century that the word

“eclampsia” was termed. (Ong, 2004). In the ancient times (16th-18th century), the treatments available were bloodletting, using opiates, using cold water on the face and hurrying the delivery(Bell, 2010). 5

The definition of preeclampsia continued evolving in the 18th and 19th century and several pioneering physicians and scientists identified the other symptoms of PE.

In 1797, Demanet recognized edema in PE women(Bell, 2010; Chesley, 1985).

In 1840, Pierre Rayer observed high levels of protein in PE women and in

1843{Thoms:hYXmnQiJ}; John Lever demonstrated that proteinuria was particular to PE women and not a distinct kidney disorder(Johns, 1843).

Eventually, Vaquez and Nobecourt recognized high blood pressure as a symptom of PE in 1897(Bell, 2010; Chesley, 1985). The link between pregnancy and symptoms like swelling and edema were finally recognized as a distinct pregnancy syndrome called preeclampsia (Chesley, 1984).

During the 20th century, the post-delivery examination of placenta was becoming common and further aided in understanding the pathophysiology of PE. In 1903,

Seitz compared the uterus of two women who died of PE and observed the tissue under a microscope and reported lesions and infarctions on both the uteri(Robertson et al., 1967). Several other groups reported that women with hypertension develop vascular lesions on the placenta and the placental blood vessels(Williams, 1915) (Hertig, 1945). The foundations of classifying PE as a hypertensive disorder is credited to Dixon and Robertson, who in

1958(Robertson et al., 1967) (Dixon and Robertson, 1958).

It is disheartening that, in spite of the advances in understanding the molecular basis of PE, the treatment and cure options for women with PE has unchanged.

Even in the 21st century, the cure is delivery of the placenta. PE is a clinically unmet challenge, both in USA and around the world.

1.4 Pathophysiology of preeclampsia

In the past, PE was often termed the “disease of theories(Ness and Roberts,

1996). The current theory is that it is a two-staged process that starts in early pregnancy even though the clinical manifestation do not arise until much later in gestation.(Borzychowski et al., 2006; Ness and Roberts, 1996; Roberts and

Hubel, 2009)Some of the most common causes are summarized below.

In the first stage, the cytotrophoblasts fail to invade the maternal arteries leading to shallow placentation followed by the second stage, which is the clinical manifestation of endothelial dysfunction or maternal syndrome (Figure 1.2). As seen in the Figure 1.3 schematic, during the normal pregnancy, the cytotrophoblasts from the fetus invade the maternal decidua/uterus and bring about vasodilation of the maternal spiral arteries(Lyall et al., 2001). This vasodilation leads to an increased flow of maternal blood to the placenta. The cytotrophoblasts undergo Epithelial-Mesenchymal Transformation (EMT)(E

Davies et al., 2016) and acquire endothelial characteristics which are needed for invading the mother’s uterine wall. In PE pregnancies, the EMT is incomplete 7

causing shallow placentation leading to reduced flow of blood and oxygen to the placenta. This reduced flow of blood/oxygen is believed to cause a hypoxic and anti-angiogenic environment in the placenta, where in there is oxidative damage and inflammation(Munaut et al., 2008; Pennington et al., 2011; Powe et al., 2011;

Saito and Nakashima, 2014; Tsatsaris et al., 2003; Webster and Myatt, 2007;

Young et al., 2010). The endothelial syndrome has a multi-system affect and if the symptoms are unchecked can lead to HELLP syndrome.

1.4.1 Angiogenic Factors and their receptors

Formation of new blood vessels from existing blood vessels, a process

termed angiogenesis is absolutely essential for pregnancy(Ferrara et al.,

1996). Several studies have reported that malfunctioning angiogenesis is a

hallmark feature of PE(Maynard and Karumanchi, 2011; Maynard et al., 2003;

Torry et al., 1998).

VEGF mediated FLT1 signaling pathway

Vascular Endothelial Growth Factor (VEGF) and Placenta Growth Factor

(PGF) ligands bind to their receptors called Vascular endothelial growth factor

receptor-1 (VEGFR-1) or “fms like tyrosine kinase” (Flt1) and sFlt1. Flt1 is an

example of an receptor tyrosine kinase protein family(Vorlová et al., 2011)

that plays an important role in the angiogenesis pathway. Binding of VEGF

and PGF to membrane-bound Flt1 (mFlt1) leads to receptor dimerization 8

followed by autophosphorylation and signal transduction pathways like angiogenesis and cell proliferation and migration (Figure 1.4).

Flt1 is expressed in several human tissues(Jebbink et al., 2011) as well as cells including endothelial cells, trophoblasts and macrophages(Clark et al.,

1996; Dumont et al., 1995). Flt1 is reported to be transcriptionally upregulated by Hypoxia inducible factor 1-a (HIF1a)(Gerber et al., 1997). In mice embryos, Flt1 is needed for vasculogenesis but not for endothelial cell differentiation(Fong et al., 1995). Additionally in mice, Flt1 is not an essential gene and on Flt1 knockdown the mice survives(Hirashima et al., 2003).

Role of sFlt1 under normal physiological and preeclamptic conditions

sFlt1 is an anti-angiogenic protein and natural inhibitor of VEGF mediated angiogenesis pathway as an antagonist to m-Flt1. sFlt1 arises because of alternative polyadenylation of the Flt1 pre-mRNA transcript. sFlt1 mRNAs include unique 3’ sequence not present in m-Flt1 mRNA. 9

Figure 1.1 VEGF mediated FLT1 signaling pathway 10

Under normal physiological condition, sFlt1 is expressed in cells like endothelial cells, in body fluids like the amniotic fluid (Hornig et al., 2000) and tissues where vascularization is not needed. For example, the corneal tissue expresses sFlt1 that maintains a avascular state which is needed for a clear eyesight(Ambati et al., 2006). It has also been reported to be expressed in the endometrium of women undergoing menstruation(Graubert et al., 2001).

Graubert et al demonstrate that sFlt1 regulates the bioavailability of VEGF aiding to pause the VEGF-mediated vascular repair (angiogenesis) during the menstrual phase in the endometrium. sFlt1 contributes to the pathology of preeclampsia as high levels of sFlt1 of placental origin and reduced levels of its ligands VEGF and PGF are seen in the serum of pregnant women with preeclampsia. It is interesting to note that patients receiving anti-VEGF therapies for inflammatory disorders such as rheumatoid arthritis and other diseases develop PE like symptoms of high blood pressure and proteinuria.

1.4.2 Oxidative stress

The placenta plays a central role in establishing the exchange of nutrients

between the maternal and fetal blood pools. It is well understood that the

generation of Reactive Oxygen Species (ROS) is a normal physiological

response during pregnancy, however, increased levels prove damaging to

the lipids and proteins(Tiwari et al., 2010). 11

a Proximal polyA sites Distal polyA site

13 14 15b 15a 15 30 mRNA Previous names New name NM_002019.4 Flt1 mFlt1

NM_001160030.1 sFlt1_v2,sFlt1-14 sFlt1-e15a

NA sFlt1_v3 sFlt1-e15b

NA sFlt1_v4 sFlt1-i14

NM_001159920.1* sFlt1-i13l sFlt1-i13 long

NM_001159920.1* sFlt1_v1, sFlt1-i13s sFlt1-i13 short

* same NM# for both S and L forms of sFlt1-i13 isoform

Figure 1.2 FLT1 RNA and its isoforms 12

It has been reported that the increase in oxygen tension from 8 week to 12 week of gestation is synchronized perfectly to the time of maternal spiral artery remodeling(Jauniaux et al., 2010). The lack or slow rate of maternal spiral artery remodeling leads to a high pressure and the maternal blood enters the intervillous space at a faster rate. This sudden change in pressure and oxygen concentrations leads to a hypoxic environment, which in turn causes severe oxidative stress(Huppertz et al., 2003).

Oxidative stress generates ROS through various mechanisms causing ischemia, inflammation and damage to the placenta(MYATT and

WEBSTER, 2009).

Several studies have reported increase in glutathione peroxidase, superoxide dismutase, thioredoxin reductase, thioredoxin that are markers of oxidative stress(Jauniaux et al., 2010; Vanderlelie et al., 2005;

Zusterzeel et al., 2001). Additionally, the PE placentae shows evidence of lipid and protein oxidation(Llurba et al., 2004; Vanderlelie et al., 2005).

Oxidative stress also leads to necrosis in placenta causing trophoblast shedding(Huppertz et al., 2003). Using immunolocalization techniques, the authors studied trophoblast turnover and showed that placental hypoxia leads to necrotic shedding in PE as opposed to apoptotic shedding in normal placenta(Huppertz et al., 2003). These circulating trophoblast 13

debris also lead to inflammation and transfer of oxidative damage in the placenta to widespread endothelial organs of the mother.

1.4.3 Immune response

The maternal-fetal interface plays an important role in adapting mother’s immune response to the fetus, which is a mix of both maternal and paternal antigens. These paternal antigens are seen as foreign by the mother’s immune system. The placenta is broadly composed of two types of trophoblasts – villous trophoblasts which lacks the MHC class I and

MHC class II molecules and hence does not cause a major immune response and extra villous trophoblasts which express a mix of MHC class

I molecules. These molecules interact with their receptors present on uterine natural killer (NK) cells. NK cells secrete multiple cytokines like IL-

8 and Interferon gamma-induced protein 10 and angiogenic growth factors like VEGF, PGF and Transforming Growth Factor –β (TGF- β) that influence extravillous trophoblasts for vascular remodeling(Fukui et al.,

2012; Yagel, 2009).

NK cells present at the maternal-fetal interface play an important role in building an immune tolerance to the paternal antigens and ensuring favorable pregnancy conditions(Yagel, 2009). It has been shown that the uterine or decidual NK cells play an important role of in regulating trophoblast invasion and remodeling of mother’s spiral arteries(Hanna et 14

al., 2006). Thus, perturbations in the NK cell signaling contribute to abnormal placentation causing PE.

Interestingly, lack of immune tolerance plays a significant role in causing

PE. The paternal antigen contribution is involved in abnormal placentation and PE(Dekker, 2002).Women who have been exposed to paternal sperm have a lower risk of developing PE as opposed to women who have never been exposed to paternal sperm due to use of barrier contraceptives(Klonoff-Cohen et al., 1989). The risk of developing PE is high in first pregnancies or in a pregnancy after a change in sexual partner(Tubbergen et al., 1999). The exposure to paternal semen during the first pregnancy seems to have a protective role during the second pregnancy with the same partner. Similarly the risk of developing PE is higher after a long interpregnancy interval(Skjærven et al., 2002). Women who undergo artificial insemination using donated semen have also been reported to have a higher risk of developing PE(Need et al., 1983; Smith et al., 1997), though it has not been reported in cases of women undergoing in vitro fertilization(Watanabe et al., 2014).

1.4.4 Other factors that cause PE

Other factors like upregulated levels of angiotensin-II receptor and soluble

Endoglin(Levine et al., 2006; Venkatesha et al., 2006) have been linked to 15

PE. There is some evidence to support the familial increase of susceptibility to develop PE. However, studies have not resulted in any unanimous single gene or a loci, which indicates that multiple genetic factors are involved in the development of PE(Arngrimsson et al., 1990;

Chappell and Morgan, 2006).

Table 1.1 Symptoms of PE Adapted from (American College of Obstetricians and GynecologistsTask Force on Hypertension in Pregnancy, 2013)

1.5 Risk Factors of preeclampsia

Duckitt et al summarized the risk that predict that the woman will develop

PE(Duckitt, 2005) which is meant as a guideline to be used at prenatal

visits for both the physician and the pregnant woman. However, there are

not many studies that predict the risk specifically of EO-PE. Some of the

factors are age, previous history of PE, family history of PE, multiple

pregnancies, pre-existing medical conditions, high BMI. A caveat of all

these studies that only one factor was studied at a time. Therefore, the

impact of the association between risk factors on pregnancy is unknown.

Perhaps, future studies can be designed keeping this point in mind. Here I

list more details on risk factors that pertain to this thesis research.

BMI, High Blood Pressure and sFlt1 : PlGF ratio

In general, women with high BMI before pregnancy have double the risk of

developing PE (Bianco et al., 1998; Fields et al., 1996; Thadhani et al.,

1999). Another study predicts this risk to be as high as 4-fold(Thadhani et

al., 1999). Several reasons exist to explain this phenomenon such as

these women are advanced in age (age>34 years) or suffer from chronic

hypertension. It is interesting to note that older women also suffer from

decreased ovarian reserves(Tanbo et al., 1995), which is also a risk factor

for developing cardiovascular diseases later in life(Chu, 2003;

Kalantaridou et al., 2004). Another study demonstrates that women who 18

have IVF pregnancies generally suffer from low ovarian reserves and have a 2.7-fold risk of developing PE(Shevell et al., 2005; Woldringh et al.,

2006). Additionally IVF babies are also at a risk of being born Small-for-

Gestational-Age (SGA)(Koudstaal, 2000). High blood pressure

Hypertension in PE is defined as systolic blood pressure greater than or equal to 140 mmHg and diastolic blood pressure greater that or equal to

90 mmHg. There are two distinct types of hypertension scenarios– chronic hypertension (high blood pressure existing before pregnancy) and gestational hypertension (high blood pressure that did not exist before getting pregnant). Studies suggest that when both systolic blood pressure and diastolic blood pressure are higher in the first trimester, the women go on to develop PE later during pregnancy(Reiss et al., 1987; Sibai et al.,

1997). A gene wide association study on Norwegian population report that women with cardiovascular risk factors are disposed to developing

PE(Magnussen et al., 2007).

Another important criteria which is now getting common in diagnosing women at risk of PE, is the ratio of sFlt to it’s ligand PlGF(Rana et al.,

2012; Verlohren et al., 2010; 2012a; 2012b). This ratio is particularly important, because not all women who develop PE show symptoms of proteinuria. This ratio is thus inclusive of women who present PE, but do not show signs of proteinuria. PlGF unlike VEGF binds exclusively to Flt1 19

receptor. It has been also reported that when sFlt1 levels rise, the levels of free PlGF falls suggesting that the ratio of the sFlt1 to its ligand PlGF correlates with the severity of PE (Karumanchi, 2016) (Levine et al., 2005;

Noori et al., 2010). In several large multi-center studies, this ratio has shown accurate prediction rate(Malshe and Sibai, 2017; Rana et al., 2012)

(Chappell et al., 2013) (Duckworth et al., 2016) (Andersen et al., 2016).

In spite of these known risk factors, the fact is that majority of the women are symptom free. For example, a woman with a high BMI will not show symptoms like high blood pressure or proteinuria. This often causes delayed diagnosis.

We were very stringent in our patient selection criteria and only women with no pre-existing diabetes or hypertension. We further used two main selection criteria- systolic blood pressures ≥ 140 mm Hg and circulating sFlt1:PlGF ratios ≥ 85, an accepted diagnostic measure of PE.

Table 1.2 Risk Factors for PE Adapted from (American College of Obstetricians and GynecologistsTask Force on Hypertension in Pregnancy, 2013)

1.6 Long term effects of preeclampsia on mother and baby

PE has been shown to have several long-term problems for both the mother

and the fetus. Even after delivering the placenta and in turn the fetus, these

mother with preeclamptic pregnancies and their babies suffer for many years

after.

1.6.1 Long-term effects on mother

Women who have preeclamptic pregnancy double their risk of developing

long-term cardiovascular diseases(McDonald et al., 2008). Other reports that

these women have a seven-fold risk of developing future cardiovascular

diseases(Rodie et al., 2004; Valdiviezo et al., 2012). Additionally, their risk of

developing cardiovascular diseases in the short term also increases(Edlow et

al., 2009). Another study has shown that 20% of women with PE pregnancies

develop hypertension or microalbuminuria within seven years of a PE

pregnancy(Nisell et al., 1995). These women also develop kidney disorders

but no association was found between women with PE and developing cancer

in the future(Bellamy et al., 2007).

1.6.2 Long-term effects on fetus

The fetus in a PE pregnancy is at a risk of being born small for its gestational

age when compared to a fetus from a CTRL pregnancy. The fetus is often

delivered prematurely which has long term effects on the neurodevelopment

of these infants(Moster et al., 2008; Saigal and Doyle, 2008). Children born to

PE women have been shown to develop psychiatric disorders and depression 22

related disorders(Tuovinen et al., 2010; 2012). It was also found that the

children of PE mother have cognitive problems. For example, in a Western

Australian Pregnancy Cohort, it was found that the verbal ability of a 10-year

old born to be PE mother had a lower mean score than compared to a 10-

year old born to off a CTRL pregnancy(Whitehouse et al., 2012).

1.7 FLT1 and PE

1.7.1 Role of Flt1 in PE

Over the last decade, several publications (Ahmed, 1997; Heydarian et al.,

2009; Kendall and Thomas, 1993; Kita and Mitsushita, 2008; Kleinrouweler et

al., 2013; Maynard et al., 2003; Sela et al., 2008; Thomas et al., 2010; 2009a)

have shown that PE patients show elevated levels of Fms-related tyrosine

kinase 1 (FLT1) gene expression than in CTRL pregnant patients. Maternal

symptoms of PE are a primary consequence of excessive levels of sFlt1

protein in the maternal blood supply. FLT1 gene located on chromosome 13

gives rise to truncated mRNA isoforms (sFlt1s) in addition to the full-length or

membrane Flt1 (mFlt1) (Figure1.5). Both mFlt1 and sFlt1 mRNAs share a

common transcription site. There have been at least five sFlt1 mRNAs

reported that arise from usage of alternative splicing and alternative

polyadenylation (APA)(Heydarian et al., 2009; Maynard et al., 2003; Sela et

al., 2008; Thomas et al., 2010; 2009a; 2007). All the sFlt1 mRNAs have

distinct 3’ prime ends and give rise to distinct sFlt1 proteins due to coding

region APA. mFlt1 consists of 30 exons and the sFlt1-i13 S or L isoform 23

shares the first 13 exons and lacks the downstream exons. sFlt1-e15a, sFlt1-

e15b and sFlt1-i14 share the first 14 exons and lack the downstream exons.

Thus, APA leads to formation of multiple protein isoforms: (a) transmembrane

bound Flt1 (mFlt1), a tyrosine kinase receptor participating in vascular

endothelial growth factor (VEGF) signaling; and (b) truncated and blood

circulating (soluble) forms, sFlt1s, which lack the transmembrane anchoring

domain and act as antagonists to mFlt1 (Figure 1.6). Relative production of

these isoforms results from polyadenylation sites within exon 30 (mFlt1) or

within introns 13 and 14 (sFlt1s) (Figure 1.5).

1.7.2 sFlt1: Anti-angiogenic effect on mFlt1 and its signaling pathways

sFlt1 protein can interfere with the VEGF signaling in the following ways. As

the ligand binding domains are shared between the sFlt1 and mFlt1, sFlt2

sequesters the VEGF and PGF ligands away from the mFlt1. Additionally, it

has also been reported that sFlt1 protein binds to VEGF with a higher affinity

than the mFlt1(Thomas et al., 2007). On ligand binding the dimerization

domain comes together to form a dimer that leads to downstream tyrosine

kinase signaling. This dimerization domain is also shared and therefore sFlt1

can form inactive receptor dimers or dominant-negative receptor with no

downstream signaling.

There is rich evidence demonstrating that there is a down regulation of free

VEGF and free PGF ligand levels when the sFlt1 receptor is upregulated in

serum of PE patients(Ahmad and Ahmed, 2004; Kendall and Thomas, 1993;

Maynard et al., 2003; McKeeman et al., 2004).

Overexpression of sFlt1 in rats leads to hypertension, proteinuria and

glomerular endotheliosis (Maynard et al., 2003; 2005) the classic symptoms

of preeclampsia. Maynard et al used serum from CTRL and PE patients to

induce endothelial tube formation (in vitro model of angiogenesis) in HUVEC

cells. They found that there was impaired angiogenesis in HUVECs treated

with serum of PE patients(Maynard et al., 2003).

1.8 Current treatments for preeclampsia

There is no current cure for PE, except delivery of the baby. Depending on

the gestational delivery week of PE diagnosis, i.e. either EO-PE (GDW<34) or

LO-PE (GDW>34), different strategies can be used. In case of EO-PE, before

the onset of labor, maternal and fetal health is monitored closely. In case of

the EO-PE women, who have to deliver the baby pre-maturely (<34 weeks),

steroid is commonly administered to ensure fetal lung development. For

women with LO-PE, monitoring the mother and fetus is usually

recommended. In case of an adverse event, delivery is recommended.

Traditionally, women with severe PE have been treated with magnesium 25

sulfate, an anticonvulsant that prevents seizures in women with PE(Magpie

Trial Follow-Up Study Collaborative Group, 2006).

An encouraging report by Thadhani et al have shown that reducing sFlt1 levels in PE women is possible using dextran sulfate columns(Thadhani,

2010; Thadhani et al., 2011). sFlt1 has a negative isoelectric point and the maternal blood is circulated through the positively charged dextran cellulose columns, where the negatively charged proteins bind to the positively charged column, and then circulated back in the mother. They treated five severe PE women and all of them showed reduced sFlt1 levels for a range of 14 to 21 days. Additionally they observed reduced proteinuria and blood pressure, both classic symptoms of PE. These women could extend the pregnancy by 2 to 4 weeks. Most importantly, the fetus was normal and showed no abnormality. However, this was a pilot study and is an expensive treatment option, especially for the developing countries.

Thus, instead of managing PE, I see a potential to treat PE. In this thesis I propose that down regulating sFlt1 is an ideal target. One way of down regulating sFlt1 protein is by targeting the mRNA encoding it.

1.9 siRNA therapeutics

Short interfering RiboNucleic Acid (siRNA) is a type of RNA molecule, which

is 20-25 nt in length. siRNAs are part of the naturally occurring RNAi (RNA

interference) pathway(Lam et al., 2015) in eukaryotic cells. RNAi is a process

by which gene can be silenced by knocking down mRNAs. RNAi has been

shown to be an effective and safe as a therapeutic in several disease areas

(Burnett et al., 2011). Clinical trials for siRNAs in kidney disorders and solid

tumor are currently in process (Burnett et al., 2011) (Davis, 2009) (Zuckerman

and Davis, 2015)

Endogenously occurring siRNAs or siRNA expression vector can both be

used to silence mRNAs (gene). siRNAs are associated with the RNA-induced

silencing complex (RISC complex) and guide the complex to bind with the

target mRNA. On binding of the siRNA-RISC to the mRNA, it leads to

cleavage of the mRNA by the complex. siRNAs need complete sequence

complementarity to the target mRNA or gene. The silencing effect usually

lasts for a few days and is easier to deliver compared to other types of RNAi

molecules like shRNA and miRNA (Lam et al., 2015).

Some of the issues with siRNAs as therapeutic agent are activation of

interferon pathway, stability in tissues, delivery to targeted cells and the

routes of administration. However, in the last several years, there has been 27

tremendous research that tackles all the above-mentioned issues. The two main ways to avoid the immune response activation is to directly injecting synthetic siRNAs or replace the natural RNA moiety by non-natural moiety.

Using the synthetic siRNAs instead of the traditional long double stranded

RNA skips the Dicer processing part, thus avoiding activation of interferon pathway. In order to increase the stability of siRNAs in the human body, the backbone of the siRNA is replaced by 2’-O-Methyl RNA or 2’-O- Fluoro RNA, thus increasing the half-life of the siRNAs. Several modifications can be made to the sugar, base and backbone of the siRNA to increase stability. Addition of conjugates to the siRNAs aids in the delivery of siRNA to various tissues. siRNAs can be manufactured at low cost, are long-lived with high heat stability(Haussecker, 2012), and can be administered via subcutaneous (SC) injection.

Thus, given the above-mentioned advances in the siRNA therapeutic world, I propose that siRNAs have strong potential to be developed as a therapy for women suffering from PE. In Chapter III and Chapter IV, I demonstrate that siRNAs can knockdown sFlt1, which leads to improvement of PE symptoms in an animal model.

SUMMARY

In summary, PE is a disorder that affects 2-16% pregnancies worldwide with high mortality rates for both mother and fetus, especially in developing countries.

Clinical signs like high blood pressure and proteinuria develop in the mother after the 20th week of pregnancy. Maternal complications include seizures, HELLP syndrome and even death, while the fetuses of mothers with preeclampsia are prone to be SGA, suffer from hypoxia induced injury and death.

As seen in Chapter I, one major contributing factor in the pathogenesis of preeclampsia is increased level of proteins called soluble Fms-related tyrosine kinase 1 (sFlt1). sFlt1 is encoded by the gene Fms-related tyrosine kinase 1

(FLT1), which produces two major protein isoforms: (1) a transmembrane bound full–length Flt1 (mFlt1), which relays angiogenic signals through the vascular endothelial growth factor (VEGF) signaling cascade; and (2) multiple truncated and blood circulating forms, sFlt1, which lacks the transmembrane domain and acts as an antagonist to mFlt1 protein. The production of these Flt1 isoforms are the result of alternative polyadenylation, either in the 3′-UTR to create the mRNA encoding mFlt1 or within upstream introns (intron 13 and intron 14) to create the mRNAs encoding sFlt1. Many studies have reported high circulating sFlt1 expression levels in patients with preeclampsia compared to women who have a normal pregnancy. Furthermore, rats injected with an exogenous vector 29

expressing sFlt1, they developed signs of preeclampsia–high blood pressure and proteinuria. In preeclamptic women, the levels of sFlt1 drop to normal once the placenta is removed, indicating that placenta is the main source of sFlt1s. The only cure for preeclampsia currently is delivery of the placenta, irrespective of the gestational age of the fetus. Significant resources and health care expenditure are dedicated to the identification and management of preeclampsia. The costs related to preterm delivery from early-onset preeclampsia (<34 weeks) are 40-

100 times when compared to uncomplicated term delivery.

In spite of tremendous progress in the last few years, the mechanism by which sFlt1 is upregulated is unclear. Additionally, despite the knowledge that sFlt1s are generated by alternative polyadenylation, the genome-wide role of alternative polyadenylation is unclear. I address the following unanswered questions in this thesis:

(1) Why do sFlt1 levels increase during preeclamptic pregnancies? (2) What other genes are differentially expressed during preeclamptic pregnancy as opposed to normal pregnancies? (3) Can a reduction in the levels of circulating sFlt1 protein through RNAi alleviate the clinical manifestations of preeclampsia?

Furthermore, this thesis delves into (4) is there an overall increase in FLT1 gene expression? and/or (5) is there a preferential switch in polyadenylation sites from those in the 3′-UTR to those in the upstream introns? 30

In my thesis research, I developed and made polyadenylation site sequencing

(PAS-Seq) libraries from RNA of normal and PE human placentae to measure gene expression in a transcriptome-wide manner and investigate alternative polyadenylation signatures associated with early-onset PE (EO-PE; symptom onset <34 weeks) and late-onset PE (LO-PE; symptom onset >34 weeks) cohorts.

To investigate if sFlt1 levels increase by alternative polyadenylation or splicing and to identify other differentially expressed genes in PE, I analyzed the bioinformatics data generated from PAS-Seq (Chapter II) and RNA-Seq

(Appendix A) on human placental RNA samples. I found no general shift in alternative polyadenylation associated with PE. However, I observe three main sFlt1 mRNA isoforms that account for >94% of all transcripts, with increased transcription of the entire locus driving FLT1 upregulation in both EO-PE and LO-

PE (Chapter II). I find that EO-PE and LO-PE cohorts do exhibit gene expression profiles distinct from both each other and from normal placentae (Chapter II). I observed only two genes upregulated across all transcriptome-wide PE analyses to date; namely NRIP1 (RIP140), a transcriptional co-regulator linked to metabolic syndromes associated with obesity, and FLT1 (Chapter II).

The sFlt1 isoforms identified in Chapter II represent potential targets for therapeutic anti-sFlt1 siRNAs in EO-PE and LO-PE. In a proof-of-concept study, I identified anti-sFlt1 siRNAs that efficiently knockdown sFlt1 (both mammalian and mice sFlt1 isoforms) in NIH-3T3 cells (Chapter III). These anti-sFlt1 siRNAs are further chemically modified for efficient uptake and injected in mice to test for pharmacodynamics and pharmacokinetics studies and test if the siRNAs can reach the placenta and reach the fetus (Chapter IV). Furthermore, in collaboration with Dr. Anastasia’s Khvorova, Dr. Ananth Karumanchi and Dr.

Annemarie Hennessey, we test these siRNAs in a baboon-model of preeclampsia and demonstrate that these siRNA injected baboons show improvement in preeclamptic symptoms like high blood pressure and proteinuria

(Chapter IV).

The results of this thesis have advanced our understanding of the molecular changes occurring in preeclampsia (Chapter II) as well as demonstrated the value of sFlt1 as a therapeutic target to treat this disease (Chapter III and

Chapter IV).

Both appendices B and C represent other contributions I have made in terms of experiments, method development and bioinformatics data analysis during the duration of my PhD thesis.

2. CHAPTER II: FLT1 and transcriptome-wide polyadenylation site (PAS) analysis in preeclampsia

Chapter II was taken verbatim from:

Patel A.A., Kaymaz Y., Rajakumar A., Bailey J.A., Karumanchi S.A., Moore M.J.

FLT1 and transcriptome-wide polyadenylation site (PAS) analysis in preeclampsia (Manuscript under review at Nature Scientific Reports)

Author contributions statement:

M.J.M conceived of the project, A.R. collected the samples and prepared total

RNA, and A.A.P made PAS-Seq libraries and performed data analysis. Y.K. wrote custom pipelines for analyzing PAS-Seq data. M.J.M. and A.A.P wrote the main manuscript text and A.A.P prepared figures. J.A.B and S.A.K provided material support, expert advice and participated in manuscript editing. All authors reviewed the manuscript.

For information not contained in this chapter (i.e. supplemental tables), please refer online after the paper is published.

2.1 Introduction

Preeclampsia (PE) is one of the three leading causes of premature birth (Ananth et al., 2013). Characterized by hypertension and proteinuria in the third trimester, PE complicates up to 10% of pregnancies, and is estimated to result in more than 76,000 maternal and 500,000 infant deaths each year worldwide

(www.preeclampsia.org). In the developed world, significant resources are dedicated to the identification and management of preeclampsia, with US health care costs estimated to be 40-100 times higher for early-onset preeclamptic deliveries than for uncomplicated term deliveries(Shih et al., 2016). PE is a highly heterogeneous disease that initiates in the placenta and is likely driven by multiple mechanisms (Chaiworapongsa et al., 2004; Kang et al., 2011). In particular, early onset PE (EO-PE; defined by symptom onset <34 weeks) is thought to be clinically distinct from late onset PE (LO-PE; defined by symptom onset >34 weeks) (Dadelszen et al., 2003; Kusanovic et al., 2009; Romero et al.,

2009; Soto et al., 2011; Várkonyi et al., 2011). EO-PE is thought to be driven by poor placentation during early development, whereas LO-PE is considered to be a more maternal syndrome(Junus et al., 2012a).

Although the root causes of PE are not fully understood, it is now well established that the clinical manifestations are due to excess anti-angiogenic proteins, primarily "soluble fms-like tyrosine kinase 1" proteins (sFlt1s) in the mother’s bloodstream (Rana et al., 2012). sFlt1s are truncated forms of the 35

membrane-bound vascular endothelial growth factor (VEGF) receptor FLT1 (aka,

VEGFR1) that normally function to buffer placental growth factor (PlGF) and

VEGF signaling. sFlt1s contain the extracellular binding domain present in full length or membrane-bound FLT1 (mFlt1), but are secreted because they lack the transmembrane and intracellular tyrosine kinase domains. When sFlt1s are abnormally high in the mother's circulatory system, they interfere with her body's ability to respond to VEGF. Among other functions, VEGF is required for maintenance of the hepatic sinusoidal vasculature and other fenestrated vascular beds in the body (Kamba et al., 2006). Breakdown of these structures impairs maternal kidney function, leading to hypertension, proteinuria and cerebral edema (Eremina et al., 2008; 2003; Young et al., 2010) – classic features of PE and eclampsia.

In mammals, FLT1 is predominantly expressed in the placenta, with human placental Flt1 mRNA levels being 40-50 times higher than those observed in any other adult tissue (Jebbink et al., 2011) (Figure 2.1 a). Whereas the full-length isoform predominates in most adult tissues, placental and liver expression are dominated by truncated isoforms due to alternative polyadenylation (Figure 2.1 b). These alternative polyadenylation sites occur within introns downstream of the exons encoding the extracellular binding domain but upstream of those encoding the transmembrane and intracellular signaling domains (Thomas et al.,

2010; 2007; Vorlová et al., 2011). Five different sFlt1 mRNA isoforms have been 36

Figure 2.1 FLT1 expression and patient selection

a. Mean RNA-Seq FLT1 gene expression in various human tissues. Expression values are given as Fragments Per Kilobase of transcript per Million reads (FPKM) (whisker: standard deviation). b. RNA-Seq data for FLT1 gene illustrating isoform differences across various tissues. All data in a. and b. are from http://v14.proteinatlas.org./ENSG00000102755-FLT1/tissue. c. Patient segmentation into three groups by maternal systolic blood pressure and serum sFlt1: PlGF: Normal

(CTRL, green, n=6), Early-Onset PE (EO-PE, red, n=8), and Late-Onset PE (LO-PE, blue, n=3).

reported to date(Heydarian et al., 2009). Two (sFlt1-i13-short and sFLT-i13-long) result from alternative polyadenylation at different sites in intron 13 to yield mRNAs encoding the same 687 amino acid sFlt1 protein isoform, but with either a 17 or 4146 nt 3'-UTR, respectively. Three others (sFlt1-i14, sFlt1-e15a and sFlt1-e15b) are due to alternative polyadenylation within intron 14. sFlt1-e15a and sFlt1-e15b result from activation of two different cryptic splice acceptors

(cryptic 3' splice sites) upon use of alternative polyadenylation sites in the latter half of intron 14. In all, these five sFlt1 mRNAs encode four different sFlt1 proteins with different C-termini.

In women with PE, circulating sFlt1 protein levels are 5 to 7 times higher than those observed in normal pregnancies. One potential therapeutic approach would be to selectively knock down sFlt1 expression by RNA silencing (RNA interference, RNAi) using short interfering RNAs (siRNAs) targeting sequences unique to the sFlt1 mRNA isoforms and not contained in mFlt1 mRNA (Turanov et al.). Success of this strategy requires knowing exactly which sFlt1 mRNA isoforms to target. Further, while circulating sFlt1 levels increase in both EO-PE and LO-PE, this increase starts earlier in EO-PE and often with a more aggressive trajectory (Noori et al., 2010). This raises the question of whether

EO-PE placentae express a different pattern of sFlt1 mRNA isoforms than LO-PE placentae. That is, is it possible to define a common set of sFlt1 mRNA isoforms suitable for therapeutic intervention in both EO-PE and LO-PE patients? 39

Here we applied polyadenylation site sequencing (PAS-Seq) method on RNA isolated from normal (CTRL) and PE human placentas to identify the most abundant sFlt1 mRNA isoforms. Because PAS-Seq specifically interrogates polyadenylation sites it provides a much more accurate assessment of transcript

3' ends than RNA-Seq(Derti et al., 2012; Shepard et al., 2011), which is generally biased toward transcript 5' ends and is a better for investigating alternative splicing pattern(Zheng et al., 2011). Like RNA-Seq, PAS-Seq can be used for differential gene expression analysis by combining all isoforms detected at an individual gene locus into a single gene expression number(Derti et al., 2012).

Therefore, in addition to quantifying alternative polyadenylation both at the Flt1 locus and transcriptome-wide, we also used our PAS-Seq data to investigate overall gene expression changes between our patient groups.

2.2 Results 2.2.1 Placental PAS-Seq datasets

Several risk factors are associated with PE(WHO, 2011). In an attempt to minimize confounding variables by other associated conditions, we used highly selective criteria for patient inclusion. First, we limited our sample set to singleton pregnancies, and excluded any patients with preexisting hypertension or diabetes mellitus. Second, all of our control set (CTRL; n = 6) delivered at term with no other pregnancy complications (such as preterm labor or premature rupture of membranes). Finally, we limited our PE set (n=11) to patients with 40

systolic blood pressures ≥ 140 mm Hg and circulating sFlt1:PlGF ratios ≥ 85, an accepted diagnostic measure of PE associated with adverse PE related maternal/fetal outcome(Verlohren et al., 2010; 2012b) (Figure 2.1 c and Table

2.1). Gestational delivery week (GDW), ranged from 37 to 40 weeks in the CTRL set and from 28 to 39 weeks in the PE set. Based on the GDW, we further divided the PE patients into Early-Onset (EO-PE; GDW <34; n=8) and Late-

Onset (LO-PE; GDW >34; n=3) subgroups. In all cases, placental tissue was collected within 30 minutes of delivery and flash frozen in liquid nitrogen. Using these frozen samples, we isolated polyA tailed RNA from the villous tissue, which is the spongy layer in between the decidua (maternal side) and the chorion (fetal side), and the main source of sFlt1(Rajakumar, 2011; Rajakumar et al., 2012)

PAS-Seq patient characteristics Variables CTRL (n=6) EO-PE (n=8) LO-PE (n=3) Systolic BP (mmHg) 125 ± 10 183 ± 14 *** 154 ± 4 ** Diastolic BP (mmHg) 79 ± 8 109 ± 11** 98 ± 15 Maternal plasma sFlt1 (pg/mL) 9238 ± 4257 29672 ± 10420*** 26243 ± 13805* Maternal plasma PGF (ng/mL) 331 ± 262 98 ± 55* 158 ± 77 sFlt1 : PlGF ratio 40 ± 24 405 ± 256*** 166 ± 14* Maternal age (years) 35 ± 8 31 ± 5 27 ± 6 Maternal pre-pregnancy BMI (kg/m2) 26 ± 5 34 ± 5* 18 ± 8 Gestational delivery week 39 ± 1 31 ±2*** 37 ± 2 Proteinuria NA 100% 100% Newborn gender (F/M) 3/3 4/4 2/1 Mode of delivery (C-sec/Vaginal) 6/0 8/0 3/0 Induction of labor (yes/no) 0/6 1/7 2/1

Mean statistic ± standard deviation. Patient set comparisons were performed using the Mann-Whitney test; *P=0.05, **P=0.001, ***P=0.0002 Table 2.1 Clinical characteristics of patients

To analyze APA transcriptome-wide, we developed a simple, kit-free method by combining elements of a published PAS-Seq protocol(Shepard et al., 2011) with a circularization strategy used in our lab for making short RNA fragment libraries(Heyer et al., 2015b) (Figure 2.2 a, see Methods). Single-end sequencing on the Illumina HiSeq 2000 platform yielded 30-70 million (in average) high quality reads per library, of which 70-85% mapped uniquely to the human genome (hg19) (Table 2.2). Using a custom pipeline (Figure 2.2 b and Methods), peaks were identified as clusters of ≥15 reads (all libraries combined) whose 3' ends mapped within a 40 nt window (to allow for microheterogeneity in PAS use)(Tian et al., 2005). Peaks representing true PAS (i.e., those at which a polyA tail was added post-transcriptionally) were distinguished from peaks due to oligoT internal priming of transcripts at genomically-encoded adenosine-rich regions using a Naives Bayes classifier based probabilistic method (Shepard et al., 2011). True PAS were then assigned to genes using GENCODE (v19) annotations and quantified as counts per million (CPM). Custom UCSC genome browser tracks enabled visualization of read clusters and true PAS sites (Figure

2.2 c). In all, we detected 31,615 PAS on or within 5 kb downstream of annotated transcripts representing 13,328 genes and 3,482 PAS in intergenic regions. The majority of the annotated set was dominated by protein-coding genes (n=27,395

PAS; 86.6%), but also contained hundreds of snRNAs, lncRNAs, miRNAs and anti-sense transcripts (Figure 2.2 d).

Libraries Samples # of Total # of reads before # of reads after # of reads mapped % uniquely Reads filtering PCR filtering PCR to hg19 genome mapped duplicates duplicates reads CTRL N1 23,511,578 23,511,578 21,673,106 13,806,565 63.7 CTRL N2 29,672,955 29,672,955 28,318,378 18,126,488 64.0 CTRL N3 22,490,543 22,490,543 19,599,350 13,344,396 68.1 CTRL N4 39,123,865 39,123,865 32,433,406 21,920,746 67.6 CTRL N5 11,599,110 11,599,110 10,130,195 7,126,542 70.3 CTRL N6 23,200,677 23,200,677 22,826,611 14,095,599 61.8 EO-PE P1 24,601,335 24,601,335 23,261,667 14,995,294 64.5 EO-PE P2 23,555,010 23,555,010 20,145,239 13,266,060 65.9 EO-PE P3 7,308,017 7,308,017 6,715,379 4,847,280 72.2 LO-PE P4 36,170,272 36,170,272 24,955,305 17,184,747 68.9 LO-PE P5 28,092,206 28,092,206 27,274,495 18,112,595 66.4 EO-PE P6 41,604,215 41,604,215 35,597,242 24,044,206 67.5 EO-PE P7 33,294,603 33,294,603 30,896,129 21,125,542 68.4 EO-PE P8 33,199,502 33,199,502 29,467,774 17,419,272 59.1 EO-PE P9 6,642,588 6,642,588 6,396,058 3,606,549 56.4 EO-PE P10 6,807,878 6,807,878 6,544,426 3,530,627 53.9 LO-PE P11 8,249,282 8,249,282 7,791,478 4,438,303 57.0 Table 2.2 PAS-Seq read mapping statistics

By summing all true PAS reads mapping across individual loci, PAS-Seq data can be used to assess overall gene expression. That is, the data can be collapsed into a single gene expression number (counts per million, CPM) that encompasses all polyadenylated isoforms derived from each gene (Derti et al.,

2012). The top 500 genes in our samples were highly enriched for those expressed in placenta (p≤8.4 e-43) and syncytiotrophoblasts (p≤6.5 e-42) (see

Methods). Consistent with a previous RNA-Seq report (Sõber et al., 2015), nearly all of the 20 most highly expressed genes in the CTRL, EO-PE and LO-PE samples (Figure 2.3 a, b, c and 2.7 a) were placental or pregnancy-related (Table

2.3). This highest abundance set was dominated by hormones (e.g., CSH1,

CSH2 and CGA), positive and negative regulators of cell growth (e.g., GDF15 and TFPI2) and hemoglobin genes (HBA1, HBA2, HBB and HBG2); also included were two high abundance lncRNAs, H19 and NEAT1. 45

Gene Gene Name Protein Atlas (V14) Summary ADAM12 ADAM metallopeptidase domain 12 Placenta enriched This gene encodes a member of a family of proteins that are structurally related to snake venom disintegrins and have been implicated in a variety of biological processes involving cell-cell and cell-matrix interactions, including fertilization, muscle development, and neurogenesis. Expression of this gene has been used as a maternal serum marker for pre-natal development. Alternative splicing results in multiple transcript variants encoding different isoforms. Shorter isoforms are secreted, while longer isoforms are membrane-bound form. (provided by RefSeq, Jan 2014) CGA Glycoprotein hormones, alpha polypeptide Placenta enriched The four human glycoprotein hormones chorionic gonadotropin (CG), luteinizing hormone (LH), follicle stimulating hormone (FSH), and thyroid stimulating hormone (TSH) are dimers consisting of alpha and beta subunits that are associated noncovalently. The alpha subunits of these hormones are identical, however, their beta chains are unique and confer biological specificity. The protein encoded by this gene is the alpha subunit and belongs to the glycoprotein hormones alpha chain family. Two transcript variants encoding different isoforms have been found for this gene. (provided by RefSeq, Nov 2011) CRH Corticotropin releasing hormone Placenta enriched Corticotropin-releasing hormone is secreted by the paraventricular nucleus (PVN) of the hypothalamus in response to stress. Marked reduction in this protein has been observed in association with Alzheimer disease and autosomal recessive hypothalamic corticotropin deficiency has multiple and potentially fatal metabolic consequences including hypoglycemia and hepatitis. In addition to production in the hypothalamus, this protein is also synthesized in peripheral tissues, such as T lymphocytes and is highly expressed in the placenta. In the placenta it is a marker that determines the length of gestation and the timing of parturition and delivery. A rapid increase in circulating levels of the hormone occurs at the onset of parturition, suggesting that, in addition to its metabolic functions, this protein may act as a trigger for parturition. (provided by RefSeq, Apr 2010) CSH1 Chorionic somatomammotropin hormone 1 (placental lactogen) Placenta enriched The protein encoded by this gene is a member of the somatotropin/prolactin family of hormones and plays an important role in growth control. The gene is located at the growth hormone locus on chromosome 17 along with four other related genes in the same transcriptional orientation; an arrangement which is thought to have evolved by a series of gene duplications. Although the five genes share a remarkably high degree of sequence identity, they are expressed selectively in different tissues. Alternative splicing generates additional isoforms of each of the five growth hormones, leading to further diversity and potential for specialization. This 46

particular family member is expressed mainly in the placenta and utilizes multiple transcription initiation sites. Expression of the identical mature proteins for chorionic somatomammotropin hormones 1 and 2 is upregulated during development, although the ratio of 1 to 2 increases by term. Mutations in this gene result in placental lactogen deficiency and Silver-Russell syndrome. (provided by RefSeq, Jul 2008)

CSH2 Chorionic somatomammotropin hormone 2 Placenta enriched The protein encoded by this gene is a member of the somatotropin/prolactin family of hormones and plays an important role in growth control. The gene is located at the growth hormone locus on chromosome 17 along with four other related genes in the same transcriptional orientation; an arrangement which is thought to have evolved by a series of gene duplications. Although the five genes share a remarkably high degree of sequence identity, they are expressed selectively in different tissues. Alternative splicing generates additional isoforms of each of the five growth hormones. This particular family member is expressed mainly in the placenta and utilizes multiple transcription initiation sites. Expression of the identical mature proteins for chorionic somatomammotropin hormones 1 and 2 is upregulated during development, while the ratio of 1 to 2 increases by term. Structural and expression differences provide avenues for developmental regulation and tissue specificity. (provided by RefSeq, Jul 2008) CYP19A1 Cytochrome P450, family 19, subfamily A, polypeptide 1 Placenta enriched This gene encodes a member of the cytochrome P450 superfamily of enzymes. The cytochrome P450 proteins are monooxygenases which catalyze many reactions involved in drug metabolism and synthesis of cholesterol, steroids and other lipids. This protein localizes to the endoplasmic reticulum and catalyzes the last steps of estrogen biosynthesis, three successive hydroxylations of the A ring of androgens. Mutations in this gene can result in either increased or decreased aromatase activity; the associated phenotypes suggest that estrogen functions both as a sex steroid hormone and in growth or differentiation. The gene expresses two transcript variants. (provided by RefSeq, Jul 2008) FBLN1 Fibulin 1 – Fibulin 1 is a secreted glycoprotein that becomes incorporated into a fibrillar extracellular matrix. Calcium-binding is apparently required to mediate its binding to laminin and nidogen. It mediates platelet adhesion via binding fibrinogen. Four splice variants which differ in the 3' end have been identified. Each variant encodes a different isoform, but no functional distinctions have been identified among the four variants. (provided by RefSeq, Jul 2008) 47

FLT1 Fms-related tyrosine kinase 1 Placenta enriched This gene encodes a member of the vascular endothelial growth factor receptor (VEGFR) family. VEGFR family members are receptor tyrosine kinases (RTKs) which contain an extracellular ligand-binding region with seven immunoglobulin (Ig)-like domains, a transmembrane segment, and a tyrosine kinase (TK) domain within the cytoplasmic domain. This protein binds to VEGFR-A, VEGFR-B and placental growth factor and plays an important role in angiogenesis and vasculogenesis. Expression of this receptor is found in vascular endothelial cells, placental trophoblast cells and peripheral blood monocytes. Multiple transcript variants encoding different isoforms have been found for this gene. Isoforms include a full-length transmembrane receptor isoform and shortened, soluble isoforms. The soluble isoforms are associated with the onset of preeclampsia.(provided by RefSeq, May 2009) GDF15 Growth differentiation factor 15 Placenta enhanced This gene encodes a secreted ligand of the TGF-beta (transforming growth factor-beta) superfamily of proteins. Ligands of this family bind various TGF-beta receptors leading to recruitment and activation of SMAD family transcription factors that regulate gene expression. The encoded preproprotein is proteolytically processed to generate each subunit of the disulfide-linked homodimer. The protein is expressed in a broad range of cell types, acts as a pleiotropic cytokine and is involved in the stress response program of cells after cellular injury. Increased protein levels are associated with disease states such as tissue hypoxia, inflammation, acute injury and oxidative stress. [provided by RefSeq, Aug 2016] H19 Imprinted Maternally Expressed Transcript – This gene is located in an imprinted region of chromosome 11 near the insulin-like growth factor 2 (IGF2) gene. This gene is only expressed from the maternally-inherited chromosome, whereas IGF2 is only expressed from the paternally-inherited chromosome. The product of this gene is a long non-coding RNA which functions as a tumor suppressor. Mutations in this gene have been associated with Beckwith-Wiedemann Syndrome and Wilms tumorigenesis. Alternative splicing results in multiple transcript variants. [provided by RefSeq, Apr 2015] HBA1 Hemoglobin Subunit Alpha 1 – The human alpha globin gene cluster located on chromosome 16 spans about 30 kb and includes seven loci: 5'- zeta - pseudozeta - mu - pseudoalpha-1 - alpha-2 - alpha-1 - theta - 3'. The alpha-2 (HBA2) and alpha-1 (HBA1) coding sequences are identical. These genes differ slightly over the 5' untranslated regions and the introns, but they differ significantly over the 3' untranslated regions. Two alpha chains plus two beta chains constitute HbA, which in normal adult life comprises about 97% of the total hemoglobin; alpha chains combine with delta chains to constitute HbA-2, which with HbF (fetal hemoglobin) makes up the remaining 3% of adult hemoglobin. Alpha thalassemias result from deletions of each of the alpha genes as well as deletions of both HBA2 and HBA1; some nondeletion alpha 48

thalassemias have also been reported. [provided by RefSeq, Jul 2008]

HBA2 Hemoglobin Subunit Alpha 2 – The human alpha globin gene cluster located on chromosome 16 spans about 30 kb and includes seven loci: 5'- zeta - pseudozeta - mu - pseudoalpha-1 - alpha-2 - alpha-1 - theta - 3'. The alpha-2 (HBA2) and alpha-1 (HBA1) coding sequences are identical. These genes differ slightly over the 5' untranslated regions and the introns, but they differ significantly over the 3' untranslated regions. Two alpha chains plus two beta chains constitute HbA, which in normal adult life comprises about 97% of the total hemoglobin; alpha chains combine with delta chains to constitute HbA-2, which with HbF (fetal hemoglobin) makes up the remaining 3% of adult hemoglobin. Alpha thalassemias result from deletions of each of the alpha genes as well as deletions of both HBA2 and HBA1; some nondeletion alpha thalassemias have also been reported. [provided by RefSeq, Jul 2008] HBB Hemoglobin Subunit Beta – The alpha (HBA) and beta (HBB) loci determine the structure of the 2 types of polypeptide chains in adult hemoglobin, Hb A. The normal adult hemoglobin tetramer consists of two alpha chains and two beta chains. Mutant beta globin causes sickle cell anemia. Absence of beta chain causes beta-zero-thalassemia. Reduced amounts of detectable beta globin causes beta-plus-thalassemia. The order of the genes in the beta-globin cluster is 5'-epsilon -- gamma-G -- gamma-A -- delta -- beta--3'. [provided by RefSeq, Jul 2008] HBG2 Hemoglobin, gamma G Placenta enriched The gamma globin genes (HBG1 and HBG2) are normally expressed in the fetal liver, spleen and bone marrow. Two gamma chains together with two alpha chains constitute fetal hemoglobin (HbF) which is normally replaced by adult hemoglobin (HbA) at birth. In some beta-thalassemias and related conditions, gamma chain production continues into adulthood. The two types of gamma chains differ at residue 136 where glycine is found in the G-gamma product (HBG2) and alanine is found in the A-gamma product (HBG1). The former is predominant at birth. The order of the genes in the beta-globin cluster is: 5'- epsilon -- gamma-G -- gamma-A -- delta -- beta--3'. [provided by RefSeq, Jul 2008] 49

IGF2 Insulin-like growth factor 2 Placenta enriched This gene encodes a member of the insulin family of polypeptide growth factors, which are involved in development and growth. It is an imprinted gene, expressed only from the paternal allele, and epigenetic changes at this locus are associated with Wilms tumour, Beckwith-Wiedemann syndrome, rhabdomyosarcoma, and Silver- Russell syndrome. A read-through INS-IGF2 gene exists, whose 5' region overlaps the INS gene and the 3' region overlaps this gene. Alternatively spliced transcript variants encoding different isoforms have been found for this gene. [provided by RefSeq, Oct 2010] KISS1 KiSS-1 metastasis-suppressor Placenta enriched This gene is a metastasis suppressor gene that suppresses metastases of melanomas and breast carcinomas without affecting tumorigenicity. The encoded protein may inhibit chemotaxis and invasion and thereby attenuate metastasis in malignant melanomas. Studies suggest a putative role in the regulation of events downstream of cell-matrix adhesion, perhaps involving cytoskeletal reorganization. A protein product of this gene, kisspeptin, stimulates gonadotropin-releasing hormone (GnRH)-induced gonadotropin secretion and regulates the pubertal activation of GnRH nuerons. A polymorphism in the terminal exon of this mRNA results in two protein isoforms. An adenosine present at the polymorphic site represents the third position in a stop codon. When the adenosine is absent, a downstream stop codon is utilized and the encoded protein extends for an additional seven amino acid residues. [provided by RefSeq, Mar 2012] NEAT1* nuclear paraspeckle assembly transcript 1/ trophoblast derived – This gene produces a long non-coding RNA (lncRNA) transcribed non-protein coding RNA from the multiple endocrine neoplasia locus. This lncRNA is retained in the nucleus where it forms the core structural component of the paraspeckle sub-organelles. It may act as a transcriptional regulator for numerous genes, including some genes involved in cancer progression. [provided by RefSeq, Mar 2015] PAPPA Pregnancy-associated plasma protein A, pappalysin 1 Placenta enriched This gene encodes a secreted metalloproteinase which cleaves insulin-like growth factor binding proteins (IGFBPs). It is thought to be involved in local proliferative processes such as wound healing and bone remodeling. Low plasma level of this protein has been suggested as a biochemical marker for pregnancies with aneuploid fetuses. [provided by RefSeq, Jul 2008] PAPPA2 Pappalysin 2 Placenta enriched This gene encodes a member of the pappalysin family of metzincin metalloproteinases. The encoded protein cleaves insulin-like growth factor-binding protein 5 and is thought to be a local regulator of insulin-like growth factor (IGF) bioavailability. Alternative splicing results in multiple transcript variants. [provided by RefSeq, Jul 2010] 50

PRG2 Proteoglycan 2 Placenta enriched The protein encoded by this gene is the predominant constituent of the crystalline core of the eosinophil granule. High levels of the proform of this protein are also present in placenta and pregnancy serum, where it exists as a complex with several other proteins including pregnancy-associated plasma protein A (PAPPA), angiotensinogen (AGT), and C3dg. This protein may be involved in antiparasitic defense mechanisms as a cytotoxin and helminthotoxin, and in immune hypersensitivity reactions. The encoded protein contains a peptide that displays potent antimicrobial activity against Gram-positive bacteria, Gram-negative bacteria, and fungi. It is directly implicated in epithelial cell damage, exfoliation, and bronchospasm in allergic diseases. Alternatively spliced transcript variants encoding different isoforms have been found for this gene. [provided by RefSeq, Nov 2014] PSG2 Pregnancy specific beta-1-glycoprotein 2 Placenta enriched The human pregnancy-specific glycoproteins (PSGs) are a family of proteins that are synthesized in large amounts by placental trophoblasts and released into the maternal circulation during pregnancy. Molecular cloning and analysis of several PSG genes has indicated that the PSGs form a subgroup of the carcinoembryonic antigen (CEA) gene family, which belongs to the immunoglobulin superfamily of genes. Members of the CEA family consist of a single N domain, with structural similarity to the immunoglobulin variable domains, followed by a variable number of immunoglobulin constant-like A and/or B domains. Most PSGs have an arg-gly-asp (RGD) motif, which has been shown to function as an adhesion recognition signal for several integrins, in the N-terminal domain (summary by Teglund et al., 1994 [PubMed 7851896]). For additional general information about the PSG gene family, see PSG1 (MIM 176390).[supplied by OMIM, Oct 2009] S100P S100 Calcium Binding Protein P – The protein encoded by this gene is a member of the S100 family of proteins containing 2 EF-hand calcium-binding motifs. S100 proteins are localized in the cytoplasm and/or nucleus of a wide range of cells, and involved in the regulation of a number of cellular processes such as cell cycle progression and differentiation. S100 genes include at least 13 members which are located as a cluster on chromosome 1q21; however, this gene is located at 4p16. This protein, in addition to binding Ca2+, also binds Zn2+ and Mg2+. This protein may play a role in the etiology of prostate cancer. [provided by RefSeq, Jul 2008] TFPI2 Tissue factor pathway inhibitor 2 Placenta enriched This gene encodes a member of the Kunitz-type serine proteinase inhibitor family. The protein can inhibit a variety of serine proteases including factor VIIa/tissue factor, factor Xa, plasmin, trypsin, chymotryspin and plasma kallikrein. This gene has been identified as a tumor suppressor gene in several types of cancer. Alternative splicing results in multiple transcript variants. [provided by RefSeq, Aug 2012] 51

TGM2 Transglutaminase 2 – Transglutaminases are enzymes that catalyze the crosslinking of proteins by epsilon-gamma glutamyl lysine isopeptide bonds. While the primary structure of transglutaminases is not conserved, they all have the same amino acid sequence at their active sites and their activity is calcium-dependent. The protein encoded by this gene acts as a monomer, is induced by retinoic acid, and appears to be involved in apoptosis. Finally, the encoded protein is the autoantigen implicated in celiac disease. Two transcript variants encoding different isoforms have been found for this gene. [provided by RefSeq, Jul 2008] *Repression of major histocompatibility complex genes by a human trophoblast ribonucleic acid. Peyman JA et al. Biol. Reprod. 1999 Jan;60(1):23-31 PMID: 9858482 Europe PMC Pubmed Human trophoblast noncoding RNA suppresses CIITA promoter III activity in murine B-lymphocytes. Geirsson A et al. Biochem. Biophys. Res. Commun. 2003 Feb;301(3):718-724 PMID: 12565840 Europe PMC Pubmed Table 2.3 Top 20 most abundant genes in placenta

Figure 2.2 Overview of bioinformatics analyses

a. Schematic overview of PAS-Seq library preparation workflow. b. PAS-Seq bioinformatics pipeline schematic. c. Representative view of read alignment locations

(top track) and called PAS peaks (bottom track) on the ADAM12 3'-UTR. Our algorithm is able to remove background noise, filter out internal priming sites and accurately identify true polyadenylation sites. d. Distribution of PAS peaks among named gene classes and intergenic regions.

2.2.2 EO-PE and LO-PE placentas exhibit distinct gene expression signature

To determine whether the different patient sets had distinct gene expression patterns, we next performed Principal Component Analysis with all detected genes with sufficient read coverage (total read count across all 17 libraries ≥10; n

=15,765). By assessing variability across all samples independent of any externally imposed subsets, Principal Component Analysis (aka PCA) allows one to determine how closely related individual samples are to one another. For the top 2000 most variable genes across all libraries, Principal Component 1, 2 and 3 accounted for 21.8, 15.4 and 10.4 percent of the variance, respectively. In a plot of Principal Component 1 vs. Principal Component 2, the CTRL samples formed a coherent cluster readily distinguishable from the PE samples (Figure 2.3 d).

Further, the EO-PE and LO-PE samples formed distinct but overlapping clusters primarily defined by PC1, with the EO-PE cluster being more distant from CTRL than the LO-PE cluster. This suggested that EO-PE and LO-PE had overlapping but distinct gene expression signatures.

To define the sets of differentially expressed genes between our three groups, we performed pairwise differential gene expression analyses between CTRL,

EO-PE and LO-PE using DESeq2 (Figure 2.3 e,f, g). In agreement with the PCA analysis, the pair exhibiting the greatest differential expression profile was between CTRL and EO-PE (n= 701, padj ≤0.05). The most highly upregulated gene in EO-PE compared to CTRL was FLT1 (4.3-fold up), followed by Myosin 55

7B (MYO7B, 3.4-fold up) and Endoglin (ENG, 3-fold up). Other upregulated genes previously linked to PE included Inhibin Alpha (INHA, 2.8-fold up), Inhibin

Beta A (INHBA, 2.5-fold up), Vascular Endothelial Growth Factor A (VEGFA, 2.2- fold up), MicroRNA 210 Host Gene (MIR210HG, 2.4-fold up) and Pappalysin 2

(PAPPA2, 2.6-fold up). Genes exhibiting the greatest downregulation were

Somatostatin Receptor Type 1 (SSRT1, 4.3-fold down), Signaling Lymphocytic

Activation Molecule Family Member 1 (SLAMF1, 4.1-fold down), Hyaluronan And

Proteoglycan Link Protein 1 (HAPLN1, 3.2-fold down) and Cell Adhesion

Molecule 3 (CADM3; 2.8-fold down).

Overall, 349 genes were differentially expressed at a padj ≤0.01 between CTRL and EO-PE (Figure 2.3 e). Using a variety of online tools like

GeneCoDis(Nogales-Cadenas et al., 2009; Tabas-Madrid et al., 2012a)

(Carmona-Saez et al., 2007)and Genomatix, we performed extensive gene ontology analyses of this set, using all other expressed 13,299 genes as background (table S2 online). Recurring themes (tables S4 and S5 online) were

Biological/Cellular Adhesion and the HIF1a, HIF2a, Interleukin 6, Angiogenesis,

Cadherin, and Wnt signaling pathways, all of which have been previously linked to PE (Knöfler and Pollheimer, 2013; Maynard and Karumanchi, 2011; Sitras et al., 2009). Also enriched were Pol II transcription factors and genes with LEF1 and TATA transcription factor binding sites (padj ≤0.01).

Consistent with the PCA, fewer genes were differentially expressed between

CTRL and LO-PE, with only 40 reaching the padj ≤ 0.01 threshold (Figure 2.3 f).

Those exhibiting the greatest fold change in LO-PE were Perilipin 2 (PLIN2; 2.1- fold up), Zinc Finger Protein 175 (ZNF175, 2.1-fold up), Acyl-CoA Synthetase

Long-Chain Family Member 1 (ACSL1, 2.1-fold up), Dual Specificity

Phosphatase 15 (DUSP15, 2.3-fold down) and Forkhead Box S1 (FOXS1, 2.3- fold down). The 97 differentially expressed genes (padj ≤ 0.05; 13,265 background gene set) were highly enriched for Pol II transcription factors and components of the HIF2a signaling pathway (tables S3, S6 and S7 online).

Comparison of the differentially expressed genes in EO-PE and LO-PE revealed that they have overlapping but clearly distinct gene expression profiles (Figure

2.7 c). Gene ontology terms associated with the 50 differentially expressed genes common to both EO-PE and LO-PE (padj ≤0.05) includes the HIF1a,

HIF2a and VEGFR1 (FLT1) signaling pathways (table S8 online). Differential expression analysis between EO-PE and LO-PE sets revealed 19 genes reaching the padj ≤ 0.05 threshold (Figure 2.3 g), with the genes exhibiting the greatest differences being Family With Sequence Similarity 19 Member A2

(FAM19A2, 2.5-fold higher in EO-PE), Pleiomorphic Adenoma Gene-Like 2

(PLAGL2f, 2.3-fold higher in EO-PE), TNF Receptor Superfamily Member 10a

(TNFRSF10A, 1.9-fold higher in LO-PE) and HAPLN1 (1.9-fold higher in LO-PE). 57

Thus, while EO-PE and LO-PE have similar clinical presentations, their distinct gene expression signatures suggest different mechanistic drivers.

Figure 2.3 Analysis of differentially expressed genes

Box-plots of top twenty most abundant transcripts in a. Normal, b. EO-PE, and c. LO-PE patients. Gene expression measured as Counts Per Million (CPM) mapped reads. Dark bar: median; box: 25-75% interquartile; whiskers and dots: range. d. Principal component analysis (PCA) of highly variable genes. Axes are principal components 1 and 2 (PC1 and PC2). Each point represents a single patient, with colors indicating the patient type as in Figure 2.1 c Volcano plots showing differential gene expression for e.

CTRL vs. EO-PE, f. CTRL vs. LO-PE, and g. EO-PE vs. LO-PE. X axis: log2 Fold

Change (FC) between samples; Y axis: -log10 false discovery rate after false positive removal (padj). Each dot is a gene: purple, padj≤0.01; black, padj≤0.05; n, number of genes in each group. h. Venn diagram illustrating overlap of differentially expressed genes (padj cutoffs as indicated) between our EO-PE and LO-PE PAS-Seq datasets, two published RNA-Seq datasets(Kaartokallio et al., 2015; Sõber et al., 2015)and a large meta analysis of available microarray data(Leavey et al., 2015). i. Scatter plot of FLT1

RNA expression (CPM) vs. Systolic Blood Pressure (mmHg). Each point represents a single patient, with colors indicating the patient type as in Figure 2.1c.

2.2.3 Comparison to previous RNA-Seq and microarray studies

Numerous other studies have employed microarrays or RNA-Seq to investigate differential gene expression in PE. We therefore compared our differentially- expressed EO-PE and LO-PE gene sets (padj≤0.05) to similar gene sets from two recent RNA-Seq studies (Kaartokallio et al., 2015; Sõber et al., 2015) and a large scale meta analysis of microarray datasets (Leavey et al., 2015) (Figure 2.3 h).

As expected from the heterogeneous and multifactorial nature of PE, the gene sets reported in the three previous studies were largely non-overlapping (Figure

2.3 h and 2.7 b). Our gene lists were most similar to the microarray list (Leavey et al., 2015), with 289 of 701 (41%) differentially-expressed EO-PE genes and 32 of 97 (33%) differentially-expressed LO-PE genes being among the 1295 genes

(padj≤0.01) in the microarray set. Notably, only one gene, NRIP1 (nuclear receptor-interacting protein 1; aka RIP140), intersected all five lists (Figure 2.3 h). Upregulated in all datasets, NRIP1 is a ubiquitously-expressed nuclear protein that acts as a co-activator and/or co-repressor for numerous nuclear receptor transcription factors (Nautiyal et al., 2013).

The only other differentially-expressed gene reported in ours and all previous studies was FLT1. Although FLT1 did not appear in the list of differentially expressed genes in the Kaartokallio et al. study (Kaartokallio et al., 2015), it was stated in the text as being differentially expressed. The magnitude of FLT1 overexpression in PE varies greatly among published reports – whereas the 61

microarray and RNA-Seq values ranged from 1.5 to 2.0-fold (Leavey et al., 2015;

Sõber et al., 2015), qPCR analyses specifically interrogating the FLT1 locus range from 3.86-fold to 4.5-fold above normal(Jebbink et al., 2011; Sõber et al.,

2015). In our PAS-Seq data, FLT1 was the most highly upregulated gene in the

EO-PE patient set, exhibiting a 4.3-fold increase relative to the CTRL set

(padj=2.21E-13; Figure 2.3 e). FLT1 was also upregulated the LO-PE set, but only by 1.9-fold (padj ≤ 0.01; Figure 2.3 f). Further, consistent with Flt1 overexpression being a significant driver of clinical symptoms, total Flt1 gene expression strongly correlated with systolic blood pressure over all samples (r =

0.84) (Figure 2.3 i). Flt1 also correlated with body mass index (BMI), but to a much lesser degree (r = 0.51, Figure 2.7 d). Thus our PAS-Seq data validate previous findings that Flt1 overexpression is central to PE.

Figure 2.4 FLT1 alternative polyadenylation isoform analyses

a. Schematic of FLT1 gene and six previously identified mRNA isoforms arising from alternative polyadenylation. Thick boxes: protein coding exons; thin boxes: 3'-UTRs; lines: introns; arrows: sFlt1 and mFlt1 polyadenylation sites. Table shows all previous isoform names and new systematic names. b. UCSC genome browser view showing

PAS-Seq peak distribution on the FLT1 gene for a representative CTRL (top track), EO-

PE (middle-track) or LO-PE (bottom track) patient. Arrows: sFlt1 and mFlt1 polyadenylation sites for isoforms in a. c. Bar graph showing mean expression (counts per million; CPM) across all libraries for all 16 detected mRNA isoforms at the FLT1 locus.

2.2.4 Non-coding RNAs and intergenic regions

In addition to protein-coding genes, our PAS-Seq data also provided expression values for numerous non-coding RNAs (Supplementary Table S3). For example, many miRNAs (miRs) derive from long non-coding polyA+ primary transcripts

(pri-miRNAs)(Lee et al., 2004). Pri-miRNAs exhibiting particularly high placental expression include the chromosome 14 and 19 miR clusters (C14MC and

C19MC, respectively), the miR-371-3 cluster and the mir210 host gene

(mir210HG)(Mouillet et al., 2015). C19MC is transcribed by Pol III (Borchert et al., 2006), so was not represented in our PAS-Seq data. Of the remaining three, only the mir210HG exhibited differential expression among our patient sets, being 2.4 and 1.8-fold increased in EO-PE and LO-PE, respectively, relative to

CTRL. Previously shown to be upregulated in PE (Enquobahrie et al., 2008), miR210 is known to downregulate the THSD7A gene (Luo et al., 2016).

Consistent with this, THSD7A was 1.5-fold downregulated in EO-PE (padj≤0.043), and 1.3-fold downregulated in LO-PE (although not statistically significant; padj≤0.58).

Other differentially-expressed noncoding transcripts with embedded miRNAs were DNM3OS (miR214; 1.7-fold down in both EO-PE and LO-PE) and RP6-

99M1.2 (miR221/miR222; 1.6-fold down in EO-PE). Finally, some miRNAs derive from the introns or 3'-UTRs of protein coding genes, including SREBF1 (miR33b and miR6777; 1.6-fold up in EO-PE, 1.5-fold up in LO-PE), PDCD4 (miR4680; 65

1.5-fold up in EO-PE), and FAM172A (miR2277; 1.4-fold down in EO-PE).

Among this set, only miR214 has been previously linked to PE, with microarray data indicating its downregulation (Xu et al., 2014). Mir214 is thought to regulate both PLGF (Gonsalves et al., 2015) and components of the β-catenin pathway

(Xia et al., 2012), with its own expression being modulated by HIF1a (Gonsalves et al., 2015).

Our datasets also included PAS for 170 lncRNAs, 112 anti-sense transcripts and

133 "processed transcripts" (defined in GenCode as a transcript with no ORF).

Although nine of these were differentially expressed in EO-PE, LO-PE or both

(table S3 online), careful examination of the associated PAS peaks on the UCSC genome browser revealed that nearly all could be explained by differential expression of an overlapping protein coding gene. Thus we conclude that differential lncRNA expression is not a general hallmark of PE. The sole exception was TP73-AS1 (downstream of and antisense to TP73, expression of which was undetectable in our dataset), a lowly expressed transcript (base mean

= 62.5 CPM) that was 1.4 fold-downregulated (padj= 0.02) in EO-PE. TP73-AS1 is differentially expressed in multiple cancers (Zang et al., 2016), and recent evidence suggests that it functions to positively regulate expression of BDH2, a cytosolic type 2 hydroxybutyrate dehydrogenase involved in mitochondrial function (Zang et al., 2016). While BDH2 expression was 1.4-fold reduced in our

EO-PE set, this change was not statistically significant (padj=0.15). 66

Finally, we examined the 3,482 intergenic PAS. Compared to the mean CPM distribution for genic PAS, the vast majority of intergenic PAS were lowly expressed (Figure 2.5 a), and DESeq2 analysis yielded only 13 and 8 differentially expressed intergenic PAS in EO-PE and LO-PE, respectively (table

S9 online). Many of these were associated with LINE and LTR repeat elements, suggesting that activation of select endogenous retroviral elements might occur in PE. For example, three intergenic PAS upregulated in EO-PE lie between the

LTF and CCRL2 genes in a region containing numerous retroviral elements, and an RNA-Seq track of multimapping reads confirmed the existence of an independent multiple LTR-containing transcript (Figure 2.5 b).

Location bas log2F pval padj log1 fold E annotation Repeatma Cl eM oldCh ue 0pad cha x sker as ean ange j nge p. element s name chr2_47090 6.7 2.19 7.13 3.15 11.5 4.6 U Downstrea MLT1L LT 314_47090 9 E-16 E-12 0114 p m of R 315 848 MCFD2 chr14_1030 9.3 2.18 9.90 3.29 11.4 4.5 U Upstream None – 57585_103 9 E-16 E-12 8331 p of RCOR1 057586 026 chr6_11160 9.1 1.57 3.62 1.79 5.74 3.0 U Downstrea Harlequin- LT 7625_1116 7 E-09 E-06 7871 p m of int R 07626 613 REV3L chr3_46458 23. 1.33 2.50 4.68 4.33 2.5 U Downstrea MER50 LT 014_46458 71 E-07 E-05 0186 p m of R 015 795 CCRL2 chr6_11159 38. 1.13 1.74 0.00 2.86 2.2 U Intron of downstrea LT 9538_1115 16 E-05 1377 0985 p RP5- m of R 99539 255 562 1112D6.4 THE1B chr6_14855 8.8 1.09 1.28 0.00 2.96 2.1 U Downstrea downstrea LI 0728_1485 3 E-05 1083 5294 p m of RP11- m of NE 50729 191 858 631F7.1 L1MB7 chr13_2884 9.7 1.04 2.66 0.00 2.76 2.1 U Intron of None – 4400_2884 8 E-05 1736 0226 p PAN3 4401 894 619 chr9_85685 23. 0.97 0.00 0.01 1.99 2.0 U US of L1ME1 LT 242_85685 97 0272 0052 7742 p RASEF R 243 538 119 366 chr3_46473 39. 0.93 0.00 0.00 2.12 1.9 U betwn LTF MER11A LT 944_46473 01 0182 7535 2904 p CCRL2 R 945 734 213 455 68

chr15_2058 41. -0.89 8.96 0.00 2.36 1.9 D Downstrea X6A_LINE LI 6733_2058 84 E-05 4296 6922 o m of NE 6734 13 63 w HERC2P3 n chr3_46477 27. 0.85 0.00 0.01 1.71 1.8 U Downstrea LTR36 LT 143_46477 30 0673 9474 0532 p m of LTF R 144 959 533 954 chr6_11950 21. 0.83 0.00 0.04 1.39 1.8 U Intron of L1MA9 LI 2940_1195 45 1870 0055 7336 p MAN1A1 NE 02941 35 66 11 chr6_11950 25. 0.82 0.00 0.04 1.39 1.8 U antisense L1M3 LI 3832_1195 33 1891 0278 4925 p to MAN1A1 NE 03833 4 656 027

Table 2.4 Differentially expressed intergenic PAS in EO-PE

Figure 2.5 Frequency distribution of called peaks and called intergenic PAS peaks

a. Frequency distribution of called PAS peaks (mean values across all datasets) in called genes and intergenic regions. b. UCSC genome browser view of PAS-Seq reads (bottom track) mapping to intergenic region between the LTF and CCRL2 genes

(top track). Also shown from top to bottom are repeat elements (Repeat Masker track), an RNA-Seq track of multi-mapping reads from an EO-PE patient, called genic PAS peaks, uniquely mapping PAS-Seq reads from a CTRL patient, called intergenic PAS peaks, and uniquely mapping PAS-Seq reads from an EO-PE patient.

Location base log2Fol pvalu padj log10p fold annotation Repeatmaske Cl Mea dChang e adj chan r elements as n e ge s chr1_1518288 12.1 - 6.45E 0.003 2.471 1.9 Upstream of LTR61 LT 29_151828830 0.89780 -06 03459 THEM5 R 6753 5 chr1_1721081 29.5 - 5.89E 0.017 1.779 1.8 Intron of DNM3 None – 91_172108192 0.85534 -05 40568 observed 7122 8 chr16_855880 5.3 0.56211 3.98E 0.012 1.896 1.5 Upstream of RP11- None – 37_85588038 9321 -05 69998 19697 118F19.1 and observed GSE1 chr5_1799102 5.1 0.46703 0.000 0.037 1.421 1.4 Upstream of MLT1J2 LT 92_179910293 4317 2296 87258 67502 CNOT6 R 4 7 9 Table 2.5 Differentially expressed intergenic PAS in LO-PE

2.2.5 All Flt1 isoforms are upregulated in PE

A major goal of this study was to determine which sFlt1 mRNA isoform(s) to target with therapeutic siRNAs (Turanov et al., in preparation). Our PAS-Seq data revealed a total of 16 PAS in the Flt1 gene (Figure 2.4 b and c; table S10 online), including all 6 previously reported Flt1 mRNA isoforms. Thirteen are low abundance, each representing <0.6% of true PAS reads mapping to the Flt1 locus. Two previously reported isoforms, sFlt1-i14 (previously sFlt1-v4) and sFlt1-e15b (previously sFlt1-v3), were among this low abundance set, so contribute little to sFlt1 expression. The four high abundance species are sFlt1- i13S (previously sFlt1_v1 short 3'UTR), sFlt1-i13L (previously sFlt1_v1 long

3'UTR), sFlt1-e15a (previously sFlt1_v2) and m-Flt1 (full-length Flt1 mRNA)

(Figure 2.4 a and b). Among these, sFlt1-e15a is the most abundant, accounting for ~50% of Flt1 PAS reads in all samples (Figure 2.4 b and c). The sFlt1-i13 short and long isoforms each accounted for 20 to 25% of PAS reads, whereas full-length m-Flt1 mRNA represented only ~5% of PAS reads. Thus, at time of delivery, sFlt1-i13 short, sFlt1-i13 long and sFlt1-e15a comprise the vast majority

(88 to 97%) of placental Flt1 mRNAs.

Comparison among datasets revealed that all four high-abundance Flt1 isoforms are upregulated in EO-PE vs. CTRL (p ≤0.05; two-way t-test) (Figure 2.6 a). A similar trend is also observed comparing LO-PE to CTRL, but only the sFlt1-i13 long isoform reached statistical significance (p ≤ 0.05; two-way t-test), likely due 73

to the small sample size. To assess whether this general upregulation of the Flt1 locus was accompanied by any shift in PAS usage, we examined how the fraction of total reads mapping to each of the four major PAS differed among the

EO-PE, LO-PE and CTRL datasets (Figure 2.6 b). The only statistically significant (two-way t-test) differences were a 1.5-fold increase in sFlt1-i13 short and a 2.0-fold reduction in mFlt1 in EO-PE compared to CTRL. Thus in EO-PE, use of the sFlt1-i13 short PAS increases at the expense of the mFlt1 PAS.

Nonetheless, the predominant feature associated with both EO-PE and LO-PE is transcriptional upregulation of the entire locus Flt1 locus.

Figure 2.6 Differential PAS isoform analyses on FLT1, SDC1 and ADAM12 genes

a. Superimposed scatter plots showing expression (counts per million mapped reads;

CPM) of the four most abundant FLT1 mRNA isoforms. b. Superimposed scatter plot comparing fractions of total PAS reads mapping to the FLT1 locus represented by each indicated mRNA isoform. For a. and b., each point represents an individual CTRL, EO-

PE or LO-PE patient with the color indicating patient type as described in Figure 2.1c. A two-way t-test was used for statistical comparisons between indicated groups.

Horizontal line: mean; n.s.: not statistically significant. c. UCSC genome browser view of

SDC1 gene illustrating differential PAS isoform abundance (Distal-to-Proximal Switch) for a representative CTRL (top track), EO-PE (middle-track) or LO-PE (bottom track) patient. d. Same as c., but ADAM12 gene (Proximal-to-Distal Switch).

2.2.6 Other genes with altered isoform abundances

The altered Flt1 mRNA isoform abundance associated with EO-PE prompted us to ask whether this phenomenon was detectable at any other gene loci. Using one-vs.-others model (see Methods), we observed PAS switches (padj≤0.05) in only eight genes in EO-PE compared to CTRL (ADAM12, FLT1, NEAT1, TIMP2,

CRH, PAPPA, PSAP and SDC1) and only four genes in LO-PE compared to

CTRL (EIF2A, PSG1, SDC1 and PHLDB2) (online table S11), with SDC1 being common to both. The switches were almost equally divided between proximal-to- distal (n=6) and distal-to-proximal (n=5) isoform shifts. Most of the altered isoform abundances were limited to shifts within untranslated regions and so were not expected to alter protein isoform ratios. The exceptions were FLT1

(discussed above), SDC1 and ADAM12. In all three cases, the affected mRNA isoforms change the ratio of secreted and membrane-bound isoforms. The SDC1 gene encodes short (secreted or "shed") and long (membrane-bound) forms of the major transmembrane heparan sulfate proteoglycan syndecan 1 from mRNA isoforms having different PAS (Figure 2.6 c). In both EO-PE and LO-PE, the isoform encoding membrane-bound Sdc1 was preferentially upregulated.

Likewise, ADAM12 expresses two alternate PAS mRNA isoforms, ADAM12-S and ADAM12-L, encoding short (secreted) and long (membrane-bound) forms of metalloproteinase-disintegrin 12 (Biadasiewicz et al., 2014) (Christians and

Beristain, 2016) (Figure 2.6 d). Whereas the entire gene is 1.6-fold upregulated 77

in EO-PE, the ADAM12-L isoform increases more than the ADAM12-S isoform

(1.8-fold vs. 1.3-fold, respectively).

N1 N2 N3 N4 N5 N6 P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P11

Chro pA pA N Pre EST Human Acces CTR CTR CTR CTR CTR CTR EO- EO- EO- LO- LO- EO- EO- EO- EO- EO- LO- moso star sto Loc e vio s mRNAs sion L L L L L L PE PE PE PE PE PE PE PE PE PE PE me t p atio w us numbe site site n na na r m me e s FLT chr13 290 290 Flt1 – non none BC02984 – 1.58 1.34 2.25 3.10 1.60 3.59 9.22 3.17 19.9 7.96 32.0 23.1 25.7 5.05 15.4 75.6 23.7 1_1 406 406 , e 9 323 427 919 951 086 827 262 689 737 837 14 941 882 969 096 668 518 00 01 intr on 3 FLT chr13 289 289 Flt1 i1 V1 BG5 U01134, – 106. 71.6 115. 239. 206. 147. 872. 411. 2576 220. 355. 4370 1341 595. 1712 1656 776. 1_2 638 638 , 3 7188 AF33982 4155 9424 8465 8203 9693 1690 6188 5983 .942 8797 6782 .675 .970 0558 .233 .924 6851 03 22 intr sh 5, 2, 2 8 16 28 32 2 7 67 07 94 2 22 9 37 36 1 on or BC0 AK29293 13 t 3900 6 7 FLT chr13 289 289 Flt1 – non – none – 5.88 2.95 3.01 8.93 6.17 6.71 22.3 11.5 44.7 12.2 15.2 41.6 21.3 16.7 22.4 65.6 16.2 1_3 628 628 , e 056 739 226 983 476 678 164 6388 686 4408 7274 915 3938 1505 5397 386 3043 95 98 intr on 13 FLT chr13 289 289 Flt1 – – – – – 23.6 12.6 21.4 37.0 29.9 29.2 152. 60.1 160. 59.4 87.9 259. 124. 91.9 143. 254. 98.5 1_4 598 598 , 3535 3612 6238 8091 5904 6598 7994 0684 4785 712 162 6145 1153 78 9699 8055 701 53 59 intr on 13 FLT chr13 289 289 Flt1 i1 sFlt – AK09247 – 143. 89.3 136. 229. 223. 284. 783. 351. 1210 412. 633. 1761 886. 651. 901. 1988 736. 1_5 597 598 , 3 1- 2 8475 4894 5557 7152 4351 1433 0109 2365 .473 0225 6229 .088 452 3447 6827 .307 7032 79 07 intr lo i13 56 7 53 75 25 94 9 26 16 8 7 3 3 11 3 on ng long 13 3'U TR FLT chr13 289 289 Flt1 – non – none – 39.1 22.4 53.4 67.7 35.2 26.2 295. 150. 238. 67.7 125. 390. 229. 120. 224. 507. 180. 1_6 596 596 , e 2838 0446 6756 8730 1896 6735 8068 4575 9950 3115 7064 9761 3800 5290 9803 7877 1180 83 91 intr 7 6 1 7 6 8 3 1 1 7 3 3 7 5 8 on 13 FLT chr13 289 289 Flt1 – non – none – 0 0 0 0 1.14 0 2.73 2.03 3.09 0 1.07 7.29 4.82 1.71 4.40 5.01 2.77 1_7 590 590 , e 347 263 321 936 692 783 587 668 275 406 105 14 15 intr on 14 FLT chr13 289 289 Flt1 i1 V4 – BG43585 – 1.80 1.79 1.63 3.26 5.03 6.47 7.85 8.13 18.9 2.91 3.42 34.6 29.6 8.94 18.4 13.6 16.6 1_8 571 571 , 4 2 94 236 164 498 129 689 631 284 405 526 658 828 339 481 915 747 263 67 68 intr on 14 FLT chr13 289 289 Flt1 – non – none – 0 0 0 0 1.14 0 2.04 2.16 7.57 1.16 0 10.1 7.76 2.52 6.60 2.27 6.33 1_9 571 571 , e 347 947 029 622 61 881 663 985 412 912 382 25 26 intr on 14 FLT chr13 289 289 Flt1 e1 V3 – BF06103 – 0 0 0 2.09 1.14 0 3.64 1.77 3.44 1.16 1.56 10.3 6.71 2.43 5.28 5.01 5.93 1_1 543 543 , 5b 9 from 892 347 35 906 374 61 644 326 098 949 33 406 796 0 83 84 intr heydarian on paper 14 FLT chr13 289 289 Flt1 e1 V2 AI18 EU36060 NM_00 325. 243. 356. 585. 649. 652. 1990 1391 4205 708. 722. 6512 3805 1574 3917 2994 2696 1_1 422 422 , 5a 8382 0, 11600 0132 4019 4516 9093 2656 7256 .150 .604 .492 7964 9093 .626 .207 .106 .564 .304 .622 1 20 75 intr EU36883 30 16 42 77 97 9 89 68 48 01 04 98 94 23 91 27 89 04 on 0, 14 AK30039 2,AK3099 01, LQ48791 6 FLT chr13 289 289 Flt1 – – – – – 1.01 0 1.12 2.72 1.82 3.59 6.60 4.32 8.95 1.84 3.23 21.0 7.91 4.42 11.0 12.7 7.12 1_1 419 419 , 779 96 082 956 827 385 057 371 633 077 987 744 723 069 631 555 2 84 85 intr on 14 79

FLT chr13 288 288 Flt1 – – – – – 1.92 1.16 1.12 1.86 2.05 2.87 10.3 6.73 22.0 4.37 7.53 18.2 18.8 3.97 16.2 17.7 5.14 1_1 771 771 , 2491 5032 9597 5701 8255 862 6122 501 3986 288 847 0844 5103 547 9017 7713 624 3 92 99 exo n 30 FLT chr13 288 288 Flt1 – – – – – 1.69 1.34 1.63 3.49 1.60 1.91 5.69 6.60 10.3 1.26 6.55 11.4 10.8 3.79 18.4 15.9 5.54 1_1 765 765 , 632 427 164 82 086 908 298 793 312 328 945 887 582 477 915 538 21 4 72 73 exo n 30 FLT chr13 288 288 Flt1 – – – – – 4.07 1.70 1.50 4.50 2.97 6.83 12.7 8.00 19.9 4.95 7.24 9.24 22.0 4.42 16.2 20.0 9.10 1_1 752 752 , 116 274 613 879 303 672 523 576 737 594 476 873 934 723 902 562 487 5 54 55 exo n 30 FLT chr13 288 288 Flt1 m full- – – NM_00 71.2 41.8 58.2 69.4 73.6 75.8 191. 129. 261. 77.9 127. 220. 205. 162. 182. 386. 154. 1_1 744 744 , Flt leng 2019 4533 5147 3706 1973 3974 0352 0558 7442 38 3453 7623 7414 4761 4527 2734 9945 7828 6 81 92 exo 1 th 1 2 7 4 n 30 Table 2.6 FLT1 isoforms

Figure 2.7 Top 20 abundant genes in normal, EO-PE and LO-PE and analyses

a. Box-plots of the most abundant transcripts in CTRL, EO-PE, and LO-PE patients.

Gene expression measured as counts per million (CPM). b. Venn diagram illustrating the overlap of differentially expressed genes between the three previously published data sets. c. Venn diagram demonstrating overlap of differentially expressed genes between EO-PE and LO-PE in our PAS-Seq data. d. Scatter plot showing correlation of

FLT1 mRNA expression vs. body mass index (BMI).

2.3 Methods 2.3.1 Human Subjects

Placentas were collected from CTRL and PE subjects (Table 2.1) in accordance with all institutional policies and with approval of the institutional review board at the Beth Israel Deaconess Medical Center (Boston, MA, USA). The diagnosis of

PE was based on the updated criteria of American College of Obstetrician and

Gynecology Task force on Hypertension in Pregnancy(American College of

Obstetricians and GynecologistsTask Force on Hypertension in Pregnancy,

2013). Patients with history of diabetes, chronic hypertension, renal disease or multiple gestations were excluded. Pre-delivery maternal plasma samples were used to measure circulating sFlt1 and PlGF using commercial ELISA kits (R & D systems, MN) as described elsewhere(Goel et al., 2015)

2.3.2 Ethics

The study was approved by the Beth Israel Deaconess Medical Center

Institutional Review Board (IRB), #2008P-000061. All subjects provided written informed consent for use of placental and blood samples for research.

2.3.3 Tissue collection, storage and RNA isolation

We excised placental biopsies (2x2 cm) without basal and chorionic plates and wiped with cotton gauze to remove blood and debris. The villous tissue, the spongy layer that is in between the decidua (maternal side) and the chorion (fetal 83

side) was collected since it is the main source of sFlt1(Rajakumar et al., 2012).

These tissue samples were flash-frozen in liquid nitrogen within 30 min of placental delivery.

2.3.4 PAS-Seq library preparation

For preparation of PAS-Seq libraries, we adapted a previously published method for making libraries from RNA fragments (Heyer et al., 2015a); (Figure 2.2 a).

Total RNA integrity was first confirmed by agarose electrophoresis (data not shown) and then polyA+ RNA enriched by oligoT hybridization. PolyA+ RNA samples were then fragmented to 60-80 nt via chemical hydrolysis and reverse transcribed with one of twelve anchored oligoT oligonucleotides containing forward and reverse Illumina sequencing primer sites separated by a hexa- ethyleneglycol spacer (Sp18) linker. At the 5′ end, each oligonucleotide began with 5'p-GG to promote ligation (Heyer et al., 2015a), followed by 5 random nucleotides (unique molecular index, UMI) to enable PCR duplicate removal.

Each primer also harbored a unique 5 nt Hamming barcode (BC), allowing for sample multiplexing (Table 2.7). Following cDNA circularization with CircLigase

I, libraries were PCR amplified (12-14 cycles) and subjected to single end 100 nt sequencing on the Illumina HiSeq platform in the UMass Medical School Deep

Sequencing Core.

2.3.5 PAS-Seq bioinformatics analysis

After confirming overall library quality with FASTQC

(http://www.bioinformatics.babraham.ac.uk/projects/fastqc/, S.Andrews, 2011), we removed duplicate reads with the same UMI. We then used Cutadapt (Martin,

2011) to trim residual adaptor/spacer sequences and poly-A stretches from read

3' ends and mapped the reads to hg19 reference genome using Bowtie2 (with parameters –m –best –p4). Only uniquely mapping reads with high quality scores

(>20) were used for further analysis. Such reads represented 70-85% of total reads for all libraries.

We defined the polyadenylation cleavage site as the base closest to the poly-A stretch captured within read sequence, thus, very last base at the 3’-end site after the trimming. Based on this, we calculated read coverages of every genomic location considering only 3’-end of the aligned reads. This forms a sharp peak of read coverage around the cleavage sites with an average width 40nt

(ranging between 1nt-120nt). We then clustered sites that are in proximity closer than 40nt window and summed the counts of clustered sites. In order to determine all possible PAS, which might possibly differ from sample to sample, we pooled all candidate locations of all samples and reiterated the clustering using the same window length. Since genomic DNA containing poly-A stretches can be hybridized and pulled with the oligo-T primers, aka internal priming, we used a Naïve Bayes classifier(Sheppard et al., 2013) based software to 85

determine the likelihood of all sites being false priming sites. In addition to filtering internal priming locations based on this, we also removed background noise caused by widespread base level read alignments by fitting the count distribution of each gene to Poisson distribution. After determining precise PA locations, we then calculated expression/read counts of every PAS for each sample and annotated using Gencode v19. We also marked known polyA sites with PolyA DB (http://exon.umdnj.edu/polya_db/). For differential gene expression tests, we used sum of all PAS counts of each gene and calculated library sizes based on this. The raw read counts for genes were used as an input for differential gene expression analyses using DESeq2 (Love et al., 2014) in R

(https://www.R-project.org). The default normalization using

'estimateSizeFactors' function was used. Adjusted p-values were calculated using the Benjamini and Hochberg method (Benjamini and Hochberg, 1995). In order to determine significance levels of alternative polyA site switches, we constructed a 2 by 2 contingency table composed of mean normalized read counts of one site and the mean of rest of the sites, if there are multiple, in each of the condition. We tested the significance with Chi-square test for given site and iterated through all sites of the gene. We then corrected p-values for multiple hypotheses testing using Benjamini and Hochberg method.

Primer name Sequence AA-HiSeq RT /5Phos/GGNNNNNATCACAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/i oligodT BC 1 Sp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTTT TTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNCGATGAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC 2 iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT TTTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNTAGCTAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/i oligodT BC 3 Sp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTTT TTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNGCTCCAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC 4 iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT TTTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNACAGTAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC 5 iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT TTTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNCAGATAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC 6 iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT TTTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNTCCCGAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC 7 iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT TTTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNGGCTAAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC 8 iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT TTTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNAGTCAAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC 9 iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT TTTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNCTTGTAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/i oligodT BC Sp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTTT 10 TTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNTGAATAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/i oligodT BC Sp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTTT 11 TTTVN-3' AA-HiSeq RT /5Phos/GGNNNNNGTAGAAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT/ oligodT BC iSp18/CTCGGCATTCCTGCTGAACCGCTCTTCCGATCTTTTTTTTTTTTTTTTT 12 TTTTVN-3' Table 2.7 PAS-Seq RT primer sequences

Red represents the Barcodes (BC), V is either base A or C or G, N is any base.

Sp18 is a spacer.

2.4 Discussion Here we present comprehensive differential gene expression and PAS isoform abundance analysis of placental RNA from subjects having an abnormal plasma angiogenic profile accompanying either EO-PE or LO-PE. We find that EO-PE and LO-PE are distinct conditions with different but related gene expression signatures. Common to both is overexpression of the anti-angiogenic factor sFlt1, expression of which strongly correlates with systolic blood pressure (Fig.

3i). Our PAS-Seq analysis confirms previous reports identifying sFlt1-e15a as the predominant sFlt1 mRNA isoform in both normal and PE placentae

(Heydarian et al., 2009; Palmer et al., 2015; Sela et al., 2008; Thomas et al.,

2009a), with the next most abundant isoforms being sFlt1-i13 short and long

(Kendall and Thomas, 1993). Together sFlt1-i13-short, sFlt1-i13-long and sFlt1- e15a account for >94% of all sFlt1 expression in both normal and PE placentae.

This information enabled us to identify siRNAs that decrease sFlt1 expression as a potential therapeutic for PE (Turanov et al.). Our observation that all Flt1 isoforms increase in PE indicates that Flt1 upregulation is primarily a transcriptional response. Intriguingly, the only other gene differentially expressed in PE across all transcriptome-wide analyses to date is the transcriptional coregulator NRIP1 (RIP140)(Nautiyal, 2017; Nautiyal et al., 2013).

Because sFlt1 is a key negative regulator of angiogenesis in both normal (Ambati et al., 2006) and cancerous tissues (Shibuya, 2011), how its production is 88

modulated has been the topic of multiple studies outside the placenta. One condition known to alter sFlt1 expression is hypoxia (Munaut et al., 2008;

Nagamatsu et al., 2004). In human microvascular endothelial cells (HMVECs), the predominant sFlt1 mRNA isoform is sFlt1-i13-short. Under hypoxic conditions, sFlt1-i13-short levels decrease, but mFlt1 levels are unaffected suggestive of a post-transcriptional mechanism (Ikeda et al., 2011). More recent work (Ikeda et al., 2016) identified heterogeneous nuclear ribonucleoprotein D

(hnRNP D) as a negative regulator of sFlt1-i13-short production, possibly acting through binding to an AU-rich element near the sFlt1-i13-short PAS and thereby blocking access of the polyadenylation machinery to this location. That study further showed that this inhibitory activity was a function of both overall hnRNP D levels and the methylation status of a specific hnRNP D arginine residue. In EO-

PE placental samples, however, we observed the opposite trend – a small increase in sFlt1-i13 short at the expense of mFlt1 (Fig. 5b). We also observed no differential gene expression for arginine methyltransferase (PRMT1), arginine demethylase (JMJD6), hnRNP D or any other hnRNP protein (data not shown).

Thus sFlt1 upregulation in PE appears to be mechanistically unrelated to the hypoxia-dependent modulation of sFlt1-i13-short production in HMVECs.

Our data clearly show that all FLT1 mRNA isoforms increase in abundance in PE

(Fig. 5a), with no general shift toward the more promoter proximal PAS (Fig. 5b).

When we considered the entire transcriptome, we detected no general 3'-UTR 89

shortening (data not shown) and only eleven statistically significant PAS isoform abundance changes overall (Supplementary Table S11 and Fig. 5c, d). Among these, there was no consistent proximal-to-distal or distal-to-proximal switch pattern. Thus there is no general polyadenylation pattern shift associated with

PE, and alternative polyadenylation is not a major contributor to sFlt1 upregulation. Rather our data indicate that the increased sFlt1 expression in PE is primarily due to increased transcription of the entire FLT1 locus.

What is driving this increased FLT1 transcription? Because of its central role in regulating angiogenesis, FLT1 transcriptional regulation has been intensely investigated in other tissues and several upstream transcription factors including

EPAS1, PAX3, p53 and ETS have been characterized (Barber et al., 2002;

Ciribilli et al., 2010; Das et al., 2005; Jinnin et al., 2008; Koyano-Nakagawa et al.,

2015; Menendez et al., 2007). Notably, the differentially expressed gene sets

(padj≤0.05) in both EO-PE and LO-PE included numerous Pol II transcription factors. Common to both sets were CEBPA, SP3, DLX5, BHLHE40 and

SREBF1, all of which had increased expression, and ZEB2, which had decreased expression. The CEBPA (CCAAT/enhancer binding protein alpha) pathway is particularly interesting, as CEBPA is expressed in the labyrinthine trophoblasts (the same cells that express Flt1) and has been shown to regulate placental development (Begay et al., 2004).

Intriguingly, the only other gene identified by every transcriptome-wide gene expression analysis method to date (microarray, RNA-Seq and PAS-Seq) as being differentially-expressed in PE is NRIP1 (nuclear receptor-interacting protein 1; aka RIP140) (Fig.2h). In all cases, NRIP1 was upregulated. A ubiquitously-expressed nuclear protein, NRIP1 modulates the activities of numerous nuclear receptor transcription factors (Nautiyal et al., 2013). Its best understood role is as a regulator of energy expenditure in adipose and muscle tissues, but it has also been linked to ovarian fertility and maintenance of a pregnancy state (White et al., 2007). Whether NRIP1 contributes to the metabolic syndrome often associated with PE or it directly regulates sFlt1 expression remains to be explored. We note that NRIP1 has been implicated in the pathogenesis of obesity(Nautiyal, 2017; Nautiyal et al., 2013)a major risk factor for PE. It is possible that upregulated NRIP1 may synergize with anti- angiogenic factors to induce the endothelial dysfunction ultimately leading to PE.

Consistent with previous findings (Junus et al., 2012b), our data suggest that EO-

PE and LO-PE are discrete conditions with distinct gene expression signatures.

EO-PE was associated with many more differentially expressed genes than LO-

PE, although the small size of our LO-PE dataset (n=3) may have precluded all but the most consistently differentially expressed genes from reaching statistical significance. An additional caveat is that some gene expression differences could be due to gestational age differences, particularly between the EO-PE 91

(gestational delivery week <34) and CTRL (gestational delivery week >37-40) groups. Nonetheless the HIF1a, HIF2a and VEGFR1 (FLT1) signaling pathways were differentially expressed in both groups, consistent with hypoxia and altered angiogenesis being generally associated with PE (Nevo, 2006; Pratt et al., 2014;

Rajakumar et al., 2001; Soleymanlou et al., 2005). But unique to the EO-PE set were genes involved in Biological/Cellular Adhesion and the Interleukin 6,

Cadherin, and Wnt signaling pathways. This suggests that EO-PE placentae may have defects in epithelial to mesenchymal transition which may lead to defective trophoblast invasion noted in this syndrome (E Davies et al., 2016).

Due to its highly heterogeneous presentation, definitive diagnosis of PE based solely on maternal signs and symptoms can be erroneous (Luef et al., 2016;

Malshe and Sibai, 2017). A variety of underlying conditions can serve as mechanistic drivers, confounding gene expression analysis of diverse patient populations. Differences in sample collection methodologies across institutions can also introduce variability. These factors likely explain the relatively poor overlap among prior transcriptome-wide analyses of placental gene expression in

PE (Fig. 2h). For the current study, we aimed to eliminate as many variables as possible. In addition to using highly selective criteria for PE diagnosis based solely on quantitative measures (Fig. 1c) and excluding known confounding drivers (e.g., multi-parity, preexisting hypertension and gestational diabetes mellitus), all samples were collected by the same personnel in a single hospital, 92

all libraries were prepared by a single researcher, and all sequencing was performed on a single instrument. Therefore our libraries should represent highly coherent sets. Nonetheless, our study does have its own limitations. Because all but one PE subject delivered at ≤36 weeks gestational age, whereas all CTRL patients delivered closer to term (≥37 weeks), PE and CTRL placentas were not gestationally matched. Therefore, additional studies using placentas delivered for other medical reasons (e.g., preterm labor) will be needed to evaluate which gene expression changes, if any, are simply attributable to earlier gestational age in the PE subjects. As stated above, our study was also limited by small sample size particularly for the LO-PE group (n=3). Future studies with adequate sample numbers are need to fully characterize the transcriptional signature specific to

LO-PE.

In addition to assessing protein-coding genes, we interrogated our PAS-Seq data with regard to non-coding and intergenic transcripts. While we found little evidence of differential lncRNA expression in PE, our data do suggest a potential association between EO-PE and endogenous retroviral element activation. Thus it may be of interest to interrogate repeat-associated transcripts in existing PE placenta RNA-Seq data (Sõber et al., 2015) (Kaartokallio et al., 2015) Our PAS-

Seq data also confirm upregulation of mir210 and downregulation of mir214 in

PE (Enquobahrie et al., 2008). Thus, some gene expression differences associated with PE are likely driven by post-transcriptional mechanisms. 93

3. CHAPTER III: PILOT SCREENING of siRNAs AGAINST Homo sapiens and Mus musculus sFlt1 mRNAs

PREFACE:

This work is unpublished and was conducted by me in Dr. Melissa J. Moore’s laboratory. The work presented here is a proof-of-concept study to knockdown sFlt1-i13 in cells using cloning, transfection and RNAi technology.

Additionally, this chapter also describes the siRNA sequences that effectively knock down both mammalian sFlt1 and mice sFlt1. In Chapter III, I lay the foundation for Chapter IV. In Chapter IV, the siRNAs selected from Chapter III were chemically modified for efficient uptake and animal studies in collaboration with Dr. Anastasia Khvorova.

3.1. Introduction

Alternative polyadenylation (APA) of human mRNAs is emerging as a major source of posttranscriptional diversity. APA generates mRNA isoforms that can even encode functionally different proteins. It has been estimated that around

20% of human genes undergo AP events in introns, termed intronic polyadenylation (IPA)(Di Giammartino et al., 2011; Tian et al., 2005). A striking example of this phenomenon is Fms like tyrosine kinase receptor (Flt1) where the switch from a canonical 3’UTR site to an upstream intronic site leads to expression of a soluble form of the protein called soluble Flt1 receptor protein

(sFlt1). Preeclampsia occurs in 5-8% of pregnant women in the United States and 2-16 % pregnant women worldwide, with high morbidity and mortality for both the mother and the fetus. It has been shown that prolonging gestation period during PE leads to higher chance of infant viability. Reduction in levels of sFlt1 protein in preeclamptic patients enables such an increase in gestation length. Recent data has shown that a one-time reduction of sFlt1 by 30-40% is enough to prolong PE pregnancies by 14-21 days(Thadhani, 2010; Thadhani et al., 2011; 2016). They also reported an improvement in proteinuria and blood pressure. This elongation in gestational time is especially crucial for EO-PE where the lungs of the infants are not yet completely developed.

Given all the background, we were interested and motivated to develop siRNA- based therapy for PE. We are convinced that this is an ideal approach, because as opposed to antibodies and small-molecules, siRNAs are relatively cheaper to 95

produce. siRNA’s have a stable shelf life and require no refrigeration if delivered as a lyophilized powder. In the developing world, shortage of electricity and infrastructure leads to issues with refrigeration. This problem can be tackled with siRNA therapy. The adsorption device therapy would also be very expensive and out of reach for majority of the patients in the developing world. sFlt1 is an attractive and validated target that has been shown to improve outcome for the neonate(Thadhani et al., 2011). The half-life of circulating sFlt1 in the mother’s blood-stream is ~ 6 hours(Powers et al., 2005). Additionally, it has been reported that on knockdown of Flt1 and sFlt1 in the placental cells

(trophoblasts) specifically, the mice develop normally(Hirashima et al., 2003). All this evidence suggests that siRNA based approach will reduce sFlt1 and prolong the preeclamptic pregnancy. The placenta is the main source of sFlt1 (Jebbink et al., 2011) (Figure 2.1 b)and we hypothesize that reducing sFlt1 is less likely to affect other organs of the mother and fetus. Also, sFlt1 is not required in the mice model and therefore reducing sFlt1 is not going to affect the fetal development.

Currently there are several therapeutic siRNAs in clinic for various diseases(Khvorova, 2017; Tiemann and Rossi, 2009) siRNAs produce sequence-specific silencing and have several advantages over other rRNA based approaches like shRNA or miRNA. Their in vivo half-lives are enough to produce a silencing effect 2-3 weeks post-injection. We therefore hypothesize that modulation of sFlt1 pre-mRNA using siRNAs could be a potential therapeutic 96

approach to alleviate PE.

My goal was to identify siRNAs that would reduce the sFlt1-i13 isoform. Such a manipulation should reduce levels of sFlt1 mRNAs and in turn the soluble sFlt1 protein. Given the fact that the 3’UTR of sFlt1-i13 is distinct from the mFlt1, we designed siRNAs that targeted the 3’UTR of sFlt1-i13 isoform. From Chapter II, we have seen that the two isoforms sFlt1-i13 S and sFlt1-i13 L share a common region. We hypothesized that we design siRNAs targeting that region of overlap; we would be able to reduce both the isoforms with a common siRNA. Due to use of a polyadenylation site internal to intron 13, the sFlt1-i13 mRNA includes intronic sequences not present in full-length Flt1 mRNA.

As reported previously (Thadhani et al., 2010) the goal here was to knockdown sFlt1-i13 levels by 30-40% and not complete 100% knockdown. We wanted to aim for transient silencing (6-8 hours) and aiming for long-term (weeks) silencing.

By using synthesized siRNAs we were at advantage as we skipped the intermediate siRNA-processing step by Dicer. Designing the siRNAs was done by first conducting a BLAT search to ensure that there was no match between the sequence of interest and other genes. Then the best siRNAs were selected based on various criteria (see Methods).

Additionally, we also wanted to design siRNAs against the sFlt1-e15a isoform.

However, at the time of this experiment, we wanted to test the siRNAs in mice.

Mice do not express the sFlt1-e15a isoform, so we initiated the pilot study by 97

targeting sFlt1-i13 S and sFlt1-i13 L isoforms.

To this end, a luciferase reporter system was used to clone the 3’-ends of sFlt1- i13. Co-transfection assays were used where the plasmid containing the cloned sFlt1-i13 3’-UTR and the siRNA were transfected in HEK-393 cells using lipofectamine.

Figure 3.1 Position of tested siRNAs on sFlt1-i13 S Red lines represent the target sequence. The sequence in black represents the sFlt1-i13 3’UTR. siRNAs targeting the sequence are labeled.

3.2. Results 3.2.1 Pilot screening of anti-human-sFlt1 siRNAs in 393-D1 cells

In order to find an anti-sFlt1-i13 siRNA and determine the efficiency of the siRNAs, we first cloned the sFlt1-3’UTR (Figure 3.1) in a psiCheckTM Vector. A total of ten siRNAs were designed to target and tiled across the sFlt1-i13-3’-UTR

(Figure 3.1). In a co-transfection experiment, each anti-sFlt1 siRNA and the plasmid carrying 3’-UTR was transfected in HEK-393 cells. These siRNAs were transfected “naked” i.e. the siRNAs were not chemically modified and did not have any conjugations added to their 5’ or 3’ends. The advantage of using the psiCHECKTM reporter vectors is the presence of both firefly and renilla luciferase.

The sFlt1-i13 3’UTR was cloned downstream of the renilla luciferase gene and firefly is was used as a normalization control. The luciferase activity was measured by a luminometer and IC50 values (Table 3.3) were determined from dose response curves (Figure 3.2).

3.2.2 Two highly efficient anti-sFlt1 siRNAs identified

Given their potency and low IC50 values, siRNAs H29 and H33 were chosen for further study in fresh trophoblast villous cultures derived from preeclamptic human placentas (Chapter IV). These siRNAs gave efficient knockdown of sFlt1.

They were further used in mice model. For delivery in the animal, the siRNAs have to be chemically modified for uptake and delivery. The siRNAs used in this pilot study were without any modifications or conjugations and therefore not 100

suitable for studies in mice.

3.2.3 Pilot testing of anti-mice-sFlt1 siRNAs in 393-D1 cells

TM Using the psiCHECK Vector system, I cloned the mice sFlt1-i13 3’-UTR.

Using the same methods as for human sFlt1 siRNAs, I also identified siRNAs capable of decreasing expression of mouse sFlt1 (Table 3.2, sense strand sequences: M504, 5'- AGUAACAGUUGUCUCUUAUUU-3'; and M506, 5'-

ACUUUCAGGCCCAGAGGAAUU-3'). These siRNAs were used in Chapter IV to assess in vivo efficacy of this silencing approach.

101

Number siRNA Sense Strand Sequence 1 H27 5' -GGAGAUGAUAGCAGUAAUAUU -3' 2 H29 5'- CCACACAAAGUAAUGUAAAUU -3' 3 H31 5'- CGGAGAUGAUAGCAGUAAUUU -3' 4 H33 5'- UCUCGGAUCUCCAAAUUUAUU -3' 5 H35 5'- AGUAACAGUUGUCUCAUAUUU -3' 6 H37 5'- CCAAAAUGAGUUCGGAGAUUU -3' 7 H39 5'- GUUUGUUGCUCUUGCUGUUUU -3' 8 H41 5'- CCCAAAAUGAGUUCGGAGAUU -3' 9 H43 5'- UGUCACUGUUGCUAACUUUUU -3' 10 H45 5'- GAGAUGAUAGCAGUAAUAAUU -3'

Table 3.1 Sequence information of tested human siRNAs

Number siRNA Sense Strand Sequence 1 M504 5'- AGUAACAGUUGUCUCUUAUUU -3' 2 M506 5'- ACUUUCAGGCCCAGAGGAAUU -3'

3 M508 5'- UGGACAUGAUAACGUAAUAUU -3' 4 M510 5'- CCGGUGUGUGUUUGGAUUUUU -3' 5 M512 5'- UGCCAAAGACGGACAUGAAUU -3' 6 M514 5'- GUAAAUAGUUCCUGGAUAAUU -3' 7 M516 5'- UGUGUUUCCUGCUGUGUUAUU -3' 8 M518 5'- CACCAAGCUUUGAUGGCAUUU -3' 9 M520 5'- AGUGGAGAUUUAAUGCAAUUU -3' 10 M522 5'- ACCCAUGAGAGAAAGAAAAUU -3'

Table 3.2 Sequence information of mice siRNAs

102

103

Figure 3.2 siRNAs against human sFlt1-i13 isoforms Different concentrations of the human anti-sFlt1-i13 siRNAs were tested and renilla and firefly activity was measured. X-axis is the concentration of siRNA (in nM). Y-axis is the normalized renilla to internal firefly luciferase ratio.

104

Figure 3.3 siRNAs against mouse sFlt1-i13 isoform Different concentrations of the mice anti-sFlt1-i13 siRNAs were tested and renilla and firefly activity was measured. X-axis is the concentration of siRNA (in nM). Y- axis is the normalized renilla to internal firefly luciferase ratio.

105

siRNA IC50 [nM] Efficiency

H27 0.46 ± 0.05 35%

H29 1.2 ± 0.1 88%

H31 1.5 ± 1.1 51%

H33 0.8 ± 0.3 95%

H35 2.8 ± 0.8 26%

H37 Not detected NA

H39 1.5 ± 0.1 66%

H41 2±8 62%

H43 0.09 ± 0.1 13%

H45 0.6 ± 0.4 30%

M504 0.10 ± 0.03 89%

M506 0.5 ± 0.2 76%

Table 3.3 siRNA IC50s and efficiencies H denotes human and M denotes mice

106

3.3. Methods 3.3.1. Plasmids psiCHECKTM reporter vectors (Promega, USA) were used to optimize the RNAi experiment. The vector is especially designed to quantify the effect of RNAi. The vector consists of 2 luciferases- Firefly and Renilla. The firefly is used as a normalization control and the sequence of interest is cloned under renilla luciferase. The insertion sites for sFlt1-i13 3’UTR were constructed by digesting the vector with XhoI and NotI and inserting the 3′ UTR target site-containing

XhoI–NotI sites downstream of renilla reporter gene. The new plasmid was called psiCHECK-sFlt-i13-3’UTR. Dual Reporter Luciferase assays were conducted using Dual-Luciferase Reporter Assay reagents (Promega) in a Veritas

Microplate Luminometer according to the manufacturers’ directions.

3.3.2. Cell Culture and Transfection

NIH-3T3 cell cultures were maintained at 37ºC and 5% CO2 in DMEM supplemented with 10% FBS and 50 U/ml ampicillin. Briefly, to test the efficacy of siRNAs targeting this region, we inserted the appropriate sequences from both human (Figure 3.1) and mouse (not shown) downstream of Renilla luciferase in

TM the psiCHECK Vector system (Promega). This construct was transiently transfected into NIH3T3 cells along with 0-20 nM of each test siRNA, and then

Renilla and Firefly luciferase activities measured 48 hours later. Cells were seeded at a density of 1.1 × 106 cells per well in 24 well plates in DMEM 107

(Invitrogen) containing 10% FBS (Invitrogen) twenty-four hours before the day of transfection. On the day of transfection experiment, 25ng/ul of psiCHECK-sFlt- i13-3’UTR in DMEM and lipofectamine and siRNAs ranging from 0 to 20nM concentrations were incubated for 20 minutes. After 20-minute incubation period the mixture was added to each well individually. The cells were then incubated at

37°C in a CO2 incubator for 4 hours. Luciferase was measured 48 hours post- transfection. A control siRNA (scrambles sFlt1-3’UTR sequence) was used to equalize the total amount of siRNA in each transfection.

3.3.3. Luciferase Assays

The screening of siRNAs was performed using the luciferase reporter assays.

Cells were washed once in 500 µl PBS and lysed in 100 ml of Passive Lysis

Buffer (Promega) at room temperature for 20 min in 24 well plates. For each well

10 µl 55 lysate was read in triplicate using dual luciferase reagents (Promega) in the luminometer. The renilla and control firefly luciferase activities were measured 40 hours post-transfection. The normalized reporter activity was measured by dividing renilla luciferase by firefly luciferase in the same sample.

This allowed to also normalize for transfection efficiencies.

3.3.4. Data Analysis

The individual biological replicates for normalized Renilla luciferase activity versus siRNA concentration was fit using Igor Pro 6.10 software. The Hill equation was used to determine IC50. 108

3.4. Discussion

The role of sFlt1 as a causative factor of PE symptoms is well established. sFlt1 is a validated target and studies report that a 30-40% reduction in sFlt1 levels ameliorates high blood pressure and proteinuria in PE patients. The approach of apheresis is a proof-of-concept study and it uses expensive machine. Our goal here was to develop a PE therapeutic, which would be affordable not only in developed countries but also in developing countries siRNAs have a lot of advantages like relative cheap cost of manufacturing, stable shelf-life, no need for refrigeration and ease of delivery in the body. Here, we aimed to identify efficient anti-sFlt1 siRNAs.

I performed in vitro screening of synthetic, “naked” siRNAs using a luciferase- reporter assay in NIH3T3 cells. Two sequences against human sFlt1-i13 were identified which were expanded further in Chapter IV. The work here demonstrates that knockdown of sFlt1-i13- 3’UTR is an effective way to knockdown the mRNA laying the foundation for further RNAi therapeutics to treat

PE. 109

4. Chapter IV: DEVELOPING anti-sFlt1 siRNAs to alleviate preeclampsia

Chapter IV was taken verbatim from the following manuscript, which is preparation during the time of this thesis submission.

Anton A. Turanov, Matthew R. Hassler, Ami Ashar-Patel, Angela Makris, Agnes

Lo, Julia F. Alterman, Andrew H. Coles, Reka A. Haraszti, Loic Roux, Bruno

MDC Godinho, Dimas Echeverria, Annemarie Hennessy, S. Ananth Karumanchi,

Melissa J. Moore, and Anastasia Khvorova

Developing anti-sFlt1 siRNAs to alleviate preeclampsia

Author contributions statement:

A.A.T., M.R.H. and J.F.A designed and M.R.H and D.E synthesized all compounds. A.A.T., A.H.C., B.DMC.G. and A.L. conducted all in vivo experiments and imaging. A.A.T., A.AP. and J.F.A conducted all in vitro screening and data analysis. R.A.H. developed and conducted PNA analysis.

A.K., M.J.M, A.A.T., S.A.K. came up with the concept and wrote the manuscript.

110

4.1. Introduction The most notable characteristics of PE are hypertension, edema and excess protein in the urine (proteinuria) after the 20th week of pregnancy. Consequences for the fetus can be grave, ranging from small-for-gestational-age infancy to hypoxia-induced neurologic injury (e.g., cerebral palsy) to death. Maternal complications include renal failure, HELLP syndrome (hemolysis, elevated liver enzymes, and low platelets), seizures, stroke, and death. PE and related hypertensive disorders are causing close to 76,000 maternal and 500,000 infant deaths globally each year. In the US, PE is responsible for 100,000 premature births and 10,500 infant deaths yearly, at a yearly cost of roughly 7 billion dollars

(3 billion for maternal disabilities and 4 billion related to infant morbidity) to the

US health care system (www.preeclampsia.org). Thus PE represents a highly significant unmet public health need.

Overwhelming evidence from epidemiological and experimental studies indicates that PE is caused by elevated levels of "soluble decoy" proteins (soluble FLT1s; sFlt1s) in the mother's blood stream expressed from the Flt1 gene (aka,

VEGFR1)(Heydarian et al., 2009; Levine et al., 2004; Maynard et al., 2003;

Young et al., 2010) (Aubuchon et al., 2011). FLT1 is a receptor tyrosine kinase

(RTK) predominantly expressed in the placenta. A general mechanism for RTK modulation is production of truncated, secreted forms of the receptor that act as dominant negative regulators of the overall signaling pathway. Ligand 111

sequestration by such soluble decoys inhibits intracellular signaling by the full- length receptor, thereby desensitizing the system to ligand concentration(Vorlová et al., 2011). In the case of FLT1, the soluble decoys are expressed from truncated mRNAs generated by polyadenylation sites within two introns (sFlt1-i13 and sFlt1-e15a) upstream of the exons encoding the fl-FLT1 transmembrane

(TM) and kinase domains.

FLT1 is predominantly expressed in the placenta, with human placental FLT1 mRNA levels being 10-100 times higher in other adult tissues(Cerdeira and

Karumanchi, 2012). The full-length isoform predominates in all tissues in non- pregnant adult humans(Cerdeira and Karumanchi, 2012), whereas placental expression is dominated by three truncated isoforms, sFlt1-i13 short, sFlt1-i13 long and sFlt1-e15a, all of which encode sFlt1 proteins (Figure 4.1.A). This same pattern of high FLT1 in placenta and low expression in other non-pregnant adult tissues is observed in rodents. However, because rodents lack the intron

14 polyadenylation site, they only express a single soluble decoy form: sFlt1- i13(Sela et al., 2008). In PE, both full-length and truncated FLT1 mRNAs accumulate to higher levels in the placenta than in normal pregnancies, with the truncated isoforms being even more pronounced (Figure 4.1.B). These changes at the mRNA level likely explain the significant rise in sFlt1 proteins in the maternal bloodstream during PE.

112

sFlt1 level have been recently identified and as a biomarker for PE, helping to discriminate between other cause of high blood pressure and proteinuria.

Thus selective modulation of sFlt1 is believed to be a viable clinical approach. As difference between sFlt1 and full length FLT1 on a protein level is minimal, attempts to develop antibodies and small molecules selective for the sFlt1 were not successful.

The alternative strategy of apheresis of the PE patient's blood with dextran column which eliminates sFlt1 (as well as other circulating factors) was shown to extend pregnancies and control blood pressure and proteinuria, clinically validating sFlt1 as a therapeutic target(Thadhani et al., 2011; 2016). While difference on the protein level is minimal, sFlt1-i13 and sFlt1-e15a isoforms have relatively large RNA regions, which are unique to sFlt1 producing mRNAs, and thus can be targeted for selective modulation of sFlt1 expression. siRNA are new class of therapeutic modalities, which directly target mRNA in a sequence selective manner and are ideal for differential modulation of FLT1.

4.2. Results

4.2.1. sFlt1-i13 and sFlt1-e15a isoforms of FLT1 are overexpressed in human placentas

Overexpression of sFlt1 is one of the key molecular mechanisms behind PE.

There are several alternative polyadenylation variants exists which produce the same sFlt1 protein(Thomas et al., 2010; 2009a). The major isoform are based on 113

use of two polyadenylation signals after exon 13 and 15 generating sFlt1-i13 short and sFlt1-i13 long isoform and sFlt1-e15a truncated mRNAs. All of these truncated variants are responsible for expression of soluble FLT1, which lack the transmembrane domain of full length FLT1 (fl-Flt1). To evaluate which of these isoforms are clinically relevant we compared sFlt1-i13 and sFlt1-e15a expression in placentas derived from CTRL and PE pregnancies. By performing polyadenylation site sequencing (PAS-Seq) on total RNA from multiple normal and PE placentas, we have shown that PE placentas overexpress sFlt1-i13 and sFlt1-e15a mRNAs with sFlt1-i13 being responsible for ~45% of reads and sFlt1- e15a ~55% (Chapter II, Figure 4.1.A-B). The intrinsic variability in isoform ratios in different samples suggests that targeting both isoforms might be the best option to cover the majority of PE patients. The common region between sFlt1- i13 short and sFlt1-i13 long and sFlt1-e15a are unique and present only in truncated sFTL1 mRNAs and their target allows selective modulation of sFlt1 without effect on full length FLT1 expression, which is critical for the safety of future therapy.

4.2.2. Screen to identify hsiRNAs selectively silencing sFlt1-i13 and i15 sFlt1 isoforms

Using standard bioinformatics criteria we have designed and synthesize all possible siRNAs targeting sFlt1-i13 S and sFlt1-e15a isoforms( Chapter III and

(Figure 4.1 C-E) show the screen of compounds targeting sFlt1-i13 and sFlt1- e15a unique regions with numbers denoting the siRNA targeting positions. The 114

exact chemical configuration and sequences of all oligonucleotides used in the study are shown in Supplemental Table1. For top five compounds from each screen, the concentration curves were generated using luciferase reporter constructs in HeLa cells or directly in primary cytotrophoblast (CTB) and WM-

115 melanoma cells. Based on this data the compounds 2283 and 2318 targeting sFlt1-i13 and 2519 and 2585 targeting i15a were selected for detailed evaluation (Figure 4.1 E-F). Figure 4.1.G-J shows concentration curve dose response using reporter assay in HeLa cells and after the treatment of primary

CTB on mRNA level. sFlt1 mRNA silencing was also responsible for reducing soluble Flt1 protein in CTB conditioned media as well (Figure 4.1 K). Although initial screen was performed on mRNA sequences for human sFlt1-i13 isoforms, target sequence for compound 2283 shares hundred percent identity between mouse, baboon and human. Having such compound as a lead candidate facilitates in-vivo validation in pre-clinical rodent models. Thus pair of compounds

2283 and 2519 was further selected for in-vivo evaluation with 2318/2585 serving as a backup.

115

Figure 4.1 Development of hydrophobically modified chemically stabilized siRNA compounds against sFlt1 116

a. Schematic representation of exon-intron structure of sFlt1 i13 and sFlt1-e15a isoforms. b. Identification of alternative FLT1 mRNA polyadenylation sites in human placentas. Total RNA was isolated from 5 normal and 6 preeclamptic human placentas and analyzed by Polyadenylation Site sequencing (PAS-Seq). c-d.

Systematic screen for identification of hsiRNA targeting sFlt1-i13 and sFlt1-e15a in luciferase reporter assay in HeLa cells (Dual-Glo, Promega). HeLa cells were transfected with sFlt1 reporter plasmid (psiCheck-2, Promega) and treated with hsiRNAs (2.5 mmol/l, passive uptake) and luciferase expression was measured 72- hours after treatment with siRNA. (n=3, mean + SD) e. sFlt1-i13 and sFlt1-e15a mRNA sequences. Locations of leading siRNAs hits are indicated (underlined). Stop codons are shown in red. f. hsiRNAs targeting sFlt1-i13 and sFlt1-e15a. g-j. hsiRNAsFlt1 validation and activity in vitro. HeLa, transfected with sFlt1 reporter plasmid or CTB cells were incubated with siRNA at concentrations shown for 72h. mRNA levels were measured using Dual-Glo System (Promega) normalized to background gene (hluc+) (HeLa, g, h) or QuantiGene® (Affymetrix) normalized to housekeeping gene (YWHAZ) (CTB, i, j) (n=3, mean + SD). k. sFlt1 protein knockdown in human trophoblasts. sFlt1-i13-2283 siRNAsFlt1 and NTC siRNA were added to CTBs at concentration shown. Level of sFlt1 protein was measured by

ELISA (#MVR100, R&D systems) in conditioned culture medium after 72 h. CTB – cytotrophoblast cells (from placenta); UNT – untreated cells; NTC – non-targeting control with matching chemistry.

117

4.2.3. Efficient delivery of fully chemically stabilized hydrophobically modified siRNAs to placental labyrinth

To achieve placenta delivery, we have used an asymmetric siRNAs, with a short duplex region (15 base-pairs) with single-stranded, fully phosphorothioated tail, where all bases are fully modified using alternating 2'-fluoro and 2'-O-methyl modification pattern, and the 3' end of the passenger strand is conjugated to a bioactive hydrophobic conjugate (TEG-Cholesterol). The cholesterol enables quick membrane association, while the single-stranded phosphorothioated tail is essential for cellular internalization by a mechanism similar to that used by conventional antisense oligonucleotides ((Alterman et al., 2015),(Hassler and Khvorova) (Navaroli et al.)) (Figure 4.2.A). We and others have screened a wide range of chemical variants and identified that an alternating 2’- fluoro/2'-O-methyl pattern optimally configures the guide strand to adopt a geometry that closely mimics that of an individual strand in a A-form RNA duplex, is recognized by the RISC . Modified siRNAs are as effective as "naked" (unmodified) siRNA in RISC entry and represent the complete chemical modification pattern with no negative impact on RISC function. Fully chemical stabilization is absolutely essential for non-formulated siRNA delivery upon systemic administration.

118

Figure 4.2 Modified siRNA enables selective delivery to placental labyrinth

119

a. Structure of chemically modified siRNA. b. siRNAsFlt1 distribution in mouse placenta.

C57BL/6J mice (E15) were injected with Cy3-labeled 2283 siRNAsFlt1 (red) (10 mg/kg; IV tail vein or SC interscapular). Twenty-four hours later mice were sacrificed, tissues fixed and 4 mm sections were used for microscopy. Nuclei were stained with DAPI (blue).

Matched slides were stained with hematoxylin and eosin (HE). Images were captured using Leica DM5500B fluorescent microscope with 10x and 63x objectives. All images were acquired at identical settings. Images were processed with LAS AF software package (Leica). Scale bars - 1 mm (upper panels) and 25 mm (lower panels). c. siRNAsFlt1 distribution in mouse fetus. C57BL/6J mice (E15) were injected with Cy3- labeled 2283 siRNAsFlt1 (red) (10 mg/kg; IV tail vein). Twenty-four hours later mice were sacrificed, tissues fixed and 4 mm sections were used for microscopy. Nuclei were stained with DAPI (blue). Representative image of two experiments. d-e. Amount of siRNAsFlt1 accumulated in tissue measured by PNA hybridization assay in mice from b.

(n=3, mean + SD). TGC –Trophoblast Giant Cells; SpT –SpongioTrophoblast; L –

Labyrinth; Jz – Junctional zone; D – Decidua.

120

4.2.4. Systemic administration of siRNA in vivo resulted in efficient delivery to placental labyrinth

To evaluate siRNAs potential for placenta targeting, wild-type (WT) pregnant mice (gestation day 15) were injected with Cy3-labeled siRNAsFlt1 2283 and distribution examined by two independent assays. Fluorescence microscopy revealed that most of the siRNAsFlt1 accumulated to three tissues: liver endothelium, kidney endothelium and placental labyrinth (Figure 4.2.B). This distribution profile is most likely defined by a combination of blood flow/filtration rate. In placenta we observe efficient delivery to CTB cells in labyrinth with effectively no uptake in a decidua. Placenta labyrinth mostly consists of CTB trophoblasts, primary cell type responsible for sFlt1 overexpression. Efficient siRNAsFlt1 delivery to placental labyrinth consistent with its primary function as a barrier (blood-placental-barrier) and active nutrient transporting organ between mothers and fetus blood (Figure 4.2.B).

While we have observed efficient delivery to placenta, the compounds do not cross into the fetus (Figure 4.2.C). To quantitatively estimate oligonucleotides tissues accumulation we evaluated guides strand tissue concentration using recently developed PNA-hybridization assay (ref and see methods). While majority of cholesterol-conjugated siRNA goes to liver, we observe around 7% injected dose accumulated in placentas. With same assay, the siRNAs concentration fetus lever fell below the detection level. We observe 121

approximately 10-20 ng/mg placenta accumulation at 24 hours, which exceed concentrations theoretically necessary to induce silencing (Figure 4.2.D-E).

4.2.5. Efficient in vivo modulation of sFlt1 in placenta have no impact on mouse progeny survival and ability to strive

Mice express only one truncated form of sFlt1- i13. The siRNAsFlt1 2283 identified in initial screen hit the site fully complementary in mice, human and NHP allowing evaluation of compound efficacy in multiple animal models. The primary source of circulating sFlt1 is placenta with level of placental expression exceeding maternal tissues by orders of magnitude. To estimate ability of systemically delivered siRNAsFlt1 2283 to silence sFlt1 in placenta WT pregnant mice were injected twice with 20 mg/kg at gestational days 14 and15 with siRNAs (Figure 4.3.A). Animals were sacrificed and levels of sFlt1 silencing was evaluated in placenta (Figure 4.3.B), maternal tissues and fetal livers).

We observed statistically significant silencing in mouse placenta. The effect was specific to siRNAsFlt1 2283 where as non targeting control (NTC), siRNA of the same chemical composition, was inactive. The productive silencing in placenta was only observed with cholesterol conjugated siRNAs, as non- conjugated compounds (NoC- siRNAsFlt1 2283) shown no significant efficacy. In addition there was no silencing in the fetal livers consistent with lack of delivery. Levels of fetal sFlt1 expression are extremely low making reliable quantification problematic (data not present). In parallel study (when mice are allowed to deliver) we looked at circulating sFlt1 protein using ELISA and show statistically 122

significant silencing of sFlt1 at day 17. Again, the effect was specific for sFlt1 targeting compound as NTC treated animals had the same levels of circulated sFlt1 as PBS treated (Figure 4.3.D).

We decided to further evaluate kinetic of sFlt1 silencing and separate groups of mice was injected with PBS or siRNAsFlt1 2283 and blood samples collected at day 15, 17, 19. We observe silencing at all time points with day 19 reaching statistical significance (Figure 4.3.E).

The animals from this study were allowed to deliver. The sFlt1 silencing has no impact on newborn pubs number, pubs weight at birth and pups ability to strive

(Figure 4.3. F-H). The levels of liver transaminases ALT and AST were all normal

(Figure 4.3.K) indicating no liver toxicity upon treatment with siRNAsFlt1 2283.

These data demonstrate that systemic administration anti-sFlt1 siRNA can efficiently modulate sFlt1 placental mRNAs, circulating sFlt1 protein and has no detectable impact on progeny.

123

Figure 4.3 Efficient silencing of sFlt1 mRNA and protein by sFlt1-i13_2283 siRNA in pregnant mice

124

a. Experimental design for in-vivo testing of siRNA. Pregnant CD1 mice were injected at E14 and E15 with 2X20 mg/kg of siRNAsFlt1 (IV, tail vein) and delivered at E20 or sacrificed at E19. b. sFlt1_i13 mRNA levels were measured using QuantiGene® assay (Affymetrix), normalized to housekeeping gene (Flt1), and presented as % of PBS control (n=6, mean +SD). c. Amount of hsiRNA accumulated in tissues five days after first injection measured by PNA-hybridization assay in mice form b. d. sFlt1 protein levels in mice serum measured by ELISA (RD Biosystems) after 2X20 mg/kg dose (n=12 for siRNA and PBS, n=6 for NTC, mean +SD). e. sFlt1 protein levels measured by

ELISA (RD Biosystems) for 2X20 mg/kg dose (n=9, mean +SD). f-h. Newborn pups number and weights from mice injected with siRNAsFlt1. k. ALT/AST enzymes activities in mice injected with 20X2 mg/kg of siRNAsFlt1 at E17 measured by Activity Assay Kits

(Sigma).

125

4.2.6. Combinatorial silencing of sFlt1-i13 and i15a isoforms in vitro and in vivo

Both sFlt1-i13 and sFlt1-e15a sFlt1 isoforms are expressed in placentas in PE

(Figure 4.1.A-B). There is no homology between unique regions present in sFlt1- i13 and sFlt1-e15a. To silencing both isoform we decided to use the mixture of two compounds.

We treated WM-115 cells, which are expressing both isoforms with 2283, 2519 and 2283/2519 siRNAsFlt1 mixture. siRNAsFlt1 2283 and the mixture of 2283/2519 efficiently silencing sFlt1-i13 (Figure 4.4.A). While 2519 siRNAsFlt1 and mixture of

2283/2519 efficiently silencing sFlt1-e15a (Figure 4.4.B). None of the these compounds along or in mixture effect expression of full length FLT1 (Fig. 4.4.C).

For both isoforms the potency of the 2283/2519 compounds mixture was higher than individual compounds with effect being more pronounced for sFlt1-i13 isoform. This experiment has been repeated several times with similar results and is not an experimental artifact. We are studying this phenomena, which is outside the scope of this paper.

In summary, siRNAsFlt1 mixture of 2283/2519 support efficient modulation of sFlt1-i13 and sFlt1-e15a sFlt1 isoforms without impact on full length of FLT1 and was selected for further in-vivo analysis.

Consistent with data presented in Figure 3 injection of 2283/2519 equimolar mixture in mice resulted in efficient modulation of sFlt1 mRNA in placenta and other parental tissues (Figure 4.4. D). The PK/PD properties of therapeutic oligonucleotides are mainly depend on the chemical configuration and expected 126

to be similar for different sequences. Indeed when we evaluated placenta accumulation of 2283 and 2519 and it was nearly identical with no detectable transfer to the fetus (Figure 4.4.E). Subsequent decrease of sFlt1protein level after siRNAsFlt1 administration had no impact on progeny numbers, weight, ability to strive (data not present) and no impact on the liver enzymes (Figure

4.4.F-G).

127

Figure 4.4 Efficient silencing of sFlt1 by siRNA mixture (sFlt1- i13_2283/sFlt1-e15a_2519) in vitro and in vivo a-c. siRNAsFlt1 (i13_2283/e15a_2519) equimolar mixture activity in vitro. WM-115 cells were incubated with hsiRNA mixture at concentrations shown for 72h. mRNA levels were measured using QuantiGene® (Affymetrix) normalized to housekeeping gene

(FLT1) (n=3, mean +SD). d. CD1 mice were injected at E14 with 20 mg/kg of siRNAsFlt1 equimolar mixture (IV, tail vein) and sacrificed at E19. mRNA levels were measured

QuantiGene® (Affymetrix) normalized to housekeeping gene (Flt1) and presented as % of PBS control (n=8, mean +SD). e. Amount of hsiRNA accumulated in tissues five days after injection measured by PNA-hybridization assay in mice from d. f. sFlt1 protein levels were measured by ELISA (RD Biosystems) (n=8, mean +SD). g. ALT/AST enzymes activities in mice injected with siRNAsFlt1 equimolar mixture at E17 measured by Activity Assay Kits (Sigma).

128

4.3. Methods 4.3.1 siRNA Design

We designed and synthesized a panel of 47 siRNA compounds targeting

the human sFlt1 i13 and sFlt1-e15a. These sequences span the unique

genes sequences and were selected to comply with standard siRNA

design parameters(Cerdeira and Karumanchi, 2012) including

assessment of GC content, specificity and low seed compliment

frequency(Fedorov et al., 2006), elimination of sequences containing

miRNA seeds, and examination of thermodynamic bias.

4.3.2 Oligonucleotide synthesis, deprotection and purification

Oligonucleotides were synthesized using standard phosphoramidite, solid-

phase synthesis conditions on a 0.2-1-µmole scale using a Mermaid 192

and Expedite DNA/RNA synthesizer. Oligonucleotides with unmodified 3′

ends were synthesized on controlled pore glass (CPG) functionalized with

long-chain alkyl amine and a Unylinker® terminus. Oligonucleotides with

3′-Cholesterol modifications were synthesized on modified solid support.

Phosphoramidites solutions were prepared at 0.15 M in acetonitrile for 2′–

TBDMS, 2′–O–Me, and Cy3 modifications or 0.13 M for 2′–fluoro.

Phosphoramidites were activated in 0.25 M 4,5-dicyanoimidazole in

acetonitrile. Detritylation was performed in 3% dichloroacetic acid in

dichloromethane for 80 seconds. Capping was performed in 16% N-

methylimidazole in THF and acetic anhydride:pyridine:THF, (1:2:2, v/v/v) 129

for 15 seconds. Oxidation was performed using 0.1 M iodine in pyridine:water:THF (1:2:10, v/v/v).

The CPG is removed from the solid phase column and placed in a polypropylene screw cap vial. Dimethylsulfoxide (DMSO) (100 µL) and

40% methylamine (250 µL) are added directly to the CPG and shaken gently at 65°C for exactly 16 min. The vial is cooled on dry ice before the cap is removed. The supernatant is transferred to another polypropylene screw cap vial, and the CPG is rinsed with two 150 µL portions of DMSO, which are combined with original supernatant. Oligonucleotides without

2′–TBDMS protecting groups were lyophilized. Oligonucleotides with 2′–

TBDMS protecting groups were desilylated by adding 375 µL triethylamine trihydrofluoride (~1.5 volumes relative to 40% methylamine) and incubating for exactly 16 min at 65°C with gentle shaking. Sample were quenched by transferring to a 15 ml conical tube containing 2 mL of 2 M triethylammonium acetate buffer (pH 7.0). The sample was stored at –

80°C until HPLC purification.

Oligonucleotides were purified by reverse-phase HPLC on a Hamilton

PRP-C18 column using an Agilent Prostar 325 HPLC. Buffer A 0.05M tetraethylammonium acetate with 5% acetonitrile, Buffer B 100% acetonitrile. Purified oligonucleotides were lyophilized to dryness, 130

reconstituted in water and passed over a Hi-Trap cation exchange column to exchange the tetraethylammonium counter-ion with sodium.

4.3.3 Cell Culture

HeLa and WM-115 cells were purchased from ATCC (ATCC, Manassas,

VA; #CCL-2, #CLR-165). HeLa cells were maintained in Dulbecco's

Modified Eagle's Medium (Cellgro, Corning, NY; #10-013CV) supplemented with 10% fetal bovine serum (FBS; Gibco, Carlsbad, CA;

#26140) and 100 U/ml penicillin/streptomycin (Invitrogen, Carlsbad, CA;

#15140) and grown at 37°C and 5% CO2. Cells were split every 2 to 5 days, and discarded after fifteen passages. WM-115 cells were maintained in Eagle's Minimum Essential Medium (Gibco, Carlsbad, CA;

#11095) supplemented with 2mM Glutamine, 1% Non Essential Amino

Acids, 1% Sodium Pyruvate, 10% % fetal bovine serum (all form Gibco) and 100 U/mL penicillin/streptomycin (Invitrogen) and grown at 37°C and

5% CO2. Cells were split every 2 to 4 days. Cytotrophoblasts isolated from human placenta were provided by S. A. Karumanchi and were cultivated as described previously(Anderson et al., 2008; Rajakumar et al., 2005).

Isolated CTB cells were maintained in M199 medium (Gibco, Carlsbad,

CA; #11043023).

131

4.3.4 Direct Delivery (passive uptake) of oligonucleotides

For initial siRNA library screening cDNA sequences corresponding to unique regions of human sFlt1 i13 and sFlt1-e15a were cloned into psiCheck-2 vector (Promega, Madison, WI; #C8021) and DualGlo luciferase assay system was used (Promega, Madison, WI; #E2920) according to manufacturer's manual.

Briefly, HeLa cells on 10cm dish were transfected with psiCheck-2-i13 or psiCheck-2-e15a plasmids using Lipofectamine 2000 (Invitrogen,

Carlsbad, CA; #11668019) according to manufacturer's manual. 24 hours later cells were plated in DMEM containing 6% FBS at 10,000 cells per well in 96-well cell culture plates. siRNA was diluted to twice the final concentration in OptiMEM (Carlsbad, CA; #31985-088), and 50 µL diluted siRNA was added to 50 µL of cells, resulting in 3% FBS final. Cells were incubated for 72 hours at 37°C and 5% CO2. The primary screen for active siRNAs was performed at 3-1.5 µM compounds, which also served as the maximal dose for in vitro dose response assays. To test siRNAs in CTB and WM-115 cells, cells were plated at 20,000-30,000 cells per well in 96- well plates in M199 or EMEM media containing 6% FBS and treated with siRNAs for 72 hours. 132

4.3.5 mRNA quantification in cells and tissue punches mRNA was quantified from both cells and tissue punches using the

QuantiGene® 2.0 DNA Assay (Affymetrix, # QS0011) as described previously30, 31. Cells were lysed in 250 µl diluted lysis mixture composed of 1 part lysis mixture (Affymetrix, #13228), 2 parts H2O, and 0.167 µg/µl proteinase K (Affymetrix,# QS0103) for 30 minutes at 55°C. Cell lysates were mixed thoroughly, and 40-80 µl of each lysate was added per well of a capture plate with 20 µl diluted lysis mixture without proteinase K. Probe sets for Human FLT1, sFlt1-i13, sFlt1-e15a , HPRT, PPIB and YWHAZ

(Affymetrix; # ) or mouse Flt1, sFlt1, Hprt and Ppib were diluted and used according to the manufacturer’s recommended protocol.

For analysis of mRNA in mouse tissues, tissue punches (2-5 mg) were homogenized in 300-400 µl of Homogenizing Buffer (Affymetrix, #) containing 2 µg/µl proteinase K in 96-well plate format on a QIAGEN

TissueLyser II, and 20-60 µl of each lysate was added to a bDNA capture plate. Diluted probe sets were added to each well of the capture plate for a final volume of 100 µL. Signal was amplified according to the Affymetrix protocol. Luminescence was detected on a Veritas Luminometer

(Promega, Madison, WI; #998–9100) or Tecan M1000 (Tecan,

Morrisville, NC).

133

4.3.6 Fluorescence Imaging

Mice were injected intravenously (tail vein) or subcutaneously

(intrascapular) with 10 mg/kg of Cy3-labeled siRNA. After 24 hours, mice were sacrificed and tissues were removed, embedded in paraffin, and sliced into 4-µm sections that were mounted on glass slides. Sections were deparaffinized by incubating in Xylene twice for 8 minutes. Sections were rehydrated in serial ethanol dilutions (100%, 95%, and 80%) for 4 minutes each, and then washed twice for 2 minutes with PBS, stained with

DAPI. All fluorescent images were acquired with a Leica DMi8 inverted tiling microscope (Leica Microsystems). Images were processed using the

Leica software suite (LAS X and LAS X Lite).

4.3.7 PNA Hybridization assay siRNA guide (antisense) strand accumulation in mouse tissues was quantified using a PNA hybridization assay, as described previously31.

Briefly, tissue punches were homogenized in MasterPure Tissue Lysis

Solution (EpiCentre) with added proteinase K (2 mg/mL, Invitrogen) and homogenized using a TissueLyser II (Qiagen), using 100 µL of lysis solution per 10 mg tissue. Following lysis, sodium dodecyl sulfate (SDS) was precipitated with KCl (3 mol/l) and cleared supernatant was hybridized to a Cy3-labeled PNA oligonucleotide fully complementary to the guide strand (PNABio) (Supp. Table 2). This mixture was analyzed by 134

high-performance liquid chromatography (HPLC) . Cy3-labeled peaks

were integrated and plotted on an internal calibration curve. Mobile phase

for HPLC was 50% water, 50% acetonitrile, 25 mmol/l Tris-HCl (pH 8.5)

and 1 mmol/l ethylenediamine-tetraacetate. The salt gradient was 0-800

mmol/l NaClO4.

4.3.8 Statistical Analysis

Data were analyzed using GraphPad Prism 7 software (GraphPad

Software, Inc., San Diego, CA). Concentration-dependent IC50 curves

were fitted using a log(inhibitor) vs. response – variable slope (four

parameters). The lower limit of the curve was set at zero, and the upper

limit of the curve was set at 100. For each independent mouse

experiment, the level of knockdown at each dose was normalized to the

mean of the control group (PBS or untreated groups). ). In vivo data were

analyzed using a Kruskal-Wallis one-way ANOVA with Dunn’s post-hoc

analysis. Differences in all comparisons were considered significant at P-

values less than 0.05.

4.4. Discussion The overall goal of this study was to explore RNA interference (RNAi) as a

novel therapeutic approach for the treatment of preeclampsia (PE). PE is a

major unmet public health need that worldwide causes >500,000 maternal 135

and infant deaths annually. PE is characterized by high maternal blood pressure, proteinuria and edema after the 20th week of pregnancy. It is now well established that these symptoms are due to overproduction by the placenta of a protein called soluble FLT1 (sFlt1). This protein is secreted into the maternal bloodstream and causes maternal kidney and liver dysfunction. Currently, the sole treatment option for PE is to deliver the placenta (and therefore the immature fetus) regardless of gestational age when maternal symptoms become life threatening; once the placenta is removed, maternal symptoms abate within 24-48 hours due to the short half-life of circulating sFlt1 protein (~4 hrs). Although it is possible to lower circulating sFlt1 in the maternal bloodstream by apheresis (blood washing) and thereby prolong PE pregnancies (Thadhani et al., 2011; 2016), this procedure is still in early stage clinical trials. It is also expensive with high medical infrastructure requirement, so an alternative approach is needed.

Our aim is to develop small interfering RNAs (siRNAs) that would target and destroy the placental mRNAs that encode sFlt1 protein. siRNAs can be manufactured at low cost, are long-lived with high heat stability, and can be administered via subcutaneous (SC) injection. Recent demonstration of multi-month effect duration from a single SC injection with negligible side effects signals that siRNA therapeutics have now come of age (Alnylam,

NEJM, 2017; Khvorova, NEJM, 2017). Therefore, this emerging class of 136

"informational drugs" represents a unique opportunity to develop a simple and cost-effective treatment for PE.

A siRNA-based therapeutic might be an ideal solution for the treatment of

PE. First, both preclinical and clinical data support lowering sFlt1 levels as a valid therapeutic strategy for prolonging PE pregnancies. Second, the unique region specific to each sFlt1 protein is very small, with only a handful of unique amino acids being appended to each C-terminus. This small target size hinders development of conventional drugs (small molecules and antibodies) that target only sFlt1s and not fl-FLT1. On the other hand, the target window at the RNA level is much larger, with the sFlt1-i13 and sFlt1-e15a mRNA isoforms having 435 and 567 unique bases, respectively, none of which are present in fl-Flt1 mRNA. Because

RNAi requires a target size of only 19-22 nts, this is more than sufficient nucleotide space in which to design multiple isoform-selective siRNAs.

Further, siRNA half-life in vivo is of sufficient duration that a single intravenous dose is well suited for a 2-6 week inhibition of sFlt1 production.

Finally, from a clinical perspective, the possibility that a single dose delivered subcutaneously will be sufficient to prevent runaway sFlt1 expression for several weeks could make treatment simple and affordable.

137

The desired level of sFlt1 silencing is only 30-40%, and a higher degree of silencing might be disadvantageous. Our preliminary data indicate that a

20-40 mg/kg dose produces > 50% silencing in mice, so lesser silencing may simply be achieved with lower dosing. But because the desired product profile is a one-time injection, higher doses might be required to extend effect duration. Thus we might also consider selecting sFlt1-i13 or sFlt1-e15a alone as a clinical candidate and make the final decision on the exact composition of our preclinical candidate for IND after completion of mouse and non-human primate efficacy studies.

We have evaluated the efficiency of siRNA uptake in CTB and have seen efficient uptake by all cells upon addition of Cy3-labeled compound to the media (data not shown).

The siRNAs are asymmetric compounds, with a short duplex region (15 base-pairs) and single-stranded fully phosphorothioated tail, where all bases are fully modified using alternating 2'-fluoro and 2'-O-methyl modification pattern (providing stabilization and avoidance of PKR response), and the 3' end of the passenger strand is conjugated to TEG-

Cholesterol (Figure 4.2.A). The cholesterol enables quick membrane association, while the single-stranded phosphorothioated tail is essential for cellular internalization by a mechanism similar to that used by conventional antisense oligonucleotides (data not shown). Addition of Cy3- 138

labeled siRNA to any cultured cell type shows quick and efficient internalization through an EE1 related part of the endocytosis pathway.

One of the major advances the Khvorova lab has made in recent years is to develop a chemical modification pattern that does not interfere with primary RISC entry. We and others have screened a wide range of chemical variations and identified that an alternating 2’-fluoro/2'-O-methyl pattern optimally configures the guide strand to adopt a geometry that closely mimics that of an individual strand in a A-form RNA duplex, is recognized by the RISC complex and supports proper positioning of the target mRNA within the cleavage site. By starting the alternating pattern with a 5'-phosphorylated 2’-O-methyl ribose (a 5' phosphate is necessary for PIWI domain interaction); this fixes placement of the 2’-fluoro modifications in even numbered positions 2-14; positions 2 and 14 were previously shown to be intolerant of bulkier 2'-ribose modifications,

These fully chemically stabilized compounds are at least as or more effective as naked siRNA in RISC entry and represent the first complete chemical modification pattern with no negative impact on RISC function.

This discovery was transformative for the PE project, as complete chemical stabilization is absolutely essential for tissue accumulation upon systemic administration. Another benefit of non-RNA containing siRNAs is ease of manufacturing – their DNA-like chemistry with no necessity for orthogonal 139

ribose protection shortens de-protection procedures and increases coupling efficiencies. Finally, complete elimination of all 2'-OH groups helps with avoidance of the innate immune response, which relies mainly on 2'-

OH interactions.

140

5. CHAPTER V: DISCUSSION AND FUTURE PERSPECTIVES

Preface:

The conclusions and discussions in this chapter are adapted from the discussion sections in Chapters II and IV.

141

5.1. Discussion

PE is a life-threatening disorder with increased levels of sFlt1 protein that

causes maternal symptoms of HBP and proteinuria to manifest from 20th

week onwards of pregnancy. The overall goals of this thesis were to build

a foundation to understand the pathophysiology of PE by characterizing

the transcriptome of normal and PE patients, characterize FLT1 isoforms

and transcriptome-wide polyadenylation site changes between normal and

PE patients and identify the most abundant sFl1t isoform to target as a

therapeutic in PE patients. This study is the first PAS-Seq study in PE and

an important step towards characterizing the PAS sites and transcriptome

changes in PE placentae.

The first question that I set out to ask with my thesis was which

mechanism regulates sFlt1 mRNA upregulation in PE? Is it due to

transcriptional upregulation of the entire gene or because of APA and

preferential proximal PAS site usage that gave rise to the sFlt1 isoforms?

The second aim of this thesis was to find other differentially expressed

genes between normal, EO-PE and LO-PE placentae. The third aim was

to identify the most abundant sFlt1 isoform expressed in the PE placentae

in order to develop novel siRNAs as a potential therapeutic.

142

Several studies have reported that FLT1 gene is upregulated in PE patients compared to the control. However, I wanted to take a step further and understand FLT1 expression in PE with a clinical lens. We therefore divided our PE samples into early-onset (EO-PE; GDW < 34) and late- onset (LO-PE; GDW > 34) groups. To answer the first question, I used total RNA from CTRL, EO-PE and LO-PE placenta to construct deep sequencing libraries. To understand whether sFlt1 isoforms upregulated is due to increase in full length Flt1 or because of APA and alternative splicing, I prepared two sets of deep sequencing libraries from the same patient sample: PAS–Seq and RNA–Seq. PAS-seq is a transcriptome- wide, 3’-sequencing methodology which detects changes in the various

APA isoforms arising from the same gene. I developed a kit-free PAS-seq by optimizing the published RNA-Seq protocol (Heyer et al., 2015b).

Furthermore, PAS-Seq is also a great proxy for calculating differential gene expression. My results indicate that the entire FLT1 locus and therefore both the m-Flt1 and sFlt1 (encompasses sFlt1-i13 S, sFlt1-i13 L and sFlt1-e15a) isoforms are significantly upregulated in EO-PE (Chapter

II). I also find that sFlt1-e15a isoform is the most abundant isoform in PE placentae, with ~50% of the FLT1 reads mapping to it. The remaining reads map to sFlt1-i13 S (19%) followed by sFlt1-i13 L (18%) isoforms.

Few reads corresponding to the sFlt1-i14 and sFlt1-e15b isoforms were 143

mapped, indicating that these are lowly expressed sFlt1 isoforms in the

PE placentae and may not play a major role in the pathophysiology of PE.

In order to address the second question, we did differential gene expression analysis and report 349 (padj ≤0.01) differentially expressed genes in EO-PE and 40 (padj ≤0.01) in LO-PE (Chapter II). We find 50 genes (padj ≤0.05) common between EO-PE and LO-PE including FLT1

(4.3-fold up in EO-PE and 1.9-fold upregulated in LO-PE). The gene ontology terms associated with these 50 genes includes the HIF1α, HIF2α and VEGFR1 (FLT1) signaling pathways. It is interesting to note that FLT1 gene is upregulated under hypoxic conditions (Ferrara, 2004; Goldman et al., 1998; Nevo, 2006). Using PCA analysis and differential gene expression this dissertation also agrees with the clinical phenotypes and extends the idea that there is distinct gene signature for EO-PE and LO-

PE. Thus PAS-Seq has been able to confirm (distinct EO-PE and LO-PE subsets) bioinformatically what has been previously reported clinically.

There is a significant enrichment of differentially expressed genes associated with transcription factors, again indicating that there is a highly transcriptional response in PE as previously reported(Sõber et al., 2015;

Vaiman et al., 2013). In particular, there are several studies that FLT1 is transcriptionally regulated by HIF-1α under hypoxic conditions. Additional studies have reported multiple transcription factors that regulate FLT1 144

expression. It is thus not a surprise that cellular stresses like hypoxia and oxidative stress in the PE placenta activates the transcription factors leading to wide spread transcriptional changes. However, future studies are needed highlight our understanding of what triggers the upregulation of FLT1 in PE and in particular the sFlt1-e15a isoform.

Nuclear Receptor-Interacting Protein (NRIP1) is upregulated in the EO-PE and LO-PE samples compared to the controls. Interestingly, analysis of other published datasets show a similar upregulation (Kaartokallio et al.,

2015; Leavey et al., 2015; Sõber et al., 2015). NRIP1 has been shown to play an important role in metabolic disorders and lipid imbalances

(Nautiyal, 2017; Nautiyal et al., 2013) indicating a change in the cellular metabolic pathways in patients with PE. Women with a high BMI have been shown to have lipid disturbances and often suffer from low-grade inflammation and endothelial dysfunction (Thadhani et al., 1999). Further

NRIP1 was found to be differentially expressed when comparing the birth weight and gestational age of the infant in PE when compared to normal samples (Sõber et al., 2015). Interestingly, genome-wide association studies in women suffering from endometriosis identified SNP variants in

NRIP1, which may have functional relevance (Caballero et al., 2005).

Similar genome-wide and functional studies in PE patient samples will help shed light on the role of NRIP1 in the disease. 145

In addition to analyzing the protein-coding gene expression, we also used the novel PAS-Seq method to annotate non-coding RNAs in normal and

PE pregnancies. This dissertation demonstrates that several PAS peaks were annotated on retrotransposon elements. In particular we found PAS- peaks distributed on long interspersed elements (LINEs) and non-long terminal repeats (LTRs). These elements have been shown to play an important role in regulating gene expression. Several studies report that these elements play an important role in Drosophila neurodevelopment

(Perrat et al., 2013), central nervous system diseases (Reilly et al., 2013) and autoimmune disorders in humans (Balada et al., 2010). There is scare data available on the role of repeat elements in pregnancy disorders.

Additionally, we also found RNA-Seq reads (Chapter II) corresponding to the same regions as PAS-Seq reads on the LTR and LINE elements in PE patients. Interestingly, sFlt1-e15a isoform has a unique Alu retrotransposon within its 3’-UTR leading a unique polyserine stretch in the C-terminal end of the sFlt1-e15a protein. Studying the role of Alu,

LINE and LTR elements in future PE studies may help shed light on the biological regulation of sFlt1-e15a isoform in PE.

The third aim of this dissertation was to identify the sFlt1 isoform that was most abundantly expressed in the PE placenta. In order to address this 146

question, I conducted a pilot screening of siRNAs targeting the unique

3’UTR of sFlt1-i13 S (Chapter III). I along with our collaborators demonstrate that targeting sFlt1-i13 S, sFlt1-sFlt1-i13 L and sFlt1-e15a mRNAs is a viable therapeutic approach to alleviate PE (Chapter III and

Chapter IV). We also provide evidence that siRNAs can be used in animal models to knockdown various sFlt1 isoforms effectively without affecting the placenta or the fetus (Chapter IV).

Some of the caveats in this study are as follows. We had a small sample size for the LO-PE group. In future studies, we will need to include a higher number of the LO-PE patients. Another caveat of this study

(Chapter II) is that the libraries are made from placental tissue, which is a mix of both maternal and fetal cells. Even though complete effort was made to collect only the villous tissue, which is the main source of sFlt1, the possibility of contamination from surrounding decidual cells (maternal cells) cannot be completely eliminated. There are a couple of ways to manage this inherent tissue variation. One is to make next-generation sequencing libraries at a cell level (ex: Single-Cell RNA-Seq) or second to make next-generation sequencing libraries at a tissue level (ex: PAS-Seq and RNA-Seq from various placental compartments). This approach will aid in distinguishing the noise from the real data.

147

Single-Cell RNA-Seq or laser microdissection as has been shown recently for normal placentas (Pavličev et al., 2017). Studying gene expression from different sections of the placenta has been done before for microarrays (Sood et al., 2006) and RNA-Seq (Kim et al., 2012) from normal placenta. Similar strategy can be applied to making next- generation libraries from these sections like the decidua (maternal side of the placenta), amnion and chorion (fetal side of the placenta). In order to further refine the PE placental transcriptome, future experiments can be designed to make PAS-Seq also from placental cells or tissue sections as described above.

5.2. Future Directions

PE is a difficult disease to study mainly because the molecular changes are likely to occur prior to symptom manifestation, which occurs after the

20th week of pregnancy. This makes it impossible to study this disease in vivo because of difficulty in getting the early stage placenta due to ethical limitations. Additionally, there are limited numbers of animal models because the placentation process differs between humans and mice or rats. As the two stage-models suggest that something goes awry in the placentation stage, studying placental cells i.e. primary trophoblasts during placentation has been a common approach to studying PE. I believe that 148

in the future, one could use stem cells derived from the embryo to differentiate them into trophoblasts, the earliest cells derived from the placenta (Chang and Parast, 2017) (Horii et al., 2016). This would allow us to study PE and understand the pathophysiology of PE as early as in the placentation stage (Horii et al., 2015).

Another technology that can help dissect the molecular players of PE is called Method for Analyzing RNA following Intracellular Sorting (MARIS).

Placenta is a complex tissue with cells coming from the mother (decidual cells) and cells coming from the fetus (chorion and amnion). In order to study the various cells coming from the different compartments MARIS technique can be applied. In this, cells from the placenta can be separated in two separate fractions; cells can be collected from the maternal and the fetal side. Furthermore, we can separate the various cells within each fraction. For example, in the decidual side, we could use fluorescent antibodies to label the lymphocytes, macrophages etc. On the fetal side, we could use antibodies for the cytotrophoblasts, syncytiotrophoblasts.

These cells are further sorted using Fluorescence-activated cell sorting

(FACS). The RNA can then be extracted for each the maternal cells and fetal cells and used for further deep sequencing. This method provides more depth than single-cell RNA-Seq (Hrvatin et al., 2014).

149

This thesis has demonstrated that in PE conditions, the entire FLT1 gene is transcriptionally upregulated. To further our understanding of the transcriptional regulation of FLT1 gene, future studies should include

Chromatin Immunoprecipitation followed by deep sequencing (ChIP-Seq) experiments. ChIP-Seq allows for identifying the binding sites of various proteins including transcription factors in a precise manner. For example, the DNA-bound protein can be isolated from normal, EO-PE and LO-PE pregnancies. The DNA can be further purified and sequenced. This would allow us to compare the binding sites of enhancers, transcription factors and other regulatory proteins.

In the future, I predict that diagnostic tests for PE will be used commonly than at present. Genome-wide studies for heterogeneous disorders like

PE have grown in popularity over the last two years (Kaartokallio et al.,

2015; Lekva et al., 2016; McGinnis et al., 2017; Sõber et al., 2015). I predict that studies similar to given in this thesis (Chapter II) and elsewhere will shed light in to the pathophysiology of PE. Studying layers of gene regulation by understanding the RNA splicing, RNA polyadenylation and RNA stability in PE will be the future, as I believe a wealth of information can be extracted by these genome-wide studies.

150

5.3. Conclusions

In conclusion, I developed a new 3’-end sequencing protocol that allows simultaneously studying alternative polyadenylation and gene expression from placental tissues. Using PAS-Seq, I first identified that sFlt1-e5a of the most abundantly expressed alternative polyadenylated sFlt1 isoform followed by sFlt1-i13S and sFlt1-i13 L in both EO-PE and LO-PE.

Secondly, I also provide evidence for a potential new therapeutic using anti-sFlt1 siRNAs. This work is further expanded in Dr. Anastasia

Khvorova’s lab where they use chemically modified siRNAs to alleviate PE in animal models of PE. Thirdly, I have utilized PAS-Seq to profile transcriptome-wide polyadenylation changes between normal, EO-PE and

LO-PE pregnancies. Lastly, I have studied the differential gene expression changes between normal, EO-PE and LO-PE pregnancies.

Furthermore, the findings in this thesis underscore the impact of genome- wide deep sequencing studies on multifactorial pregnancy disorder like

PE. In addition, this study also validates the idea that EO-PE and LO-PE are distinct sub-types of the PE spectrum.

The data provided here gives us a better understanding of the pathophysiology of PE and potentially aid in further development of RNAi based drugs for pregnancy disorders. 151

An exciting promising study (McGinnis et al., 2017) was published during the preparation of this thesis, wherein they found SNP variants near the gene FLT1 that has been the focus of Chapter II. These SNPs have been found near the FLT1 promoter region which can is an important implication to understand FLT1 gene regulation (McGinnis et al., 2017). This opens an exciting avenue for understanding how sFlt1’s are upregulated in PE.

152

APPENDIX A: RNA SEQUENCING Introduction

In addition to constructing PAS-Seq libraries from placental samples, we also constructed RNA-Seq libraries from the same patient samples

(CTRL, EO-PE and LO-PE) as PAS-Seq. There were multiple aims to address with RNA-Seq libraries. One of the main aims was to study the

RNA processing like splicing, alternate 5’ promoter usage, differential isoform and gene analysis. RNA-Seq would be another way to compare and contrast our PAS-Seq results. Also, at the time when this project began and ongoing (September 2011- August 2015), there were published

RNA-Seq datasets only for CTRL placenta(Kim et al., 2012; Sood et al.,

2006) and none from PE placenta.

Results Placental RNA-Seq samples

RNA-Seq libraries were made from the same samples as we used in

Chapter II. Using highly selective criteria for patient inclusion, we limited our sample set to singleton pregnancies, and excluded any patients with preexisting hypertension or diabetes mellitus. We had a control set

(CTRL; n = 6) delivered at term with no other pregnancy and our PE set

(n=11) included patients with systolic blood pressures ≥ 140 mm Hg and circulating sFlt1:PlGF ratios ≥ 85, an accepted diagnostic measure of PE 153

associated with adverse PE related maternal/fetal outcome(Verlohren et al., 2010; 2012b) (Figure 2.1 c and Table 2.1). Based on the Gestational

Delivery Week, we further divided the PE patients into Early-Onset (EO-

PE; GDW <34; n=8) and Late-Onset (LO-PE; GDW >34; n=3) subgroups.

The placental tissue was collected in the same manner as in Chapter II.

The placenta was collected within 30 minutes of delivery and flash frozen in liquid nitrogen. Using these frozen samples, we isolated total RNA from the villous tissue, which is the spongy layer in between the decidua

(maternal side) and the chorion (fetal side), and the main source of sFlt1(Rajakumar, 2011; Rajakumar et al., 2012). The rRNA was depleted from the total RNA (see Figure A.1 and Methods). The remainder rRNA minus fraction contained all the polyA+ RNA plus the snRNAs, snoRNAs and other non-coding RNAs. Briefly, this RNA pool was used to reverse transcribe using RT primers (Table A.1), followed by circularization and

PCR amplification. 154

Figure A.1 RNA-Seq method flowchart 155

Bioinformatics analyses

Before mapping the sequencing reads to the human transcriptome, we

checked the initial quality of the data using FASTQC program. Out of the

sixteen libraries, we got mixed results in terms of quality. Some of the

libraries (Figure A.2.B-D) had very poor quality. Some of the reasons for

bad quality are that the RNA had degraded over the time of the library

preparation, issues with sequence length distribution or a problem with the

sequencing run.

We decided to continue analyzing these libraries and used the Dolphin

pipeline to further filter reads to any non-mRNA reads before mapping to

hg19 genome. We found that a wide range (4%-80%) of our reads

mapped to the rRNA (Table A.2). This was concerning as it could mean

two things, either the rRNA depletion step was inadequate or the RNA

was degraded causing degradation of the rRNA and therefore the rRNA

depletion step was inefficient. We decided not to analyze the data, but just

out of curiosity we wanted to explore the differential gene expression in

these samples.

RNA-Seq library issues

Given the two observations as follows 1. Bad quality reads and 2.

Difference in size selection, we chose not to move forward with further analyses of the RNA-Seq data. 156

b a

Figure A.2 BoxWhisker plot showing the quality score across all bases on every position in the fastq file a b. Examples of CTRL libraries c. d. Example of EO-PE and LO-PE library respectively.

X-axis represent the position in read (bp) The Y-axis shows the quality score. Higher number on the Y-axis represents good read calls, with the background color representing very good (green), medium (orange) and bad (pink) quality calls. The yellow boxes shows the inter-quartile range (25-75%), the blue line running through the plot represents the mean quality and the upper and lower whiskers represent 10-90% points. 157

Methods Isolation of mRNA for deep sequencing

As rRNA makes up 80% of the total RNA contents, the total RNA was depleted of rRNA using the Ribo-Zero Magnetic Gold kit

(Human/Mouse/Rat) (Illumina # MRZG12324). This allowed to enrich for mRNA and other RNAs (tRNAs, snRNAs etc.). The rRNA depleted

RNA samples were further fragmented using RNA Fragmentation Reagent

(Ambion #AM8740). RNAs were size selected for 100-150 nts and dephosphorylated as above.

Library Preparation

The dephosphorylated RNA was fragmented and a preadenylated adaptor

(sequence–TGGAATTCTCGGGTGCCAAGG) was ligated to RNA 3' ends, after which the ligated RNAs were reverse transcribed. The RT product was gel purified, circularized, and PCR amplified prior to sequencing as described in (Heyer et al., 2015b).

RNA-Seq Bioinformatics analysis

The reads were demultiplexed using the barcodes (Table A.1). After separating the reads in different libraries, the FASTQC program was used to look at the quality of the reads. The adapter (sequence–

TGGAATTCTCGGGTGCCAAGG) at the 3′end of the reads was trimmed 158

using Trimmomatic-0.32 (parameters: adapter.fa:1:30:5 MINLEN:15), allow 1 mismatch, min length-drop reads below 15 nt)

rRNA, miRNA, tRNA, snRNA and repeat elements were filtered out using and put away in a separate folder (parameters: bowtie2-2.2.3/bowtie2 -p 2

-N 1 --no-unal) (-p=threads, -N allows one mismatch, no-unal=Suppress

SAM records for reads that failed to align)

Reads were mapped to human transcriptome (hg19) using rsem-calculate- expression function in RSEM (parameters: bowtie-0.12.9, -p 4 --bowtie-e

70 --bowtie-chunkmbs 100 --output-genome-bam --calc-ci). P= number of threads, bowtie e=Bowtie parameter max sum of mismatch quality scores across the alignment, chunkmbs=memory allocated for best first alignment calculation, calc-ci=Calculate 95% credibility intervals and posterior mean estimates, output bam=Generate a BAM file,

'sample_name.genome.bam', with alignments mapped to genomic coordinates and annotated with their posterior probabilities. In addition,

RSEM will call Samtools (included in RSEM package) to sort and index the bam file.)

159

BC1: 5' - pGGATCACAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC2: 5' - pGGCGATGAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC3: 5' - pGGTAGCTAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC4: 5' - pGGGCTCCAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC5: 5' - pGGACAGTAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC6: 5' - pGGCAGATAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC7: 5' - pGGTCCCGAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC8: 5' - pGGGGCTAAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC9: 5' - pGGAGTCAAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC10: 5' - pGGCTTGTAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC11: 5' - pGGTGAATAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3' BC12: 5' - pGGGTAGAAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-SP18- CTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-CCTTGGCACCCGAGAATTCCA – 3'

Table A.1 RNA-Seq RT primer sequences

160

RNA-Seq Differential gene expression

The reads were analyzed using the DEBrowser package in R

(https://github.com/UMMS-Biocore/debrowser). The raw counts for each gene in each library were uploaded to the DEBrowser. Default normalization parameters were used for further DESeq2 analyses.

161

Figure A.3 Scatter plot showing differential gene expression between CTRL and EO-PE samples.

Red dot indicates genes, which are downregulated in EO-PE, and green dot indicates gene, which are upregulated in EO-PE. Grey dots indicate genes whose expression is unchanged between CTRL and EO-PE.

162

Figure A.4 Scatter plot showing differential gene expression between CTRL and LO-PE samples

Green dot indicates gene, which are upregulated in LO-PE. Grey dots indicate genes whose expression is unchanged between CTRL and LO-PE. 163

Reads Uniquely Libr Total ERCC Reads After aligned to Multimap mapped ary Reads (spike-in) rRNA miRNA tRNA snRNA rmsk Filtering hg19 ped reads reads 19,570 24 (0.00 3,269,546 1,056 8,287 11,271 (0.06 19,672 (0.10 16,291,673 110,909 17,364 684,329 (3.50 N1 ,586 %) (16.71 %) (0.01 %) (0.04 %) %) %) (83.25 %) (0.57 %) (0.09 %) %) 10,568 17,327 2,803,870 16,804 176,298 560,960 (5.31 749,310 7,554,304 1,425,757 198,770 2,895,039 N2 ,603 (0.16 %) (26.53 %) (0.16 %) (1.67 %) %) (7.09 %) (71.48 %) (13.49 %) (1.88 %) (27.39 %) 16,181 8,354 7,170,726 11,729 116,533 360,496 (2.23 500,125 8,874,408 960,698 106,972 1,862,545 N3 ,750 (0.05 %) (44.31 %) (0.07 %) (0.72 %) %) (3.09 %) (54.84 %) (5.94 %) (0.66 %) (11.51 %) 21,279 180 (0.00 15,690,967 9,808 16,798 3,287 (0.02 36,877 (0.17 5,561,607 107,413 14,803 280,348 (1.32 N4 ,360 %) (73.74 %) (0.05 %) (0.08 %) %) %) (26.14 %) (0.50 %) (0.07 %) %) 23,439 23,404 7,315,275 25,659 106,427 1,159,018 1,440,305 15,968,823 2,124,144 278,064 4,413,167 N5 ,588 (0.10 %) (31.21 %) (0.11 %) (0.45 %) (4.94 %) (6.14 %) (68.13 %) (9.06 %) (1.19 %) (18.83 %) 16,227 264 (0.00 12,847,331 6,875 10,488 5,140 (0.03 20,180 (0.12 3,363,040 66,358 (0.41 10,914 160,947 (0.99 N6 ,998 %) (79.17 %) (0.04 %) (0.06 %) %) %) (20.72 %) %) (0.07 %) %) 32,155 840 (0.00 335,993 (1.04 74,730 513,122 7,069,547 3,457,736 31,230,772 6,146,897 946,475 14,632,112 P1 ,457 %) %) (0.23 %) (1.60 %) (21.99 %) (10.75 %) (97.12 %) (19.12 %) (2.94 %) (45.50 %) 14,633 36,734 2,691,646 18,636 179,778 1,321,161 1,623,860 11,706,876 2,212,957 371,069 4,800,926 P2 ,670 (0.25 %) (18.39 %) (0.13 %) (1.23 %) (9.03 %) (11.10 %) (80.00 %) (15.12 %) (2.54 %) (32.81 %) 33,763 79,365 4,352,764 50,304 414,933 11,073,492 3,170,995 28,865,757 4,236,960 663,234 10,422,001 P3 ,123 (0.24 %) (12.89 %) (0.15 %) (1.23 %) (32.80 %) (9.39 %) (85.49 %) (12.55 %) (1.96 %) (30.87 %) 19,837 11,050 7,667,206 32,845 527,574 553,624 (2.79 840,700 11,599,018 1,523,540 276,501 3,074,130 P4 ,693 (0.06 %) (38.65 %) (0.17 %) (2.66 %) %) (4.24 %) (58.47 %) (7.68 %) (1.39 %) (15.50 %) 32,452 47,487 10,147,410 34,921 267,470 2,185,547 2,064,083 21,955,346 4,387,649 562,014 9,543,242 P5 ,634 (0.15 %) (31.27 %) (0.11 %) (0.82 %) (6.73 %) (6.36 %) (67.65 %) (13.52 %) (1.73 %) (29.41 %) 10,285 282 (0.00 4,317,988 5,783 16,049 62,063 (0.60 214,222 5,945,087 517,257 80,753 1,209,500 P6 ,189 %) (41.98 %) (0.06 %) (0.16 %) %) (2.08 %) (57.80 %) (5.03 %) (0.79 %) (11.76 %) 8,630, 76 (0.00 402,324 (4.66 6,504 16,956 1,802,679 999,105 8,204,850 1,967,059 226,670 3,845,987 P7 710 %) %) (0.08 %) (0.20 %) (20.89 %) (11.58 %) (95.07 %) (22.79 %) (2.63 %) (44.56 %) 8,386, 6 (0.00 138,242 (1.65 5,354 9,822 1,211,001 1,002,209 8,233,196 2,584,394 503,719 5,510,431 P8 620 %) %) (0.06 %) (0.12 %) (14.44 %) (11.95 %) (98.17 %) (30.82 %) (6.01 %) (65.71 %) 8,620, 53 (0.00 3,818,003 3,688 10,953 123,836 (1.44 292,975 4,787,736 780,837 119,685 1,563,694 P9 433 %) (44.29 %) (0.04 %) (0.13 %) %) (3.40 %) (55.54 %) (9.06 %) (1.39 %) (18.14 %) 6,890, 32 (0.00 677,982 (9.84 4,616 18,087 1,076,087 720,012 6,190,141 1,528,518 233,917 3,087,346 P10 858 %) %) (0.07 %) (0.26 %) (15.62 %) (10.45 %) (89.83 %) (22.18 %) (3.39 %) (44.80 %) 6,327, 119 (0.00 1,351,666 5,858 19,979 345,705 (5.46 579,081 4,949,982 1,165,342 216,395 2,254,833 P11 604 %) (21.36 %) (0.09 %) (0.32 %) %) (9.15 %) (78.23 %) (18.42 %) (3.42 %) (35.63 %)

Table A.2 RNA-Seq mapping statistics 164

rRNA sub-units Library 28S 18S 5.8S 5S N1 2255886 1013373 258 29 N2 2494010 308777 980 103 N3 6384688 785096 847 95 N4 15594215 96095 567 90 N5 6923320 391370 538 47 N6 12808152 38770 357 52 P1 213268 120545 2053 127 P2 2492543 197452 1542 109 P3 4292320 59154 880 410 P4 7382889 278698 5197 422 P5 10027692 117905 1371 442 P6 2912451 1402369 2936 232 P7 164705 236815 798 6 P8 122241 15977 24 0 P9 1962586 1855024 373 20 P10 355448 321322 1199 13 P11 665984 684124 1520 38 Table A.3 RNA-Seq reads mapping to rRNA sub-units

165

Discussion

During the analyses of the libraries it was observed that our data suffered with rRNA contamination, which made any analyses difficult to conduct.

As we were aware of the limitations of our libraries, we decided to limit the analyses of our data. Even though we could not make any biologically relevant conclusions, we decided to instead present here the problems we encountered in our data so future researchers can be aware of these problems.

The rRNA contamination we observed could either be 1. the rRNA removal kit was freeze-thawed multiple times and the reagents were degraded, 2. the total RNA was degraded prior to making libraries or 3. . the total RNA was degraded during the library preparation. Additionally when batch effect analysis (data not shown) was conducted on the RNA-

Seq libraries, the samples showed bias in library preparation method.

166

APPENDIX B: Staufen1 senses overall transcript secondary structure to regulate translation

Preface

Appendix B was taken verbatim from the following published manuscript:

Staufen1 senses overall transcript secondary structure to regulate translation

Emiliano P Ricci, Alper Kucukural, Can Cenik, Blandine C Mercier, Guramrit Singh, Erin E Heyer, Ami Ashar-Patel, Lingtao Peng & Melissa J Moore

(2014). Staufen1 senses overall transcript secondary structure to regulate translation. Nature Structural & Molecular Biology, 21(1), 26–35. http://doi.org/10.1038/nsmb.2739

Author Contributions E.P.R. and M.J.M. conceived the study, designed the experiments and wrote the manuscript. E.P.R. performed the experiments. A.K. conducted most bioinformatics analyses. C.C. designed and implemented the ribo-seq analysis. E.P.R. and A.K. designed and performed GC-content and secondary structure– prediction analyses. B.C.M. and G.S. contributed with tandem-affinity purification of Stau1 complexes. E.E.H. contributed with cDNA library preparation for high- throughput sequencing. A.A.P. prepared PAS-Seq cDNA libraries. L.P. participated in quantitative PCR analysis.

Supplemental tables, primary accessions and referenced accessions are not provided in this thesis due to size. Please refer to the online supplemental material http://doi.org/10.1038/nsmb.2739

167

Abstract Human Staufen (Stau) is a double-stranded RNA (dsRNA)-binding protein implicated in multiple post-transcriptional gene- regulatory processes. Here we combined RNA immunoprecipitation in tandem (RIPiT) with RNase footprinting, formaldehyde cross-linking, sonication-mediated RNA fragmentation and deep sequencing to map Staufen -binding sites transcriptome wide. We find that Stau binds complex secondary structures containing multiple short helices, many of which are formed by inverted Alu elements in annotated 3′ untranslated regions

(UTRs) or in ‘strongly distal’ 3′ UTRs. Stau also interacts with actively translating ribosomes and with mRNA coding sequences (CDSs) and 3′ UTRs in proportion to their GC content and propensity to form internal secondary structure. On mRNAs with high CDS GC content, higher Stau levels lead to greater ribosome densities, thus suggesting a general role for Stau in modulating translation elongation through structured CDS regions. Our results also indicate that Stau regulates translation of transcription-regulatory proteins.

168

Introduction Staufen proteins are highly conserved dsRNA-binding proteins (dsRBPs) found in most bilateral animals(Kerner et al., 2011). Mammals contain two Staufen paralogs encoded by different loci. Stau1, expressed in most tissues, has a microtubule-binding domain, a dimerization domain and four conserved dsRNA- binding domains (dsRBDs), only two of which (dsRBDs 3 and 4) are necessary for dsRNA binding (LUO et al., 2002a). Within cells, Stau1 can make direct interactions both with itself and with Stau2, the more tissue-specific paralog

(Martel et al., 2010). Functionally, Staufen proteins are involved in multiple post- transcriptional regulatory processes. In flies, 3′ UTR–bound Staufen is required for proper localization and translational control of bicoid and prospero mRNAs during oogenesis(St Johnston et al., 1991) (Ferrandon, 1997). In mammals,

Stau1 has been implicated in mRNA transport to neuronal dendrites, (Köhrmann et al., 1999) regulation of translation via physical interaction with the ribosome(Dugré-Brisson et al., 2005), a form of translation-dependent mRNA degradation known as Staufen-mediated decay (SMD)(Gleghorn et al., 2013;

Kim et al., 2005; Park et al., 2013) (Gong and Maquat, 2011), regulation of stress-granule homeostasis(Thomas et al., 2009b), alternative splicing, nuclear export and translation of a gene containing 3′-UTR CUG-repeat expansions(Ravel-Chapuis et al., 2012a). Although Stau1 is not essential for mammalian development, neurons lacking Stau1 have dendritic spine- morphogenesis defects in vitro, and knockout mice have locomotor-activity deficits(Vessey et al., 2008). 169

Crucial for the understanding of how Stau1 regulates gene expression is comprehensive knowledge of its intracellular RNA-binding sites. Although mammalian Stau1- and Drosophila Staufen-associated mRNAs were identified by microarray analysis after native RNA immunoprecipitation (RIP)(Furic et al.,

2008; Kusek et al.; Laver et al., 2013; Maher-Laporte and DesGROSEILLERS,

2010), those studies were unable to directly map any individual Stau1-binding site, and subsequent bioinformatics analysis yielded no clear consensus for identified mammalian targets(Laver et al., 2013). Thus, with the exception of a few well-characterized binding sites validated by mutagenesis(Ferrandon et al.,

1994; Kim et al., 2007), the exact target sites and RNA structures recognized by mammalian Stau1 remain to be determined. To address this, we here undertook a tandem affinity purification strategy (RIPiT(Singh et al., 2014)) to map Stau1- binding sites transcriptome wide in human tissue-cultured cells. We also knocked down and overexpressed Stau1 to measure functional consequences on target- mRNA levels and translation efficiency. Our results revealed a new role for Stau1 in regulating translation of GC-rich mRNAs by ‘sensing’ overall transcript secondary structure.

Results

Transcriptome-wide mapping of Stau-binding sites

Using the Flp-In system and a tetracycline promoter, we generated HEK293 cells 170

that inducibly expressed a single Flag-tagged copy of either the Stau1 65-kDa spliced isoform (Stau1-WT) or a mutant version (Stau1-mut) containing point mutations in dsRBDs 3 and 4 known to disrupt binding to dsRNA(LUO et al.,

2002b) (Figure B.1.A). Consistently with its propensity to bind dsRNA through the sugar-phosphate backbone(Ramos et al., 2000) and with a previous report suggesting poor UV-cross-linking ability (Liu et al., 1996), we found that Stau1 cross-linked with very poor efficiency to poly(A)+ RNA upon shortwave UV irradiation of living cells. Therefore we used a RIPiT approach wherein initial immunoprecipitation (IP) with anti-Flag antibody was followed by affinity elution with Flag peptide and then a second IP with a polyclonal anti- Stau1 antibody.

RIPiT was performed under two different regimens: (i) To finely-map stable Stau1 footprints, we extensively digested samples with RNase I in between native anti-

Flag and native anti-Stau1 IPs, generating 30- to 50-nt Stau1-bound RNA fragments (FOOT libraries; Figure B.1.B). However, many of these short reads derived from Alu repeat elements (described below) and so were not uniquely mappable. Further, under native conditions, Stau1 can make new dsRNA associations after cell lysis. (ii) Therefore, we also subjected cells to formaldehyde cross-linking before lysis, extensively sonicated the lysates to shear long RNAs into 200- to 300-nt fragments (thereby increasing their ability to be mapped) and performed a denaturing anti-Flag IP and then a native anti-

Stau1 IP (CROSS libraries; Figure B.1.B). Cross-linking and subsequent denaturation should both preserve weak in situ interactions that might otherwise 171

dissociate during sample workup and prevent formation of any new interactions after cell lysis.

We sequenced all libraries constructed by 3′-adaptor ligation to RNA fragments, reverse transcription and circularization on GAII or HiSeq 2000

Illumina platforms and then mapped them to HG18 by using RefSeq gene annotations. Biological replicates of WT and mut CROSS and WT FOOT libraries exhibited extremely high correlations (r >0.98), thus indicating the reproducibility of the approach.

172

Figure B.1 Mapping of Stau1 RNA-binding sites reveal coding regions and 3'UTRs as major occupancy sites

173

(a) Scheme of the tagged WT and mut Stau1 proteins used in this study. (b) Scheme of tandem affinity purifications for footprinting (FOOT) and cross-linking (CROSS) library construction. (c) Pie charts showing the distribution of sequencing reads for each library.

(d) Example of Stau1 cross-linking signal across the CDS of ALDOA (NM_184041.2). (e)

Composite plot of the distribution of sequencing reads across the 5′ UTR, CDS and 3′

UTR of all genes for RNA-seq (red), Stau1-WT CROSS (blue) and Stau1-mut CROSS

(black) libraries. (f) Per-gene scatter plot of CDS ribo-seq read density versus CDS

Stau1-WT CROSS read density with the associated Spearman correlation and calculated P value (n = 2 biological replicates).

174

Stau associates with translating ribosomes

In contrast to our previous exon junction complex (EJC) RIPiT libraries(Singh et al., 2012), all Stau1 FOOT libraries (WT and mut) were dominated by rRNA- mapping reads (14–30% versus 74–83%, respectively; Supplementary Fig. 2).

Further, despite attempts to specifically deplete rRNA fragments during CROSS- library preparation, WT and mut CROSS libraries also contained abundant rRNA- mapping reads (Supplementary Fig. 2). These findings are consistent with a previous report that Stau1 cosediments with 60S ribosomal subunits via interactions independent of the functionality of dsRBDs 3 and 4 (ref. 2).

To further investigate this ribosome association, we performed sucrose sedimentation in the presence of inhibitors that either block elongation

(cycloheximide) or initiation (harringtonine) (Supplementary Fig. 3a). In the presence of cycloheximide, both endogenous Stau1 and Flag–Stau1-WT cosedimented with 60S subunits, 80S monosomes and polysomes, with very little Stau1 observable in ribosome-free fractions at the top of the gradient.

However, when lysates were treated with RNase before sedimentation, ~60% of

Stau1 sedimented at the top of the gradient, with the remainder cosedimenting with 60S and 80S ribosomes (Supplementary Fig. 3a). This suggests that dsRBP-independent interactions with the ribosome are not the sole factor driving

Stau1 polysome association. Finally, when translation initiation was blocked with harringtonine and elongating ribosomes allowed to complete translation (i.e., run off the mRNAs) before cell lysis, Stau1 sedimentation mirrored that of RPL26, an 175

integral 60S protein. Both Stau1 and RPL26 rapidly shifted from heavy polysomal to 80S ribosome fractions upon inhibition of translation initiation (Supplementary

Fig. 3b), suggesting that Stau1 associates with actively translating ribosomes.

Consistently with our ribosome-association data, approximately half of mRNA- mapping WT and mut CROSS reads (49% and 54%, respectively) mapped to coding exons (CDS regions; Figure B.1.C–E). To test whether these CDS- mapping reads were due to Stau1 association with translating ribosomes, we compared their density to the density of ribosome footprints (ribo-seq; Figure

B.1.F). For both Stau1-WT (Figure B.1.F; Spearman correlation = 0.89) and

Stau1-mut (data not shown), CROSS-read density strongly correlated with ribosome density in CDS regions. This correlation held for the entire gene population, thus suggesting that Stau1 generally associates with elongating ribosomes.

In sum, our data indicate that Stau1 is generally associated with the 60S ribosomal subunit, both on and off mRNA. Further, this ribosome association does not require dsRBD functionality but is partially dependent on RNA integrity.

Last, Stau1 appears to associate with actively translating, not stalled, ribosomes.

Stau binds paired Alu elements in 3′ UTRs

Whereas WT and mut libraries were quite similar in their rRNA content, they were quite different with regard to Alu repeat– mapping reads. Alu repeats are 176

~300-nt primate-specific mobile elements in the short interspersed nuclear element family; the human genome contains ~1 million Alu elements, primarily in intergenic regions, introns and 3′ UTRs. Reads mapping to Alu repeats constituted 42% and 28% of non-rRNA–mapping reads in WT FOOT and

CROSS libraries, respectively, but only 19% and 14% in the corresponding mut libraries (Supplementary Fig. 2). Greater Alu enrichment in WT libraries suggested that their interaction depended on Stau1’s ability to bind dsRNA.

Consistently with this, WT CROSS reads were often highly enriched over and adjacent to closely spaced Alu pairs likely to form dsRNA secondary structures.

We detected such Alu-pair Stau1-binding sites on only two large intergenic noncoding RNAs (NR_026757 and NR_026999) and minimally in introns

(Supplementary Fig. 4). Conversely, they were highly enriched in 3′ UTRs (Figure

B.2.A and Supplementary Fig. 4) and in select ‘intergenic’ regions immediately

3′ to annotated 3′ UTRs (Figure B.2.B and Supplementary Fig. 4).

Polyadenylation-site sequencing (PAS-seq) revealed the intergenic regions to represent strongly distal 3′ UTRs(Miura et al., 2013) (Figure.B2.B). Overall, we detected 515 strongly distal 3′ UTRs enriched for Stau1-WT CROSS reads

(Supplementary Table 1), most of which contained multiple Alu pairs.

To identify those 3′ UTRs most enriched for dsRBD-dependent Stau1 binding, we called peaks in the WT CROSS libraries by using ASPeak (an expression- sensitive peak-calling algorithm(Kucukural et al., 2013)). We then compared, for each gene, the cumulative read counts under peak positions in the WT and mut 177

CROSS libraries (Figure B.2.C-D). Overall, the data sets were highly correlated (r

= 0.83). Nonetheless, an outlier population (n = 574; Supplementary Table 2) exhibited much higher cross-linking (by a factor of 2.7) in WT than in mut (Figure

B. 2.C-D); these outliers constitute a set of high-confidence 3′ UTRs displaying dsRNA-dependent Stau1 binding.

We next investigated the structural features of these targets. To identify those containing Alu pairs, we wrote an algorithm to identify, transcriptome wide, pairs of full-length Alu elements in the same (tandem) or opposite (inverted) orientation. Overlaying the inter-Alu distance for tandem Alu pairs on the WT versus mut CROSS scatter plot (Figure B.2.C) revealed no specific relationship between tandem pairs and Stau1 cross-linking. However, the inverted Alu-pair over- lay revealed a striking coincidence with the above outlier population (Figure

B.2.D). Further, inverted Alu elements separated by the least distance were the most outlying (Figure B.2.D).

We confirmed the inverse relationship between dsRBD-dependent Stau1 cross- linking efficiency and inverted-pair inter-Alu distance in composite plots

(FigureB.3.A and Supplementary Fig. 5). We found similar, but less striking, results for inverted pairs containing partial Alu elements and for inverted pairs in introns and strongly distal 3′ UTRs (Supplementary Fig. 5a). As expected, we observed no specific mapping of WT reads on tandem Alu pairs or mapping of mut reads on Alu pairs in either orientation (Figure B.3.A). The inverse 178

Figure B.2 Inverted Alu pairs are an important class of Stau1 binding sites

179

(a,b) Distribution of sequencing reads obtained from RNA-seq (green), Stau1-WT

CROSS (yellow), Stau1-mut CROSS (blue), Stau1-WT FOOT (brown), Stau1-mut FOOT

(violet) and PAS-seq libraries (black) for the 3′ UTR of PAICS (NM_001079524) (a) and the strongly distal 3′ UTR of BRI3BP (NM_080626.5) (b). (c,d) Per-gene scatter plots of

Stau1-WT CROSS and Stau1-mut CROSS read counts under called 3′-UTR Stau1-WT

CROSS peak positions with associated Spearman correlation and calculated P values (n

= 2 biological replicates). Genes containing a 3′ UTR Alu pair are colored with respect to the distance between each tandem Alu pair (c) or inverted Alu pair (d). The dashed

180

correlation between Stau1-WT cross-linking efficiency and inverted-pair inter-Alu distance is consistent with the expectation that secondary-structure formation should inversely correlate with pairing-partner distance. Because intact Alu elements are ~300 nt, inverted pairs containing full-length Alu elements could potentially form very long helices. However, individual elements in pairs exhibiting the highest WT cross- linking signal were often from different Alu families unlikely to be fully complementary. Consistently with this, in silico folding of an inverted Alu pair exhibiting one of the strongest Stau1-WT occupancies suggests the presence of many short helices interrupted by small loops (Figure

B.3.B). To assess the generalizability of this, we folded in silico all-full-length, 3′-

UTR inverted Alu pairs highly enriched for Stau1-WT cross-linking and compared them to 3′-UTR sequences of similar length randomly chosen from nontarget genes. Histograms of predicted helix and loop lengths (Figure B. 3.C-D revealed that Stau1-interacting Alu pairs tend to form structures with multiple helices containing<30 interrupted base pairs, spaced by 2- to 10-nt loops. Conversely, nontarget 3′ UTRs were predicted to have significantly shorter paired stretches

(Wilcoxon rank-sum test, P = 0.04) interrupted by longer loops (Wilcoxon rank- sum test, P = 6 × 10−5).

181

Non-Alu 3′-UTR targets

Among the outlier population in Figure B.2.D, 201 of 574 contained clearly identifiable inverted Alu pairs (Alu targets), and 373 of 574 did not (non-Alu targets; Supplementary Table 2). Many of these non-Alu targets had clearly defined WT footprints in regions with high base-pairing probability (Figure B.4.A).

A few contained a single strong footprint corresponding to a short stem-loop structure (Figure B.4.A); others resembled inverted Alu pairs with many consecutive helices separated by short loops (Figure B.4.B). The largest set, however, consisted of complex structures covering a few hundred nucleotides within which Stau1 footprints could be observed on multiple 7- to 40-bp helices

(Figure B.4.C-D). WT footprints were also present on the Arf1 3′ UTR, for which the precise Stau1-binding site was previously mapped by mutagenesis

(Supplementary Fig. 6a)(Kim et al., 2005).

Comparison of FOOT and CROSS reads mapping to individual 3′ UTRs revealed that CROSS reads generally extended over much more of the 3′ UTR than did

FOOT reads (for example, Figure B.4.C). We could even observe extensive

CROSS read coverage for many 3′ UTRs having no detectable footprints

(Supplementary Fig. 6b). Greater abundance of such CROSS reads in WT libraries than in mut libraries indicated that they depended on Stau1’s ability to bind dsRNA. This suggested that the kinetically stable Stau1-binding sites revealed by native footprinting represent only a small subset of RNA-interaction sites occurring within cells. Supporting the notion of many low-affinity Stau1- 182

interaction sites in vivo, we observed a strong correlation (r = 0.63, P < 2.2 ×

10−16) over all expressed genes between average per-nucleotide predicted secondary- structure strength (∆G of the minimum free-energy structure/3′-UTR length) and the ratio of total WT/mut CROSS reads per 3′ UTR (Figure B.5.A).

We observed a similarly strong correlation (r = 0.55, P < 2.2 × 10−16) between this ratio and 3′-UTR GC content in all expressed genes (Figure B.5.B).

We conclude that some Stau1-binding sites in 3′ UTRs consist of highly defined structures containing multiple short helices to which Stau1 binding is kinetically stable. Other binding sites, however, are more kinetically labile, and the extent of

Stau1 occupancy on these sites is a function of overall 3′-UTR secondary structure–forming propensity, often driven by high GC content.

183

Figure B.3 Characterization of the structural features of Stau1 Alu-binding sites

184

(a) Composite plot of Stau1-WT CROSS, Stau1-mut CROSS and RNA-seq read counts around tandem or inverted Alu pairs. Read counts normalized to host gene reads per kilobase per million mapped reads (RPKM) were determined for a region spanning 1 kb up- and downstream of the paired Alu elements separated by the indicated distance (n =

2 biological replicates). (b) Example of a 3′-UTR inverted Alu-binding site in C11orf58

(NM_014267.5) showing read counts per million mapped (pmm) reads for RNA-seq,

Stau1-WT CROSS and Stau1-mut CROSS libraries. The centroid secondary structure for this Alu pair predicted with the Vienna folding package is shown below center. (c,d)

Length distribution of predicted helices (c) and loop (d) regions within secondary structures of 3′-UTR inverted Alu pairs or 3′-UTR sequences of identical size randomly picked from nontarget genes. P values corresponding to the comparison of helix and loop length distributions between Stau1 Alu targets and random 3′ UTRs were calculated with the Wilcoxon rank-sum test (n = 2 biological replicates).

185

dsRBD-dependent binding of Stau to CDS regions

As previously discussed, both WT and mut CROSS reads mirrored ribosome density across CDS regions transcriptome wide (Figure B.1.F). WT and mut reads were also similarly distributed relative to start and stop codons (Figure

B.1.E). In contrast to the general population, however, our 373 non-Alu 3′-UTR target genes had significantly greater CROSS reads in CDS regions for WT than for mut (Figure B.6.A). This strong relationship between 3′ UTR and CDS

WT/mut cross-linking initially suggested to us that dsRBD-dependent Stau1 binding within the 3′ UTR increases its association with CDS-bound ribosomes.

Consistently with this, the correlation between preferential WT cross- linking in 3′-

UTR and CDS regions held true for the entire mRNA population (Figure B.6.B; r

= 0.61, P < 2.2 × 10−16), with our identified 3′-UTR target genes simply being strongly skewed toward the higher end of both ratios. However, we also found that predicted per-nucleotide secondary structure–forming propensity and GC content were strongly correlated (r = 0.55 and 0.73, respectively, P < 2.2 ×

10−16) between the 3′-UTR and CDS regions of individual genes (Figure B.6. C-

D); that is, the genes with high 3′-UTR secondary structure–forming propensity and GC content also tend to have high CDS secondary structure forming propensity and GC content. Consistently with this, preferential WT cross-linking in CDS regions strongly correlated with both predicted CDS secondary structure and GC content (Figure B.6 and e, r = 0.62 and 0.65, respectively, P < 2.2 ×

10−16) and the 5′ UTR (r = 0.2; data not shown). These analyses suggest that 186

enhanced Stau1-WT binding within CDS regions is primarily driven by GC content and secondary structure–forming propensity of the CDS itself, rather than by interactions of 3′ UTR–bound Stau1 with CDS-bound ribosomes.

From the above data, we conclude that Stau1 interacts to varying extents with the CDS and 3′-UTR regions of all cellular mRNAs in a manner dependent on their secondary structure–forming propensities. Further, the observed correlation between dsRBD-dependent Stau1 occupancy in CDS and 3′-UTR regions mainly reflects similar GC content between the CDS and 3′ UTR in individual genes rather than any direct effect of 3′-UTR binding on CDS binding. Instead, Stau1-

WT occupancy on CDS regions appears to be driven by a combination of direct interactions with CDS secondary structures and its dsRBD-independent association with actively translating ribosomes.

187

188

Figure B.4 Examples of 3'-UTR non-Alu Staufen-binding sites

(a–d) Read distributions for RNA-seq (green), Stau1-WT CROSS (yellow) and Stau1-

WT FOOT (brown) libraries on the 3′ UTRs of LMBR1 (NM_022458.3) (a), TEP1

(NM_007110.4) (b), IGF2BP1 (NM_006546.3) (c) and MDM2 (NM_002392) (d) (left) together with the corresponding centroid secondary structure colored for base-pairing probability as predicted by the Vienna folding package (right). Numbers below the Stau1

WT FOOT track and in the predicted secondary structure correspond to Stau1-binding sites.

189

Gene ontology analysis

To assess whether any particular gene classes were specifically enriched for dsRBD-dependent Stau1 binding, we performed gene ontology analysis using

GeneCodis(Carmona-Saez et al., 2007; Nogales-Cadenas et al., 2009; Tabas-

Madrid et al., 2012b)(Supplementary Table 3). We obtained the most significant associations for the 469 genes having the highest WT/mut CDS cross-linking ratios (>1.9) and the 515 genes exhibiting high WT cross-linking to strongly distal

3′ UTRs. Both sets were highly enriched in transcription-regulatory proteins (P =

7.1 × 10−13 and P = 1.1 × 10−13, respectively). Among transcription- factor types, C2H2 zinc-finger proteins were the most enriched (P = 4.5 × 10−6), with homeobox and high-mobility group (HMG) proteins following close behind (P =

1.3 × 10−5 and 6.9 × 10−5, respectively). Consistently with the strong correlation between Stau1 CDS and 3′-UTR occupancy, transcription-regulatory proteins were also highly enriched among our 373 non-Alu 3′-UTR targets (P = 6.9 ×

10−7). Thus Stau1 may have a role in post-transcriptional regulation of transcription factors. Also enriched in the non-Alu and extended 3′-UTR targets

(P = 0.001 and P = 5.0 × 10−5, respectively), but not in the 469 high CDS targets, were proteins involved in cell-cycle control.

190

Figure B.5 Stau1 binding to 3'-UTRs correlates with GC content and predicted secondary-structure free energy

191

(a) Per-gene scatter plot of 3′-UTR predicted secondary-structure free energy normalized by the length of the 3′ UTR (kcal/mol/nucleotide) against Stau1 WT/mut 3′-

UTR ratio (log2). (b) Per-gene scatter plot of 3′-UTR GC content (%) against Stau1

WT/mut 3′-UTR ratio (log2) with associated Spearman correlation and calculated P values. Red and yellow dots correspond to called Alu and non-Alu binding sites, respectively (n = 2 biological replicates).

192

Functional consequences of varying Stau protein levels

To directly test the functional consequences of Stau1 binding, we next varied intracellular Stau1 concentration (Figure B.7A-B). Transduction of HEK293 FLP- in cells with a lentivirus expressing an anti-Stau1 short hairpin RNA (shRNA) stably reduced endogenous Stau1 to ~20% normal levels (UNDER, Figure

B.7.A). Incubation of our stably integrated Flag–Stau1-WT cells overnight (16 h) with a high level of doxycycline induced transgene overexpression by 300–400% relative to endogenous Stau1 (OVER). We then assessed effects of Stau1 depletion or overexpression by preparing cytoplasmic poly(A)+ RNA-seq and ribo-seq libraries.

RNA-seq and ribo-seq read counts on individual genes were highly correlated both between biological replicates (r ≥ 0.98; E.P.R., unpublished data) and between UNDER and OVER samples (r ≥ 0.98; Supplementary Fig. 7). Other than STAU1 itself, there were no clear outlier genes between UNDER and OVER conditions for either RNA-Seq or ribo-seq. Further, no significant changes in alternative- splicing patterns could be detected (data not shown); thus, at least in

HEK cells, binding of Stau1 in introns is of little apparent consequence for pre- mRNA splicing. However, small negative correlations between RNA levels and

Stau1 levels could be detected when we ordered transcripts by CDS GC content or preferential Stau1-WT CDS cross-linking (Figure B.7.C-D). That is, transcripts with high CDS GC content (which drives greater Stau1-WT CDS binding) exhibited slightly lower cytoplasmic mRNA abundances when Stau1 was 193

overexpressed than underexpressed.

The strongest observable effect of varying Stau1 concentration was on ribosome occupancy. Cumulative histograms revealed positive relationships between ribosome occupancy and both Stau1-WT CDS cross-linking and CDS GC content across the entire transcriptome (Figure B.7.C-D; Spearman correlation r

= 0.21 and r = 0.34, respectively, P < 2.2 × 10−16). That is, genes with higher

Stau1 CDS occupancy and higher CDS GC content exhibited increased ribosome occupancy upon Stau1 overexpression compared to Stau1 knockdown; conversely, genes with lower Stau1 CDS occupancy and CDS GC content exhibited decreased ribosome occupancy upon Stau1 overexpression compared to Stau1 knockdown. This suggests that higher Stau1 protein levels increase ribosome occupancy on high-GC-content transcripts at the expense of low-GC- content transcripts. Ontology analysis of the 400 genes exhibiting the greatest increase in ribosome occupancy between UNDER and OVER conditions revealed significant enrichments for transcription-regulatory proteins (P = 0.004) and zinc-binding proteins (P = 1.1 × 10−6; Supplementary Table 3), the same terms obtained above for genes exhibiting the highest CDS and extended 3′-UTR

Stau1 occupancies.

194

Figure B.6 Stau1 occupancy on the CDS strongly correlates with GC content and predicted secondary-structure free energy

195

(a) Composite plot of the distribution of sequencing reads across the 5′ UTR, CDS and 3′

UTR of called Stau1-target genes for RNA-seq (red), Stau1-WT CROSS (blue) and

Stau1-mut CROSS (black) libraries. (b) Per-gene scatter plot (log10) of total CDS Stau1-

WT CROSS read counts/total CDS Stau1-mut CROSS reads counts (CDS ratio) versus the ratio of Stau1-WT CROSS and Stau1-mut CROSS read counts under called 3′-UTR

Stau1-WT CROSS peak positions (3′-UTR peak ratio). Red and yellow dots correspond to called Alu- and non-Alu–binding sites, respectively. (c) Per-gene scatter plot of 3′ UTR against CDS predicted secondary-structure free energy normalized by the length of the

3′ UTR (kcal/mol/nucleotide). (d) Per-gene scatter plot of 3′ UTR against CDS GC content (%). (e) Per-gene scatter plot of CDS predicted secondary-structure free energy normalized by the length of the CDS (kcal/mol/nucleotide) against Stau1 WT/mut 3′-UTR ratio (log2). (f) Per-gene scatter plot of CDS GC content (%) against Stau1 WT/mut 3′-

UTR ratio (log2). All correlation coefficients and P values were calculated with the

Spearman rank correlation (n = 2 biological replicates) throughout figure.

196

Although we observed the strongest effects of varying Stau1 protein levels for genes with high Stau1 CDS occupancy, we also examined the effects of Stau1 over- and underexpression on our 3′-UTR non- Alu and Alu target sets.

Ribosome occupancy increased slightly on non-Alu 3′-UTR targets (10% change from UNDER to OVER; P = 0.00005) when compared to the total population, whereas their mRNA levels decreased slightly (−2% change from UNDER to

OVER; P = 0.01). Thus, non-Alu targets behaved like high-GC-content mRNAs.

Conversely, Alu targets exhibited no significant change in ribosome occupancy, but their cytoplasmic mRNA levels increased upon Stau1 upregulation (+8% change from UNDER to OVER; P = 0.03; Figure B.7.E). Therefore, mRNAs containing 3′-UTR inverted Alu pairs behave differently from other cellular mRNAs in response to Stau1 abundance. For the strongly distal 3′-UTR Stau1- binding sites, we detected no significant effect of Stau1 expression on either mRNA levels or ribosome occupancy (E.P.R., unpublished data), possibly because such isoforms represent only a minor fraction of transcripts from individual loci.

To confirm that changes in Stau1 levels are of little consequence for levels of mRNAs with 3′ UTR–binding sites, we performed quantitative reverse- transcription PCR (qRT-PCR) on several Alu and non- Alu 3′-UTR target mRNAs including Arf1, a previously identified SMD target (Supplementary Fig. 8).

Consistently with our RNA-seq results, neither downregulation nor overexpression of Stau1 had a significant impact on the abundance of tested 197

targets (Supplementary Fig. 8a). Experiments performed in two other cell lines

(Huh7 and SK-Hep1) in which either Stau1 or Stau2 or both were downregulated yielded similar results (Supplementary Fig. 8b,c).

Taken together, our results indicate that Stau1 binding to the CDS results in increased ribosome occupancy and in decreased mRNA levels proportionate to both the amount of bound Stau1 and the GC content of the target mRNA.

Further, at least in the cell lines we tested, Stau1 binding within the 3′ UTR appears to be of little or no consequence for translation efficiency or steady-state mRNA levels.

198

Figure B.7 Consequences of Stau1 binding on RNA levels and ribosome density

199

(a) Inset, western blot of lysates from control cells, cells expressing anti-Stau1 shRNA and cells overexpressing Flag–Stau1-WT. Bar graph, quantitation of inset blot (n = 2).

Uncropped gel image is shown in Supplementary Figure 7a. (b) Workflow for ribosome profiling and RNA-seq analysis for cells expressing anti-Stau1 shRNA or overexpressing

Flag–Stau1-WT. (c) Cumulative plots of cytoplasmic RNA levels (left) and ribosome density (right) fold change (log2) between cells overexpressing Flag–Stau1-WT and cells expressing anti-Stau1 shRNA, based on Stau1 WT/mut CDS ratio. (d) Same as c but based on CDS GC content. (e) Box-plot representation of mRNA levels (left) and translation efficiency (right) fold change (log2 scale) between cells overexpressing Flag–

Stau1-WT and cells expressing anti-Stau1 shRNA for genes lacking Stau1 3′ UTR– binding sites (nontargets), genes with 3′-UTR inverted Alu target sites (Stau1 Alu targets) and genes with 3′-UTR non-Alu Stau1-target sites (Stau1 non-Alu targets). The upper and lower 'hinges' correspond to the first and third quartiles. The upper whisker extends from the hinge to the highest value within 1.5× interquartile range (or distance between the first and third quartiles) of the hinge. The lowest whisker extends from the hinge to the lowest value within 1.5× interquartile range of the hinge. Data points beyond the end of the whisker correspond to outliers. All P values were calculated with the

Kolmogorov–Smirnov test (n = 2 biological replicates) throughout figure.

200

Figure B.8 Model of Stau1 RNA binding and its functional role in translation

Stau1 interacts with both actively translating ribosomes and secondary structures in

CDS and 3′-UTR regions. Some 3′ UTRs contain highly complex secondary structures

(for example, inverted Alu pairs) that serve as kinetically stable Stau1-binding sites.

However, Stau1 also makes transient interactions with smaller secondary structures throughout CDS and 3′-UTR regions. Formation of these structures is a function of overall CDS and 3′-UTR GC content. Whereas interaction of Stau1 with 3′-UTR Alu pairs has a small positive effect on cytoplasmic mRNA levels, high Stau1 CDS occupancy both increases ribosome density and slightly decreases cytoplasmic mRNA levels.

201

Methods

Plasmids and cell lines pcDNA5-TetO-Flag was previously described(Ivanov et al., 2008). A cDNA encoding Stau1 (HindIII-NotI) was inserted into the polylinker of pcDNA5-TetO-

Flag. A cDNA encoding the Stau1 mutant lacking RNA-binding activity was created by PCR using primers carrying the mutations described in ref. 2, in which phenylalanines at position 216 in dsRBD3 and at position 319 in dsRBD4 were mutated into alanines.

Stable cell lines were generated as described in ref. 38. In these cells, the expression level of the stably integrated Flag-tagged protein was optimized by titration of doxycycline (Dox; 0–2,000 ng ml−1) to determine a concentration at which exogenous protein expression levels were comparable to those of endogenous counterparts.

Generation of Stau1-knockdown cell line

HEK293T LentiX cells (Clontech) were transfected with pGIPZ encoding shRNAs directed against Stau1 (Open Biosystems, CloneID: V2LHS_42695), pPAX2 and pMD2.G at a 12:9:3 ratio with Lipofectamine 2000 (Invitrogen). 2 d after transfection, the supernatant of transfected cells was collected and passed through a 0.45-µm filter. To generate the Stau1-knockdown cell line, HEK293

TRex cells (Invitrogen) were transduced with 7 mL of the supernatant of lentiviral- producing cells in the presence of 10 µg ml−1 polybrene for 6 h. Transduced cells were then selected in the presence of puromycin (3 µg ml−1) for 2 weeks. 202

Stau1 RIPiT

The procedure was performed essentially as described in (ref.(Ivanov et al.,

2008)). For each Staufen purification, TRex-HEK293 cells containing a stable copy of Flag-tagged Stau proteins (Stau1-WT and Stau1-mut) or control cells

(expressing Flag tag only) were grown in four 15-cm plates. Expression of the

Flag-tagged protein was induced with doxycycline for 16 h. 1 h before cell harvesting, cycloheximide (CHX) was added to 100 µg ml−1. The monolayer was rinsed and harvested in phosphate-buffered saline (PBS) containing 100 µg ml−1

CHX. The cells were lysed in 3 ml hypotonic lysis buffer (20 mM Tris-HCl, pH 7.5,

15 mM NaCl, 10 mM EDTA, 0.5% NP-40, 0.1% Triton X-100, 1× EDTA-free protease inhibitor cocktail (ROCHE), and 100 µg ml−1 CHX) for 10 min on ice.

The suspension was sonicated (Branson Digital Sonifier-250) at 40% amplitude with a Microtip for a total of 20 s (in 2-s bursts with 10-s intervals). NaCl was adjusted to 150 mM, and the lysate was cleared by centrifugation at 15,000g for

10 min at 4 °C. The lysate was incubated for 2 h at 4 °C with 420 µl of anti-Flag agarose beads (50% slurry, Sigma) prewashed twice with 10 ml isotonic wash buffer (IsoWB) (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.1% NP-40). The

RNA–protein (RNP) complexes captured on beads were washed four times (4 ×

10 ml) with 10 ml IsoWB. After the fourth wash, bound RNP complexes were incubated with one bed volume of IsoWB containing 1 U µl−1 of RNase I for 10 min at 37 °C with intermittent shaking. RNP complexes were again washed four times with 10 ml IsoWB. Flag epitope–containing complexes were affinity eluted 203

from the beads in one bed volume of IsoWB containing 250 µg.ml−1 Flag peptide with gentle shaking at 4 °C for 2 h. To prepare the recovered elution for input into a second IP, its volume was adjusted to 400 µl and its composition adjusted to that of the lysis buffer above with NaCl at 150 mM. The suspension was incubated with 7 µg of anti-Stau1 antibody (ab105398, Abcam, validation of IP in

Supplementary Fig. 1c,d) precoupled to 35 µl of Protein G Dynabeads

(Invitrogen) according to the manufacturer's instructions. Immunoprecipitation was carried out at 4 °C for 2 h. Captured RNP complexes were washed six times with 1 ml of ice-cold IsoWB and eluted with 200 µl of clear sample buffer (100 mM Tris-HCl, pH 6.8, 4% SDS, 10 mM EDTA and 100 mM DTT) at 25 °C for 5 min and subsequently at 95 °C for 2 min.

For IPs under protein–cross-linking conditions, cells were collected and rinsed once in PBS + CHX and then resuspended in PBS + CHX. Formaldehyde was added to 0.1%, and the suspension was gently mixed at RT for 10 min. A one- tenth volume of quenching buffer (2.5 M glycine, and 25 mM Tris base) was added. Cells were pelleted and lysed in hypotonic lysis buffer supplemented with

0.1% SDS and 0.1% sodium deoxycholate. Sonication after cell lysis was performed at 40% amplitude with a Microtip for a total of 90 s (in 5-s bursts with

30-s intervals). After Flag IP as described above, IP samples were washed twice with IsoWB + 0.1% SDS and 0.1% sodium deoxycholate and then with IsoWB.

All subsequent steps were as above with omission of RNase I treatment.

204

RIPiT RNA extraction

The volume of RIPiT elution was extracted as described in ref. (Ivanov et al.,

2008). For Stau1 cross-linked RIPiT experiments, extracted RNAs were depleted of rRNA before cDNA library construction with the Ribozero rRNA-removal kit from Epicentre.

Preparation of samples for RNA-Seq

75 µg of total RNA were poly(A)-selected with the Dynabeads mRNA-purification kit (Invitrogen). After poly(A) selection, mRNAs were fragmented with RNA- fragmentation buffer (Ambion) for 4 min and 30 s at 70 °C to obtain fragments

100–125 nt long. After fragmentation, RNAs were precipitated in three volumes of 100% ethanol at −20 °C overnight. After a wash with 70% ethanol, RNA was resuspended in 5 µl of water and the 3′ ends dephosphorylated with PNK (New

England BioLabs) for 1 h at 37 °C. After this, RNAs were subjected to cDNA library preparation.

Poly(A)-site sequencing (PAS-Seq)

75 µg of total RNA were poly(A)-selected with the Dynabeads mRNA purification kit (Invitrogen). After poly(A) selection, mRNAs were fragmented with RNA fragmentation buffer (Ambion) for 5 min at 70 °C to obtain fragments 60–80 nt long. After fragmentation, RNAs were reverse transcribed with an anchored and barcoded oligo(dT)21VN (where V corresponds to A, C or G residues) containing 205

the sequences complementary to Illumina's paired-end primers (PE1.0 and

PE2.0) separated by a polyethylene glycol (PEG) spacer. After reverse transcription (RT) with Superscript III (Invitrogen), cDNAs were size-selected and circularized with Circligase I (Epicentre) for 4 h at 60 °C. This was followed by inactivation at 80 °C for 10 min. After circularization, cDNAs were PCR-amplified with Illumina's PE1.0 and 2.0 primers for a total of 14 cycles. PCR products were size-selected on a nondenaturing polyacrylamide gel and sent for high- throughput sequencing on Illumina's HiSeq2000 platform.

Quantitative RT-PCR

Total RNA from HEK293, Huh7 and HepG2 cells transfected with control shRNA or a Stau1-targeting shRNA construct was reverse transcribed with the Vilo RT kit from Life Technologies. Obtained cDNAs were then used as input for quantitative PCR analysis as described in ref. 48 with the following primers:

GAPDH forward, 5′ tccaccaccctgttgctgtag 3′ and reverse,

5′acccactcctccacctttgac3′; Arf1 forward, 5′ atcttcgcctcccgactc 3′ and reverse,

5′atgcttgtggacaggtgga3′; C11orf58 forward, 5′ cagacgacgatctgggatct 3′ and reverse, 5′ tgatctcctataacaagacgaccag 3′; PAICS forward, 5′ aaggaaaagctgcaatctcaa 3′ and reverse, 5′ ccccacattttctggtgaag 3′; MDM2 forward, 5′ catgcctgcccactttaga 3′ and reverse, 5′ ggaggctcccaactgctt 3′; MDM4 forward, 5′ agggatgaaatgcttcttgg 3′ and reverse, 5′ aaggttgctatgaggtctaccttg 3′.

206

Sucrose-gradient sedimentation of Flag–Stau1-WT cells.

HEK293 cells were plated at 5 million in a 150-mm2 plate and Flag–Stau1-WT expression induced overnight with doxycycline at 0.5 ng ml−1. 16 h after induction, cells were either incubated with cycloheximide (100 µg ml−1) for 10 min or with harringtonine (2 µg ml−1) for either 3, 10 or 40 min. This was followed by incubation with cycloheximide (100 µg ml−1) for 10 min. Cells were then washed in PBS + cycloheximide (100 µg ml−1) or PBS + harringtonine (2 µg ml−1) + cycloheximide (100 µg ml−1) and scraped. Cells were then lysed in 1 ml of lysis buffer (10 mM Tris-HCl, pH 7.5, 5 mM MgCl2, 100 mM KCl, 1% Triton X-100, 2 mM DTT, 100 µg/ml cycloheximide and 1× Protease-Inhibitor Cocktail EDTA-free

(Roche). Lysate was homogenized by gentle pipetting up and down with a P1000 pipettor for a total of eight strokes and incubated at 4 °C for 10 min. The lysate was centrifuged at 1,300g for 10 min at 4 °C; the supernatant was recovered.

After this, samples were loaded on top of a 10–50% (w/v) sucrose gradient (20 mM HEPES-KOH, pH 7.4, 5 mM MgCl2, 100 mM KCl, 2 mM DTT, and 100 µg ml−1 cycloheximide) and centrifuged in a SW-40ti rotor at 35,000 r.p.m. for 2 h 40 min at 4 °C. Samples were fractionated into 14 individual samples that were subjected to SDS-PAGE to monitor Stau1 (ab105398, Abcam, 1:1,000 dilution) and Rpl26 (Bethyl, A300-685A, 1:1,000 dilution) levels by western-blotting.

Experimental validation of antibodies used for western blots can be found at the manufacturers' websites. 207

Oligo(dT) pulldown of poly(A) RNAs after UV exposure of living cells

HEK293 Flp-In cells were exposed to 0, 0.2, 0.4 and 0.8 J cm−2 of 254-nm UV light. Cells were then scraped and lysed with binding buffer (0.5 M NaCl, 10 mM

Tris-HCl, pH 7.5, 0.5% SDS, 0.1 mM EDTA and protease inhibitor cocktail). Cell lysates were passed through a 22-gauge syringe needle five times and spun at

15,000g for 10 min. Cleared cell lysates were added to oligo(dT) cellulose beads

(Ambion, AM10020) previously washed in binding buffer. After 1 h of incubation at room temperature, beads were washed three times with binding buffer and once with nondenaturing wash buffer (0.5 M NaCl, 10 mM Tris-HCl, pH 7.5, 0.1%

NP-40, 0.1% Triton-X 100 and 0.2 mM EDTA). Finally, beads were resuspended in elution buffer (10 mM Tris-HCl, pH 7.5, and 1 mM EDTA) complemented with

RNase A/T1 cocktail (Ambion AM2286) and incubated at 37 °C for 1 h. The supernatant was finally recovered and loaded on a 12% SDS-PAGE for western- blotting of hnRNPA1 and Stau1.

Stau1 RNA reassociation test

HEK293 Flp-In, Flag–Stau1-WT or Flag–Stau1-mut cells were harvested and lysed as described in the Stau1 RIPiT section (above). After cell lysis, 0.1 pmol of a radiolabeled in vitro–transcribed Arf1 3′-UTR sequence (labeled with [α-

32P]UTP) was added for each 150-mm2 plate. The remaining procedure is identical to that described in the Stau1 RIPiT section until the Flag elution step. 208

Flag eluates on each sample were monitored for radioactivity with liquid scintillation.

Ribosome profiling

HEK293 Flag–Stau1-WT cells incubated in the presence of 1 ng ml−1 of doxycycline (Stau1 overexpression) or not (control) as well as HEK293 cells expressing the shRNAs against Stau1 (Stau1 knockdown) were seeded at 5 million cells in a 150-mm dish. After 16 h of culture, cycloheximide was added to

100 µg ml−1 for 10 min. Cells were then washed two times in ice-cold PBS + cycloheximide (100 µg ml−1) and scraped in 1 ml of PBS + cycloheximide (100 µg ml−1). Cells were pelleted at 500g for 5 min at 4 °C and lysed in 1 ml of lysis buffer (10 mM Tris-HCl, pH 7.5, 5 mM MgCl2, 100 mM KCl, 1% Triton X-100, 2 mM DTT, 100 µg/ml cycloheximide and 1× Protease-Inhibitor Cocktail EDTA-free

(Roche)). Lysate was homogenized with a P1000 pipettor by gentle pipetting up and down for a total of eight strokes and incubated at 4 °C for 10 min. The lysate was centrifuged at 1,300g for 10 min at 4 °C, the supernatant recovered and the absorbance at 260 nm measured. For the footprinting, 5 A260 units of the cleared cell lysates were incubated with 300 units of RNase T1 (Fermentas) and

500 ng of RNase A (Ambion) for 30 min at RT. After this, samples were loaded on top of a 10–50% (w/v) linear sucrose gradient (20 mM HEPES-KOH, pH 7.4, 5 mM MgCl2, 100 mM KCl, 2 mM DTT and 100 µg ml−1 of cycloheximide) and centrifuged in a SW-40ti rotor at 35,000 r.p.m. for 2 h 40 min at 4 °C. 209

Samples were then collected from the top of the gradient while absorbance was measured at 254 nm and the fraction corresponding to 80S monosomes recovered. The collected fraction was complemented with SDS to 1% final and proteinase K (200 µg ml−1) and then incubated at 42 °C for 45 min. After proteinase K treatment, RNA was extracted with one volume of phenol (pH

4.5)/chloroform/isoamyl alcohol (25:24:1). The recovered aqueous phase was supplemented with 20 µg of glycogen, 300 mM sodium acetate, pH 5.2, and 10 mM MgCl2. RNA was precipitated with three volumes of 100% ethanol at −20 °C overnight. After a wash with 70% ethanol, RNA was resuspended in 5 µl of water and the 3′ ends dephosphorylated with PNK (New England BioLabs) in MES buffer (100 mM MES-NaOH, pH 5.5, 10 mM MgCl2, 10 mM β-mercaptoethanol and 300 mM NaCl) at 37 °C for 3 h. Dephosphorylated RNA footprints were then resolved on a 15% acrylamide (19:1), 8 M urea denaturing gel for 1 h 30 min at

35 W and fragments ranging from 26 nt to 32 nt size-selected from the gel. Size- selected RNAs were extracted from the gel slice by overnight nutation at RT in

RNA elution buffer (300 mM NaCl, and 10 mM EDTA). The recovered aqueous phase was supplemented with 20 µg of glycogen, 300 mM sodium acetate, pH

5.2, and 10 mM MgCl2. RNA was precipitated with three volumes of 100% ethanol at −20 °C overnight. After a wash with 70% ethanol, RNA was resuspended in 5 µl of water and subjected to cDNA library construction.

The remaining undigested cell lysates were extracted with an equal volume of phenol (pH 4.5)/chloroform/isoamyl alcohol (25:24:1). The recovered aqueous 210

phase was supplemented with 20 µg of glycogen, 300 mM sodium acetate, pH

5.2, and 10 mM MgCl2. RNA was precipitated with three volumes of 100% ethanol at −20 °C overnight. After a wash with 70% ethanol, RNA was resuspended in 9 µl of water and fragmented with RNA fragmentation buffer

(Ambion) at 70 °C for 4 min 30 s in order to obtain RNA fragments of 100–150 nt.

Fragmented RNAs were supplemented with 20 µg of glycogen, 300 mM sodium acetate, pH 5.2, and 10 mM MgCl2 and precipitated with three volumes of 100% ethanol at −20 °C overnight. After a wash with 70% ethanol, RNA was resuspended in 5 µl of water and dephosphorylated as described above with

PNK. After 3′-end dephosphorylation, RNA fragments were subjected to cDNA library construction.

Illumina cDNA library construction cDNA libraries were prepared with a homemade kit (E.E.H., unpublished data).

Briefly, RNA fragments with a 3′-OH were ligated to a preadenylated DNA adaptor. Following this, ligated RNAs were reverse transcribed with Superscript

III (Invitrogen) with a barcoded reverse-transcription primer that anneals to the preadenylated adaptor. After reverse transcription, cDNAs were resolved in a denaturing gel (10% acrylamide and 8 M urea) for 1 h and 45 min at 35 W. Gel- purified cDNAs were then circularized with CircLigase I (Epicentre) and PCR- amplified with Illumina's paired-end primers 1.0 and 2.0 for 5 cycles (ribosome footprints), 12 cycles (Stau1 RIPiT) or 16 cycles (RNA-Seq libraries). PCR 211

amplicons were gel-purified and submitted for sequencing on the Illumina HiSeq

2000 platform.

Mapping of high-throughput sequencing reads

First, reads were split with respect to their 5′-barcode sequence. After this, 5′- barcode and 3′-adaptor sequences were removed from reads. Reads were then aligned to University of California, Santa Cruz (UCSC) human hg18 assembly with TopHat (Trapnell et al., 2009). Unmapped reads from TopHat were then mapped with Bowtie50 to a custom set of sequences including 18S, 28S, 45S, 5S and 5.8S rRNA, repeat elements, small-nuclear RNAs (snRNAs), tRNAs, microRNAs and pre-microRNAs.

Transcript-level quantification and normalization for all high- throughput sequencing libraries

Read counts from all high-throughput sequencing libraries were normalized to the total number of mapped reads. When a single read aligned across the boundary of two different regions (for example, CDS and 3′ UTR), the read was divided proportionally to the aligned length in the given region. To quantify gene expression, reads per kilobase per million of mapped reads (RPKM) were calculated for the most abundant isoform of each gene.

212

Transcriptome-wide pairing of inverted and tandem Alu elements

Genomic coordinates for all Alu elements were obtained from the repeat-masker track of the UCSC genome browser. With the BedTools (Quinlan and Hall, 2010) intersectBed function with –s(strand) option, we obtained coordinates of all Alu elements located in 3′ UTRs, distal 3′ UTRs and introns. After this step, Alu elements in Watson and Crick strands were separated. To pair inverted Alu elements, we used the ClosestBed function between Alu elements in the Watson and Crick strands. The same steps were also performed to detect same-strand pairs on Watson-Watson and Crick-Crick pairs. In this case, tandem pairs that had an inverted Alu pair less than 2,000 nt apart were excluded from the analysis.

Definition of distal 3′-UTR regions

To define distal 3′-UTR regions, we used PAS-seq mapped reads. For this, peaks were called from the PAS-Seq library with ASPeak. With the BedIntersect function (BedTools) we found all genes that had a polyadenylation site within

10,000 nt of the canonical polyadenylation site, provided that they were upstream of the transcription start site of the downstream gene. If multiple peaks were called within that interval, the called peak most distal to the canonical polyadenylation site was selected.

213

Counting of sequencing reads for Alu pairs separated by different distances

With the genomic coordinates of tandem and inverted Alu pairs described above, we created a new file for each region (3′ UTR, distal 3′ UTR and introns) that contained the genomic coordinates of each Alu element within every pair in addition to 1,000 nt upstream of the 5′-most Alu element and 1,000 nt downstream of the 3′-most Alu element. After this, read counts from Stau1-WT

CROSS, Stau1-mut CROSS and RNA-seq were obtained for each defined region within pairs and then normalized with RNA-seq RPKM value for the same interval. Normalized read counts for each interval were added for all Alu pairs located between 0 and 200 nt, 200 and 500 nt, 500 and 1,000 nt, 1,000 and

2,000 nt, and 2,000 nt and beyond.

Secondary-structure analysis of inverted Alu pairs

To analyze the secondary structure of inverted Alu pairs, we obtained the sequence of both Alu elements, including the region separating them. As a control, same-length sequences were randomly chosen from the list of genes devoid of Stau1-target sites. After the sequences were obtained, the Vienna

Package RNAfold 2.1.1 (ref. (Gruber et al., 2008) ) was used to predict secondary structures.

214

Analysis of ribosome-profiling reads

To calculate translational-efficiency changes upon knockdown or overexpression of Stau1, we used a generalized linear model (GLM). We used the number of sequencing reads mapping to the annotated coding region (or ORF) for each

RefSeq transcript. In the GLM, we used the cell type (overexpression of

Stau1WT, Stau1-shRNA), sequence-library preparation batch and type of sequence data (RNA-seq or ribo-seq) as predictor variables of the number of mapped reads per transcript. We had two biological replicates for all conditions, which were used to estimate a biological variability in the number of counts. We used a trended dispersion estimation method, following ref. (McCarthy et al.,

2012). To extract translation-efficiency changes upon a given treatment, we used the contrast between type of sequence data, ribo-seq versus RNA-seq, in each pair of conditions.

Accession codes

High-throughput sequencing data corresponding to native and cross-linked Stau1

RIPiT experiments as well as PAS-seq, RNA-seq and ribo-seq have been deposited in the GEO database under accession number GSE52447.

215

Discussion Like many RNA-binding factors, the Drosophila and mammalian Staufen proteins have been implicated in multiple post-transcriptional processes including alternative splicing(Ravel-Chapuis et al., 2012b), RNA localization(Kiebler et al.,

1999; Macchi et al., 2004; Micklem et al., 2000; St Johnston et al., 1991), translational activation(Dugré-Brisson et al., 2005) and translation-dependent mRNA decay(Cho et al., 2012; Gleghorn et al., 2013; Kim et al., 2005; Kretz et al., 2013; Park et al., 2013; Vessey et al., 2008). Which activity is observed depends on the cellular context, the identity of the bound RNA and the location of the binding site on the target RNA. Many of Staufen’s previously documented activities parallel those of the EJC(Cho et al., 2012; Dugré-Brisson et al., 2005;

Ferrandon et al., 1994; Ghosh et al., 2012; Gleghorn et al., 2013; Kim et al.,

2005; Nott et al., 2004; Wiegand et al., 2003). To better understand EJC function, we recently determined the complete EJC RNA-binding landscape in HEK293 cells(Singh et al., 2012). Here we undertook the same analysis for Stau1.

Up to now, confirmed Staufen-binding sites were limited to a few well- characterized structures(Ferrandon et al., 1994; Kim et al., 2005). Broader identification of Staufen- associated mRNAs has been attempted in various organisms by combination of native RIP protocols with microarray analysis(Furic et al., 2008; Kusek et al.; Laver et al., 2013; Maher-Laporte and 216

DesGROSEILLERS, 2010). Unfortunately, however, such methodologies have yielded no consensus as to general features of Staufen targets. One recent study of Staufen-associated mRNAs from Drosophila oocytes reported enrichment of three different secondary-structural motifs that might explain Staufen binding specificity in flies(Laver et al., 2013). However, the authors were unable to identify similar structural motifs among human Staufen-associated mRNAs from available native mammalian Stau1 and Stau2 RIP microarray data(Furic et al.,

2008). We show here that human Stau1 generally associates with actively translating ribosomes; therefore, it is impossible to discriminate between sites of direct Stau1-mRNA interaction via dsRNA binding and sites of indirect Stau1- mRNA association via elongating ribosomes without some sort of footprinting approach. Further, because of (i) Stau1’s strong ribosome association, (ii) the prevalence of kinetically labile Stau1-binding sites in vivo and (iii) Stau1’s ability to form new interactions with dsRNA after cell lysis, native RIP experiments are likely to be biased toward both highly translated mRNAs and RNAs containing the most stable sites of direct Stau1-dsRNA interaction. Our experimental design, which combined formaldehyde cross-linking and fragmentation of Stau1- associated RNAs, using both WT and mut proteins, allowed us to both avoid binding-site reassortment after cell lysis and discriminate between binding modes that do or do not require Stau1 dsRBD functionality.

The majority of non-rRNA reads in our cross-linked libraries mapped sense to 3′

UTRs and CDS regions. Within 3′ UTRs, we identified numerous high-occupancy 217

Stau1-binding sites composed of either inverted Alu pairs (Alu targets) or sequences with extremely high secondary structure–forming propensity (non-Alu targets). Observable native Stau1 footprints showed that these structures often consist of several closely spaced helices separated by short loops.

Bioinformatics analysis of the footprints, however, failed to identify any particular enriched motif (A.K. and E.P.R., unpublished data), results consistent with the idea that Staufen recognizes dsRNA in a sequence-independent manner(Marión et al., 1999; Monshausen et al., 2001; Wickham et al., 1999).

Unexpectedly, in addition to detecting strong binding to large RNA secondary structures, we also detected extensive dsRBD-dependent Stau1 cross-linking extending throughout the entire length of 3′ UTRs and CDS regions. This cross- linking strongly correlated with both GC content and per-nucleotide predicted secondary-structure strength. Because GC content in CDS and 3′-UTR regions also correlate, mRNAs exhibiting preferential Stau1-WT 3′-UTR cross-linking also tend to exhibit preferential Stau1-WT CDS cross-linking. Inverted Alu pairs, the 3′

UTRs containing them and their associated CDS regions, however, exhibit average GC content. Despite high Stau1 occupancy on such 3′ UTRs, their CDS occupancies are close to levels that would be expected from their GC content alone. We therefore conclude that the strongest feature driving dsRBD- dependent Stau1 binding within CDS regions is the secondary structure–forming propensity of the CDS itself. Thus dsRBD-dependent Stau1 binding to 3′ UTRs appears to be functionally uncoupled from dsRBD-dependent Stau1 binding to 218

CDS regions, with the correlation between 3′ UTR and CDS cross- linking driven primarily by GC-content similarity.

Our results suggest that endogenous Stau1 RNA targets can be divided into two broad classes dependent on their structural topology. One class corresponds to stable RNA secondary structures such as inverted Alu pairs and other sequences with extremely high secondary structure–forming propensity. Such elements are capable of simultaneously binding multiple Staufen molecules whose association may be further stabilized by multimerization. Close association of multiple

Staufen-binding sites would assure continuous Staufen occupancy even though individual protein molecules might come and go. It is of note that we generally detected such binding sites in annotated 3′ UTRs and extended 3′ UTRs, the latter being particularly rich in inverted Alu pairs. Recently, extended 3′ UTRs were shown to be especially prevalent in the brain. Because Stau1 is known to have a role in dendritic mRNA targeting, these stable RNA secondary structures with their long-lived Stau1 associations could well be the functional binding sites through which Stau1 promotes proper subcellular mRNA localization in neurons.

The second class consists of smaller and more labile secondary structures as might occur in GC-rich CDS regions. Here our data indicate that transient Stau1 binding, perhaps by Stau1 molecules simultaneously interacting with elongating ribosomes has a role in regulating translation. We arrived at this conclusion by analyzing cytoplasmic poly(A)+ RNA-seq and ribo-seq data from cells under- and 219

overexpressing Stau1. This allowed us to assess the effects of varying intracellular Stau1 concentration on both cytoplasmic mRNA levels and ribosome occupancy. Observable changes in mRNA levels were extremely subtle.

Consistently with recent data indicating that Stau1 binding to mRNAs containing inverted Alu elements enhances their nucleocytoplasmic export(Elbarbary et al.,

2013), we did observe a small positive effect of increasing Stau1 on cytoplasmic mRNA levels for our 3′-UTR Alu targets. Conversely, for all other sets of mRNAs exhibiting preferential Stau1-WT cross-linking, Stau1 levels negatively influenced cytoplasmic mRNA levels proportionately to CDS Stau1 occupancy but not to 3′-

UTR occupancy. Thus we could find little evidence for SMD driven by 3′ UTR– bound Stau1, either over the entire mRNA population or for previously identified

SMD targets. Instead, higher Stau1 levels led to a preferential increase in ribosome density on high-GC-content mRNAs.

We propose a model (Fig. 8) based on these findings, wherein ribosome-bound

Stau1 molecules transiently interact with short dsRNA helices throughout the

CDS and 3′ UTR. In the CDS, such interactions somehow serve to increase ribosome density. Because Stau1 interacts with actively translating ribosomes, the increase in ribosome density may reflect increased translation efficiency. One possibility is that Stau1 helps ribosomes elongate through otherwise inhibitory secondary structures by recruiting factors such as RNA helicase A (RHA or

DHX9) to disrupt them. RHA is a positive regulator of translation on mRNAs containing 5′-UTR secondary structures(Hartman et al., 2006) and is known to 220

copurify with Stau1 (ref. (Villacé et al., 2004) and E.P.R., unpublished data).

Another abundant translational-regulatory protein that binds ribosomes and cross-links throughout CDS regions is the fragile X protein, FMRP(Comery et al.,

1997). FMRP, however, is a negative regulator of translation. Whereas deletion of either FMRP or Stau1 causes neurological defects, the phenotypes are opposite: absence of FMRP leads to dendritic spine overgrowth (Gruber et al.,

2008), whereas absence of Stau1 results in fewer spines(Vessey et al., 2008).

Thus it is possible that FMRP and Stau1 have opposing roles in synaptic protein production, with FMRP inhibiting translation and Stau1 promoting it.

Finally, mRNAs encoding transcription-regulatory proteins were recently reported as being enriched in Drosophila Staufen RIP samples(Laver et al., 2013).

Consistently with this, we found that mRNAs encoding transcription factors of the

C2H2 zinc-finger, HMG and homeobox families were highly enriched among mRNAs exhibiting the highest preferential 3′-UTR and CDS Stau1-WT occupancy. Transcription factors and zinc-binding proteins were also highly enriched among the mRNAs whose ribosome density was most positively affected by Stau1 protein levels. Thus Stau1 may have a previously unrecognized role in the translational regulation of transcription-regulatory proteins.

221

APPENDIX C: Alterations in mRNA 3′UTR isoform abundance accompany gene expression changes in human Huntington’s disease brains

Appendix C was taken verbatim from the following manuscript, which is preparation during the time of this thesis submission.

Lindsay Romo, Ami Ashar-Patel, Edith Pfister, and Neil Aronin

Author contributions statement:

L.R. and N.A. conceived the project, L.R. and E.P. designed experiments, L.R. performed experiments and interpreted results, A.A.P. designed the PAS-Seq method, and L.R. wrote the manuscript with comments from all authors.

222

Introduction

Huntington’s disease (HD) is a neurodegenerative disorder caused by expansion of the CAG repeat in huntingtin (HTT) exon 1 (MACDONALD, 1993).

The expansion is dominantly inherited and fully penetrant. Individuals usually develop symptoms in their mid-thirties, and death occurs within two decades

(reviewed in {Suchowersky:k4GDVAry}). Neurodegeneration starts in the striatum and motor cortex, but spreads to other brain regions by the time of death

(Vonsattel et al., 1985). Neuronal toxicity is largely due to the effect of mutant

HTT protein during aging, although mutant messenger RNA (mRNA) may also be toxic. Many emerging therapies aim to decrease mutant HTT mRNA (Hu et al.,

2010) (Pfister et al., 2009).

There are three known isoforms of HTT mRNA. A truncated (7.9 kb) isoform is generated via termination at a cryptic polyadenylation (polyA) signal in the first intron (Sathasivam et al., 2013). This isoform is of low abundance. Its presence in HD patients and mouse models, but not controls, suggests it may be associated with disease (Sathasivam et al., 2013). The two predominant HTT mRNA isoforms are generated by alternative polyadenylation in the 3′ untranslated region (3′UTR) (Lin et al., 1993). The short (10.3kb) 3′UTR isoform predominates in dividing cells such as lymphoblasts, whereas the long (13.7kb) isoform predominates in nondividing cells such as neurons(Lin et al., 1993) . It has not been established whether HTT mRNA 3′UTR isoforms are metabolized 223

differently, or whether their relative abundance changes in HD. Although many genes exhibit expression and splicing differences in HD, 3′UTR isoform expression changes have not been investigated (Hodges et al., 2006) (Labadorf et al., 2015) (Lin et al., 2016) (Mykowska et al., 2011).

3′UTR length is important for mRNA localization, stability, and translation

(Di Giammartino et al., 2011). Many mRNAs use 3′UTR alternative polyadenylation to achieve tissue-specific expression and function (Lianoglou et al., 2013; Smibert et al., 2012; Zhang et al., 2005). Long mRNA isoforms contain binding sites for trans-acting factors such as microRNAs that exert dynamic changes in steady-state mRNA levels (Sandberg et al., 2008). Ubiquitously expressed genes, like HTT, are the most likely to use alternative polyadenylation to alter mRNA expression in the brain (Lianoglou et al., 2013; Lin et al., 1993).

Changes in mRNA 3′UTR length occur in Parkinson’s disease and cancer (Mayr and Bartel, 2009; Rhinn et al., 2012). We sought to determine if alterations in

3′UTR length are a feature of HD.

We provide evidence that the abundance of HTT mRNA 3′UTR isoforms differs between HD patients and non-HD controls, including a novel mid-length

3′UTR isoform. We show that HTT mRNA 3′UTR isoforms have different localizations, half-lives, polyA tail lengths, and microRNA sites. Isoform alterations in HD are not unique to HTT. There are widespread changes in 224

mRNA 3′UTR isoform expression in human HD motor cortex. Genes with isoform changes are associated with pathways dysfunctional in HD. We demonstrate that knockdown of the CNOT6 RNA binding protein leads to isoform shifts similar to those in HD motor cortex. Our findings indicate that altered 3′UTR isoform expression of HTT and a subset of other genes is a feature of molecular pathology in HD motor cortex.

Results The abundance of HTT mRNA 3′UTR isoforms changes in Huntington’s disease

We identified two abundant HTT mRNA 3′UTR isoforms in public 3′ sequencing data (Figure C.1.A) (Lianoglou et al., 2013). As in previous studies, we found

HTT mRNA isoform abundance varies across normal tissues: the longer isoform predominates in the brain, neuronal precursor (NTERA2) cells, breast, and ovary, whereas the shorter isoform predominates in testes, B-cells, muscle, and human embryonic kidney (HEK293) cells (Figure C.1B). (Lin et al., 1993).

To determine whether HTT mRNA isoform abundance changes in HD, we performed qRT-PCR of the HTT long isoform normalized to total HTT expression on patient cerebellum, motor cortex, fibroblasts, and neural stem cells. HD cerebellum samples were from grade 2, 3, or 4 brains, whereas motor cortex samples were from grade 1 and 2 brains (Table S1 online). HD brain grades are based on the degree of striatal degeneration and gliosis. Grades range from 1 225

(50% neuronal loss) to 4 (95% neuronal loss) (Vonsattel et al., 1985). There is limited involvement of the cerebellum in all grades (Rüb et al., 2013; Vonsattel,

2008; Vonsattel et al., 1985). Cortical pathology is variable between samples, but typically 10% of neurons are lost in grade 1 with minimal gliosis, and up to 40% are lost in grade 2 with evident gliosis (Thu et al., 2010; Vonsattel et al., 1985).

Neuronal loss can confound gene expression studies. We chose to analyze HTT isoform expression in cortex and cerebellum rather than in the striatum because neuronal loss is more limited in these regions.

We found a 1.6-fold increase in the long 3′UTR isoform relative to total HTT mRNA in HD patient cerebellum. Conversely, there is a 2.5-fold decrease in the long isoform in patient motor cortex (Figure C.1. C). Patient fibroblasts and neural stem cells also exhibit a decrease in the long isoform (Figure 1D, 2-fold and 3.3-fold respectively). These results demonstrate the abundance of HTT mRNA 3′UTR isoforms changes in HD in a tissue and cell-specific manner.

226

Figure C.1 HTT 3'UTR isoforms change their relative amounts in HD

(A) NTERA2 cell 3′seq reads mapping to the HTT 3′UTR. (B) Quantification of HTT

3′UTR isoform 3′seq reads relative to total HTT 3′UTR reads across several human tissues and cell lines. (C) qPCR of the HTT mRNA long 3′UTR isoform (long) normalized to total HTT expression (total, via ΔΔCt method) in HD patient and control cerebellum

(CB) and motor cortex (MCx). *, **, and *** signify p<0.05, 0.005, and 0.0005. (D) qPCR as in C, using patient or control fibroblasts or neural stem cells (NSCs) differentiated from fibroblasts.

227

HTT mRNA isoform changes extend to liver and muscle and arise from both alleles

We used HD model mice to determine if disease-associated isoform changes extend beyond brain. We bred mice that lack murine Htt, but harbor full length human HTT with either 18 or 128 CAG repeats on a yeast artificial chromosome

(Yac18, Yac128) (Slow et al., 2003; Zeitlin et al., 1995). Like patients, Yac128 cerebellum exhibits an increase (1.7-fold) in the long HTT 3′UTR isoform compared to yac18 mouse cerebellum. Isoform differences extend to muscle

(quadriceps) and liver, with a 2.2-fold increase and a 1.6-fold decrease in the

HTT long mRNA isoform respectively (Figure C.2.A). These findings indicate that at least two non-brain tissues exhibit tissue-specific HTT mRNA isoform changes.

Altered HTT mRNA isoform abundance in patient brains could arise from the wild-type allele, the mutant allele, or both. To determine which allele is responsible for the changes, we performed allele-specific qPCR on human cerebellum and motor cortex samples heterozygous for single nucleotide polymorphism (SNP) rs362267 in the HTT long, but not short, 3′UTR. This SNP heterozygosity is not linked to the CAG repeat; in some patients, the C allele is expanded, whereas in others the T allele is expanded. The long isoform qPCR probe matched either the C or the T SNP. These probes efficiently and specifically amplify the matching but not the nonmatching allele. The amounts of both the mutant and wild-type HTT long 3′UTR isoform are changed in HD cerebellum and motor cortex compared to controls (Figure C.2 B, C). The 228

direction of isoform changes is consistent between alleles in each patient (Figure

C.2.D, correlation coefficient 0.83). Thus the disease-associated change in HTT mRNA isoforms arises from both alleles.

229

Figure C.2 HTT isoform changes extend to muscle and liver and both HTT alleles change 3'UTR isoform amounts in HD

230

(A) qPCR of the HTT long 3′UTR isoform (long) normalized to total HTT expression (total, via ΔΔCt method) across Yac128 and Yac18 mouse tissues.

CB=cerebellum. *, **, and *** signify p<0.05, 0.005, and 0.0005. (B) Allele- specific qPCR using SNP rs362267C/T in the HTT long 3′UTR isoform (long) normalized to total HTT expression in patient and control cerebellum. (C) Allele- specific qPCR in patient and control motor cortex, as in C. (D) Correlation of HTT allele abundance within each HD or control (con) cerebellum (CB) or motor cortex (MCx) sample; r=0.83.

231

PolyA site sequencing identified a novel conserved mid-3′UTR isoform of HTT mRNA

To identify unknown 3′UTR isoforms of HTT, we used a new polyA site sequencing method, PAS-Seq (Figure C.3.A) (Ashar-Patel et al., 2017). PAS-

Seq requires minimal RNA input to produce high quality sequences, or reads.

The PAS-Seq reverse transcription primer can anneal to mRNA polyA tails and polyA sequences encoded in the genome. We excluded genomically-primed reads from our analysis (Figure C.3.B, see Experimental Procedures). PAS-Seq confirmed there is a decrease in the long isoform (1.4-fold) in HD motor cortex, whereas there is a decrease in the short isoform (1.8-fold) in HD cerebellum

(Figure C.3.G-H,J).

Using PAS-Seq, we identified a novel 12.5kb isoform of HTT expressed in both control and HD human cerebellum and motor cortex with abundance similar to the short 3′UTR isoform (Figure C.3.C, top). We performed PAS-Seq on mice to test if the mid 3′UTR isoform is conserved. We used wild-type and Q140 HD mice, which harbor 140 CAG repeats knocked into the mouse HTT genomic locus (Menalled et al., 2003). The mid-3′UTR isoform, and the short and long

3′UTR isoforms, are present in wild-type and Q140 mice (Figure C.3.C, bottom).

A polyA signal is required for mRNA cleavage and polyadenylation. There is a putative polyA signal (AAUGAA) located twenty nucleotides upstream of the mid-

3′UTR isoform polyA site (Sheets et al., 1990; Tian et al., 2005). This polyA signal and site are conserved between mice and humans (Figure C.3.D). The 232

regions up and downstream of the isoform polyA sites are also conserved (Figure

C.3.E). The abundance of the mid-3′UTR isoform varies across tissues and increases 2.8-fold in HD motor cortex compared to control (Figure C.3.F-H).

Isoform changes can’t be explained by gene expression alterations: total HTT expression is decreased by 1.4-fold in HD motor cortex but unchanged in cerebellum compared to controls (Figure C.3.I). In addition, isoform changes are still significant after normalizing to total HTT expression (Figure C.3.J).

233

Figure C.3 PAS-Seq revealed a third HTT 3'UTR isoform whose relative abundance also changes in HD 234

(A) Schematic of PAS-seq library preparation. (B) Schematic of the PAS-seq data analysis pipeline on Tbr1, an exemplary gene with two 3′UTR isoforms. See also Figure S1. (C) PAS-seq peak coordinates mapping to the HTT 3′UTR from

HD patient and control cerebellum (CB) and motor cortex (MCx), and Q140 and wild-type (WT) mouse striatum (Str). (D) Needleman-Wunsch alignment of the human (top) and murine (bottom) HTT 3′UTR around the short, mid, and long

3′UTR isoform polyA signals (blue) and sites (yellow). (E) NCBI BLAST alignment score of the human and mouse HTT 3′UTR; arrows indicate the three polyA sites at 600, 2763, and 3903 bases into the human 3′UTR. (F) HTT mid 3′UTR isoform

3′seq reads relative to total HTT 3′UTR reads across several human tissues and cell lines. (G) Reads corresponding to HTT isoforms in HD motor cortex compared to controls, normalized to sequencing depth. * signifies p<0.05. (H)

Reads counts as in (G), in HD and conrol cerebellum. (H) Total HTT expression in HD motor cortex or cerebellum compared to controls, as in (G). (I) Reads corresponding to HTT isoforms in HD motor cortex (left) or cerebellum (right) compared to controls, normalized to sequencing depth (top) or total HTT expression (bottom).

235

The HTT mRNA 3′UTR isoforms have different localization and half- lives

3′UTR length affects mRNA stability, translation, and localization (Di

Giammartino et al., 2011). To determine how shifts in HTT isoform abundance affect HTT mRNA metabolism, we assessed the localization and stability of HTT isoforms in SH-SY5Y (human neuroblastoma) cells. These cells have neuronal characteristics and express all three HTT 3′UTR isoforms, making them a model of HTT mRNA stability in human brain (Datta et al., 2013). To assay the localization of HTT isoforms, we performed isoform specific qRT-PCR on nuclear or cytoplasmic RNA from SH-SY5Y cells. Over three times as much of the short and long isoform is localized in the cytoplasm than in the nucleus, whereas the mid isoform is more abundant in the nucleus (Figure C.4.A). To determine if isoform localization changes in disease, we assayed RNA from HD and wild-type fibroblasts. As in SH-SY5Y cells, we found little difference between the localization of the long and short isoforms; however, in wild-type fibroblasts, the mid isoform was significantly more abundant in the cytoplasm than the other isoforms. The ratio of cytoplasmic to nuclear transcripts is lower in HD fibroblasts than in wild-type fibroblasts for all three isoforms, indicating HTT mRNA shifts to the nucleus in HD, as has been reported (de Mezer et al., 2011).

To test isoform stability, we incubated SH-SY5Y cells with labeled uridine, and collected labeled mRNA 0, 3, 6, 9, and 12 hours after washing away the modified nucleotide. We performed isoform specific qRT-PCR on the RNA to determine 236

the relative stability of the HTT short and long isoforms. We found all HTT isoforms have significantly different half-lives; the short isoform is most stable, whereas the mid isoform is least stable (Figure C.4.B). This finding suggests the altered relative abundance of HTT isoforms in HD brains is accompanied by a change in total HTT half-life.

The different localization and stability of the HTT 3′UTR isoforms could be due to different polyA tail lengths, microRNA sites, or RNA binding protein sites (Di

Giammartino et al., 2011; Norbury, 2013). To determine the polyA tail length of the HTT 3′UTR isoforms, we added a universal sequence to the 3′ end of the mRNA polyA tail that was used as a priming site during isoform-specific PCR.

PCR products were resolved on a gel, enabling quantification of poly-A tail length

(Figure C.4.C). We found the HTT short isoform has a poly-A tail length of about sixty in both SH-SY5Y cells and in patient and control fibroblasts, whereas the

HTT mid and long isoforms have poly-A tail lengths of about five and ten (Figure

C.4.). Recent studies found mRNA stability and rate of translation drops as the polyA tail length decreases below twenty nucleotides (Chang et al., 2014; Park et al., 2016). The very short polyA tail length of the mid and long HTT isoforms may explain their low half-lives compared to the short HTT isoform.

To identify microRNA sites exclusive to the HTT long isoform, we used

TargetScan. This program predicts microRNA target sites based on several local 237

factors known to affect microRNA binding strength and specificity (Lewis et al.,

2003). We found several microRNAs expressed in human brain with sites exclusive to the mid and long HTT 3′UTR (Figure C.4.E). To test if these microRNAs affect HTT isoform abundance, we transfected SH-SY5Y cells with mimics of microRNAs 221, 137, or both. MicroRNA 137 exclusively binds the long isoform, while microRNA 221 has an 8-mer binding site in the long isoform and a 6-mer binding site common to the mid and long isoform. We found the microRNA 137 mimic had no effect on isoform abundance, but transfection of the microRNA 221 or microRNA 221 and 137 mimics reduced the abundance of the long and mid but not short isoform relative to total HTT (Figure C.4.F) or to

GAPDH (Figure C.4.G). These results show at least one microRNA exclusively degrades the long and mid HTT transcripts, potentially contributing to the lower stability of these isoforms compared to the short isoform.

238

Figure C.4 HTT 3'UTR isoforms are metabolized differently

239

(A) Cytoplasmic versus nuclear abundance of HTT 3′UTR isoforms across cell lines. Data points are the average of triplicates with standard deviation. *, **, and

*** signify p<0.05, 0.005, and 0.0005. (B) Abundance of labeled HTT 3′UTR isoforms in SH-SY5Y cells at various times after a 12-hour eithinyl-uridine pulse.

Data points are the average of triplicates with standard deviation. (C) The polyA

Tail Length Assay adds a universal sequence to the 3′ end of the polyA tail. HTT isoform-specific PCR is then performed. The polyA tail length is the PCR product length after subtraction of the universal primer length and the distance from the forward primer to the polyA site. (D) Gel of PCR products from the polyA tail length assay. (E) Brain-expressed microRNA target sites in the HTT 3′UTR predicted by TargetScan. (F-G) Quantitative PCR of HTT 3′UTR isoforms normalized to total HTT (F) or GAPDH (G) expression in SH-SY5Y cells after transfection of microRNA mimics (miRs) with targets exclusive to the HTT long or mid and long 3′UTR isoforms. NT (non-targeting) miR is a C. Elegans microRNA mimic

240

To identify RNA binding protein sites exclusive to the long isoform, we used the

CLIPdb and starBase CLIP-Seq databases (Li et al., 2014; Yang et al., 2015).

We found several brain-expressed RNA binding proteins adhere to the HTT

3′UTR. These RNA binding proteins may differentially affect the stability and localization of HTT isoforms. Based on these results, we expect the short isoform is more translationally efficient than the mid and long isoform due to its longer polyA tail and increased stability. Indeed, a study published during preparation of this manuscript found the short HTT 3′UTR isoform may be translated more efficiently than the long isoform (Xu et al., 2017).

Many other genes exhibit changes in isoform abundance

We studied whether alterations in mRNA isoform expression are unique to HTT, or features of many genes in HD. To assay transcriptome-wide isoform expression, we performed PAS-Seq on motor cortex from grade 1 HD brains

(n=6) versus controls (n=5), and cerebellum from grade 2-4 patient brains (n=9) versus controls (n=7). We identified genes with multiple 3′UTR isoforms, and normalized isoform expression to gene expression (see Experimental

Procedures). We compared the normalized isoform expression between HD and control samples using two thresholds for significance: false discovery rate less than 10%, or p-value less than 0.01. The false discovery rate threshold corrects for false-positives due to multiple testing, but may reject true positives. We included the more sensitive p-value threshold to compare HD motor cortex, where mRNA expression changes are substantial, to HD cerebellum, where 241

changes are smaller and less likely to pass a stringent threshold (Hodges et al.,

2006).

PAS-Seq uses oligo(dT) to capture polyA tails during reverse transcription.

Oligo(dT) also anneals to genomic polyA stretches. These genomic sites are artefacts that can lead to overestimation of the number of alternatively polyadenylated genes (Sheppard et al., 2013). We found 33% of genes in the motor cortex and 30% in the cerebellum display alternative polyadenylation. This percentage is similar to that obtained via the 3Pseq 3′ sequencing method, which avoids oligo(dT) priming (Jan et al., 2011). Consistency with 3Pseq data indicates PAS-seq accurately captured reads primed from mRNA polyA tails while excluding reads primed from genomic polyA regions. As reported, we found long isoforms are more abundant than short isoforms in cerebellum and motor cortex (Miura et al., 2013).

In the motor cortex, 11% of alternatively polyadenylated genes showed a significant (false discovery rate<10%) change in abundance of at least one

3′UTR isoform in HD, while an additional 2% passed the p-value threshold

(p<0.01) (Figure C.5.A, red and orange). Many isoform relative abundances changed over 25% in HD motor cortex versus controls, indicating a major shift in isoform expression for the corresponding gene (Figure C.5.B). In the cerebellum, none of the alternatively polyadenylated genes showed a significant (false discovery rate <10%) change in abundance of at least one isoform, and just 5% 242

passed the p-value threshold (p<0.01) (Figure 5C, green). Isoform abundance changes in the cerebellum were not as large as in the motor cortex (Figure

C.5.D). These results indicate that there are widespread isoform changes in HD motor cortex that alter the 3’UTR isoform abundances for many genes, whereas changes in the cerebellum are more modest.

Neurons die and glia proliferate in the HD brain (Vonsattel and DiFiglia, 1998).

Neuronal loss and gliosis are limited in motor cortex from grade 1 brains (Thu et al., 2010). However, it is possible some or all isoform changes we see are due to changes in cell populations. We found isoform abundance is highly correlated

(r=0.87) between the motor cortex and cerebellum despite the different cell milieu and trace neuronal loss in the cerebellum. Previous studies that used HD caudate, which suffers extensive neuronal loss (50-95%), have demonstrated mRNA expression is similar in HD caudate tissue and HD caudate neurons

(Hodges et al., 2006). These results suggest the isoform changes we observe in

HD motor cortex are not solely due to neuronal loss.

To determine whether genes with significant (p<0.01) isoform changes shifted towards longer or shorter mRNA 3′UTR isoforms, we calculated a weighted change (see Experimental Procedures). If the weighted change is positive, the gene shifts towards longer mRNA isoforms in HD patients, whereas if it is negative, the gene shifts towards shorter isoforms. We found 129 genes shift 243

towards longer isoforms in HD motor cortex, whereas 110 genes shift towards shorter isoforms (Figure C.5.E). In the cerebellum, we found 31 genes shifted to the longer isoform, whereas 83 shifted to the shorter isoform (Figure C.5.F). Thus isoform shifts are gene-specific, and there is no generalized shift towards longer or shorter 3′UTR isoforms in HD.

244

Figure C.5 Transcriptome-wide PAS-Seq analysis identifies a large subset of genes with 3'UTR isoform changes in HD motor cortex

245

(A) Scatter plot of PAS-seq reads from control (n=5) and HD (n=6) motor cortex. Each dot represents a unique 3’UTR isoform expressed in at least five HD and control samples. Shown is the number of reads mapping to the isoform divided by the total reads mapping to the gene 3′UTR (isoform fraction). Orange dots indicate isoforms with p<0.01, whereas red dots indicate isoforms with adjusted p<0.1 (false discovery rate<10%). (B) Volcano plot of reads from control and HD motor cortex, colors as in A.

The percent change is calculated as the difference in the isoform fraction between HD and control samples multiplied by 100. (C) Scatter plot of PAS-seq reads from control

(n=7) and HD (n=9) cerebellum, as in A. Green dots indicate isoforms with p<0.01. (D)

Volcano plot of reads from control and HD cerebellum, colors as in C. (E) Heat map of the weighted change (see methods) of genes with one or more isoform differentially expressed in patient motor cortex (p<0.01). A positive number indicates the gene shifts towards longer isoforms in HD, whereas a negative number indicates the gene shifts towards shorter isoforms in HD. (F) Heat map of genes with one or more isoform significantly differentially expressed in patient cerebellum, as in E. (G) Comparison of gene weighted changes between patient versus control motor cortex (MCx) and cerebellum (CB). Green dots have significant changes in HD cerebellum, orange dots have significant changes in HD motor cortex, and purple dots have both (p<0.01). (H)

Gene ontology analysis of genes exhibiting significant (p<0.01) shifts to longer (long) or shorter (short) isoforms in HD motor cortex (MCx) and cerebellum (CB).

246

We compared significant (p<0.01) isoform shifts in the motor cortex and cerebellum to look for common changes. Only 11 genes exhibit mRNA 3′UTR isoform shifts in both the motor cortex and cerebellum (Figure C.5.G, purple). Of these, 9 showed opposite changes in the motor cortex, with 7 of the 9 shifting towards shorter isoforms in the cerebellum and longer isoforms in the motor cortex. This finding indicates that isoform changes are region-specific.

To identify affected pathways, we performed gene ontology analysis on genes with significant (p<0.01) isoform shifts (Figure C.5.H). We found several pathways that have been reported to be enriched among genes differentially expressed in HD, including cytokine signaling and production, RNA pol2 transcription, translation, calcium signaling, vesicle-mediated transport, microtubule organization, and DNA binding (Hodges et al., 2006; Labadorf et al.,

2015). Some enriched pathways are thought to be involved in HD pathogenesis, including calcium signaling, cytokine signaling, histone acetylation, transcription factor binding, axonal transport, synaptic vesicle localization, dendritic spine morphogenesis, microtubule organizing, and phagocytic vesicles (Ellrichmann et al., 2013; Martinez-Vicente et al., 2010; Moumné et al., 2013; Trushina et al.,

2003; Zuccato et al., 2010). These results suggest aberrant 3′UTR isoform expression may influence pathogenesis in these disease pathways.

247

Figure C.6 Gene expression analysis shows most genes with 3'UTR isoform changes do not exhibit expression changes in HD

248

(A) Scatter plot of PAS-seq reads from control (n=5) and HD (n=6) motor cortex. Each dot represents the normalized expression of a gene. Orange dots indicate isoforms with p<0.01, whereas red dots indicate isoforms with adjusted p<0.1 (false discovery rate<10%). (B) Scatter plot of PAS-seq reads from control (n=7) and HD (n=9) cerebellum, as in A. Green dots indicate isoforms with p<0.01, whereas blue dots indicate isoforms with adjusted p<0.1 (false discovery rate<10%). (C) Heat map of genes significantly differentially expressed in patient motor cortex (false discovery rate<10%).

Shown are fold changes in normalized gene expression. (D) Heat map of genes significantly differentially expressed in HD cerebellum (false discovery rate<10%), as in

C. (E) Comparison of gene isoform weighted changes (see methods) and differential expression (DE) log 2 fold changes in HD versus control motor cortex. Green dots have significant isoform changes in HD, orange dots have significant expression changes in

HD, and red dots have both (p<0.01). (F) Comparison of gene isoform weighted changes and expression log 2 fold changes in HD versus control cerebellum, as in E.

(G) Comparison of gene expression log 2 fold changes (in motor cortex (MCx) and cerebellum (CB). Red dots have significant expression changes in HD motor cortex, blue dots have significant changes in HD cerebellum, and light blue have both (false discovery rate<10%). (H) Gene ontology analysis of genes exhibiting significant (false discovery rate<10%) expression changes in HD motor cortex (MCx) and cerebellum

(CB).

249

Most genes with isoform changes in HD are not differentially expressed

Widespread gene expression changes are reported in the cortex of late-stage

Huntington’s disease patients (Hodges et al., 2006; Labadorf et al., 2015). We sought to determine if the extensive 3′UTR isoform changes in HD motor cortex co-occur with gene expression changes. To measure steady-state gene expression, we combined all PAS-seq reads from each gene (see Experimental

Procedures). Consistent with previous studies, many genes were differentially expressed in HD patient motor cortex relative to control motor cortex (false discovery rate<10%) (Figure C.6.A) (Hodges et al., 2006; Labadorf et al., 2015).

Quantitative PCR confirmed differential expression of three genes (Figure S5). Of differentially expressed genes, 700 exhibited increased expression in HD motor cortex, whereas 1008 exhibited decreased expression (Figure C.6.C).

Interestingly, the percent of detected genes exhibiting expression changes is identical to the percent of alternatively polyadenylated genes exhibiting isoform changes (11%). In contrast, only 1% of genes are differentially expressed (false discovery rate<10%) in patient cerebellum, with 2.5% making the p-value cutoff

(Figure C.6.B). Of differentially expressed genes (false discovery rate<10%), 89 increased expression in HD patient cerebellum whereas 64 decreased (Figure

C.6.D). Thus, isoform changes are proportional to gene expression changes in the HD motor cortex, whereas isoform and gene expression changes are minimal in the cerebellum. 250

To determine if genes with isoform changes also exhibit expression changes, we compared gene expression to isoform expression for all alternatively polyadenylated genes in the motor cortex and cerebellum using a significance cutoff of p<0.01 (Figure C.6.E-F). In the motor cortex, only 17% of genes with isoform shifts also exhibit gene expression changes (Figure C.6.E, red). Of genes exhibiting isoform and gene expression changes, we found no association between isoform shift and gene expression direction. This dissociation of isoform and gene expression is consistent with studies showing the association between

3′UTR length and isoform stability is gene-specific (Spies et al., 2013). In the cerebellum, only 4% of genes with significant isoform changes are also differentially expressed (Figure C.6.F, red). Of these, all are shifted towards shorter isoforms, with two exhibiting increased expression and two exhibiting decreased expression. These results indicate that most genes with isoform changes are not expressed differently in HD motor cortex and cerebellum.

We compared gene expression between the motor cortex and cerebellum to see if there is overlap in differentially expressed genes. We found sixteen (11%) of the genes expressed differently in HD cerebellum are also expressed differently in HD motor cortex (Figure C.6.G, light blue). We performed gene ontology analysis to identify pathways enriched (p<0.01) among genes with increased or decrease expression in HD motor cortex, cerebellum, or both (Figure C.6.H). Our analysis identified several pathways previously reported to be enriched among genes differentially expressed in HD motor cortex: CNS development, protein 251

localization, extracellular exosome, fatty acid biosynthesis, and regulation of TNF signaling (Hodges et al., 2006; Labadorf et al., 2015). Unlike previous studies, we separated genes into categories based on region and direction of change. We found many pathways are enriched between genes with decreased expression exclusively in motor cortex, and genes with increased expression exclusively in cerebellum. Some of these pathways, including protein folding and ubiquitination, are vital for clearance of the mutant HTT protein. Region-specific regulation of these processes may be related to region-specific pathology.

We sought to determine if isoform and gene expression changes are associated with common cell pathways. We found several ontology categories enriched among genes with isoform changes and genes with expression changes including mRNA 3′UTR binding, cell junction, ubiquitination, mRNA binding, and protein transport (Figure C.5.H, Figure C.6.H). Within each category, genes with isoform differences and genes with expression changes do not overlap; i.e., the same gene does not exhibit aberrant isoform and gene expression. These results indicate that isoform alterations and gene expression changes are separate, but they may affect common processes in HD brains.

Decreasing expression of the RNA binding protein CNOT6 leads some genes to change isoform abundance

A change in the expression of an RNA binding protein may lead to changes in isoform abundance by either altering isoform production rates during alternative polyadenylation, or by affecting isoform decay rates post-transcriptionally. To 252

identify candidate RNA binding proteins, we searched for genes differently expressed in HD motor cortex with the RNA binding protein GO-term (Figure

C.7.A). Of those, twenty were involved in mRNA alternative polyadenylation or stability. We reasoned an RNA binding protein responsible for HTT isoform shifts would show opposite expression changes in HD motor cortex and cerebellum. Of nine fitting that description, two (CNOT6 and CNOT7) are subunits of the Ccr4- not complex that plays a widespread role in mRNA metabolism (Mittal et al.,

2011; Shirai et al., 2014). CNOT6, the subunit that catalyzes deadenylation of mRNAs, is most differently expressed in HD motor cortex and cerebellum

(Yamashita et al., 2005).

To determine if changes in CNOT6 expression exert changes in isoform expression, we transfected control human fibroblasts with siRNAs targeting

CNOT6 or an off-target control (MAP4K4). The CNOT6 siRNAs reduced CNOT6 mRNA expression over 80% but did not affect total HTT mRNA levels (Figure

C.7.B-C). We chose HTT and SECISBP2L (SECIS Binding Protein 2 Like) as candidate CNOT6 targets. HTT and SECISBP2L shift to longer 3′UTR isoforms in

HD cerebellum and shorter isoforms in HD motor cortex. We collected RNA from transfected cells for qPCR analysis of the HTT long 3′UTR isoform versus total

HTT, and of the SECISBP2L long 3′UTR isoform versus total SECISBP2L. We found knock down of CNOT6 resulted in a decrease in the HTT and SECISBP2L long isoforms similar to that seen in HD patient motor cortex and in HD 253

fibroblasts (Figure C.7.D). HTT isoform-specific qPCR revealed that the abundance of the short isoform didn’t change while abundance of the long isoform decreased by almost two-fold, as in HD patient motor cortex (Figure

C.7.E). These results suggest changes in the expression of CNOT6 may influence the abundance of some 3′UTR isoforms in HD patient motor cortex and cerebellum.

254

Figure C.7 Changes in the expression of RNA binding proteins may influence alterations in isoform abundances

255

(A) Algorithm to identify proteins that might cause 3′UTR isoform changes in HD patient brains identified CNOT6 and CNOT7, members of the multi-functional Ccr4-not complex.

Numbers are fold changes (FC) in mRNA expression between HD and control motor cortex (MCx) or cerebellum (CB). (B) qRT-PCR quantification CNOT6 mRNA compared to GAPDH mRNA after transfection of CNOT6 siRNA (gray), no siRNA (white), or non- targeting (NT) MAP4K4 control siRNA (black) into wild-type fibroblasts. *, **, and *** signify p<0.05, 0.005, and 0.0005. Shown is the average of at least three transfections with standard deviation. (C) qRT-PCR quantification HTT (HTT) mRNA compared to

GAPDH mRNA after transfection as in (B). (C) qRT-PCR quantification of the HTT long isoform (long) compared to total HTT expression (total) or of the SECISBP2L long isoform (long) compared to total SECISBP2L expression (total) after transfection as in

(B). Red and blue bars are untreated fibroblasts, as in Figure 1D. (D) Isoform-specific qRT-PCR quantification of each HTT long isoform (long) compared to total HTT expression after transfection of CNOT6 siRNA (colored), no siRNA (white), or non- targeting (NT) MAP4K4 control siRNA (black) into wild-type fibroblasts.

256

Methods Samples

Human samples were supplied by the New York and Neurological Foundation of

New Zealand Brain Banks. Fibroblast lines were obtained from the Coriell

Repository. The M. DiFiglia lab (Massachusetts General Hospital) supplied neural stem cell pellets. For mouse experiments, 7-9 month male and female mice were sacrificed, and brain regions were dissected and stored in RNAlater

(Sigma Aldrich) prior to RNA extraction.

Public 3′Seq data analysis

Raw 3′ sequencing data was downloaded from the NCBI short sequence read archive, accession SRP029953 (Lianoglou et al., 2013), and analyzed as below.

RNA extraction

Total RNA was extracted with Trizol (Ambion), cleaned on Clean and

Concentrator columns (Zymo), analyzed via Bioanalyzer (Agilent), and treated with TurboDNAse (Ambion).

qRT-PCR

RNA was reverse transcribed by SuperScript IV (Invitrogen). Quantitative PCR was performed by Qiagen QuantiFast SYBR green master mix (non allele- 257

specific), or Qiagen 1x Type-it Fast SNP PCR master mix (allele-specific).

Product amounts were calculated using the ΔΔCt method. We compared group averages with unpaired two-tailed t-tests and considered p<0.05 significant.

Primer and probe sequences are listed in Table S2.

Nuclear and cytoplasmic isoform abundance

SH-SY5Y nuclear and cytoplasmic RNA fractions were collected as described in

Supplemental Experimental Procedures, and RNA was submitted to qRT-PCr as above.

In vitro transcription of ethinyl-uridine (EU) pulse chase spike-in

Firefly luciferase with the T7 promoter was in vitro transcribed as described in the

Supplemental Experimental Procedures. In vitro transcripts were purified on

Centrispin 20 columns (VWR), precipitated by phenol extraction and ethanol precipitation, and Bioanalyzed (Agilent) to ensure a single product.

EU pulse chase and isoform-specific qRT-PCR

SH-SY5Y cells (ATCC) were pulsed in 200µM EU for 14 hours, washed in PBS, and chased with unmodified media. We isolated labeled RNA using the Click-iT

Nascent RNA Capture Kit (Life Technologies) as detailed in the Supplemental

Experimental Procedures. SuperScript IV (Invitrogen) reverse transcribed labeled 258

RNA. qPCR of the HTT short and long isoforms was performed as above, and

Isoform amounts were normalized to those of a firefly luciferase spike-in.

PolyA tail length assay

RNA was extracted as above. The HTT short and long isoform polyA tails were reverse transcribed and amplified by the Affymetrix PolyA Tail Length Assay

(primers are listed in Table S2). PCR products were resolved on a 2% agarose,

1x TAE gel.

MicroRNA and RNA binding protein target site prediction

We used the CLIPdb and starBase v2.0 browsers to identify RNA binding proteins that interact with the HTT 3′UTR (Li et al., 2014; Yang et al., 2015).

Brain expression was confirmed with the human protein atlas, www.proteinatlas.org (Uhlen et al., 2015). We included binding sites represented by more than five reads. To identify predicted microRNA binding sites, we used

TargetScanHuman version 7.0 (Lewis et al., 2003). We identified microRNAs with predicted 7mer-1A, 7mer-m8, 8mer binding sites. Brain expression was confirmed with miRmine Human miRNA Expression Database (Panwar et al.,

2017).

MicroRNA transfection

SH-SY5Y cells were transfected with microRNA mimics 137, 221, or both 259

(Dharmacon) using Lipofectamine RNAimax transfection reagent (Invitrogen) per manufacturer’s instructions. After 72 hours, RNA was extracted and submitted to qRT-PCR as above.

PAS-seq library preparation

For the detailed PAS-seq method, see Supplemental Experimental Procedures.

Briefly, extracted RNA was fragmented and reverse-transcribed by Superscript III

(Invitrogen) from the PAS-seq oligo d(T) primer. cDNA fragments 160-200 nucleotides long were circularized by incubation with CircLigase II (Epicentre), and Phusion polymerase (NEB) amplified the cDNA and adaptors with PE 1.0 and PE 2.0 primers (Illumina). The 215-254 nucleotide library was size-selected via gel electrophoresis and subjected to single-end 100 base-pair sequencing.

PAS-Seq analysis

For the detailed PAS-Seq analysis and bioinformatics pipeline, see the supplemental methods. Briefly, high-quality reads were mapped to gene ORFs or 3′UTRs in the hg19 or mm10 genome, and reads corresponding to the same isoform were combined. We removed genomically-primed reads and normalized isoform expression to gene expression. Normalized isoform abundance was compared between HD and control samples using two-tailed t-tests with or without correction for multiple-hypothesis testing. We considered an adjusted p- value <0.1 to be significant unless otherwise stated (false discovery rate<10%). 260

Only isoforms represented in at least five disease and control samples were considered.

PAS-seq analysis: isoform weighted change

We weighed each isoform ratio by the isoform number multiplied by ten. For instance, for a gene transcribed into two isoforms of equal abundance (isoform ratios of 0.5), the shorter isoform weighted ratio is 5 (0.5 x 10 x 1) whereas the longer isoform weighted ratio is 10 (0.5 x 10 x 2). We then calculated the difference in the average weighted isoform ratios between HD and control samples for the gene. A positive value indicates a shift towards longer isoforms in HD samples, whereas a negative value indicates a shift towards shorter isoforms in HD samples.

PAS-seq analysis: HD versus control gene expression comparison

The total reads mapping to each gene were compared between patients and controls using DESeq2 (Love et al., 2014a). We considered genes with a change greater than 50% and false discovery rate <10% significantly differently expressed.

Gene ontology analysis

The list of genes with significant changes in total expression (false discovery 261

rate<10%) or 3′UTR isoform expression (p<0.01) in HD motor cortex or cerebellum were compared to the background total or alternatively polyadenylated expressed genes in each region using Panther gene ontology analysis (Mi et al., 2013). We considered categories enriched if they had five or more genes, enrichment greater than 1.5 fold, and p<0.01.

CNOT6 siRNA transfection

Wild-type fibroblasts were transfected with three siRNAs targeting CNOT6

(Qiagen FlexiTube) via Lipofectamine RNAimax (Invitrogen). After 72 hours, RNA was extracted and submitted to qRT-PCR as above.

Discussion

Widespread changes in isoform abundance may lead to aberrant mRNA metabolism in HD. Some tandem 3′UTR isoforms exhibit unique localization and function. The long but not short 3′UTR isoform of brain-derived neurotrophic factor mRNA is translated in dendrites, where the protein is vital for pruning and spine enlargement (An et al., 2008). The long isoform of CD47 mRNA acts as a scaffold for proteins that translocate the CD47 protein to the cell membrane, whereas the short isoform localizes the protein to the endoplasmic reticulum

(Berkovits and Mayr, 2015). Loss of normal mRNA isoform abundance can contribute to disease. The alpha synuclein long 3′UTR isoform is preferentially localized to mitochondria; increased abundance of the extended alpha synuclein 262

mRNA may contribute to mitochondrial dysfunction in Parkinson’s disease (Rhinn et al., 2012). Many genes in cancer cells shift towards more stable short isoforms, increasing translation of tumorigenic proteins (Mayr and Bartel, 2009).

In myotonic dystrophy, triplet repleat RNAs sequester muscleblind, a regulator of

3′UTR length, resulting in an altered polyadenylation profile (Batra et al., 2014).

We demonstrate that genes with isoform changes in HD motor cortex are associated with pathways disrupted in HD including calcium signaling, cytokine signaling, histone acetylation, transcription factor binding, axonal transport, synaptic vesicle localization, dendritic spine morphogenesis, microtubule organizing, and phagocytic vesicles. Altered metabolism of mRNAs in disease pathways could result in aberrant localization or function of proteins and contribute to HD pathogenesis.

Our results point to a model in which changes in the expression of RNA binding proteins in the HD brain lead to changes in the expression of mRNA 3′UTR isoforms. RNA binding proteins may interact with nascent mRNA during transcription and alternative polyadenylation, leading to changes in polyA site selection and isoform production. For instance, aberrant expression of 3′ end modulatory factors known to correlate to polyA site selection may lead to isoform abundance changes (Gruber et al., 2014). We found three 3′ end processing factors are differentially expressed in HD grade 1 motor cortex: CPSF2, PCBP2, and THOC5. In addition to interacting with transcripts during alternative 263

polyadenylation, RNA binding proteins may adhere to isoforms post- transcriptionally, affecting the stability and decay of isoforms and leading to changes in isoform abundance. We show a decrease in the CNOT6 deadenylase is associated with a decrease in the HTT long isoform, suggesting CNOT6 stabilizes the long isoform or regulates its metabolism via some other mechanism.

Further studies are necessary to determine the mechanism of 3′UTR isoform changes in HD. We found HTT and SECISBP2L isoform amounts are responsive to changes in CNOT6 mRNA levels. However, altered CNOT6 expression in HD cannot explain all isoform changes we identified: isoform differences are largely limited to the motor cortex, whereas CNOT6 is down-regulated in motor cortex and up-regulated in cerebellum. Other RNA binding proteins, such as 3′ end processing factors, likely affect isoform abundance in HD. Exploration of the causes of isoform changes in HD motor cortex may result in novel therapeutic targets.

Our findings have important implications for HD and may apply to other diseases.

New therapies aim to disrupt the mutant but not wild-type HTT mRNA by targeting SNPs heterozygous in patients (Østergaard et al., 2013; Pfister et al.,

2009). Our results suggest therapies targeting SNPs exclusive to the long isoform may be less effective, as HTT long isoform abundance is reduced in 264

Huntington’s patient motor cortex. A study published during preparation of this manuscript found transfected HTT exon 1-3′UTR constructs are localized differently depending on the 3′UTR length, and the short 3′UTR construct forms more aggregates (Xu et al., 2017). The short isoform, which increases its relative abundance in HD motor cortex, may be the ideal target for allele-specific therapies. Further research exploring 3′UTR isoform changes in other diseases is necessary to determine if isoform alterations are unique to HD, or if they frequently accompany gene expression changes in human disease.

265

REFERENCES

ACOG Committee on Practice Bulletins--Obstetrics (2002). ACOG practice bulletin. Diagnosis and management of preeclampsia and eclampsia. Number 33, January 2002. Obstet Gynecol 99, 159–167.

Ahmad, S., and Ahmed, A. (2004). Elevated placental soluble vascular endothelial growth factor receptor-1 inhibits angiogenesis in preeclampsia. Circulation Research 95, 884–891.

Ahmed, A. (1997). Heparin-binding angiogenic growth factors in pregnancy: a review. Placenta.

Alterman, J.F., Hall, L.M., Coles, A.H., Hassler, M.R., Didiot, M.-C., Chase, K., Abraham, J., Sottosanti, E., Johnson, E., Sapp, E., et al. (2015). Hydrophobically Modified siRNAs Silence Huntingtin mRNA in Primary Neurons and Mouse Brain. Mol Ther Nucleic Acids 4, e266.

Ambati, B.K., Nozaki, M., Singh, N., Takeda, A., Jani, P.D., Suthar, T., Albuquerque, R.J.C., Richter, E., Sakurai, E., Newcomb, M.T., et al. (2006). Corneal avascularity is due to soluble VEGF receptor-1. Nature 443, 993–997.

American College of Obstetricians and Gynecologists, Task Force on Hypertension in Pregnancy (2013). Hypertension in pregnancy. Report of the American College of Obstetricians and Gynecologists’ Task Force on Hypertension in Pregnancy. pp. 1122–1131.

An, J.J., Gharami, K., Liao, G.-Y., Woo, N.H., Lau, A.G., Vanevski, F., Torre, E.R., Jones, K.R., Feng, Y., Lu, B., et al. (2008). Distinct role of long 3’ UTR BDNF mRNA in spine morphology and synaptic plasticity in hippocampal neurons. Cell 134, 175–187.

Ananth, C.V., Keyes, K.M., and Wapner, R.J. (2013). Pre-eclampsia rates in the United States, 1980-2010: age-period-cohort analysis. Bmj 347, f6564–f6564.

Andersen, L.B., Dechend, R., Jørgensen, J.S., Luef, B.M., Nielsen, J., Barington, T., and Christesen, H.T. (2016). Prediction of preeclampsia with angiogenic biomarkers. Results from the prospective Odense Child Cohort. Hypertens Pregnancy 35, 405–419.

Anderson, E., Boese, Q., Khvorova, A., and Karpilow, J. (2008). Identifying 266

siRNA-induced off-targets by microarray analysis. Methods Mol. Biol. 442, 45– 63.

Arngrimsson, R., Björnsson, S., Geirsson, R.T., Björnsson, H., Walker, J.J., and Snaedal, G. (1990). Genetic and familial predisposition to eclampsia and preeclampsia in a defined population. British Journal of Obstetrics and Gynaecology 97, 762–769.

Aubuchon, M., Schulz, L.C., and Schust, D.J. (2011). Preeclampsia: animal models for a human cure. Proceedings of the National Academy of Sciences of the United States of America 108, 1197–1198.

Balada, E., Vilardell-Tarrés, M., and Ordi-Ros, J. (2010). Implication of human endogenous retroviruses in the development of autoimmune diseases. Int. Rev. Immunol. 29, 351–370.

Bañez-Coronel, M., Ayhan, F., Tarabochia, A.D., Zu, T., Perez, B.A., Tusi, S.K., Pletnikova, O., Borchelt, D.R., Ross, C.A., Margolis, R.L., et al. (2015). RAN Translation in Huntington Disease. Neuron 88, 667–677.

Barber, T.D., Barber, M.C., Tomescu, O., Barr, F.G., Ruben, S., and Friedman, T.B. (2002). Identification of Target Genes Regulated by PAX3 and PAX3–FKHR in Embryogenesis and Alveolar Rhabdomyosarcoma. Genomics 79, 278–284.

Batra, R., Charizanis, K., Manchanda, M., Mohan, A., Li, M., Finn, D.J., Goodwin, M., Zhang, C., Sobczak, K., Thornton, C.A., et al. (2014). Loss of MBNL leads to disruption of developmentally regulated alternative polyadenylation in RNA- mediated disease. Mol. Cell 56, 311–322.

Begay, V., Smink, J., and Leutz, A. (2004). Essential Requirement of CCAAT/Enhancer Binding Proteins in Embryogenesis. Molecular and Cellular Biology 24, 9744–9751.

Bell, M.J. (2010). A historical overview of preeclampsia-eclampsia. J Obstet Gynecol Neonatal Nurs 39, 510–518.

Bellamy, L., Casas, J.-P., Hingorani, A.D., and Williams, D.J. (2007). Pre- eclampsia and risk of cardiovascular disease and cancer in later life: systematic review and meta-analysis. Bmj 335, 974–974.

Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B …. 267

Berkovits, B.D., and Mayr, C. (2015). Alternative 3′ UTRs act as scaffolds to regulate membrane protein localization. Nature 522, 363–367.

Biadasiewicz, K., Fock, V., Dekan, S., Proestling, K., Velicky, P., Haider, S., Knöfler, M., Fröhlich, C., and Pollheimer, J. (2014). Extravillous trophoblast- associated ADAM12 exerts pro-invasive properties, including induction of integrin beta 1-mediated cellular spreading. Biology of Reproduction 90, 101.

Bianco, A.T., Smilen, S.W., Davis, Y., and Lopez, S. (1998). Pregnancy outcome and weight gain recommendations for the morbidly obese woman. Obstet Gynecol 91, 97–102.

Borchert, G.M., Lanier, W., and Davidson, B.L. (2006). RNA polymerase III transcribes human microRNAs. Nat Struct Mol Biol 13, 1097–1101.

Borzychowski, A.M., Sargent, I.L., and Redman, C.W.G. (2006). Inflammation and pre-eclampsia. Semin Fetal Neonatal Med 11, 309–316.

Buchbinder, A., Sibai, B.M., Caritis, S., MacPherson, C., Hauth, J., Lindheimer, M.D., Klebanoff, M., VanDorsten, P., Landon, M., Paul, R., et al. (2002). Adverse perinatal outcomes are significantly higher in severe gestational hypertension than in mild preeclampsia. American Journal of Obstetrics and Gynecology 186, 66–71.

Burnett, J.C., Rossi, J.J., and Tiemann, K. (2011). Current progress of siRNA/shRNA therapeutics in clinical trials. Biotechnol J 6, 1130–1146.

Caballero, V., Ruiz, R., Sainz, J.A., Cruz, M., López-Nevot, M.A., Galán, J.J., Real, L.M., de Castro, F., López-Villaverde, V., and Ruiz, A. (2005). Preliminary molecular genetic analysis of the Receptor Interacting Protein 140 (RIP140) in women affected by endometriosis. J. Exp. Clin. Assist. Reprod. 2, 11.

Carmona-Saez, P., Chagoyen, M., Tirado, F., Carazo, J.M., and Pascual- Montano, A. (2007). GENECODIS: a web-based tool for finding significant concurrent annotations in gene lists. Genome Biology 8, R3.

Cerdeira, A.S., and Karumanchi, S.A. (2012). Angiogenic factors in preeclampsia and related disorders. Cold Spring Harb Perspect Med 2, a006585–a006585.

Chaiworapongsa, T., Romero, R., Espinoza, J., Bujold, E., Mee Kim, Y., Gonçalves, L.F., Gomez, R., and Edwin, S. (2004). Evidence supporting a role for blockade of the vascular endothelial growth factor system in the pathophysiology of preeclampsia. Young Investigator Award. Ymob 190, 1541– 7–discussion1547–50. 268

Chang, H., Lim, J., Ha, M., and Kim, V.N. (2014). TAIL-seq: Genome-wide determination of poly(A) tail length and 3’ end modifications. Mol. Cell 53, 1044– 1052.

Chang, C.-W., and Parast, M.M. (2017). Human trophoblast stem cells: Real or not real? Placenta.

Chappell, L.C., Duckworth, S., Seed, P.T., Griffin, M., Myers, J., Mackillop, L., Simpson, N., Waugh, J., Anumba, D., Kenny, L.C., et al. (2013). Diagnostic accuracy of placental growth factor in women with suspected preeclampsia: a prospective multicenter study. Circulation 128, 2121–2131.

Chappell, S., and Morgan, L. (2006). Searching for genetic clues to the causes of pre-eclampsia. Clinical Science (London, England : 1979) 110, 443–458.

Chesley, L.C. (1984). OCLC WorldShare Managed-User Sign-In. Clin Obstet Gynecol.

Chesley, L.C. (1985). Hypertensive disorders in pregnancy. J Nurse Midwifery 30, 99–104.

Cho, H., Kim, K.M., Han, S., Choe, J., Park, S.G., Choi, S.S., and Kim, Y.K. (2012). Staufen1-mediated mRNA decay functions in adipogenesis. Molecular Cell 46, 495–506.

Christians, J.K., and Beristain, A.G. (2016). ADAM12 and PAPP-A: Candidate regulators of trophoblast invasion and first trimester markers of healthy trophoblasts. Cell Adh Migr 10, 147–153.

Chu, M.C. (2003). Elevated basal FSH in normal cycling women is associated with unfavourable lipid levels and increased cardiovascular risk. Human Reproduction 18, 1570–1573.

Ciribilli, Y., Andreotti, V., Menendez, D., Langen, J.-S., Schoenfelder, G., Resnick, M.A., and Inga, A. (2010). The Coordinated P53 and Estrogen Receptor Cis-Regulation at an FLT1 Promoter SNP Is Specific to Genotoxic Stress and Estrogenic Compound. PLoS ONE 5, e10236–15.

Clark, D.E., Smith, S.K., Sharkey, A.M., and Charnock-Jones, D.S. (1996). Localization of VEGF and expression of its receptors flt and KDR in human placenta throughout pregnancy. Hum. Reprod. 11, 1090–1098.

Comery, T.A., Harris, J.B., Willems, P.J., Oostra, B.A., Irwin, S.A., Weiler, I.J., and Greenough, W.T. (1997). Abnormal dendritic spines in fragile X knockout 269

mice: maturation and pruning deficits. Proceedings of the National Academy of Sciences of the United States of America 94, 5401–5404.

Dadelszen, von, P., Payne, B., Li, J., and Ansermino, J.M. (2011). Prediction of adverse maternal outcomes in pre-eclampsia: development and validation of the fullPIERS model. The Lancet.

Dadelszen, von, P., Magee, L.A., and Roberts, J.M. (2003). Subclassification of preeclampsia. Hypertens Pregnancy 22, 143–148.

Das, B., Yeger, H., Tsuchida, R., Torkin, R., Gee, M.F.W., Thorner, P.S., Shibuya, M., Malkin, D., and Baruchel, S. (2005). A Hypoxia-Driven Vascular Endothelial Growth Factor/Flt1 Autocrine Loop Interacts with Hypoxia-Inducible Factor-1 through Mitogen-Activated Protein Kinase/Extracellular Signal- Regulated Kinase 1/2 Pathway in Neuroblastoma. Cancer Research 65, 7267– 7275.

Datta, P.K., Bonner, J.F., Haas, C.J., and Fischer, I. (2013). Considerations for the Use of SH-SY5Y Neuroblastoma Cells in Neurobiology. Neuronal Cell Cult. Methods Protoc. 1078, 35–44.

Davis, M.E. (2009). The first targeted delivery of siRNA in humans via a self- assembling, cyclodextrin polymer-based nanoparticle: from concept to clinic. Mol. Pharm. 6, 659–668.

Dekker, G. (2002). The partner's role in the etiology of preeclampsia. J. Reprod. Immunol. 57, 203–215.

Derti, A., Garrett-Engele, P., Macisaac, K.D., Stevens, R.C., Sriram, S., Chen, R., Rohl, C.A., Johnson, J.M., and Babak, T. (2012). A quantitative atlas of polyadenylation in five mammals. Genome Research 22, 1173–1183.

DiFiglia, M., Sapp, E., Chase, K.O., Davies, S.W., Bates, G.P., Vonsattel, J.P., and Aronin, N. (1997). Aggregation of huntingtin in neuronal intranuclear inclusions and dystrophic neurites in brain. Science 277, 1990–1993.

Di Giammartino, D.C.D., Nishida, K.K., and Manley, J.L.J. (2011). Mechanisms and consequences of alternative polyadenylation. Molecular Cell 43, 853–866.

Dixon, H.G., and Robertson, W.B. (1958). A STUDY OF THE VESSELS OF THE PLACENTAL BED IN NORMOTENSIVE AND HYPERTENSIVE WOMEN. BJOG: an International Journal of Obstetrics &Amp; Gynaecology 65, 803–809. 270

Duckitt, K. (2005). Risk factors for pre-eclampsia at antenatal booking: systematic review of controlled studies. Bmj 330, 565–0.

Duckworth, S., Griffin, M., Seed, P.T., North, R., Myers, J., Mackillop, L., Simpson, N., Waugh, J., Anumba, D., Kenny, L.C., et al. (2016). Diagnostic Biomarkers in Women With Suspected Preeclampsia in a Prospective Multicenter Study. Obstet Gynecol 128, 245–252.

Dugré-Brisson, S., Elvira, G., Boulay, K., Chatel-Chaix, L., Mouland, A.J., and DesGROSEILLERS, L. (2005). Interaction of Staufen1 with the 5′ end of mRNA facilitates translation of these RNAs. Nucleic Acids Research 33, 4797.

Dumont, D.J., Fong, G.H., Puri, M.C., Gradwohl, G., Alitalo, K., and Breitman, M.L. (1995). Vascularization of the mouse embryo: a study of flk-1, tek, tie, and vascular endothelial growth factor expression during development. Dev. Dyn. 203, 80–92.

E Davies, J., Pollheimer, J., Yong, H.E.J., Kokkinos, M.I., Kalionis, B., Knöfler, M., and Murthi, P. (2016). Epithelial-mesenchymal transition during extravillous trophoblast differentiation. Cell Adh Migr 10, 310–321.

Edlow, A.G., Srinivas, S.K., and Elovitz, M.A. (2009). Investigating the risk of hypertension shortly after pregnancies complicated by preeclampsia. Ymob 200, e60–e62.

Elbarbary, R.A., Li, W., Tian, B., and Maquat, L.E. (2013). STAU1 binding 3' UTR IRAlus complements nuclear retention to protect cells from PKR-mediated translational shutdown. Genes & Development 27, 1495–1510.

Ellrichmann, G., Reick, C., Saft, C., and Linker, R.A. (2013). The Role of the Immune System in Huntington’s Disease. Clin. Dev. Immunol. 2013, doi:10.1155/2013/541259.

Enquobahrie, D.A., Meller, M., Rice, K., Psaty, B.M., Siscovick, D.S., and Williams, M.A. (2008). Differential placental gene expression in preeclampsia. American Journal of Obstetrics and Gynecology 199, 566.e1–566.e11.

Eremina, V., Jefferson, J.A., Kowalewska, J., Hochster, H., Haas, M., Weisstuch, J., Richardson, C., Kopp, J.B., Kabir, M.G., Backx, P.H., et al. (2008). VEGF inhibition and renal thrombotic microangiopathy. The New England Journal of Medicine 358, 1129–1136.

Eremina, V., Sood, M., Haigh, J., Nagy, A., Lajoie, G., Ferrara, N., Gerber, H.-P., Kikkawa, Y., Miner, J.H., and Quaggin, S.E. (2003). Glomerular-specific 271

alterations of VEGF-A expression lead to distinct congenital and acquired renal diseases. J. Clin. Invest. 111, 707–716.

Fedorov, Y., Anderson, E.M., Birmingham, A., Reynolds, A., Karpilow, J., Robinson, K., Leake, D., Marshall, W.S., and Khvorova, A. (2006). Off-target effects by siRNA can induce toxic phenotype. Rna 12, 1188–1196.

Ferrandon, D. (1997). RNA-RNA interaction is required for the formation of specific bicoid mRNA 3' UTR-STAUFEN ribonucleoprotein particles. Embo J 16, 1751–1758.

Ferrandon, D., Elphick, L., Nüsslein-Volhard, C., and St Johnston, D. (1994). Staufen protein associates with the 3′UTR of bicoid mRNA to form particles that move in a microtubule-dependent manner. Cell 79, 1221–1232.

Ferrara, N., Carver-Moore, K., Chen, H., Dowd, M., Lu, L., O'Shea, K.S., Powell- Braxton, L., Hillan, K.J., and Moore, M.W. (1996). Heterozygous embryonic lethality induced by targeted inactivation of the VEGF gene. Nature 380, 439– 442.

Ferrara, N. (2004). Vascular endothelial growth factor as a target for anticancer therapy. Oncologist 9 Suppl 1, 2–10.

Fields, S.J., Vainder, M., Livshits, G., Merlob, P., and Sirotta, L. (1996). Obesity and the risk of toxaemia of pregnancy. Annals of Human Biology 23, 353–362.

Fong, G.H., Rossant, J., Gertsenstein, M., and Breitman, M.L. (1995). Role of the Flt-1 receptor tyrosine kinase in regulating the assembly of vascular endothelium. Nature 376, 66–70.

Fukui, A., Yokota, M., Funamizu, A., Nakamua, R., Fukuhara, R., Yamada, K., Kimura, H., Fukuyama, A., Kamoi, M., Tanaka, K., et al. (2012). Changes of NK Cells in Preeclampsia. Am J Reprod Immunol 67, 278–286.

Furic, L., Maher-Laporte, M., and DesGROSEILLERS, L. (2008). A genome-wide approach identifies distinct but overlapping subsets of cellular mRNAs associated with Staufen1- and Staufen2-containing ribonucleoprotein complexes. Rna 14, 324–335.

Gerber, H.P., Condorelli, F., Park, J., and Ferrara, N. (1997). Differential transcriptional regulation of the two vascular endothelial growth factor receptor genes. Flt-1, but not Flk-1/KDR, is up-regulated by hypoxia. J. Biol. Chem. 272, 23659–23667.

Ghosh, S., Marchand, V., Gáspár, I., and Ephrussi, A. (2012). Control of RNP 272

motility and localization by a splicing-dependent structure in oskar mRNA. Nat Struct Mol Biol 19, 441–449.

Gleghorn, M.L., Gong, C., Kielkopf, C.L., and Maquat, L.E. (2013). Staufen1 dimerizes through a conserved motif and a degenerate dsRNA-binding domain to promote mRNA decay. Nat Struct Mol Biol 20, 515–524.

Goel, A., Maski, M.R., Bajracharya, S., Wenger, J.B., Zhang, D., Salahuddin, S., Shahul, S.S., Thadhani, R., Seely, E.W., Karumanchi, S.A., et al. (2015). Epidemiology and Mechanisms of De Novo and Persistent Hypertension in the Postpartum Period. Circulation 132, 1726–1733.

Goldman, C.K., Kendall, R.L., Cabrera, G., Soroceanu, L., Heike, Y., Gillespie, G.Y., Siegal, G.P., Mao, X., Bett, A.J., Huckle, W.R., et al. (1998). Paracrine expression of a native soluble vascular endothelial growth factor receptor inhibits tumor growth, metastasis, and mortality rate. Proceedings of the National Academy of Sciences of the United States of America 95, 8795–8800.

Gong, C., and Maquat, L.E. (2011). lncRNAs transactivate STAU1-mediated mRNA decay by duplexing with 3? UTRs via Alu elements. Nature 470, 284–288.

Gonsalves, C.S., Li, C., Mpollo, M.-S.E.M., Pullarkat, V., Malik, P., Tahara, S.M., and Kalra, V.K. (2015). Erythropoietin-mediated expression of placenta growth factor is regulated via activation of hypoxia-inducible factor-1α and post- transcriptionally by miR-214 in sickle cell disease. Biochem. J. 468, 409–423.

Graubert, M.D., Ortega, M.A., Kessel, B., Mortola, J.F., and Iruela-Arispe, M.L. (2001). Vascular repair after menstruation involves regulation of vascular endothelial growth factor-receptor phosphorylation by sFLT-1. Am. J. Pathol. 158, 1399–1410.

Gruber, A.R., Lorenz, R., Bernhart, S.H., Neuböck, R., and Hofacker, I.L. (2008). The Vienna RNA websuite. Nucleic Acids Research 36, W70–W74.

Gruber, A.R., Martin, G., Keller, W., and Zavolan, M. (2014). Means to an end: Mechanisms of alternative polyadenylation of messenger RNA precursors. Wiley Interdiscip. Rev. RNA 5, 183–196.

Hanna, J., Goldman-Wohl, D., Hamani, Y., Avraham, I., Greenfield, C., Natanson-Yaron, S., Prus, D., Cohen-Daniel, L., Arnon, T.I., Manaster, I., et al. (2006). Decidual NK cells regulate key developmental processes at the human fetal-maternal interface. Nat. Med. 12, 1065–1074.

Haram, K., Svendsen, E., and Abildgaard, U. (2009). The HELLP syndrome: 273

clinical issues and management. A Review. BMC Pregnancy and Childbirth 9, 8.

Hartman, T.R., Qian, S., Bolinger, C., Fernandez, S., Schoenberg, D.R., and Boris-Lawrie, K. (2006). RNA helicase A is necessary for translation of selected messenger RNAs. Nat Struct Mol Biol 13, 509–516.

Hassler, M.R., and Khvorova, A. Full chemical stabilization is essential for conjugate-mediated siRNA delivery in vivo.

Haussecker, D. (2012). The Business of RNAi Therapeutics in 2012. Mol Ther Nucleic Acids 1, e8.

Hertig, A.T. (1945). Vascular pathology in hypertensive albuminic toxemias of pregnancy.

Heydarian, M., McCaffrey, T., Florea, L., Yang, Z., Ross, M.M., Zhou, W., and Maynard, S.E. (2009). Novel splice variants of sFlt1 are upregulated in preeclampsia. Placenta 30, 250–255.

Heyer, E.E., Ozadam, H., Ricci, E.P., Cenik, C., and Moore, M.J. (2015a). An optimized kit-free method for making strand-specific deep sequencing libraries from RNA fragments. Nucleic Acids Research 43, e2–e2.

Heyer, E.E., Ozadam, H., Ricci, E.P., Cenik, C., and Moore, M.J. (2015b). An optimized kit-free method for making strand-specific deep sequencing libraries from RNA fragments. Nucleic Acids Research 43, e2.

Hirashima, M., Lu, Y., Byers, L., and Rossant, J. (2003). Trophoblast expression of fms-like tyrosine kinase 1 is not required for the establishment of the maternal- fetal interface in the mouse placenta. Proceedings of the National Academy of Sciences of the United States of America 100, 15637–15642.

Hodges, A., Strand, A.D., Aragaki, A.K., Kuhn, A., Sengstag, T., Hughes, G., Elliston, L.A., Hartog, C., Goldstein, D.R., Thu, D., et al. (2006). Regional and cellular gene expression changes in human Huntington's disease brain. Human Molecular Genetics 15, 965–977.

Horii, M., Li, Y., Wakeland, A.K., Pizzo, D.P., Nelson, K.K., Sabatini, K., Laurent, L.C., Liu, Y., and Parast, M.M. (2016). Human pluripotent stem cells as a model of trophoblast differentiation in both normal development and disease. Proceedings of the National Academy of Sciences of the United States of America 113, E3882–E3891.

Horii, M., Moretto-Zita, M., Nelson, K.K., Li, Y., and Parast, M.M. (2015). MTA3 regulates differentiation of human cytotrophoblast stem cells. Placenta 36, 974– 274

980.

Hornig, C., Barleon, B., Ahmad, S., Vuorela, P., Ahmed, A., and Weich, H.A. (2000). Release and complex formation of soluble VEGFR-1 from endothelial cells and biological fluids. Lab. Invest. 80, 443–454.

Hrvatin, S., Deng, F., O'Donnell, C.W., Gifford, D.K., and Melton, D.A. (2014). MARIS: method for analyzing RNA following intracellular sorting. PLoS ONE 9, e89459.

Hu, J., Liu, J., and Corey, D.R. (2010). Allele-selective inhibition of huntingtin expression by switching to an miRNA-like RNAi mechanism. Chem. Biol. 17, 1183–1188.

Huppertz, B., Kingdom, J., Caniggia, I., Desoye, G., Black, S., Korr, H., and Kaufmann, P. (2003). Hypoxia Favours Necrotic Versus Apoptotic Shedding of Placental Syncytiotrophoblast into the Maternal Circulation. Placenta 24, 181– 190.

Ikeda, T., Sun, L., Tsuruoka, N., Ishigaki, Y., Yoshitomi, Y., Yoshitake, Y., and Yonekura, H. (2011). Hypoxia down-regulates sFlt-1 (sVEGFR-1) expression in human microvascular endothelial cells by a mechanism involving mRNA alternative processing. Biochem. J. 436, 399–407.

Ikeda, T., Yoshitomi, Y., Saito, H., Shimasaki, T., Yamaya, H., Kobata, T., Ishigaki, Y., Tomosugi, N., Yoshitake, Y., and Yonekura, H. (2016). Regulation of soluble Flt-1 (VEGFR-1) production by hnRNP D and protein arginine methylation. Molecular and Cellular Biochemistry 1–10.

Ivanov, P.V., Gehring, N.H., Kunz, J.B., Hentze, M.W., and Kulozik, A.E. (2008). Interactions between UPF1, eRFs, PABP and the exon junction complex suggest an integrated model for mammalian NMD pathways. Embo J 27, 736–747.

Jan, C.H., Friedman, R.C., Ruby, J.G., and Bartel, D.P. (2011). Formation, regulation and evolution of Caenorhabditis elegans 3’UTRs. Nature 469, 97–101.

Jauniaux, E., Watson, A.L., Hempstock, J., Bao, Y.-P., Skepper, J.N., and Burton, G.J. (2010). Onset of Maternal Arterial Blood Flow and Placental Oxidative Stress. Am. J. Pathol. 157, 2111–2122.

Jebbink, J., Keijser, R., Veenboer, G., van der Post, J., Ris-Stalpers, C., and Afink, G. (2011). Expression of placental FLT1 transcript variants relates to both gestational hypertensive disease and fetal growth. Hypertension 58, 70–76. 275

Jinnin, M., Medici, D., Park, L., Limaye, N., Liu, Y., Boscolo, E., Bischoff, J., Vikkula, M., Boye, E., and Olsen, B.R. (2008). Suppressed NFAT-dependent VEGFR1 expression and constitutive VEGFR2 signaling in infantile hemangioma. Nat. Med. 14, 1236–1246.

Johns, R. (1843). Observations on puerperal convulsions. The Dublin Journal of Medical Science 24, 101–115.

Junus, K., Centlow, M., Wikstrom, A.K., Larsson, I., Hansson, S.R., and Olovsson, M. (2012a). Gene expression profiling of placentae from women with early- and late-onset pre-eclampsia: down-regulation of the angiogenesis-related genes ACVRL1 and EGFL7 in early-onset disease. Molecular Human Reproduction 18, 146–155.

Junus, K., Centlow, M., Wikstrom, A.K., Larsson, I., Hansson, S.R., and Olovsson, M. (2012b). Gene expression profiling of placentae from women with early- and late-onset pre-eclampsia: down-regulation of the angiogenesis-related genes ACVRL1 and EGFL7 in early-onset disease. Molecular Human Reproduction 18, 146–155.

Kaartokallio, T., Cervera, A., Kyllönen, A., Laivuori, K., FINNPEC Core Investigator Group (2015). Gene expression profiling of pre-eclamptic placentae by RNA sequencing. Sci Rep 5, 14107.

Kalantaridou, S.N., Naka, K.K., Papanikolaou, E., Kazakos, N., Kravariti, M., Calis, K.A., Paraskevaidis, E.A., Sideris, D.A., Tsatsoulis, A., Chrousos, G.P., et al. (2004). Impaired endothelial function in young women with premature ovarian failure: normalization with hormone therapy. J. Clin. Endocrinol. Metab. 89, 3907–3913.

Kamba, T., Tam, B.Y.Y., Hashizume, H., Haskell, A., Sennino, B., Mancuso, M.R., Norberg, S.M., O'Brien, S.M., Davis, R.B., Gowen, L.C., et al. (2006). VEGF-dependent plasticity of fenestrated capillaries in the normal adult microvasculature. Am. J. Physiol. Heart Circ. Physiol. 290, H560–H576.

Kang, J.H., Song, H., Yoon, J.A., Park, D.Y., Kim, S.H., Lee, K.J., Farina, A., Cho, Y.K., Kim, Y.N., Park, S.W., et al. (2011). Preeclampsia leads to dysregulation of various signaling pathways in placenta. Journal of Hypertension 29, 928–936.

Karumanchi, S.A. (2016). Angiogenic Factors in Preeclampsia: From Diagnosis to Therapy. Hypertension 67, 1072–1079.

Kendall, R.L., and Thomas, K.A. (1993). Inhibition of vascular endothelial cell growth factor activity by an endogenously encoded soluble receptor. Proceedings 276

of the National Academy of Sciences of the United States of America 90, 10705– 10709.

Kerner, P., Degnan, S.M., Marchand, L., Degnan, B.M., and Vervoort, M. (2011). Evolution of RNA-Binding Proteins in Animals: Insights from Genome-Wide Analysis in the Sponge Amphimedon queenslandica. Molecular Biology and Evolution 28, 2289–2303.

Khvorova, A. (2017). Oligonucleotide Therapeutics - A New Class of Cholesterol- Lowering Drugs. The New England Journal of Medicine 376, 4–7.

Kiebler, M.A., Hemraj, I., Verkade, P., Köhrmann, M., Fortes, P., Marión, R.M., Ortín, J., and Dotti, C.G. (1999). The mammalian staufen protein localizes to the somatodendritic domain of cultured hippocampal neurons: implications for its involvement in mRNA transport. J. Neurosci. 19, 288–297.

Kim, J., Zhao, K., Jiang, P., Lu, Z.-X., Wang, J., Murray, J.C., and Xing, Y. (2012). Transcriptome landscape of the human placenta. BMC Genomics 13, 115.

Kim, Y.K., Furic, L., DesGROSEILLERS, L., and Maquat, L.E. (2005). Mammalian Staufen1 Recruits Upf1 to Specific mRNA 3′UTRs so as to Elicit mRNA Decay. Cell 120, 195–208.

Kim, Y.K., Furic, L., Parisien, M., Major, F.O., DesGROSEILLERS, L., and Maquat, L.E. (2007). Staufen1 regulates diverse classes of mammalian transcripts. Embo J 26, 2670–2681.

Kita, N., and Mitsushita, J. (2008). A possible placental factor for preeclampsia: sFlt-1. Current Medicinal Chemistry 15, 711–715.

Kleinrouweler, C.E., van Uitert, M., Moerland, P.D., Ris-Stalpers, C., van der Post, J.A.M., and Afink, G.B. (2013). Differentially Expressed Genes in the Pre- Eclamptic Placenta: A Systematic Review and Meta-Analysis. PLoS ONE 8, e68991.

Klonoff-Cohen, H.S., Savitz, D.A., Cefalo, R.C., and McCann, M.F. (1989). An epidemiologic study of contraception and preeclampsia. Jama 262, 3143–3147.

Knöfler, M., and Pollheimer, J. (2013). Human placental trophoblast invasion and differentiation: a particular focus on Wnt signaling. Front Genet 4, 190.

Koudstaal, J. (2000). Obstetric outcome of singleton pregnancies after IVF: a matched control study in four Dutch university hospitals. Human Reproduction 15, 1819–1825. 277

Koyano-Nakagawa, N., Shi, X., Rasmussen, T.L., Das, S., Walter, C.A., and Garry, D.J. (2015). Feedback Mechanisms Regulate Ets Variant 2( Etv2) Gene Expression and Hematoendothelial Lineages. J. Biol. Chem. 290, 28107–28119.

Köhrmann, M., LUO, M., Kaether, C., DesGROSEILLERS, L., Dotti, C.G., and Kiebler, M.A. (1999). Microtubule-dependent Recruitment of Staufen-Green Fluorescent Protein into Large RNA-containing Granules and Subsequent Dendritic Transport in Living Hippocampal Neurons. Mol Biol Cell 10, 2945–2953.

Kretz, M., Siprashvili, Z., Chu, C., Webster, D.E., Zehnder, A., Qu, K., Lee, C.S., Flockhart, R.J., Groff, A.F., Chow, J., et al. (2013). Control of somatic tissue differentiation by the long non-coding RNA TINCR. Nature 493, 231–235.

Kucukural, A., Ozadam, H., Singh, G., Moore, M.J., and Cenik, C. (2013). ASPeak: an abundance sensitive peak detection algorithm for RIP-Seq. Bioinformatics 29, 2485–2486.

Kusanovic, J.P., Romero, R., Chaiworapongsa, T., Erez, O., Mittal, P., Vaisbuch, E., Mazaki-Tovi, S., Gotsch, F., Edwin, S.S., Gomez, R., et al. (2009). A prospective cohort study of the value of maternal plasma concentrations of angiogenic and anti-angiogenic factors in early pregnancy and midtrimester in the identification of patients destined to develop preeclampsia. J. Matern. Fetal. Neonatal. Med. 22, 1021–1038.

Kusek, G., Campbell, M., Doyle, F., Tenenbaum, S.A., Kiebler, M., and Temple, S. Asymmetric Segregation of the Double-Stranded RNA Binding Protein Staufen2 during Mammalian Neural Stem Cell Divisions Promotes Lineage Progression. Cell Stem Cell 11, 505–516.

Labadorf, A., Hoss, A.G., Lagomarsino, V., and Latourelle, J.C. (2015). RNA sequence analysis of human huntington disease brain reveals an extensive increase in inflammatory and developmental gene expression. PLoS ONE.

Lam, J.K.W., Chow, M.Y.T., Zhang, Y., and Leung, S.W.S. (2015). siRNA Versus miRNA as Therapeutics for Gene Silencing. Mol Ther Nucleic Acids 4, e252.

Langmead, B., Trapnell, C., Pop, M., and Salzberg, S. (2009). Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 10, R25.1-R25.10.

Larroque, B. (2004). Survival of very preterm infants: Epipage, a population based cohort study. Archives of Disease in Childhood - Fetal and Neonatal Edition 89, 139F–144. 278

Laver, J.D., Li, X., Ancevicius, K., Westwood, J.T., Smibert, C.A., Morris, Q.D., and Lipshitz, H.D. (2013). Genome-wide analysis of Staufen-associated mRNAs identifies secondary structures that confer target specificity. Nucleic Acids Research 41, 9438–9460.

Leavey, K., Bainbridge, S.A., and Cox, B.J. (2015). Large scale aggregate microarray analysis reveals three distinct molecular subclasses of human preeclampsia. PLoS ONE 10, e0116508.

Lee, Y., Kim, M., Han, J., Yeom, K.-H., Lee, S., Baek, S.H., and Kim, V.N. (2004). MicroRNA genes are transcribed by RNA polymerase II. Embo J 23, 4051–4060.

Lekva, T., Lyle, R., Roland, M.C.P., Friis, C., Bianchi, D.W., Jaffe, I.Z., Norwitz, E.R., Bollerslev, J., Henriksen, T., and Ueland, T. (2016). Gene expression in term placentas is regulated more by spinal or epidural anesthesia than by late- onset preeclampsia or gestational diabetes mellitus. Sci Rep 1–12.

Levine, R.J., Lam, C., Qian, C., Yu, K.F., Maynard, S.E., Sachs, B.P., Sibai, B.M., Epstein, F.H., Romero, R., Thadhani, R., et al. (2006). Soluble endoglin and other circulating antiangiogenic factors in preeclampsia. The New England Journal of Medicine 355, 992–1005.

Levine, R.J., Maynard, S.E., Qian, C., Lim, K.-H., England, L.J., Yu, K.F., Schisterman, E.F., Thadhani, R., Sachs, B.P., Epstein, F.H., et al. (2004). Circulating angiogenic factors and the risk of preeclampsia. The New England Journal of Medicine 350, 672–683.

Levine, R.J., Thadhani, R., Qian, C., Lam, C., Lim, K.-H., Yu, K.F., Blink, A.L., Sachs, B.P., Epstein, F.H., Sibai, B.M., et al. (2005). Urinary placental growth factor and risk of preeclampsia. Jama 293, 77–85.

Lewis, B.P., Shih, I., Jones-Rhoades, M.W., Bartel, D.P., and Burge, C.B. (2003). Prediction of Mammalian MicroRNA Targets. Cell 115, 787–798. Li, J.H., Liu, S., Zhou, H., Qu, L.H., and Yang, J.H. (2014). StarBase v2.0: Decoding miRNA-ceRNA, miRNA-ncRNA and protein-RNA interaction networks from large-scale CLIP-Seq data. Nucleic Acids Res. 42, 92–97.

Li, L.-B., Yu, Z., Teng, X., and Bonini, N.M. (2008). RNA toxicity is a component of ataxin-3 degeneration in Drosophila. Nature 453, 1107–1111.

Lianoglou, S., Garg, V., Yang, J.L., Leslie, C.S., and Mayr, C. (2013). Ubiquitously transcribed genes use alternative polyadenylation to achieve tissue- specific expression. Genes Dev. 27, 2380–2396. 279

Lin, B., Rommens, J.M., Graham, R.K., Kalchman, M., MacDonald, H., Nasir, J., Delaney, A., Goldberg, Y.P., and Hayden, M.R. (1993). Differential 3' polyadenylation of the Huntington disease gene results in two mRNA species with variable tissue expression. Human Molecular Genetics 2, 1541–1545.

Lin, L., Park, J.W., and Ramachandran, S. (2016). Transcriptome sequencing reveals aberrant alternative splicing in Huntington's disease. Human Molecular ….

Liu, Z.R., Wilkie, A.M., Clemens, M.J., and Smith, C.W. (1996). Detection of double-stranded RNA-protein interactions by methylene blue-mediated photo- crosslinking. Rna 2, 611–621.

Llurba, E., Gratacós, E., Martıń -Gallán, P., Cabero, L., and Dominguez, C. (2004). A comprehensive study of oxidative stress and antioxidant status in preeclampsia and normal pregnancy. Free Radical Biology and Medicine 37, 557–570.

Love, M.I., Huber, W., and Anders, S. (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology 15, 31.

Luef, B.M., Andersen, L.B., Renäult, K.M., Nohr, E.A., Jørgensen, J.S., and Christesen, H.T. (2016). Validation of hospital discharge diagnoses for hypertensive disorders of pregnancy. Acta Obstet Gynecol Scand 95, 1288– 1294.

LUO, M., DUCHA NE, T.F., and DesGROSEILLERS, L. (2002a). Molecular mapping of the determinants involved in human Staufen?ribosome association. Biochem. J. 365, 817–824.

LUO, M., DUCHAÎNE, T.F., and DesGROSEILLERS, L. (2002b). Molecular mapping of the determinants involved in human Staufen-ribosome association. Biochem. J. 365, 817–824.

Luo, R., Wang, Y., Xu, P., Cao, G., Zhao, Y., Shao, X., Li, Y.-X., Chang, C., Peng, C., and Wang, Y.-L. (2016). Hypoxia-inducible miR-210 contributes to preeclampsia via targeting thrombospondin type I domain containing 7A. Sci Rep 1–11.

Lyall, F., Bulmer, J.N., Duffie, E., Cousins, F., Theriault, A., and Robson, S.C. (2001). Human trophoblast invasion and spiral artery transformation: the role of PECAM-1 in normal pregnancy, preeclampsia, and fetal growth restriction. Am. J. Pathol. 158, 1713–1721. 280

Macchi, P., Brownawell, A.M., Grunewald, B., DesGROSEILLERS, L., Macara, I.G., and Kiebler, M.A. (2004). The brain-specific double-stranded RNA-binding protein Staufen2: nucleolar accumulation and isoform-specific exportin-5- dependent export. J. Biol. Chem. 279, 31440–31444.

MACDONALD, M. (1993). A novel gene containing a trinucleotide repeat that is expanded and unstable on Huntington's disease chromosomes. Cell 72, 971– 983.

Magnussen, E.B., Vatten, L.J., Lund-Nilsen, T.I., Salvesen, K.A., Davey Smith, G., and Romundstad, P.R. (2007). Prepregnancy cardiovascular risk factors as predictors of pre-eclampsia: population based cohort study. Bmj 335, 978–978.

Magpie Trial Follow-Up Study Collaborative Group (2006). The Magpie Trial: a randomised trial comparing magnesium sulphate with placebo for pre-eclampsia. Outcome for children at 18 months. BJOG: an International Journal of Obstetrics & Gynaecology 114, 289–299.

Maher-Laporte, M., and DesGROSEILLERS, L. (2010). Genome wide identification of Staufen2-bound mRNAs in embryonic rat brains. BMB Rep 43, 344–348.

Malshe, A.K., and Sibai, B.M. (2017). Angiogenic and Antiangiogenic Markers for Prediction and Risk Classification of Preeclampsia. Clin Obstet Gynecol 60, 134– 140.

Marión, R.M., Fortes, P., Beloso, A., Dotti, C., and Ortín, J. (1999). A Human Sequence Homologue of Staufen Is an RNA-Binding Protein That Is Associated with Polysomes and Localizes to the Rough Endoplasmic Reticulum. Molecular and Cellular Biology 19, 2212–2219.

Martel, C., Dugré-Brisson, S., Boulay, K., Breton, B., Lapointe, G., Armando, S., Trépanier, V., Duchaîne, T., Bouvier, M., and DesGROSEILLERS, L. (2010). Multimerization of Staufen1 in live cells. Rna 16, 585–597.

Martin, M. (2011). Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.Journal 17, 10.

Martinez-Vicente, M., Talloczy, Z., Wong, E., Tang, G., Koga, H., Kaushik, S., de Vries, R., Arias, E., Harris, S., Sulzer, D., et al. (2010). Cargo recognition failure is responsible for inefficient autophagy in Huntington’s disease. Nat. Neurosci. 13, 567–576.

Maynard, S.E., and Karumanchi, S.A. (2011). Angiogenic factors and 281

preeclampsia. Ysnep 31, 33–46.

Maynard, S.E., Min, J.-Y., Merchan, J., Lim, K.-H., Li, J., Mondal, S., Libermann, T.A., Morgan, J.P., Sellke, F.W., Stillman, I.E., et al. (2003). Excess placental soluble fms-like tyrosine kinase 1 (sFlt1) may contribute to endothelial dysfunction, hypertension, and proteinuria in preeclampsia. J. Clin. Invest. 111, 649–658.

Maynard, S.E., Venkatesha, S., Thadhani, R., and Karumanchi, S.A. (2005). Soluble Fms-like tyrosine kinase 1 and endothelial dysfunction in the pathogenesis of preeclampsia. Pediatric Research 57, 1R–7R.

Mayr, C., and Bartel, D.P. (2009). Widespread Shortening of 3’UTRs by Alternative Cleavage and Polyadenylation Activates Oncogenes in Cancer Cells. Cell 138, 673–684.

McCarthy, D.J., Chen, Y., and Smyth, G.K. (2012). Differential expression analysis of multifactor RNA-Seq experiments with respect to biological variation. Nucleic Acids Research 40, 4288–4297.

McDonald, S.D., Malinowski, A., Zhou, Q., Yusuf, S., and Devereaux, P.J. (2008). Cardiovascular sequelae of preeclampsia/eclampsia: A systematic review and meta-analyses. American Heart Journal 156, 918–930.

McGinnis, R., Steinthorsdottir, V., Williams, N.O., Thorleifsson, G., Shooter, S., Hjartardottir, S., Bumpstead, S., Stefansdottir, L., Hildyard, L., Sigurdsson, J.K., et al. (2017). Variants in the fetal genome near FLT1 are associated with risk of preeclampsia. Nat Genet 77, 127.

McKeeman, G.C., Ardill, J.E.S., Caldwell, C.M., Hunter, A.J., and McClure, N. (2004). Soluble vascular endothelial growth factor receptor-1 (sFlt-1) is increased throughout gestation in patients who have preeclampsia develop. Ymob 191, 1240–1246.

Menalled, L.B., Sison, J.D., Dragatsis, I., Zeitlin, S., and Chesselet, M.-F. (2003). Time course of early motor and neuropathological anomalies in a knock-in mouse model of Huntington’s disease with 140 CAG repeats. J. Comp. Neurol. 465, 11–26.

Menendez, D., Inga, A., Snipe, J., Krysiak, O., Schonfelder, G., and Resnick, M.A. (2007). A Single-Nucleotide Polymorphism in a Half-Binding Site Creates p53 and Estrogen Receptor Control of Vascular Endothelial Growth Factor Receptor 1. Molecular and Cellular Biology 27, 2590–2600. 282

de Mezer, M., Wojciechowska, M., Napierala, M., Sobczak, K., and Krzyzosiak, W.J. (2011). Mutant CAG repeats of Huntingtin transcript fold into hairpins, form nuclear foci and are targets for RNA interference. Nucleic Acids Res. 39, 3852– 3863.

Micklem, D.R., Adams, J., Grünert, S., and St Johnston, D. (2000). Distinct roles of two conserved Staufen domains in oskar mRNA localization and translation. Embo J 19, 1366–1377.

Mi, H., Muruganujan, A., and Thomas, P.D. (2013). PANTHER in 2013: Modeling the evolution of gene function, and other gene attributes, in the context of phylogenetic trees. Nucleic Acids Res. 41, 377–386.

Mittal, S., Aslam, A., Doidge, R., Medica, R., and Winkler, G.S. (2011). The Ccr4a (CNOT6) and Ccr4b (CNOT6L) deadenylase subunits of the human Ccr4- Not complex contribute to the prevention of cell death and senescence. Mol. Biol. Cell 22, 748–758.

Miura, P., Shenker, S., Andreu-Agullo, C., Westholm, J.O., and Lai, E.C. (2013). Widespread and extensive lengthening of 3' UTRs in the mammalian brain. Genome Research 23, 812–825.

Monshausen, M., Putz, U., Rehbein, M., Schweizer, M., DesGROSEILLERS, L., Kuhl Dietmar, Richter, D., and Kindler, S. (2001). Two rat brain Staufen isoforms differentially bind RNA. Journal of Neurochemistry 76, 155–165.

Moster, D., Lie, R.T., and Markestad, T. (2008). Long-term medical and social consequences of preterm birth. The New England Journal of Medicine 359, 262– 273.

Mouillet, J.-F., Ouyang, Y., Coyne, C.B., and Sadovsky, Y. (2015). MicroRNAs in placental health and disease. American Journal of Obstetrics and Gynecology 213, S163–S172.

Moumné, L., Betuing, S., and Caboche, J. (2013). Multiple Aspects of Gene Dysregulation in Huntington’s Disease. Front. Neurol. 4, 127.

Munaut, C., Lorquet, S., Pequeux, C., Blacher, S., Berndt, S., Frankenne, F., and Foidart, J.-M. (2008). Hypoxia is responsible for soluble vascular endothelial growth factor receptor-1 (VEGFR-1) but not for soluble endoglin induction in villous trophoblast. Human Reproduction 23, 1407–1415.

MYATT, L., and WEBSTER, R.P. (2009). Vascular biology of preeclampsia. J. 283

Thromb. Haemost. 7, 375–384.

Mykowska, A., Sobczak, K., and Wojciechowska, M. (2011). CAG repeats mimic CUG repeats in the misregulation of alternative splicing. Nucleic Acids ….

Nagamatsu, T., Fujii, T., Kusumi, M., Zou, L., Yamashita, T., Osuga, Y., Momoeda, M., Kozuma, S., and Taketani, Y. (2004). Cytotrophoblasts up- regulate soluble fms-like tyrosine kinase-1 expression under reduced oxygen: an implication for the placental vascular development and the pathophysiology of preeclampsia. Endocrinology 145, 4838–4845.

Nautiyal, J. (2017). Transcriptional coregulator RIP140: an essential regulator of physiology. J. Mol. Endocrinol. 58, R147–R158.

Nautiyal, J., Christian, M., and Parker, M.G. (2013). Distinct functions for RIP140 in development, inflammation, and metabolism. Trends Endocrinol. Metab. 24, 451–459.

Navaroli, D.M., Khvorova, A., Pandarinathan, L., Fogarty, K., C, S., Bellve, K., Prot, M., and Corvera, S. Self-delivering therapeutic siRNA internalization through a distinct class of early endosomes.

Need, J.A., Bell, B., Meffin, E., and Jones, W.R. (1983). Pre-eclampsia in pregnancies from donor inseminations. J. Reprod. Immunol. 5, 329–338.

Ness, R.B., and Roberts, J.M. (1996). Heterogeneous causes constituting the single syndrome of preeclampsia: a hypothesis and its implications. Ymob 175, 1365–1370.

Nevo, O. (2006). Increased expression of sFlt-1 in in vivo and in vitro models of human placental hypoxia is mediated by HIF-1. AJP: Regulatory, Integrative and Comparative Physiology 291, R1085–R1093.

Nisell, H., Lintu, H., Lunell, N.O., Möllerström, G., and Pettersson, E. (1995). Blood pressure and renal function seven years after pregnancy complicated by hypertension. BJOG: an International Journal of Obstetrics &Amp; Gynaecology 102, 876–881.

Nogales-Cadenas, R., Carmona-Saez, P., Vazquez, M., Vicente, C., Yang, X., Tirado, F., Carazo, J.M., and Pascual-Montano, A. (2009). GeneCodis: interpreting gene lists through enrichment analysis and integration of diverse biological information. Nucleic Acids Research 37, W317–W322.

Noori, M., Donald, A.E., Angelakopoulou, A., Hingorani, A.D., and Williams, D.J. (2010). Prospective Study of Placental Angiogenic Factors and Maternal 284

Vascular Function Before and After Preeclampsia and Gestational Hypertension. Circulation 122, 478–487.

Nott, A., Le Hir, H., and Moore, M.J. (2004). Splicing enhances translation in mammalian cells: an additional function of the exon junction complex. Genes & Development 18, 210–222.

Norbury, C.J. (2013). Cytoplasmic RNA: a case of the tail wagging the dog. Nat Rev Cancer 13, 643–653.

Ong, S. (2004). Pre-eclampsia: A historical perspective (Pre-eclampsia current perspectives on management. …).

Østergaard, M.E., Southwell, A.L., Kordasiewicz, H., Watt, A.T., Skotte, N.H., Doty, C.N., Vaid, K., Villanueva, E.B., Swayze, E.E., Bennett, C.F., et al. (2013). Rational design of antisense oligonucleotides targeting single nucleotide polymorphisms for potent and allele selective suppression of mutant Huntingtin in the CNS. Nucleic Acids Res. 41, 9634–9650.

Palmer, K.R., Kaitu'u-Lino, T.J., Hastie, R., Hannan, N.J., Ye, L., Binder, N., Cannon, P., Tuohey, L., Johns, T.G., Shub, A., et al. (2015). Placental-Specific sFLT-1 e15a Protein Is Increased in Preeclampsia, Antagonizes Vascular Endothelial Growth Factor Signaling, and Has Antiangiogenic Activity. Hypertension 66, 1251–1259.

Panwar, B., Omenn, G.S., and Guan, Y. (2017). miRmine: A Database of Human miRNA Expression Profiles. Bioinformatics 2017, doi: 10.1093/bioinformatics/btx019.

Park, J.E., Yi, H., Kim, Y., Chang, H., and Kim, V.N. (2016). Regulation of Poly(A) Tail and Translation during the Somatic Cell Cycle. Mol. Cell 62, 462– 471.

Park, E., Gleghorn, M.L., and Maquat, L.E. (2013). Staufen2 functions in Staufen1-mediated mRNA decay by binding to itself and its paralog and promoting UPF1 helicase but not ATPase activity. Pnas 110, 405–412.

Pfister, E.L., Kennington, L., Straubhaar, J., Wagh, S., Liu, W., DiFiglia, M., Landwehrmeyer, B., Vonsattel, J.-P., Zamore, P.D., and Aronin, N. (2009). Five siRNAs targeting three SNPs may provide therapy for three-quarters of Huntington’s disease patients. Curr. Biol. 19, 774–778.

285

Pavličev, M., Wagner, G.P., Chavan, A.R., Owens, K., Maziarz, J., Dunn- Fletcher, C., Kallapur, S.G., Muglia, L., and Jones, H. (2017). Single-cell transcriptomics of the human placenta: inferring the cell communication network of the maternal-fetal interface. Genome Research 27, 349–361.

Pennington, K.A., Schlitt, J.M., Jackson, D.L., Schulz, L.C., and Schust, D.J. (2011). Preeclampsia: multiple approaches for a multifactorial disease. Disease Models & Mechanisms 5, 9–18.

Perrat, P.N., DasGupta, S., Wang, J., Theurkauf, W., Weng, Z., Rosbash, M., and Waddell, S. (2013). Transposition-driven genomic heterogeneity in the Drosophila brain. Science 340, 91–95.

Pfister, E.L., Kennington, L., Straubhaar, J., Wagh, S., Liu, W., Difiglia, M., Landwehrmeyer, B., Vonsattel, J.-P., Zamore, P.D., and Aronin, N. (2009). Five siRNAs Targeting Three SNPs May Provide Therapy for Three-Quarters of Huntington's Disease Patients. Current Biology 19, 774–778.

Powe, C.E., Levine, R.J., and Karumanchi, S.A. (2011). Preeclampsia, a disease of the maternal endothelium: the role of antiangiogenic factors and implications for later cardiovascular disease. Circulation 123, 2856–2869.

Powers, R.W., Roberts, J.M., Cooper, K.M., Gallaher, M.J., Frank, M.P., Harger, G.F., and Ness, R.B. (2005). Maternal serum soluble fms-like tyrosine kinase 1 concentrations are not increased in early pregnancy and decrease more slowly postpartum in women who develop preeclampsia. Ymob 193, 185–191.

Pratt, A., Da Silva Costa, F., Borg, A.J., Kalionis, B., Keogh, R., and Murthi, P. (2014). Placenta-derived angiogenic proteins and their contribution to the pathogenesis of preeclampsia. Angiogenesis 18, 115–123.

Quinlan, A.R., and Hall, I.M. (2010). BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841–842.

Rajakumar, A., Michael, H.M., Rajakumar, P.A., SHIBATA, E., Hubel, C.A., Karumanchi, S.A., Thadhani, R., WOLF, M., Harger, G., and MARKOVIC, N. (2005). Extra-placental expression of vascular endothelial growth factor receptor- 1, (Flt-1) and soluble Flt-1 (sFlt-1), by peripheral blood mononuclear cells (PBMCs) in normotensive and preeclamptic pregnant women. Placenta 26, 563– 573.

Rajakumar, A., Whitelock, K.A., and Weissfeld, L.A. (2001). Selective overexpression of the hypoxia-inducible transcription factor, HIF-2α, in placentas from women with preeclampsia (Biology of …). 286

Rajakumar, A. (2011). Placental syncytial knots and aggregates may contribute to circulating sFlt1 in preeclampsia. 1–39.

Rajakumar, A., Cerdeira, A.S., Rana, S., Zsengeller, Z., Edmunds, L., Jeyabalan, A., Hubel, C.A., Stillman, I.E., Parikh, S.M., and Karumanchi, S.A. (2012). Transcriptionally active syncytial aggregates in the maternal circulation may contribute to circulating soluble fms-like tyrosine kinase 1 in preeclampsia. Hypertension 59, 256–264.

Ramos, A., Grünert, S., Adams, J., Micklem, D.R., Proctor, M.R., Freund, S., Bycroft, M., St Johnston, D., and Varani, G. (2000). RNA recognition by a Staufen double-stranded RNA-binding domain. Embo J 19, 997–1009.

Rana, S., Powe, C.E., Salahuddin, S., and Verlohren, S. (2012). Angiogenic Factors and the Risk of Adverse Outcomes in Women With Suspected PreeclampsiaClinical Perspective. Circulation.

Ravel-Chapuis, A., Bélanger, G., Yadava, R.S., Mahadevan, M.S., DesGROSEILLERS, L., Côté, J., and Jasmin, B.J. (2012a). The RNA-binding protein Staufen1 is increased in DM1 skeletal muscle and promotes alternative pre-mRNA splicing. J Cell Biol 196, 699–712.

Ravel-Chapuis, A., Bélanger, G., Yadava, R.S., Mahadevan, M.S., DesGROSEILLERS, L., Côté, J., and Jasmin, B.J. (2012b). The RNA-binding protein Staufen1 is increased in DM1 skeletal muscle and promotes alternative pre-mRNA splicing. J Cell Biol 196, 699–712.

Reilly, M.T., Faulkner, G.J., Dubnau, J., Ponomarev, I., and Gage, F.H. (2013). The Role of Transposable Elements in Health and Diseases of the Central Nervous System. J. Neurosci. 33, 17577–17586.

Reiss, R.E., O'Shaughnessy, R.W., Quilligan, T.J., and Zuspan, F.P. (1987). Retrospective comparison of blood pressure course during preeclamptic and matched control pregnancies. American Journal of Obstetrics and Gynecology 156, 894–898.

Rhinn, H., Qiang, L., Yamashita, T., Rhee, D., Zolin, A., Vanti, W., and Abeliovich, A. (2012). Alternative α-synuclein transcript usage as a convergent mechanism in Parkinson’s disease pathology. Nat. Commun. 3, 1084.

Roberts, J.M., and Hubel, C.A. (2009). The two stage model of preeclampsia: variations on the theme. Placenta 30 Suppl A, S32–S37.

Robertson, W.B., Brosens, I., and Dixon, H.G. (1967). The pathological response 287

of the vessels of the placental bed to hypertensive pregnancy. J Pathol Bacteriol 93, 581–592.

Rodie, V.A., Freeman, D.J., Sattar, N., and Greer, I.A. (2004). Pre-eclampsia and cardiovascular disease: metabolic syndrome of pregnancy? Atherosclerosis 175, 189–202.

Romero, R., Nien, J.K., Espinoza, J., Todem, D., Fu, W., Chung, H., Kusanovic, J.P., Gotsch, F., Erez, O., Mazaki-Tovi, S., et al. (2009). A longitudinal study of angiogenic (placental growth factor) and anti-angiogenic (soluble endoglin and soluble vascular endothelial growth factor receptor-1) factors in normal pregnancy and patients destined to develop preeclampsia and deliver a small for gestational age neonate. The Journal of Maternal-Fetal & Neonatal Medicine 21, 9–23.

Rüb, U., Hoche, F., Brunt, E.R., Heinsen, H., Seidel, K., Del Turco, D., Paulson, H.L., Bohl, J., Von Gall, C., Vonsattel, J.P., et al. (2013). Degeneration of the cerebellum in huntingtons disease (HD): Possible relevance for the clinical picture and potential gateway to pathological mechanisms of the disease process. Brain Pathol. 23, 165–177.

Saigal, S., and Doyle, L.W. (2008). An overview of mortality and sequelae of preterm birth from infancy to adulthood. The Lancet.

Saito, S., and Nakashima, A. (2014). A review of the mechanism for poor placentation in early-onset preeclampsia: the role of autophagy in trophoblast invasion and vascular remodeling. J. Reprod. Immunol. 101-102, 80–88.

Sandberg, R., Neilson, J.R., Sarma, A., Sharp, P.A., and Burge, C.B. (2008). Proliferating Cells Express mRNAs with Shortened 3’ Untranslated Regions and Fewer MicroRNA Target Sites. Science (80-. ). 320, 1643–1647.

Sathasivam, K., Neueder, A., Gipson, T.A., Landles, C., Benjamin, A.C., Bondulich, M.K., Smith, D.L., Faull, R.L.M., Roos, R.A.C., Howland, D., et al. (2013). Aberrant splicing of HTT generates the pathogenic exon 1 protein in Huntington disease. Proceedings of the National Academy of Sciences of the United States of America 110, 2366–2370.

Sela, S., Itin, A., Natanson-Yaron, S., Greenfield, C., Goldman-Wohl, D., Yagel, S., and Keshet, E. (2008). A novel human-specific soluble vascular endothelial growth factor receptor 1: cell-type-specific splicing and implications to vascular endothelial growth factor homeostasis and preeclampsia. Circulation Research 102, 1566–1574. 288

Shepard, P.J., Choi, E.-A., Lu, J., Flanagan, L.A., Hertel, K.J., and Shi, Y. (2011). Complex and dynamic landscape of RNA polyadenylation revealed by PAS-Seq. Rna 17, 761–772.

Sheets, M.D., Ogg, S.C., and Wickens, M.P. (1990). Point mutations in AAUAAA and the poly(A) addition site: effects on the accuracy and efficiency of cleavage and polyadenylation {Iin vitro}. Nucleic Acids Res 18, 5799–5805.

Sheppard, S., Lawson, N.D., and Zhu, L.J. (2013). Accurate identification of polyadenylation sites from 3' end deep sequencing using a naive Bayes classifier. Bioinformatics 29, 2564–2571.

Shevell, T., Malone, F.D., Vidaver, J., Porter, T.F., Luthy, D.A., Comstock, C.H., Hankins, G.D., Eddleman, K., Dolan, S., Dugoff, L., et al. (2005). Assisted reproductive technology and pregnancy outcome. Obstet Gynecol 106, 1039– 1045.

Shibuya, M. (2011). Involvement of Flt-1 (VEGF receptor-1) in cancer and preeclampsia. Proc. Jpn. Acad., Ser. B 87, 167–178.

Shih, T., Peneva, D., Xu, X., Sutton, A., Triche, E., Ehrenkranz, R., Paidas, M., and Stevens, W. (2016). The Rising Burden of Preeclampsia in the United States Impacts Both Maternal and Child Health. Amer J Perinatol 33, 329–338.

Shirai, Y.T., Suzuki, T., Morita, M., Takahashi, A., and Yamamoto, T. (2014). Multifunctional roles of the mammalian CCR4-NOT complex in physiological phenomena. Front. Genet. 5, 1–11.

Sibai, B.M., Ewell, M., Levine, R.J., Klebanoff, M.A., Esterlitz, J., Catalano, P.M., Goldenberg, R.L., and Joffe, G. (1997). Risk factors associated with preeclampsia in healthy nulliparous women. The Calcium for Preeclampsia Prevention (CPEP) Study Group. Ymob 177, 1003–1010.

Sibai, B.M. (2004). Diagnosis, controversies, and management of the syndrome of hemolysis, elevated liver enzymes, and low platelet count. Obstet Gynecol 103, 981–991.

Singh, G., Ricci, E.P., and Moore, M.J. (2014). RIPiT-Seq: a high-throughput approach for footprinting RNA: protein complexes. Methods 65, 320–332.

Singh, G., Kucukural, A., Cenik, C., Leszyk, J.D., Shaffer, S.A., Weng, Z., and Moore, M.J. (2012). The Cellular EJC Interactome Reveals Higher-Order mRNP Structure and an EJC-SR Protein Nexus. Cell. 289

Slow, E.J., van Raamsdonk, J., Rogers, D., Coleman, S.H., Graham, R.K., Deng, Y., Oh, R., Bissada, N., Hossain, S.M., Yang, Y.-Z., et al. (2003). Selective striatal neuronal loss in a YAC128 mouse model of Huntington disease. Hum. Mol. Genet. 12, 1555–1567.

Sitras, V., Paulssen, R.H., Gr o naas, H., Leirvik, J., Hanssen, T.A., V aa rtun, A., and Acharya, G. (2009). Differential placental gene expression in severe preeclampsia. Placenta 30, 424–433.

Skjærven, R., Wilcox, A.J., and Lie, R.T. (2002). The Interval between Pregnancies and the Risk of Preeclampsia. New England Journal of Medicine 346, 33–38.

Smibert, P., Miura, P., Westholm, J.O., Shenker, S., May, G., Duff, M.O., Zhang, D., Eads, B.D., Carlson, J., Brown, J.B., et al. (2012). Global Patterns of Tissue- Specific Alternative Polyadenylation in Drosophila. Cell Rep. 1, 277–289.

Smith, G.N., Walker, M., Tessier, J.L., and Millar, K.G. (1997). Increased incidence of preeclampsia in women conceiving by intrauterine insemination with donor versus partner sperm for treatment of primary infertility. American Journal of Obstetrics … 177, 455–458.

Spies, N., Burge, C.B., and Bartel, D.P. (2013). 3′ UTR-Isoform choice has limited influence on the stability and translational efficiency of most mRNAs in mouse fibroblasts. Genome Res. 23, 2078–2090.

Soleymanlou, N., Jurisica, I., Nevo, O., Ietta, F., Zhang, X., Zamudio, S., Post, M., and Caniggia, I. (2005). Molecular Evidence of Placental Hypoxia in Preeclampsia. J. Clin. Endocrinol. Metab. 90, 4299–4308.

Sood, R., Zehnder, J.L., Druzin, M.L., and Brown, P.O. (2006). Gene expression patterns in human placenta. Proceedings of the National Academy of Sciences of the United States of America 103, 5478–5483.

Soto, E., Romero, R., Kusanovic, J.P., Ogge, G., Hussein, Y., Yeo, L., Hassan, S.S., Kim, C.J., and Chaiworapongsa, T. (2011). Late-onset preeclampsia is associated with an imbalance of angiogenic and anti-angiogenic factors in patients with and without placental lesions consistent with maternal underperfusion. The Journal of Maternal-Fetal & Neonatal Medicine 25, 498–507.

Sõber, S., Reiman, M., Kikas, T., Rull, K., Inno, R., Vaas, P., Teesalu, P., Marti, J.M.L., Mattila, P., and Laan, M. (2015). Extensive shift in placental transcriptome 290

profile in preeclampsia and placental origin of adverse pregnancy outcomes. Sci Rep 5, 13336.

St Johnston, D., Beuchle, D., and Nüsslein-Volhard, C. (1991). staufen, a gene required to localize maternal RNAs in the Drosophila egg. Cell 66, 51–63.

Suchowersky, O. (2014). Huntington disease: Clinical features and diagnosis.

Tabas-Madrid, D., Nogales-Cadenas, R., and Pascual-Montano, A. (2012a). GeneCodis3: a non-redundant and modular enrichment analysis tool for functional genomics. Nucleic Acids Research 40, W478–W483.

Tabas-Madrid, D., Nogales-Cadenas, R., and Pascual-Montano, A. (2012b). GeneCodis3: a non-redundant and modular enrichment analysis tool for functional genomics. Nucleic Acids Research 40, W478–W483.

Tanbo, T., Dale, P.O., Lunde, O., Moe, N., and Åbyholm, T. (1995). Obstetric outcome in singleton pregnancies after assisted reproduction. Obstet Gynecol.

Thadhani, R., Stampfer, M.J., Hunter, D.J., Manson, J.E., Solomon, C.G., and Curhan, G.C. (1999). High body mass index and hypercholesterolemia: risk of hypertensive disorders of pregnancy. Obstet Gynecol 94, 543–550.

Thadhani, R. (2010). Inching towards a targeted therapy for preeclampsia. Hypertension 55, 238–240.

Thadhani, R., Hagmann, H., Schaarschmidt, W., Roth, B., Cingoez, T., Karumanchi, S.A., Wenger, J., Lucchesi, K.J., Tamez, H., Lindner, T., et al. (2016). Removal of Soluble Fms-Like Tyrosine Kinase-1 by Dextran Sulfate Apheresis in Preeclampsia. Journal of the American Society of Nephrology : JASN 27, 903–913.

Thadhani, R., Kisner, T., Hagmann, H., Bossung, V., Noack, S., Schaarschmidt, W., Jank, A., Kribs, A., Cornely, O.A., Kreyssig, C., et al. (2011). Pilot study of extracorporeal removal of soluble fms-like tyrosine kinase 1 in preeclampsia. Circulation 124, 940–950.

Thomas, C.P.C., Raikwar, N.S.N., Kelley, E.A.E., and Liu, K.Z.K. (2010). Alternate processing of Flt1 transcripts is directed by conserved cis-elements within an intronic region of FLT1 that reciprocally regulates splicing and polyadenylation. Nucleic Acids Research 38, 5130–5140.

Thomas, C.P., Andrews, J.I., and Liu, K.Z. (2007). Intronic polyadenylation signal sequences and alternate splicing generate human soluble Flt1 variants and 291

regulate the abundance of soluble Flt1 in the placenta. Faseb 21, 3885–3895.

Thomas, C.P., Andrews, J.I., Raikwar, N.S., Kelley, E.A., Herse, F., Dechend, R., Golos, T.G., and Liu, K.Z. (2009a). A recently evolved novel trophoblast-enriched secreted form of fms-like tyrosine kinase-1 variant is up-regulated in hypoxia and preeclampsia. J. Clin. Endocrinol. Metab. 94, 2524–2530.

Thomas, M.G., Martinez Tosar, L.J., Desbats, M.A., Leishman, C.C., and Boccaccio, G.L. (2009b). Mammalian Staufen 1 is recruited to stress granules and impairs their assembly. Journal of Cell Science 122, 563–573.

Thu, D.C. V, Oorschot, D.E., Tippett, L.J., Nana, A.L., Hogg, V.M., Synek, B.J., Luthi-Carter, R., Waldvogel, H.J., and Faull, R.L.M. (2010). Cell loss in the motor and cingulate cortex correlates with symptomatology in Huntington’s disease. Brain 133, 1094–1110.

Tian, B., Hu, J., Zhang, H., and Lutz, C.S. (2005). A large-scale analysis of mRNA polyadenylation of human and mouse genes. Nucleic Acids Research 33, 201–212.

Tiemann, K., and Rossi, J.J. (2009). RNAi-based therapeutics-current status, challenges and prospects. 1, 142–151.

Tiwari, A.K.M., Mahdi, A.A., Zahra, F., Chandyan, S., Srivastava, V.K., and Negi, M.P.S. (2010). Evaluation of oxidative stress and antioxidant status in pregnant anemic women. Indian J Clin Biochem 25, 411–418.

Torry, D.S., Wang, H.S., Wang, T.H., Caudle, M.R., and Torry, R.J. (1998). Preeclampsia is associated with reduced serum levels of placenta growth factor. Ymob 179, 1539–1544.

Trapnell, C., Pachter, L., and Salzberg, S.L. (2009). TopHat: discovering splice junctions with RNA-Seq. Bioinformatics 25, 1105–1111.

Trushina, E., Heldebrant, M.P., Perez-Terzic, C.M., Bortolon, R., Kovtun, I. V, Badger, J.D., Terzic, A., Estévez, A., Windebank, A.J., Dyer, R.B., et al. (2003). Microtubule destabilization and nuclear entry are sequential steps leading to toxicity in Huntington’s disease. Proc. Natl. Acad. Sci. U. S. A. 100, 12171– 12176.

Tsatsaris, V., Goffin, F., Munaut, C., Brichant, J.-F., Pignon, M.-R., Noel, A., Schaaps, J.-P., Cabrol, D., Frankenne, F., and Foidart, J.-M. (2003). Overexpression of the Soluble Vascular Endothelial Growth Factor Receptor in 292

Preeclamptic Patients: Pathophysiological Consequences. J. Clin. Endocrinol. Metab. 88, 5555–5563.

Tubbergen, P., Lachmeijer, A.M., Althuisius, S.M., Vlak, M.E., van Geijn, H.P., and Dekker, G.A. (1999). Change in paternity: a risk factor for preeclampsia in multiparous women? J. Reprod. Immunol. 45, 81–88.

Tuovinen, S., Räikkönen, K., Kajantie, E., Pesonen, A.K., Heinonen, K., Osmond, C., Barker, D.J.P., and Eriksson, J.G. (2010). Depressive symptoms in adulthood and intrauterine exposure to pre-eclampsia: the Helsinki Birth Cohort Study. BJOG: an International Journal of Obstetrics &Amp; Gynaecology 117, 1236– 1242.

Tuovinen, S., Räikkönen, K., Pesonen, A.-K., Lahti, M., Heinonen, K., Wahlbeck, K., Kajantie, E., Osmond, C., Barker, D.J.P., and Eriksson, J.G. (2012). Hypertensive disorders in pregnancy and risk of severe mental disorders in the offspring in adulthood: the Helsinki Birth Cohort Study. J Psychiatr Res 46, 303– 310.

Turanov, A.A., Hassler, M.R., Ashar-Patel, A., Makris, A., Lo, A., Alterman, J.F., Coles, A.H., Haraszti, R.A., Godinho, B.M., Echeverria, D., et al. Development of Therapeutic anti-sFLT1 siRNA for the Treatment of Preeclampsia.

Uhlen, M., Fagerberg, L., Hallstrom, B.M., Lindskog, C., Oksvold, P., Mardinoglu, A., Sivertsson, A., Kampf, C., Sjostedt, E., Asplund, A., et al. (2015). Tissue- based map of the human proteome. Science (80-. ). 347, 1260419–1260419.

Vaiman, D., Calicchio, R., and Miralles, F. (2013). Landscape of Transcriptional Deregulations in the Preeclamptic Placenta. PLoS ONE 8, e65498–14.

Valdiviezo, C., Garovic, V.D., and Ouyang, P. (2012). Preeclampsia and Hypertensive Disease in Pregnancy: Their Contributions to Cardiovascular Risk. Clin Cardiol 35, 160–165.

Vanderlelie, J., Venardos, K., Clifton, V.L., Gude, N.M., Clarke, F.M., and Perkins, A.V. (2005). Increased biological oxidation and reduced anti-oxidant enzyme activity in pre-eclamptic placentae. Placenta 26, 53–58.

Várkonyi, T., Nagy, B., Füle, T., Tarca, A.L., Karászi, K., Schönléber, J., Hupuczi, P., Mihalik, N., Kovalszky, I., Rigó, J., et al. (2011). Microarray profiling reveals that placental transcriptomes of early-onset HELLP syndrome and preeclampsia are similar. Placenta 32 Suppl, S21–S29.

Venkatesha, S., Toporsian, M., Lam, C., Hanai, J.-I., Mammoto, T., Kim, Y.M., 293

Bdolah, Y., Lim, K.-H., Yuan, H.-T., Libermann, T.A., et al. (2006). Soluble endoglin contributes to the pathogenesis of preeclampsia. 12, 642–649.

Verlohren, S., Galindo, A., Schlembach, D., Zeisler, H., Herraiz, I., Moertl, M.G., Pape, J., Dudenhausen, J.W., Denk, B., and Stepan, H. (2010). An automated method for the determination of the sFlt-1/PIGF ratio in the assessment of preeclampsia. American Journal of Obstetrics and Gynecology 202, 161.e1– 161.e11.

Verlohren, S., Herraiz, I., Lapaire, O., Schlembach, D., Moertl, M., Zeisler, H., Calda, P., Holzgreve, W., Galindo, A., Engels, T., et al. (2012a). The sFlt-1/PlGF ratio in different types of hypertensive pregnancy disorders and its prognostic potential in preeclamptic patients. American Journal of Obstetrics and Gynecology 206, 58.e1–.e8.

Verlohren, S., Stepan, H., and Dechend, R. (2012b). Angiogenic growth factors in the diagnosis and prediction of pre-eclampsia. Clinical Science (London, England : 1979) 122, 43–52.

Vessey, J.P., Macchi, P., Stein, J.M., Mikl, M., Hawker, K.N., Vogelsang, P., Wieczorek, K., Vendra, G., Riefler, J., Tübing, F., et al. (2008). A loss of function allele for murine Staufen1 leads to impairment of dendritic Staufen1-RNP delivery and dendritic spine morphogenesis. Proceedings of the National Academy of Sciences of the United States of America 105, 16374–16379.

Villacé, P., Marión, R.M., and Ortín, J. (2004). The composition of Staufen- containing RNA granules from human cells indicates their role in the regulated transport and translation of messenger RNAs. Nucleic Acids Research 32, 2411– 2420.

Vonsattel, J.P., Myers, R.H., and Stevens, T.J. (1985). Neuropathological classification of Huntington's disease. … Of Neuropathology & ….

Vonsattel, J.P.G. (2008). Huntington disease models and human neuropathology: Similarities and differences. Acta Neuropathol. 115, 55–69.

Vonsattel, J.P.G., and DiFiglia, M. (1998). Huntington disease. J. Neuropathol. Exp. Neurol. 57, 369–384.

Vorlová, S., Rocco, G., Lefave, C.V., Jodelka, F.M., Hess, K., Hastings, M.L., Henke, E., and Cartegni, L. (2011). Induction of antagonistic soluble decoy receptor tyrosine kinases by intronic polyA activation. Molecular Cell 43, 927– 939.

Watanabe, N., Fujiwara, T., Suzuki, T., Jwa, S.C., Taniguchi, K., Yamanobe, Y., 294

Kozuka, K., and Sago, H. (2014). Is in vitro fertilization associated with preeclampsia? A propensity score matched study. BMC Pregnancy and Childbirth 14, 1–7.

Webster, R.P., and Myatt, L. (2007). Elucidation of the molecular mechanisms of preeclampsia using proteomic technologies. 1, 1147–1155.

White, R., Morganstein, D., Christian, M., Seth, A., Herzog, B., and Parker, M.G. (2007). Role of RIP140 in metabolic tissues: Connections to disease. FEBS Letters 582, 39–45.

Whitehouse, A.J.O., Robinson, M., Newnham, J.P., and Pennell, C.E. (2012). Do hypertensive diseases of pregnancy disrupt neurocognitive development in offspring? Paediatr Perinat Epidemiol 26, 101–108.

WHO (2011). WHO Recommendations for Prevention and Treatment of Pre- Eclampsia and Eclampsia. WHO Recommendations for Prevention and Treatment of Pre-Eclampsia and Eclampsia.

Wickham, L., Duchaîne, T., LUO, M., Nabi, I.R., and DesGROSEILLERS, L. (1999). Mammalian Staufen Is a Double-Stranded-RNA- and Tubulin-Binding Protein Which Localizes to the Rough Endoplasmic Reticulum. Molecular and Cellular Biology 19, 2220–2230.

Wiegand, H.L., Lu, S., and Cullen, B.R. (2003). Exon junction complexes mediate the enhancing effect of splicing on mRNA expression. Proceedings of the National Academy of Sciences of the United States of America 100, 11327– 11332.

Williams, J.W. (1915). Premature separation of the normally implanted placenta

Woldringh, G.H., Frunt, M.H.A., Kremer, J.A.M., and Spaanderman, M.E.A. (2006). Decreased ovarian reserve relates to pre-eclampsia in IVF/ICSI pregnancies. Human Reproduction 21, 2948–2954.

Xia, H., Ooi, L.L.P.J., and Hui, K.M. (2012). MiR-214 Targets β-Catenin Pathway to Suppress Invasion, Stem-Like Traits and Recurrence of Human Hepatocellular Carcinoma. PLoS ONE 7, e44206–e44213.

Xu, H., An, J.J., and Xu, B. (2017). Distinct cellular toxicity of two mutant huntingtin mRNA variants due to translation regulation. PLoS One 1–19.

Xu, P., Zhao, Y., Liu, M., Wang, Y., Wang, H., Li, Y., and Zhu, X. (2014). Variations of MicroRNAs in Human Placentas and Plasma From Preeclamptic 295

PregnancyNovelty and Significance.

Yagel, S. (2009). The developmental role of natural killer cells at the fetal- maternal interface. American Journal of Obstetrics and Gynecology 201, 344– 350.

Yamashita, A., Chang, T.-C., Yamashita, Y., Zhu, W., Zhong, Z., Chen, C.-Y. a, and Shyu, A.-B. (2005). Concerted action of poly(A) nucleases and decapping enzyme in mammalian mRNA turnover. Nat. Struct. Mol. Biol. 12, 1054–1063.

Yang, Y.-C.T., Di, C., Hu, B., Zhou, M., Liu, Y., Song, N., Li, Y., Umetsu, J., and Lu, Z. (2015). CLIPdb: a CLIP-seq database for protein-RNA interactions. BMC Genomics 16, doi: 10.1186/s12864-015-1273-2.

Young, B.C., Levine, R.J., and Karumanchi, S.A. (2010). Pathogenesis of preeclampsia. Annual Review of Pathology: Mechanisms of Disease 5, 173–192.

Zang, W., Wang, T., Wang, Y., Chen, X., Du, Y., Sun, Q., Li, M., Dong, Z., and Zhao, G. (2016). Knockdown of long non-coding RNA TP73-AS1 inhibits cell proliferation and induces apoptosis in esophageal squamous cell carcinoma. Oncotarget 7, 19960–19974.

Zeitlin, S., Liu, J.-P., Chapman, D.L., Papaioannou, V.E., and Efstratiadis, A. (1995). Increased apoptosis and early embryonic lethality in mice nullizygous for the Huntington’s disease gene homogue. Nat. Genet. 10, 155–163.

Zhang, H., Lee, J.Y., and Tian, B. (2005). Biased alternative polyadenylation in human tissues. Genome Biol. 6, R100.

Zheng, W., Chung, L.M., and Zhao, H. (2011). Bias detection and correction in RNA-Sequencing data. BMC Bioinformatics 12, 290.

Zuckerman, J.E., and Davis, M.E. (2015). Clinical experiences with systemically administered siRNA-based therapeutics in cancer. Nat Rev Drug Discov 14, 843–856.

Zusterzeel, P., Rütten, H., Roelofs, H., and Peters, W. (2001). Protein carbonyls in decidua and placenta of pre-eclamptic women as markers for oxidative stress. Placenta 22, 213–219.

Zuccato, C., Valenza, M., and Cattaneo, E. (2010). Molecular Mechanisms and Potential Therapeutical Targets in Huntington’s Disease. Physiol. Rev. 90, 905– 981.