546 Diabetes Care Volume 40, April 2017

Extracellular RNAs Are Associated Ravi Shah,1 Venkatesh Murthy,2 Michael Pacold,3 Kirsty Danielson,1 With Insulin Resistance and Kahraman Tanriverdi,4 Martin G. Larson,5 Kristina Hanspers,6 Alexander Pico,6 Metabolic Phenotypes Eric Mick,4 Jared Reis,7 Sarah de Ferranti,8 Elizaveta Freinkman,3 Daniel Levy,7 9 10 Diabetes Care 2017;40:546–553 | DOI: 10.2337/dc16-1354 Udo Hoffmann, Stavroula Osganian, Saumya Das,1 and Jane E. Freedman4

OBJECTIVE Insulin resistance (IR) is a hallmark of obesity and metabolic disease. Circulating extracellular RNAs (ex-RNAs), stable RNA molecules in plasma, may play a role in 1 IR, though most studies on ex-RNAs in IR are small. We sought to characterize the Department of Medicine, Massachusetts Gen- eral Hospital, Harvard Medical School, Boston, relationship between ex-RNAs and metabolic phenotypes in a large community- MA based human cohort. 2Department of Medicine and Radiology, Univer- sity of Michigan-Ann Arbor, Ann Arbor, MI RESEARCH DESIGN AND METHODS 3Metabolomics Core, Whitehead Institute, Mas- sachusetts Institute of Technology, Boston, MA We measured circulating plasma ex-RNAs in 2,317 participants without diabetes 4 EPIDEMIOLOGY/HEALTH SERVICES RESEARCH fi University of Massachusetts at Worcester, in the Framingham Heart Study (FHS) Offspring Cohort at cycle 8 and de ned Worcester, MA associations between ex-RNAs and IR (measured by circulating insulin level). 5Biostatistics Department, Boston University We measured association between candidate ex-RNAs and markers of adiposity. School of Public Health, Boston, MA 6 Sensitivity analyses included individuals with diabetes. In a separate cohort of Gladstone Institutes, San Francisco, CA 7Division of Cardiovascular Sciences, National 90 overweight/obese youth, we measured selected ex-RNAs and metabolites. Heart, Lung, and Blood Institute, Bethesda, MD Biology of candidate microRNAs was investigated in silico. 8Preventative Cardiology, Department of Medi- cine and Cardiology, Boston Children’s Hospital, RESULTS Boston, MA 9 The mean age of FHS participants was 65.8 years (56% female), with average BMI Department of Radiology, Massachusetts Gen- 2 eral Hospital, Boston, MA 27.7 kg/m ; participants in the youth cohort had a mean age of 15.5 years (60% 10Department of Medicine, Division of General 2 female), with mean BMI 33.8 kg/m . In age-, sex-, and BMI-adjusted models across Pediatrics, Boston Children’s Hospital, Boston, 391 ex-RNAs in FHS, 18 ex-RNAs were associated with IR (of which 16 were microRNAs). MA miR-122 was associated with IR and regional adiposity in adults and IR in children Corresponding authors: Ravi Shah, rvshah@ (independent of metabolites). Pathway analysis revealed metabolic regulatory roles partners.org, and Venkatesh Murthy, vlmurthy@ med.umich.edu. for miR-122, including regulation of IR pathways (AMPK, target of rapamycin signaling, Received 23 June 2016 and accepted 7 January and mitogen-activated protein kinase). 2017. CONCLUSIONS This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/ These results provide translational evidence in support of an important role of suppl/doi:10.2337/dc16-1354/-/DC1. ex-RNAs as novel circulating factors implicated in IR. R.S., V.M., M.P., and K.D. contributed equally to this work. The content is solely the responsibility of the au- Insulin resistance (IR) is a hallmark of human obesity and associated with the risk of thors and does not necessarily represent the of- developing diabetes and cardiovascular disease. IR can exist across the spectrum of ficial views of Harvard Catalyst, Harvard BMI from lean (,25 kg/m2) to overweight/obese (.25 kg/m2). These findings in- University, and its affiliated academic health dicate that a BMI-centric definition of obesity may not capture its underlying biology care centers or the National Institutes of Health. (1). Investigation of clinical and molecular markers that define architecture of IR has © 2017 by the American Diabetes Association. intensified, focusing on adipose tissue distribution and function, metabolite pro- Readers may use this article as long as the work fi is properly cited, the use is educational and not les, gut microbial diversity, and epigenetic and genetic variation. Recently, RNA for profit, and the work is not altered. More infor- located outside of cellular structures (extracellular RNAs [ex-RNAs]), circulating RNA mation is available at http://www.diabetesjournals molecules that are stable in plasma, have emerged as potential novel mediators in .org/content/license. care.diabetesjournals.org Shah and Associates 547

IR, potentially orchestrating control over measured by ELISA (Roche e411; Roche measured using an electrochemilumines- networks of gene expression. Indeed, an- Diagnostics, Indianapolis, IN; intra-assay cence immunoassay (Roche Elecsys/ imal models suggest exquisite regulation coefficient of variation [CV] 2.0%). HOMA Cobas immunoassay analyzer; CV 1.1– of circulating ex-RNAs in the develop- of IR (HOMA-IR), a marker of IR, was cal- 4.9%). HOMA-IR was calculated as speci- ment and resolution of obesity and in culated as the product of insulin (mIU/mL) fied above. Blood for plasma specimen metabolic cross talk between various or- and glucose (mmol/L) divided by 22.5 (9). storage was collected in the fasting state gans involved in adipocyte dysfunction Insulin (pmol/L) was converted to mIU/mL in lithium heparin-containing tubes, (2), suggesting their importance as clinical by multiplying by 0.144, and glucose centrifuged immediately at 4°C, then and functional biomarkers. As such, stud- (mg/dL) was converted to mmol/L by mul- transferred into cryovials, and frozen ies in small groups of patients with obe- tiplying by 0.0555. Interleukin-6 (IL-6; at 280°C for long-term storage. sity and IR have identified candidate ELISA, R&D Systems, Minneapolis, MN; ex-RNAs associated with metabolic dys- intra-assay CV 3.7%) and soluble tumor Quantification of Plasma Extracellular function (3–6), though there is absence necrosis factor receptor II (TNFRII; ELISA; Circulating ex-RNAs of validation in large at-risk populations R&D Systems) were measured at the time Detailed methods for quantification of and against metabolic phenotypes (e.g., of FHS Offspring Exam Cycle 8. Total adi- ex-RNAs in FHS have been published by visceral and hepatic adiposity) known to ponectin from FHS Offspring Exam Cycle our group (14). In the initial study in FHS, impact cardiometabolic risk. 7 (1998–2001) was included, measured by plasma small RNA sequencing was per- In this study, we investigate plasma- ELISA(R&DSystems)asdescribed(10).For formed in 40 FHS participants to deter- circulating ex-RNA abundance in two sep- ex-RNA analysis, an aliquot of 170 mLplasma mine which ex-RNAs were abundantly arate cohorts across the life span of samples was transferred to our laboratory in and reliably expressed in human plasma human obesitydthe 8th Framingham March 2014 and stored at 280°C for analy- (14). A plasma microRNA (miRNA) was Heart Study (FHS) Offspring Cohort sis. A subset of participants underwent chosen to be included for validation in (adults) and the POOL study (obese/ abdominal computed tomographic (CT) im- the full FHS Offspring Exam Cycle 8 co- overweight youth)dto study ex-RNAs as- aging (June 2002 to April 2005) for quanti- hort if it was expressed at .10 reads sociated with IR. We further investigate fication of visceral and subcutaneous per kilobase transcript per million reads the relationship between these ex-RNAs adipose tissue volume and fat attenuation mapped by sequencing. Because of the and several clinical hallmarks of metabolic (with lower attenuation as a marker of fat novelty and limited understanding of dysfunction. We subsequently tested the quality [11]) and liver attenuation (a surro- the other (non-microRNA) ex-RNA tar- association of two top candidate ex-RNAs gate of hepatic steatosis [12]), with previ- gets, we included all expressed small discovered in FHS with IR in the POOL ously described methods (13). nucleolar RNAs (snoRNAs) and Piwi- study and defined a relationship between Written informed consent was obtained interacting RNAs (piRNAs). We subse- miR-122 and IR independent of adverse from all study participants, with Institu- quently included only those plasma metabolite profiles in youth. tional Review Board approval at Boston ex-RNAs detectable in at least 100 FHS University, Massachusetts General Hospi- participants for this analysis (as deter- RESEARCH DESIGN AND METHODS tal, and the University of Massachusetts. mined from the final analytic cohort Framingham Heart Study specified below; N =2,317).Ofthe The Framingham Heart Study (FHS) is a Study Population and Clinical 2,822 plasma samples from the FHS Off- community-based, prospective study of Assessment in POOL spring Exam Cycle 8 in FHS, 59 (2%) cardiovascular disease conducted in Fra- POOL is an ongoing, prospective research subjects were excluded because of lab- mingham, MA, with serial examinations ev- registry of overweight and obese youth oratory error (e.g., inaccurate volume ery 4–8 years and concomitant in-depth and young adults, 2–25 years old (Boston of plasma pipetted, N = 31; poor protein phenotyping of metabolic traits over mul- Children’s Hospital, Boston, MA). Subjects precipitation performance, N = 23; or tiple prior examinations. The study design were local residents who were over- potential contamination, N =5),result- has been published (7). Standard anthro- weight or obese at study entry (BMI ing in 2,763 subjects. We subsequently pometric indices were measured as reported more than or equal to age/sex-specific excluded individuals who were not fast- (8). Diabetes was defined as fasting 85th percentile on Centers for Disease ing at the time of the blood draw for at plasma glucose $126 mg/dL, hemoglobin Control and Prevention growth charts least 8 h (N = 54), individuals without A1c $6.5% (where measured), or treat- for those ,18 years of age or $25 insulin or fasting blood glucose mea- ment with either insulin or a hypoglycemic kg/m2 for those $18 years of age). In- sured (N = 2), and individuals with di- agent. formed consent to participate was obtained abetes, as defined above (N =390), Blood collected in Framingham for the from a parent or legal guardian for minors yielding a final analytic cohort of 2,317 FHS Offspring Exam Cycle 8 (March (,18 years of age; with participant as- study participants. 2005 to January 2008) was analyzed in sent) or from adult participants. Clinical Finally, based on our analyses of this study. Venipuncture was performed and demographic data were collected, in- ex-RNAs in FHS, we selected two miRNAs on study participants in a supine position. cluding height and weight (measured (miR-122 and miR-192) to analyze in a Blood was collected into blood collection twice to the nearest 0.1 cm and kg, re- separate study of 90 obese/overweight tubes with a liquid-buffered sodium citrate spectively). Total body fat was measured young participants (POOL). Of note, additive (0.105 mol), centrifuged, and using a plethysmographic method that miRNA quantification in POOL was per- plasma separated and frozen at 280°C uses whole-body densitometry to deter- formed by our group in a separate project within 90 min of collection. Insulin was mine fat and fat-free mass. Insulin was with a separate set of 90 ex-RNA targets. 548 Extracellular RNAs and Insulin Resistance Diabetes Care Volume 40, April 2017

We chose to direct our analysis in POOL used in the miRNA-insulin/HOMA rela- their protein elements. The strategy of toward those ex-RNAs associated with tionships above (N = 2,317). In addition, preannotating pathways with targeting metabolic phenotypes in FHS. Therefore, we performed a sensitivity analysis to miRNA is described by Godard and van we only analyzed miR-122 and miR-192 in maximize population size (and power Eyll (17). The pathways were sourced this POOL. (piRNAs and snoRNAs were to detect association) across individuals from WikiPathways (18). The code to gen- not assessed in POOL.) with available imaging, anthropometric, erate the lookup table, the tables them- or biochemical measures in the overall selves, and the Pathway Finder tool are all Metabolite Profiling in POOL cohort (N = 2,763, including diabetes). In freely available as open source code at We performed polar metabolite profil- addition to BMI and waist circumfer- https://github.com/nrnb/mirna-pathway- ing as previously described (15) (across ence, we included several imaging- finder. The output of the tool is a ranked 82 metabolites). Metabolites were nor- based measures of regional adiposity: list of pathways with miRNA–protein tar- malized to internal standards. Metabo- visceral and subcutaneous fat volume get event counts. Gene identifier mapping lites below detection limit were counted (log-transformed), visceral-to-subcutaneous was performed using BridgeDb databases as “0” in analyses. fat volume ratio (a measure of propen- (19) derived directly from Ensembl release Statistical Analysis sity to store fat viscerally, log-transformed), 83. The mappings of miRNA-Entrez Gene Clinical and demographic data are pre- hepatic attenuation (a measure of he- targets were extracted from miRTarBase sented as mean and SD, with appropriate patic steatosis), and visceral and subcu- version 6.1 (20). tests for intergroup comparison (Wilcoxon taneous fat attenuation (a measure of Sixteen miRTarBase identifiers were en- for continuous and x2 for categorical). In fat “quality” and adipose tissue meta- tered into the Target Interaction Finder tool, the absence of formal glucose tolerance bolic function). Moreover, we examined which compares miRNA lists against an testing performed alongside ex-RNA mea- the relationship of our candidate ex-RNAs XGMML representation of the miRTarBase surements, we defined the degree of IR by with circulating adiponectin, IL-6, TNFRII, database produced by CyTargetLinker (21). plasma insulin level (16) (our primary IR and triglyceride-to-HDL ratio (all bio- The database contains experimentally val- measure); HOMA-IR was used as a second- markers log-transformed). Of note, for idated miRNA–gene target interactions. ary measure of IR. We recognize that these IL-6 and TNFRII, 116 subjects (from the The tool outputs a new XGMML file that are crude measures of IR, but they are overall set of assayed samples) had sam- focused on target interactions involving rapidlyavailableandusedinclinicalprac- ples run twice; we retained the higher the input list of miRNA. The XGMML file tice. As described above, we included 391 value for this analysis arbitrarily. We ad- was then imported into Cytoscape (22) for ex-RNAs (297 miRNAs, 36 snoRNAs, and 58 justed all models for age and sex. As in further filtering and visualization. A gene piRNAs) in our analysis (Supplementary previous models, we used an FDR correc- list from the “insulin signaling” pathway Table 1). Of note, because any given tion to guard against multiple hypothesis was used to perform a selection within ex-RNA was not necessarily detectable in testing. the complete miRNA gene target net- every FHS participant, models for insulin To examine the role of selected ex-RNAs work, and first neighbors were also se- or HOMA-IR had a different number of associated with IR in FHS on cardiome- lected. This subnetwork was extracted study subjects for each ex-RNA (denoted tabolic dysfunction in a younger popu- as a representation of the mixed targeting in regression models in RESULTS). Of note, lation, we studied the association of events by the 16 miRNAs and 69 insulin we specifically chose not to perform impu- two candidate ex-RNAs consistently as- signaling pathway genes. tation (or set below–detection limit ex-RNA sociated with IR-based phenotypes in expression to “23,” the highest PCR cycle FHS (miR-122 and miR-192) with IR in RESULTS number possible on the Biomark system) to 90 obese/overweight individuals from Baseline Characteristics of the FHS avoid bias. POOL. We next analyzed association be- Offspring and POOL Cohorts Our first step was to identify ex-RNAs tween miR-122 and insulin, independent Selected clinical, demographic, and re- associated with IR. We constructed age-, of age, sex, BMI, and circulating metabo- gional adiposity characterization of our sex-, and BMI-adjusted linear models to lite profiles (as defined by principal com- analytic sample in the FHS Offspring measure association of log-transformed ponents [PCs] of 81 polar metabolites, Exam Cycle 8 (N = 2,317) and POOL study insulin (primary) or HOMA-IR (secondary) using varimax rotation). (N = 90) are shown in Table 1. Our cohort with each ex-RNA (mean-centered and All statistics were performed with SAS was elderly (mean age 66 years old), 56% standardized). Given the multiple models 9.3 software (SAS Institute, Cary, NC) or female, and overweight (median BMI constructed (one model for each ex-RNA), R (R Project, www.rproject.org) with a 27.7 kg/m2). The POOL cohort had a we used a false discovery rate (FDR) cor- two-tailed P value ,0.05 (with appro- mean age of 15.5 years (range 4.6–25.5 rection using the Benjamini-Hochberg priate FDR thresholds, as noted) consid- years; 60% female), without diabetes, and method with a threshold of 0.05 (using ered statistically significant. with an average BMI percentile of 97% PROC MULTTEST in SAS) pooling raw (mean BMI 33.8 kg/m2). P values for all ex-RNAs together. Pathway Analysis and Network We next quantified the association of Visualization Identification of ex-RNAs Associated candidate ex-RNAs (associated with IR) Sixteen miRTarBase identifiers served as With IR and IR-Related Adiposity with CT-defined regional adiposity and input to the Pathway Finder bioinfor- Phenotypes in FHS adipokines using linear models. Our pri- matics tool, which compares miRNA lists We constructed age-, sex-, and BMI- mary analysis was based on imaging against a table of pathways, their miRNA adjusted linear models to identify data available from the population elements, and the miRNAs that target ex-RNAs associated with IR. From the care.diabetesjournals.org Shah and Associates 549

Table 1—Clinical and biochemical characteristics in our study FHS Offspring Cohort POOL Youth Cohort Variable N Value N Value Age (years) 2,317 65.8 6 8.9 90 15.5 6 4.8 Female sex, n (%) 2,317 1,307 (56) 90 54 (60) Current smoking, n (%) 2,314 191 (8) dd BMI (kg/m2) 2,313 27.7 6 5.1 90 33.8 6 10.0 (percentile: 97 6 3) Waist circumference (cm) 2,302 97.1 6 14.2 dd Systolic blood pressure (mmHg) 2,315 128 6 17 90 110 6 11 Diastolic blood pressure (mmHg) 2,313 74 6 10 90 66 6 8 Glucose (mg/dL) 2,317 100 6 99080.26 6.9 Biochemical indices* Triglycerides (mg/dL) 2,317 113 6 63 90 92.3 6 55.6 HDL cholesterol (mg/dL) 2,316 59 6 18 90 47.0 6 11.5 Insulin (pmol/L) 2,317 69 6 46 90 17.8 6 13.3

Hemoglobin A1c, % (mmol/mol) 2,316 5.6 6 0.3 (38) dd IL-6 (pg/mL) 2,246 2.58 6 3.04 dd TNFRII (pg/mL) 2,314 2,592 6 1,033 dd Adiponectin (ng/mL) 1,812 10.6 6 6.3 dd HOMA-IR 2,317 2.51 6 1.82 90 3.59 6 2.95 C-reactive protein 2,315 3.2 6 7.3 90 4.5 6 9.6 Regional adiposity Percent body fat (%) dd90 38.4 6 10.3 Liver attenuation (HU) 1,089 65.9 6 9.0 dd Subcutaneous fat volume, cm3 1,061 2,961 6 1,299 dd Visceral fat volume, cm3 1,061 1,970 6 1,022 dd Subcutaneous fat attenuation (HU) 1,061 2101 6 4.7 dd Visceral fat attenuation (HU) 1,061 293.7 6 4.5 dd Data are mean 6 SD unless otherwise noted. HU, Hounsfield units. *All biochemical indices were measured at the 8th examination, except for adiponectin, which was measured at the 7th examination. Timing of adiposity measures discussed in text. overall panel of 391 ex-RNAs included in (P =5.723 1027) and quality (P = (Supplementary Table 3). Interactions our analysis, we identified 16 miRNAs, 3.92 3 1026), and lower liver attenuation among all 16 target miRNAs and genes in 1 piRNA, and 1 snoRNA associated with (P =2.513 1025), but not subcutaneous the insulin signaling pathway are shown in insulin (Table 2), our primary outcome. fat (P = 0.005; did not survive FDR) or qual- Supplementary Fig. 1, suggesting signifi- An additional two ex-RNAs were associ- ity (P = 0.26). In addition, greater miR-122 cant cross-targeting of multiple IR-related ated with HOMA-IR, our secondary out- was associated with increased TNFRII and genes by multiple miRNAs identified by as- come. Of note, we observed a stepwise triglyceride-to-HDL ratio. miR-122 was not sociation. In addition, we visualized miRNA increase in plasma abundance of miR- significantly associated with adiponectin gene expression targeting events from se- 122 across quartiles of insulin after ad- after FDR (b = 20.024 for log-adiponectin; lected pathways (as described in RESEARCH justment for age, sex, and BMI (Fig. 1). P = 0.004; did not survive FDR). DESIGN AND METHODS) to identify functional tar- We next measured association among We found similar results for miR-122 gets of miR-122 (denoted in blue), miR-192 18 insulin-associated ex-RNAs with meta- when the entire population was consid- (denoted in red), and the other 14 miRNAs bolic phenotypes to begin to discern poten- ered (Supplementary Table 2B). In addi- (denoted in gray) on insulin signaling (Fig. 2 tial mechanisms by which these functional tion, we observed that miR-192 was and Supplementary Fig. 2). We selected biomolecules may promote IR. As noted in consistently associated with several car- four specific pathways given their impor- RESEARCH DESIGN AND METHODS, for the analysis of diometabolic phenotypes, including tance in IR: 1) “insulin signaling” (shown in miRNA–phenotype associations, we in- BMI, waist circumference, and liver at- Fig. 2); 2) “factors and pathways affect- cluded individuals from our primary popu- tenuation, but not subcutaneous fat. In ing insulin-like growth factor signaling”; 3) lation (N =2,317,nodiabetes)andina addition, across the overall population, target of rapamycin signaling; and sensitivity analysis (across all N = 2,763 par- we observed greater miR-122 and miR- 4) AMPK signaling (all shown in Supple- ticipants) in whom imaging or biochemical 192 associated with lower adiponectin. mentary Fig. 2). Of note, several genes indices were available. The results of these targeted by miR-122 had been previ- models are shown in Supplementary Table miRNAs Associated With IR in the FHS ously implicated in pathogenesis of IR, 2A. In our primary analysis (from cohort Target-Relevant Signaling Pathways including protein tyrosine phosphatase, excluding diabetes), we found consistent We next performed a pathway analysis to nonreceptor type 1 (also called PTP1B) associations between a greater plasma address whether the 16 miRNAs associated (23), mitogen-activated protein (MAP) abundance of miR-122 and greater BMI with insulin in FHS target pathways were kinases (24), and AMPK (25). (P =1.323 1027), waist circumference relevant to IR. We identified “insulin signal- Given consistent association with IR, (P =8.413 1026), visceral fat quantity ing”as a pathway targeted by all 16 miRNAs adiposity, and pathways involved in IR, 550 Extracellular RNAs and Insulin Resistance Diabetes Care Volume 40, April 2017

b Table 2—Ex-RNAs associated with IR (miR-122: = 0.14, P = 0.002; and miR- b Insulin (log-transformed) HOMA-IR (log-transformed) 192: = 0.14, P = 0.02, respectively). We found three PCs that explained 52.5% b b Candidate ex-RNAs N Estimated P value Estimated P value variance in the polar metabolome as- 2 2 miR-122-5p 2,198 0.041 1.68 3 10 8 0.046 2.98 3 10 9 sayed. Of note, the first PC was highly miR-16-5p 2,278 0.022 1.68 3 1023 0.024 1.54 3 1023 loaded on several different metabolites miR-191-5p 2,225 0.033 8.40 3 1029 0.037 2.56 3 1029 implicated in IR, including leucine, iso- miR-192-5p 1,725 0.047 3.13 3 1025 0.053 1.32 3 1025 leucine, and phenylalanine (branched- miR-194-5p 2,023 0.031 3.25 3 1025 0.033 3.52 3 1025 chain and aromatic amino acids). To miR-197-3p 2,013 dd0.038 2.00 3 1023 assess whether miR-122 had a metabolism- miR-19b-3p 2,230 0.034 3.08 3 1025 0.037 2.58 3 1025 independent association with IR, we es- miR-24-3p 2,220 dd0.032 1.00 3 1023 timated the association of insulin with miR-301b-3p 1,419 0.027 5.57 3 1025 0.029 6.93 3 1025 miR-122,adjustedforage,sex,BMI,and miR-30d-5p 2,221 0.030 1.26 3 1024 0.033 7.39 3 1025 all three metabolite PCs. Greater miR- miR-320a 2,208 0.029 1.07 3 1023 0.035 2.87 3 1024 122 was associated with greater insulin b miR-320b 1,665 20.015 2.87 3 1024 20.016 1.98 3 1024 ( = 0.10 log change in insulin per two- miR-342-3p 2,246 0.037 1.09 3 1024 0.045 1.48 3 1025 fold increase in miR-122; P =0.005),in- 2 2 dependent of age, sex, BMI, or metabolite miR-4446-3p 2,104 20.032 8.84 3 10 5 20.033 1.43 3 10 4 profile. miR-486-5p 2,268 0.028 2.98 3 1024 0.030 3.25 3 1024 3 24 3 24 miR-574-3p 1,917 0.030 4.28 10 0.035 1.35 10 CONCLUSIONS miR-616-5p 1,089 0.039 1.37 3 1023 0.040 2.34 3 1023 In a large community-based population of 3 25 3 25 miR-664b-3p 1,678 0.014 1.95 10 0.015 2.23 10 adults, we identify circulating ex-RNAs 2 3 23 2 3 23 piRNA-48383 1,799 0.025 1.62 10 0.027 2.14 10 that are associated with markers of IR 23 23 snoRNA-1210 1,531 0.015 1.74 3 10 0.016 1.56 3 10 and adiposity, independent of age, sex, All models were adjusted for age, sex, and BMI. We accounted for multiple hypothesis and BMI. Specifically, miR-122 was consis- testing with an appropriate prespecified 5% FDR threshold. N denotes number of observations tently related to dysfunctional adiposity in each model. An estimated b that is listed as “d” represents an ex-RNA that did not pass FDR. Each estimated b is change in log insulin or HOMA-IR per twofold increase (one PCR phenotypes previously demonstrated to cycle change) in plasma ex-RNA concentration. influence downstream cardiometabolic risk, including visceral and hepatic fat and selected adipokines and inflamma- we carried forward miR-122 and miR- measured by insulin (b = 0.12 log change tory mediators. In a separate cohort of 192 to the POOL cohort. per twofold increase in plasma miR-122; 90 overweight/obese youth without dia- P = 0.004) and HOMA-IR (b =0.12log betes, we demonstrated that miR-122 miR-122 Is Associated With Metabolic change per twofold increase in miR-122; was associated with IR independent of Phenotypes in Youth P = 0.006). Neither miR-122 nor miR-192 metabolite profile (via metabolomics), After adjustment for age, sex, and BMI, a was associated with hs-CRP or percent age, sex, or BMI, suggesting that higher miR-122 (but not miR-192) abun- body fat in POOL, but they were associ- ex-RNAs may have a role in IR indepen- dance was associated with greater IR, as ated with triglyceride-to-HDL ratio dent of emerging metabolic markers of IR. Based on in silico pathway analyses for the 16 miRNAs found in FHS (of total 18 ex-RNAs), we found that the identified miRNAs targeted several key pathways previously implicated in IR (including mammalian target of rapamycin, insulin signaling, and AMPK), with significant cross-targeting of multiple IR-related genes by multiple miRNAs (Fig. 2). Specif- ically, miR-122 targeted several genes previously implicated in IR, including genes involved in muscle responses to in- sulin (e.g., PRKAB1, a subunit of AMPK, a critical regulator of metabolism in IR [26]). Collectively, these findings provide translational support for a role of ex-RNAs (specifically miR-122) in IR across weight class, metabolism, and age, calling for Figure 1—Age, sex, and BMI-adjusted plasma abundance of miR-122 across quartiles of circu- lating insulin. Comparisons across all quartiles were statistically significant (after Bonferroni further mechanistic investigation to correction for type 1 error), except 1st vs. 2nd quartile and 2nd vs. 3rd quartile. delineate a role for ex-RNAs in the meta- bolic architecture of IR. care.diabetesjournals.org Shah and Associates 551

Figure 2—Visualization of selected miRNA targeting events on the insulin signaling pathway (from WikiPathways). The genes targeted by miRNA per pathway, as counted in Supplementary Table 3, are visualized in this figure for selected pathways. Pathway was imported into Cytoscape from WikiPathways, and ID mapping was performed to obtain Entrez Gene identifiers for each gene. An intermediate file from the Pathway Finder tool was parsed and imported into Cytoscape to supply the mappings between Entrez Gene and the selected set of miRNAs. A visual style was defined in Cytoscape to highlight any gene targeted by these miRNAs in preferential order: miR-122 (blue), miR-192 (red), and any of the other 14 possible miRNAs (gray). Other selected pathways from WikiPathways are shown in Supplementary Fig. 1.

There is increasing recognition that Humaninvestigationinbothchildren and regional adiposity, specifically miR- specific ex-RNAs function in pathways (6) and adults (3) has demonstrated sev- 122. Using pathway analysis across all critical to obesity and metabolic disease, eral miRNAs dysregulated in obesity and 16 miRNAs with curated pathways (of including adipocyte differentiation, an- progression of cardiometabolic disease, the total 18 ex-RNAs), we found that these giogenesis, hepatic steatosis, oxidative with dynamic changes in miRNAs during miRNAs targeted genes involved in central stress, and inflammation. Furthermore, weight loss. Despite these important ad- pathways of IR, with some genes targeted adipocyte-derived miRNAs can mediate vances, most studies have been limited by multiple miRNAs. These translational cross talk with circulating macrophages by sample size, profile preselected findings are in keeping with several prior (27) or hepatocytes (28), or in muscle miRNA candidates (excluding piRNAs reports from smaller cohorts on the signif- tissue (29), altering mRNA expression or snoRNAs), and restrict their popula- icance of miR-122 in obesity and IR path- of key intermediates involved in IR. tion of interest to obesity. Given the im- ogenesis (33). miR-122 is dynamically These findings implicate ex-RNAs as portant differences between animal regulated during surgical weight loss and functional biomarkers that may orches- models of diabetes or obesity and is associated with hepatic steatosis trate high-level transcriptional and met- human disease, large-scale human trans- (32,34), with a near abolition of circulating abolic control in humans. Accordingly, lational data alongside detailed obesity- miR-122 levels after bariatric surgery (35). there has been a surge in translational related phenotypes (e.g., regional adiposity) Seminal work by Esau et al. (36) demon- investigation in this area, demonstrating are critical. strated that direct antisense-mediated si- involvement of miRNAs in brown/white In this report, we identify ex-RNAs as- lencing of miR-122 caused a reduction in fat specification and adipose tissue in- sociated with IR in the FHS, demonstrating hepatic steatosis, decreased circulating flammation (30), pancreatic b-cell func- associations with key phenotypes central cholesterol levels, and global shifts in lipid tion (31), and hepaticsteatosis(32). to IR, including adipokines, inflammation, metabolism. Taken in concert with our 552 Extracellular RNAs and Insulin Resistance Diabetes Care Volume 40, April 2017

findings of an association between miR- age-independent role for ex-RNAs in in- by microRNAs in adipose tissue. J Clin Endocri- 122 and CT-determined hepatic fat in tegrating metabolic inputs in IR. Selected nol Metab 2015;100:E1467–E1476 Framingham, these findings suggest a miRNAs associated with IR target genes 5. Ortega FJ, Mercader JM, Moreno-Navarrete JM, et al. Profiling of circulating microRNAs re- potential role for miR-122 in hepatic implicated in IR in muscle and may veals common microRNAs linked to type 2 di- steatosis, a major comorbidity involved have a functional, trans-organ role in me- abetes that change with insulin sensitization. in IR pathogenesis. In addition, in pathway diating IR. These results provide large- Diabetes Care 2014;37:1375–1383 analyses, miR-122 appeared to target scale human translational epidemiologic 6. Prats-Puig A, Ortega FJ, Mercader JM, et al. PRKAB1, a member of the AMPK pathway, data to support a role for ex-RNAs in IR Changes in circulating microRNAs are associ- ated with childhood obesity. J Clin Endocrinol a regulator of IR in muscle, suggesting that and its metabolic consequences. Future Metab 2013;98:E1655–E1660 liver-derived miR-122 may target remote investigation into specificmechanisms 7. Feinleib M, Kannel WB, Garrison RJ, metabolically active tissues as an endo- and modulation of ex-RNA biology to re- McNamara PM, Castelli WP. The Framingham crine mediator of disease. Indeed, miR- duce the burden of diabetes is warranted. Offspring Study. Design and preliminary data. 122 has been found in circulating Prev Med 1975;4:518–525 exosomes (37) that may transfer epige- 8. Mahabadi AA, Massaro JM, Rosito GA, et al. Association of pericardial fat, intrathoracic fat, Acknowledgments. The authors thank the par- netic information across tissue types. and visceral abdominal fat with cardiovascular ticipants of the POOL and FHS, as well as the Finally, the finding of an association be- disease burden: the Framingham Heart Study. countless volunteers and administrative staff with- Eur Heart J 2009;30:850–856 tween IR and non-miRNA ex-RNAs (e.g., out whom this work would not be possible. The 9. Rutter MK, Wilson PW, Sullivan LM, Fox CS, piRNA and snoRNA) is intriguing, as the authors also thank Dr. Aifeng Zhang and Dr. Fei D’Agostino RB Sr, Meigs JB. Use of alternative role of non-miRNA ex-RNAs (e.g., piRNAs) Wang (Massachusetts General Hospital, Boston, thresholds defining insulin resistance to predict in metabolic diseases is just beginning to MA) for assistance in experimentation. Data from the ex-RNAs measured in the FHS have incident type 2 diabetes mellitus and cardiovas- be clarified: several non-miRNA species – been deposited in the National Institutes of Health cular disease. Circulation 2008;117:1003 1009 (e.g., piRNAs) have been recently shown Database of Genotypes and Phenotypes. 10. Hivert MF, Manning AK, McAteer JB, et al. to transfer epigenetic information from Funding. This work was supported by the Amer- Common variants in the adiponectin gene sperm to egg (38). Ultimately, although ican Heart Association (14FTF19940000 to R.S.), (ADIPOQ) associated with plasma adiponectin these human translational findings are as- the Harvard Catalyst (UL1-TR-001102), the levels, type 2 diabetes, and diabetes-related quantitative traits: the Framingham Offspring sociational in nature, they motivate fur- Thrasher Research Fund (to R.S.), the National Institutes of Health (K23-HL-127099), the National Study. Diabetes 2008;57:3353–3359 ther mechanistic research into ex-RNA Institutes of Health Common Fund Extracellular 11. Rosenquist KJ, Pedley A, Massaro JM, et al. biogenesis and the role of ex-RNAs in RNA Communication Consortium (UH3-TR- Visceral and subcutaneous fat quality and car- cell–cell communication and target organ 900921 to J.E.F. and UH3-TR-000901 to S.D.), diometabolic risk. JACC Cardiovasc Imaging – metabolic signaling. the Milton Foundation, New Balance Foundation 2013;6:762 771 ’ 12. Shah RV, Allison MA, Lima JA, et al. Liver The limitations of our study should be Obesity Prevention Center at Boston Children s Hospital, and Boston Children’sHospital(forthe steatosis and the risk of albuminuria: the viewed in light of its design. We focused POOL study toS.O.). The FHS is funded by National multi-ethnic study of atherosclerosis. J Nephrol on ex-RNAs commonly abundant in Institutes of Health contract N01-HC-25195. 2015;28:577–584 plasma of FHS participants, limiting dis- Duality of Interest. No potential conflicts of 13. Rosenquist KJ, Massaro JM, Pedley A, et al. covery of low-abundance ex-RNAs. In ad- interest relevant to this article were reported. Fat quality and incident cardiovascular disease, Author Contributions. R.S., V.M., M.P., M.G.L., all-cause mortality and cancer mortality. J Clin dition, we included those participants and E.M. performed statistical analyses. R.S., V.M., Endocrinol Metab 2015;100:227–234 with expressed levels of each miRNA in M.P., K.T., S.D., and J.E.F. wrote the manuscript. 14. Freedman JE, Gerstein M, Mick E, et al. Diverse regressions for insulin (not “imputing” M.P. and E.F. performed metabolomics analyses. human extracellular RNAs are widely detected in values for miRNAs that were below de- K.T. and J.E.F. performed RNA quantification. K.H. human plasma. Nat Commun 2016;7:11106 tection limit). Although this limits conclu- and A.P. performed bioinformatics analyses. D.L. 15. Birsoy K, Wang T, Chen WW, Freinkman E, performed data collection within FHS. U.H. per- sions to participants in whom each Abu-Remaileh M, Sabatini DM. An essential role formed CT analysis in FHS. S.O. and S.d.F. gathered of the mitochondrial electron transport chain in miRNA was expressed, complex traits data and samples in the POOL study. S.D. and J.E.F. cell proliferation is to enable aspartate synthe- like IR are likely influenced by variation jointly supervised the work. All authors provided sis. Cell 2015;162:540–551 in common epigenetic factors. Although critical revision of the manuscript. R.S., V.M., S.D., 16. Laakso M. How good a marker is insulin CT measures or biomarkers were not pre- and J.E.F. are the guarantors of this work and, as level for insulin resistance? Am J Epidemiol such, had full access to all the data in the study and 1993;137:959–965 sent in all FHS participants with ex-RNA take responsibility for the integrity of the data and fi 17. Godard P, van Eyll J. Pathway analysis from quanti cation (a potential source of bias), the accuracy of the data analysis. lists of microRNAs: common pitfalls and alternative the observed associations are strong in strategy. Nucleic Acids Res 2015;43:3490–3497 the largest cohort studied so far in meta- References 18. Kutmon M, Riutta A, Nunes N, et al. Wiki- bolic disease. Finally, use of gold-standard 1. Neeland IJ, Turer AT, Ayers CR, et al. Dysfunc- Pathways: capturing the full diversity of path- methods to determine insulin physiology tional adiposity and the risk of prediabetes and way knowledge. Nucleic Acids Res 2016;44: – in individuals with diabetes (e.g., clamp) type 2 diabetes in obese adults. JAMA 2012;308: D488 D494 1150–1159 19. van Iersel MP, Pico AR, Kelder T, et al. The will be important in future studies to ex- 2. Arner P, Kulyte´ A. MicroRNA regulatory net- BridgeDb framework: standardized access to tend our results to diabetes. works in human adipose tissue and obesity. Nat gene, protein and metabolite identifier map- In conclusion, we identified a group of Rev Endocrinol 2015;11:276–288 ping services. BMC Bioinformatics 2010;11:5 plasma-circulating ex-RNAs related to IR 3. Ortega FJ, Mercader JM, Catalan´ V, et al. Tar- 20. Chou CH, Chang NW, Shrestha S, et al. and related adiposity, inflammatory, and geting the circulating microRNA signature of miRTarBase 2016: updates to the experimen- obesity. Clin Chem 2013;59:781–792 tally validated miRNA-target interactions data- metabolic phenotypes. These associa- 4. Ortega FJ, Mercader JM, Moreno-Navarrete base. Nucleic Acids Res 2016;44:D239–D247 tions are independent of sex and BMI JM, et al. Surgery-induced weight loss is associ- 21. Kutmon M, Kelder T, Mandaviya P, Evelo CT, and conserved across age, suggesting an ated with the downregulation of genes targeted Coort SL. CyTargetLinker: a cytoscape app to care.diabetesjournals.org Shah and Associates 553

integrate regulatory interactions in network into blood circulation. Biochem Biophys Res 33. Wang R, Hong J, Cao Y, et al. Elevated cir- analysis. PLoS One 2013;8:e82160 Commun 2010;398:723–729 culating microRNA-122 is associated with obe- 22. Shannon P, Markiel A, Ozier O, et al. Cyto- 28. Sharma H, Estep M, Birerdinc A, et al. Ex- sity and insulin resistance in young adults. Eur J scape: a software environment for integrated pression of genes for microRNA-processing en- Endocrinol 2015;172:291–300 models of biomolecular interaction networks. zymes is altered in advanced non-alcoholic fatty 34.YamadaH,SuzukiK,IchinoN,etal.Associa- Genome Res 2003;13:2498–2504 liver disease. J Gastroenterol Hepatol 2013;28: tions between circulating microRNAs (miR-21, 23. Stull AJ, Wang ZQ, Zhang XH, Yu Y, Johnson 1410–1415 miR-34a, miR-122 and miR-451) and non-alcoholic – WD, Cefalu WT. Skeletal muscle protein tyrosine 29. Wang YC, Li Y, Wang XY, et al. Circulating fatty liver. Clin Chim Acta 2013;424:99 103 phosphatase 1B regulates insulin sensitivity in Af- miR-130b mediates metabolic crosstalk be- 35. Wu Q, Li JV, Seyfried F, et al. Metabolic phenotype-microRNA data fusion analysis of rican Americans. Diabetes 2012;61:1415–1422 tween fat and muscle in overweight/obesity. the systemic consequences of Roux-en-Y gastric 24. Wang CC, Goalstone ML, Draznin B. Molecular Diabetologia 2013;56:2275–2285 bypass surgery. Int J Obes 2015;39:1126–1134 mechanisms of insulin resistance that impact car- 30. Karbiener M, Scheideler M. MicroRNA func- 36. Esau C, Davis S, Murray SF, et al. miR-122 reg- diovascular biology. Diabetes 2004;53:2735–2740 tions in brite/brown fat - novel perspectives to- ulation of lipid metabolism revealed by in vivo an- 25. Ruderman NB, Carling D, Prentki M, wards anti-obesity strategies. Comput Struct tisense targeting. Cell Metab 2006;3:87–98 – Cacicedo JM. AMPK, insulin resistance, and Biotechnol J 2014;11:101 105 37. Bala S, Petrasek J, Mundkur S, et al. Circulat- the metabolic syndrome. J Clin Invest 2013; 31. Fernandez-Valverde SL, Taft RJ, Mattick JS. ing microRNAs in exosomes indicate hepatocyte b 123:2764–2772 MicroRNAs in -cell biology, insulin resistance, injury and inflammation in alcoholic, drug- 26. Long YC, Zierath JR. AMP-activated protein diabetes and its complications. Diabetes 2011; induced, and inflammatory liver diseases. Hepa- kinase signaling in metabolic regulation. J Clin 60:1825–1831 tology 2012;56:1946–1957 Invest 2006;116:1776–1783 32. Becker PP, Rau M, Schmitt J, et al. Perfor- 38. Donkin I, Versteyhe S, Ingerslev LR, et al. 27. Ogawa R, Tanaka C, Sato M, et al. Adipocyte- mance of serum microRNAs -122, -192 and -21 Obesity and bariatric surgery drive epigenetic derived microvesicles contain RNA that is trans- as biomarkers in patients with non-alcoholic variation of spermatozoa in humans. Cell Metab ported into macrophages and might be secreted steatohepatitis. PLoS One 2015;10:e0142661 2016;23:369–378