Tissue-Specific and Genetic Regulation of Insulin Sensitivity-Associated Transcripts in African Americans

Integrative multiomics analyses of adipose and muscle tissue transcripts, _S, and genotypes revealed novel genetic regulatory mechanisms of insulin resistance in African Americans.

Abstract

Context:

Compared with European Americans, African Americans (AAs) are more insulin resistant, have a higher insulin secretion response to glucose, and develop type 2 diabetes more often. Molecular processes and/or genetic variations contributing to altered glucose homeostasis in high-risk AAs remain uncharacterized.

Objective:

Adipose and muscle transcript expression profiling and genotyping were performed in 260 AAs to identify genetic regulatory mechanisms associated with insulin sensitivity (S_I). We hypothesized that: 1) transcription profiles would reveal tissue-specific modulation of physiologic pathways with S_I, and 2) a subset of S_I-associated transcripts would be controlled by DNA sequence variants as expression quantitative traits, and these variants in turn would be associated with S_I.

Design and Settings:

The cross-sectional research study was performed in a clinical research unit.

Participants:

Unrelated nondiabetic AAs were recruited for the study.

Main Outcome Measures:

S_I was measured by frequently sampled iv glucose tolerance test.

Results:

The expression levels of 2212 transcripts in adipose and 145 transcripts in muscle were associated with S_I. Genes involved in eIF2, eIF4-p70S6K, and mTOR signaling were modulated with S_I in both tissues. Genes involved in leukocyte extravasation signaling showed adipose-specific regulation, and genes involved in oxidative phosphorylation had discordant regulation between tissues. Intersecting cis-expression quantitative trait loci results with data from transcript-S_I association analysis identified cis-regulatory single nucleotide polymorphisms for 363 and 42 S_I-associated transcripts in adipose and muscle, respectively. Cis-eSNPs for three S_I-associated adipose transcripts, NINJ1, AGA, and CLEC10A were associated with S_I. Abrogation of NINJ1 induction in THP1 macrophages modulated expression of genes in chemokine signaling, cell adhesion, and angiogenesis pathways.

Conclusion:

This study identified multiple pathways associated with S_I; particularly discordant tissue-specific regulation of the oxidative phosphorylation pathway, and adipose-specific regulation of transcripts in the leukocyte extravasation signaling pathway that seem to be important in insulin resistance. Identification of single nucleotide polymorphisms associated with S_I and with modulation of expression of S_I-associated transcripts, including NINJ1, reveals novel genetic regulatory mechanisms of insulin resistance in AAs.

Type 2 diabetes (T2D) is a heterogeneous metabolic disorder that can result from dysfunctional interactions between complex physiological processes maintaining glucose homeostasis (1). Insulin resistance in muscle, adipose, and liver, impaired pancreatic β-cell insulin secretion, and obesity typically precede the development of T2D. Although the relative contributions and timing of these processes remain unclear (2, 3), longitudinal studies in high-risk individuals suggest that insulin resistance occurs years before glucose intolerance or β-cell failure (1, 4, 5). Investigating early pathophysiological changes may help to identify the molecular processes involved in T2D.

In healthy individuals, the relationship between insulin sensitivity (S_I) and the β-cell insulin response is reciprocal and nonlinear (hyperbolic), and ethnic-specific variation is observed in its distribution (2). Compared with age-, sex-, and body mass index (BMI) –matched non-Hispanic Caucasians, African Americans (AAs) are more insulin resistant (lower S_I), but show a greater acute insulin response to iv glucose (AIR_G) and a higher disposition index (DI = S_I × AIR_G) (6, 7). Unique evolutionary histories and ancestry-specific genetic variation could contribute to ethnic-specific physiology and higher risk for T2D in AAs. Genome-wide transcript profiling analyses implicated modulation of many genes in insulin-responsive tissues, involving derangement of multiple biological pathways in insulin-resistant individuals (1). Although elucidation of these pathways advanced our understanding of altered glucose homeostasis at the molecular level, most studies did not assess individuals of African ancestry. Molecular processes and/or genetic variations that may contribute to altered glucose homeostasis in high-risk populations, including AAs, remain uncharacterized.

Thus, we systematically analyzed genome-wide transcript expression profiles of sc adipose and skeletal muscle (insulin-responsive tissues) and genome-wide SNP genotypes from a metabolically characterized cohort of 260 unrelated and nondiabetic AAs. S_I measured by the frequently sampled iv glucose tolerance test (FSIGT) was the primary physiological outcome. Correlated secondary phenotypes were also assessed. These data were used in an integrative multiomics approach (Supplemental Figure 1) to test two hypotheses: 1) transcription profiles in adipose and muscle would correlate with S_I and might reveal tissue-specific modulation of physiologic pathways, and 2) a subset of S_I-associated transcripts would be controlled by DNA sequence variants as expression quantitative traits, and these variants in turn would be associated with S_I.

Materials and Methods

Human subjects

Studies were performed at the Wake Forest School of Medicine (WFSM) Clinical Research Unit. The study was approved by the WFSM Institutional Review Board and all participants provided written informed consent. Participants were healthy, self-described AAs age 18–60 years, with a BMI between 18 and 42 kg/m². Individuals taking medications that increased bleeding risk or altered S_I and/or gene expression in insulin-responsive tissues (eg, steroids or thyroid hormone replacement) were excluded. Methods for recruitment, physical examination, physiological experiments, and performance of adipose and muscle biopsies have been reported (8–11).

Height and weight and waist and hip circumference were measured, body fat was determined by a bioelectrical impedance analyzer (RJL analyzer Quantum Desktop), and fasting blood was drawn for DNA isolation and biochemical analyses (complete metabolic panel, hemoglobin A1c, and lipid profile) at the screening visit. A standard 75-g oral glucose tolerance test (OGTT) was completed during the screening visit to exclude diabetes, and glucose and insulin were measured at baseline and 30-minute intervals for 2 hours. Of 319 screened participants, 31 were excluded based on abnormal laboratory results (13 had T2D) and 28 were lost to followup. Nondiabetic individuals (fasting plasma glucose < 126 mg/dL and 2-h OGTT glucose < 199 mg/dL) were invited to a second visit for adipose and muscle tissue biopsies under fasting conditions and an insulin-modified (0.03 U/kg) FSIGT. Abdominal sc adipose near the umbilicus and vastus lateralis skeletal muscle biopsies were obtained under local anesthesia (lidocaine). Biopsies was performed by opening a Bergstrom needle to cut 100–200-mg pieces of tissue, then applying suction using a 50-mL syringe. Tissues were immediately rinsed in sterile saline, quick-frozen in liquid nitrogen, and stored at −80°C. A total of 260 unrelated individuals completed all study visits and are referred to as the “African American Genetics of Metabolism and Expression” (AAGMEx) cohort. Adipose and muscle biopsies were collected from 256 individuals and high-quality FSIGT data were available from 235 participants.

Laboratory measurements and physiological phenotyping

Details of clinical/biochemical methods are presented in Supplemental Methods. In brief, OGTT data were analyzed by homeostatic model assessment (HOMA; http://mmatsuda.diabetes-smc.jp/MIndex.html) (12, 13). The MINMOD Millennium program was used to analyze FSIGT data to determine S_I and AIR_G (14).

Gene expression analysis and genotyping methods

Details of gene expression analysis, genotyping, and data quality control methods are presented in Supplemental Methods. In brief, genome-wide expression data were generated using HumanHT-12 v4 Expression BeadChip (Illumina) whole-genome gene expression arrays for quantitative analyses of transcript expression in adipose and muscle samples. Infinium HumanOmni5Exome-4 v1.1 DNA Analysis BeadChips (Illumina) were used to genotype DNA samples based on manufacturer instructions.

Statistical and bioinformatic methods

To test for an association between glucose homeostasis or obesity traits and expression level, a linear regression model was computed with the trait (transformed, if necessary; see Supplemental Methods) as the outcome and expression level (log₂) as the predictor. Models included age, sex, and ancestry proportion (admixture) as covariates. Benjamini-Hochberg false-discovery rate (FDR) adjusted P-values were computed by trait (15). Expression of a transcript associated with a trait at an FDR-adjusted P < .01 (1%) was considered significant. In a secondary analysis, comparison of transcript expression between participants in the top and bottom quartiles of the S_I distribution (adjusted for age, sex, and African ancestry proportion) was performed to identify transcripts with significantly different expression between groups. Standard two-class unpaired analysis as implemented in Statistical Analysis for Microarray (SAM) software was used (16). To identify expression quantitative trait loci (eQTLs), linear regression was computed with the log₂-transformed expression values as the outcome and an additive genetic model for the SNP as implemented in the R-package MatrixEQTL (17); age, sex, and African ancestry proportion were covariates. Detailed statistical and bioinformatic data analysis methods are presented in Supplemental Methods. Sample sizes in each analysis varied based on available data after quality control (Supplemental Table 1).

In vitro studies to identify functional significance of a cis-eGene

The effect of experimental manipulation of the expression of a selected (see Results) cis-eGene on cellular pathways was tested in THP1 cells, a widely used human monocytic leukemia cell line model for monocyte-macrophage differentiation (18). Gene-specific lentiviral short hairpin RNA (shRNA) was used for transduction and to silence the expression of the gene by stable RNA interference. Detailed descriptions of cell culture experiments, RNA sequencing (RNA-Seq), qRT-PCR and analytic methods are presented in Supplemental Methods.

Results

Phenotypic characteristics

Anthropometric and physiological characteristics are summarized in Table 1. Participants had a broad range of BMIs (28.9 ± 5.6 kg/m²), adiposity (waist:hip ratio = 0.91 ± 0.08; fat mass = 32.9 ± 9.7%), and glucose homeostasis-related phenotypes, including FSIGT-derived insulin sensitivity (S_I = 4.0 ± 3.5). Hemoglobin A1c levels were at least 5.7% in 113 participants, fasting plasma glucose concentrations were at least 100 mg/dL in 47 participants, and 2-h glucose values during the 75-g OGTT were at least 140 mg/dL in 28 participants. In addition, high-density lipoprotein cholesterol concentrations were less than or equal to 40 mg/dL in 32 participants, and triglycerides were at least 150 mg/dL in 21 individuals. Consistent with other studies, age- and sex-adjusted S_I showed significant negative correlation with BMI (r = −0.48; P = 4.7 × 10⁻¹⁵) and insulin resistance (HOMA-IR, r = −0.47; P = 5.5 × 10⁻¹⁴). S_I had a negative correlation with AIR_G (r = −0.58; P = 5.3 × 10⁻²²) and a positive correlation with OGTT-derived insulin sensitivity index (Matsuda insulin sensitivity index, r = 0.57; P = 1.1 × 10⁻²⁰).

Table 1.

Demographics of the Study Population: AAGMEx Cohort

Trait	N	Mean	Median	SD
Sex, M/F	139/121
Age, y	260	40.6	41.4	11.5
BMI, kg/m2	260	28.9	28.5	5.6
Waist-to-hip ratio	256	0.91	0.91	0.08
Fat mass, %	236	32.9	32.3	9.7
Systolic BP, mm Hg	260	122.1	119.8	15.8
Diastolic BP, mm Hg	260	75.3	74.3	10.6
Cholesterol, total, mg/dL	259	176.5	173.0	36.3
Triglycerides, mg/dL	259	84.0	73.0	43.3
HDL cholesterol, mg/dL	259	56.2	54.0	15.6
VLDL cholesterol, mg/dL	259	16.8	15.0	8.6
LDL cholesterol, mg/dL	259	103.5	99.0	32.3
Hemoglobin A1c, %	260	5.6	5.6	0.3
Fasting insulin, mU/L	260	10.0	8.0	8.4
2-h insulin, mU/La	260	62.2	40.1	77.5
Fasting glucose, mg/dL	260	91.3	90.0	9.4
2-h glucose, mg/dLa	259	100.7	97.0	29.6
Matsuda indexa	252	6.2	4.6	6.6
HOMA-IR	260	2.3	1.8	2.2
SI [×10−4 (mU/L)−1 · min−1]b	235	4.0	3.1	3.5
AIRG, mU.L−1 · minb	235	779.4	581.8	645.8
DIb	235	2287.5	1886.6	1519.5
SG, min−1b	235	0.019	0.017	0.010

Abbreviations: BP, blood pressure; HDL, high-density lipoprotein; LDL, low-density lipoprotein; VLDL, very low-density lipoprotein.

S_G, measure of the ability of glucose to promote its own uptake (glucose effectiveness). Units are taken from MINMOD Millennium program.

^aMetabolic measurements from OGTT.

^bMetabolic traits from FSIGT evaluation of nondiabetic individuals.

S_I-associated transcripts in adipose tissue

S_I was positively and negatively associated, respectively, with expression of 1169 and 1043 adipose transcripts (FDR < 1%) (Supplemental Table 2A). Expression levels of adipose transcripts were significantly correlated with glucose homeostasis and obesity phenotypes. The clustering of Matsuda index and S_I indicated significant overlap in direction of association of genes correlated with these insulin sensitivity traits (Supplemental Figure 2). S_I was strongly correlated with BMI; after adjustment for BMI, only 17 adipose tissue transcripts remained significantly associated (FDR < 5%) with S_I (Supplemental Table 2B).

Ingenuity pathway analysis (IPA) identified significant enrichment of biological pathways among the 2212 adipose transcript probes associated with S_I (FDR < 1%; Supplemental Table 3). Genes involved in metabolic pathways including oxidative phosphorylation (P = 2.0 × 10⁻¹²), EIF2 signaling (P = 1.4 × 10⁻⁶), regulation of eIF4 and p70S6K signaling (P = 1.8 × 10⁻⁴), and valine and isoleucine degradation (P ≤ .0002) were positively associated with S_I. Genes involved in leukocyte extravasation and toll-like receptor signaling pathways (P ≤ .002) were negatively associated with S_I (Figure 1A; Supplemental Figures 3 and 4). Furthermore, gene set enrichment analysis (GSEA), using the ranked list of all transcripts expressed in adipose tissue, also identified enrichment of S_I-associated transcripts (Supplemental Table 4). Genes involved in ribosome function (Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway:3010; P < 1.0 × 10⁻⁶) were positively associated with S_I, and genes in lysosome (KEGG: 4142) and proteasome (KEGG: 3050) pathways were negatively associated with S_I (P < .001).

Figure 1.

Common and tissue-specific characteristics of adipose and muscle tissue transcripts associated with S_I in AAs. A stacked bar graph shows enrichment of adipose (A) and muscle (B) tissue transcripts associated with S_I in IPA biological pathway analysis. A Venn diagram (C) shows overlap of adipose and muscle tissue transcripts positively or negatively associated with S_I. Scatter plot (D) shows the concordance and discordance in direction of association (effect, β from regression analysis) of transcripts associated with S_I in both adipose and muscle tissue. Visualization of running sum statistics from advanced gene set enrichment analysis (GSEA, GeneTrail) shows (E) common and tissue-specific modulation of transcripts with S_I and enrichment in biological pathways.

In a secondary analysis, comparison of adipose transcript expression between individuals in the top and bottom quartiles of the S_I distribution (adjusted for age, sex, and ancestry proportion) identified 247 differentially expressed transcripts (≥1.2-fold; FDR ≤ 5%). Those in the bottom quartile were insulin resistant (IR; S_I = 1.37 ± 0.47; N = 57) and those in the top quartile were insulin sensitive (IS; S_I = 8.17 ± 4.09, N = 58). Differentially expressed probes included 156 up-regulated and 91 down-regulated mRNA transcripts of RefSeq genes with known function (Supplemental Table 5). Genes with higher expression in IR participants included dermatopontin (DPT), NADPH dehydrogenase quinone 1 (NQO1), ubiquitin carboxyl-terminal esterase L1 (UCHL1), leptin (LEP), Fc fragment of IgG binding protein (FCGBP), integrin beta 2 (ITGB2), secreted frizzled-related protein 4 (SFRP4), lymphocyte cytosolic protein 1 (LCP1), and monocyte/macrophage serine esterase 1 and carboxylesterase 1(CES1). Genes with lower expression in IR individuals included α-2-glycoprotein 1 zinc-binding (AZGP1), lipin 1 (LPIN1), 1-acylglycerol-3-phosphate O-acyltransferase 9 (AGPAT9), and NDRG family member 4 (NDRG4). However, differential expression between IR and IS subjects for all of these transcripts were less than 2-fold.

Fewer muscle tissue transcripts were associated with S_I and showed distinct tissue-specific characteristics

Compared with adipose, fewer transcripts in muscle were positively (n = 56) or negatively (n = 89) associated with S_I (Supplemental Table 6A). Only three muscle tissue transcripts were significantly associated (FDR < 5%) with S_I after adjustment for BMI (Supplemental Table 6B).

As in adipose, muscle tissue transcripts positively associated with S_I were enriched for eIF2, eIF4-p70S6K, and mTOR signaling (P = .001–2.0 × 10⁻¹⁴; Supplemental Tables 7 and 8; Figure 1B). S_I was associated with expression levels of 42 transcripts in both adipose and muscle (FDR < 1%). However, 17 of these transcripts show discordant directions of association, indicating tissue-specific transcript modulation (Figure 1, C and D). S_I-associated transcripts in muscle were enriched for oxidative phosphorylation (P = 2.5 × 10⁻⁶). In contrast with the predominantly positive association in adipose (41 genes), the expression of transcripts in this pathway (9 genes) were negatively (inversely) associated in muscle (Supplemental Table 9A; Supplemental Figure 3). GSEA validated discordant and tissue-specific modulation of genes in the oxidative phosphorylation pathways (Figure 1E; Supplemental Table 9B).

In a secondary analysis, comparison of muscle transcript expression between those in the top and bottom quartiles of the S_I distribution identified only 57 differentially expressed transcripts using less stringent significance thresholds (≥ 1.2-fold; FDR ≤ 5%; Supplemental Table 10). Consistent with animal models and human studies in insulin resistance and T2D, genes with higher expression in IR participants included G0/G1switch 2 (G0S2) (19), dysbindin domain containing 1 (DBNDD1) (20), RAP2A member of RAS oncogene family (RAP2A) (21), and aldo-keto reductase family 1 member C3 (AKR1C3) (22). Genes down-regulated in IR individuals included lipoprotein lipase (LPL) (22) and insulin receptor substrate 1 (IRS1). Expression of G0S2 was 1.5-fold higher in IR participants in this cohort. Interestingly, Parikh et al (23) reported that the expression of G0S2 in skeletal muscle was strongly induced by insulin. The higher expression of the G0S2 gene observed in IR individuals may reflect elevated fasting concentrations of insulin compared with IS individuals. The function of this gene in muscle is poorly understood; however, a recent study showed that whole-body knockout of the G0S2 gene enhanced adipose lipolysis, attenuated gain of adiposity, and improved insulin sensitivity (24).

Replication of S_I-transcript associations

Strong concordance was observed between correlations of adipose tissue transcripts and S_I in nondiabetic AAGMEx AAs (N = 230) in this cohort and 99 European Americans from a previously published data set (see Supplemental Methods and Results; Supplemental Figure 5A) (8). The correlation between four selected muscle transcript expression and S_I identified in AAs was also validated by qRT-PCR in European American and AA nondiabetic individuals from Arkansas (N = 127) (11) (see Supplemental Methods and Results; Supplemental Figure 5, C–F).

Coexpressed adipose and muscle transcripts support novel regulatory interactions

Weighted gene coexpression network analysis (25, 26) in the AAGMEx cohort identified 28 coexpressed transcript modules in adipose and 22 modules in muscle (Supplemental Figure 6; Supplemental Tables 11 and 12). Modules are denoted by unique colors. A subset of these “modules” was correlated with S_I or BMI and enriched for genes in biological pathways (Table 2; Supplemental Tables 13 and 14; Supplemental Results).

Table 2.

S_I-associated Coexpression Modules in Adipose and Muscle Tissue Are Enriched for Biological Pathways and Processes

Module	No. of Probe in Module	Correlation (P) of Module Eigengene			Percentage of Gene Associated With SI (P < .003)		Enrichment (n, p, %FDR)	Hub Genes
		BMI	PFAT	SI	% β+	% β−
Adipose
Yellow	538	−0.44 (2e-13)	−0.43 (8e-13)	0.23 (3e-04)	27.5	8.9	hsa00230:Purine metabolism (16, 1.61e-05, 0.018)	RNASE4, PSMD14, CIDEA, CORO1C, ANG, SLBP, DNAJC27, LETMD1, ZCCHC8, TUBB4Q, PPWD1, PKM2, PABPC1, EIF3H, ADH1B, HAT1, TM7SF2, MLLT10, PHF13, CSNK2A2, PGAM4, SIN3A
Blue	704	0.6 (9e-26)	0.53 (1e-19)	−0.31 (3e-07)	29.5	31.1	hsa04142:Lysosome (16, 2.95e-05, 0.035)	CD163, ITGB5, GLB1, HTRA1, AZGP1, GRN, UCHL1, VSIG4, GPR116, NPC2, PFKFB3, EFEMP1, CRTAP, FOLR2, KIAA1598, TTYH3, IRAK1, MYO9B, TIMP1, TXN, M6PRBP1, AADACL1, STOX1, CYYR1
Red	416	0.37 (8e-10)	0.27 (1e-05)	−0.16 (0.01)	3.4	34.9	hsa04670:Leukocyte transendothelial migration (16, 1.34e-06, 0.0015)	ARHGAP30, DOCK2, LCP1, PRKCB1, WAS, RASSF2, EVI2B, FCGR2A, CSF3R, NCKAP1 L, CD3D, CD86, LTB
Brown	686	0.31 (5e-07)	0.32 (2e-07)	−0.13 (0.04)	10.5	22.6	hsa03050:Proteasome (10, 1.98e-04, 0.030)	MYL6, AKAP1, HSPA8, PSME3, SIL1, ATP6AP1, ATMIN, TAX1BP3, DPP3, NCLN, AASS, PECI, SIRT1, MAGED1, ANXA5, IFP38, CROP, AKR1C3, KIAA0556, CCM2, UBE2E2, CAV2, CD33, BOP1, TMEM138, B4GALT7, DSG2, SPARC, ARID4A, TAGLN2, LIX1 L, GMPPA, COPB2, PSMC2, TUBA1A, GBA, TUBA1C, RPRC1, TNIP1, PSMB10
Pink	372	−0.54 (2e-20)	−0.44 (1e-13)	0.32 (2e-07)	46.8	3.8	hsa00190:Oxidative phosphorylation (21, 1.40e-10, 1.60e-07)	PCCB, CS, PRDX3, ATP5A1, UQCRFS1, COX5B, CYC1, MOSC1, PRDX2, GPAM, BRP44, ETS1, DCI, SDHB, UQCRH, AGPAT2, AKAP11, RPS15, PPP1R16A
Muscle
Darkgreen	26	0.2 (0.001)	0.2 (0.001)	−0.26 (4e-05)	0.00	15.38	GO:0046320∼regulation of fatty acid oxidation (4, 4.37e-06, 0.0059)	SLC25A20, SLC25A34, MLYCD, ACSL1, PREB, ACACB, CD36
Cyan	131	−0.36 (3e-09)	−0.29 (3e-06)	0.24 (1e-04)	11.45	7.63	hsa03010:Ribosome (10, 2.76e-07, 3.5e-04)	RBM3, NT5C, RPL18A, SDHC, ENY2, UBXN1, PKIA, EIF3 L, SYF2
Red	324	0.2 (0.002)	0.18 (0.003)	−0.16 (0.01)	0.62	6.79	hsa00190:Oxidative phosphorylation (47, 4.35e-43, 4.90e-40)	ATP5A1, CKMT2, BOLA3, ATP5J, NDUFS8, UQCRC1, DECR1, ETFB, UQCRH, BRP44 L

Abbreviations: PFAT, percent fat mass; %β⁺ and %β⁻, percentages of transcripts positively and negatively associated with S_I.

Algorithm for the Reconstruction of Accurate Cellular Networks (ARACNE) analysis (27) identified significant interactions between genes in each of these S_I-associated modules and identified “hub genes” in the regulatory network (Table 2 and Supplemental Results).

Cis-eSNPs for S_I-associated transcripts

We hypothesized that the expression level of a subset of S_I-associated transcripts would be associated with genotype and manifest as eQTLs. We identified 1971 and 2078 transcripts with at least one significant cis-eQTL (top eQTL single nucleotide polymorphism [SNP] within ± 500 kb around the expressed transcript at FDR < 1%) in adipose (N = 250) and muscle (N = 247), respectively. Intersecting significant cis-eQTL results with data from transcript-S_I association analysis to test our hypothesis, identified cis-regulatory single nucleotide polymorphisms (SNPs) (FDR < 1%) for 363 and 42 S_I-associated transcripts in adipose and muscle, respectively (Supplemental Tables 15 and 16).

Cis-expression regulatory SNPs (cis-eSNPs) identified in adipose and muscle of AAs represented a prioritized set of SNPs providing statistical evidence for genotype-dependent variations in transcript abundance. To identify roles of these putatively functional SNPs in modulating insulin sensitivity, tissue eQTL data from this cohort was examined to identify whether a subset of cis-eSNPs was also associated with S_I. The association of 2 296 925 autosomal SNPs (representing 2 210 735 unique high-quality genotyped SNPs with minor allele frequency (MAF) > 0.01 and Hardy-Weinberg equilibrium P-value > 1 × 10⁻⁶) was assessed with S_I in 230 AAGMEx AAs. Of these, S_I was associated with 2325 SNPs at P < .001 (additive genetic model), including 29 associated with S_I at P ≤ 1 × 10⁻⁵. Intersecting cis-eQTLs with SNP-S_I association results revealed that among S_I-associated SNPs, 10 and 8 were cis-regulatory SNPs (at FDR < 0.01) for seven and five transcripts in adipose and muscle, respectively (Supplemental Table 17).

Finally, intersecting transcript-S_I, SNP-transcript, and SNP-S_I association data identified genetic polymorphisms in three loci modulating expression of S_I-associated transcripts in adipose tissue as cis-eSNPs and also associated with S_I in the AAGMEx cohort (Table 3). Expression of Ninjurin1 gene (NINJ1) transcript in adipose was significantly associated with S_I (P = 7.97 × 10⁻⁶; β = −1.24). A 3′-UTR SNP (rs7033638) in NINJ1 was a significant cis-eSNP in adipose and muscle (P = 1.82 × 10⁻¹¹ and 5.87 × 10⁻¹⁶). The rs7033638-C allele (MAF = 0.36) was associated with lower expression (β = −0.17 and −0.14) of NINJ1 transcript and significantly associated with increased S_I (P = 6.24 × 10⁻⁵; β = 0.26; Figure 2). Cis-eSNPs for two other S_I-associated adipose transcripts, aspartylglucosaminidase (AGA; Supplemental Figure 9) and C-type lectin domain family 10 member A (CLEC10A; Supplemental Figure 10) were also associated with S_I in AAGMEx. Formal analysis by the causal inference test (CIT) (28) identified no genome-wide significant loci. However, S_I-associated cis-eSNPs were nominally significant in CIT (P_CIT < .05) for these three genes (Table 3), indicating putative causal association between the SNPs and S_I mediated by transcript expression of these genes (SNP → mRNA → S_I). Thus, a subset of regulatory SNPs may modulate transcript expression in tissues important for glucose homeostasis and could affect insulin sensitivity.

Table 3.

Genetic Polymorphisms Modulate Expression of S_I-associated Transcripts in Adipose as cis-eSNPs and Associated With S_I

Gene	Association of Adipose Tissue Expression With SI		Association of SNP with Expression and SI
	P-Valuea (Q-value)	βb	Best cis-eSNP	SNP Chr	SNP Position (bp)	Distance to TSS	A1	A2	MAF	βc	P-Valuee	Q-Value/FDR
NINJ1 (ILMN_1815086)	7.97e-06	−1.24	In adipose
	(1.27e-04)		rs7033638	9	95 883 785	12 785	C	T	0.359	−0.17	1.82e-11	3.07e-08
			In muscle
			rs7033638							−0.14	5.87e-16	1.56e-12
			SI-associated SNP							βd	P-valuef	Model	PCIT
			rs7033638							0.261	6.24e-05	add	0.048
AGA (ILMN_1802631)	7.56e-06	−1.73	Best cis-eSNP							βc	P-valuee	FDR
	(1.23e-04)		In adipose
			rs77842019	4	178 625 097	−261 440	A	G	0.145	−0.11	6.77e-06	0.0029
			SI-associated SNP							βd	P-valuef	Model	PCIT
			rs77842019							0.34	1.02e-03	dom	4.28e-03
										0.31	8.36e-04	add
CLEC10A (ILMN_2415303)	3.11e-04	−1.38	Best cis-eSNP							βc	P-valuee	FDR
	(2.26e-03)		In adipose
			rs35318160	17	6 980 105	3495	A	G	0.026	0.28	8.11e-08	6.25e-05
			Other cis-eSNP
			rs75437894	17	6 966 496	17 104	G	A	0.016	0.27	2.92e-05	9.87e-03
			SI-associated SNP							βd	P-valuef	Model	PCIT
			rs75437894							−0.93	2.56e-04	dom	0.025
										−0.93	2.56e-04	add

Abbreviations: A1, minor allele; A2, major allele of cis-eSNP; add, addative genetic model; β^c, effect size of minor allele (A1) on expression; β^d, effect of minor allele on S_I; dom, dominant genetic model. FDR, false discovery rate; MAF, minor allele frequency; P_CIT, omnibus P-value from CIT for a SNP considering S_I as the clinical trait. P-value^a and β^b, significance of association and effect direction between S_I and transcript expression. P-value^e, significance in additive model (MatrixEQTL analysis); P-value^f, significance of association between genetic marker and S_I; TSS, transcription start site.

Figure 2.

Ninjurin1 (NINJ1) transcript is associated with S_I and cis-eSNPs for this transcript also associate with S_I in AAs. The scatter plot shows association of NINJ1 transcript expression (ILMN_1815086) in adipose tissue with S_I and BMI (A, B). The box plot shows association of NINJ1 transcript expression (ILMN_1815086) in adipose with genotype of the cis-eSNP rs7033638 (C) and association of S_I and with genotype of the SNP rs7033638 (D). LocusZoom plots show regional association of NINJ1 cis-eQTL region SNPs (genotyped and imputed) with transcript expression and S_I (E, F). Significance level (-log10 P-values) indicated in LocusZoom plots are based on score test implemented in the program SNPTEST. LD plot shows LD relationship (r²) between genotyped SNPs in the marked region.

NINJ1 modulates expression of genes in cellular pathways involved in insulin sensitivity

Among the cis-eGenes identified in this study, expression of Ninjurin 1 (NINJ1) in adipose tissue was negatively associated with S_I, and the genotype of the strongest cis-eSNP for this gene also showed association with S_I in the AAGMEx cohort. NINJ1 was further associated with BMI. Although mature adipocytes are the predominant cells in adipose tissue, preadipocytes, macrophages, and other cell types are present. To define the contribution of adipose tissue cells to the observed relationship between NINJ1 expression and S_I and BMI, expression in RNA from adipocytes and adipose stromal vascular cells was compared. NINJ1 expression was approximately six times higher in RNA isolated from the stromal vascular fraction compared with the adipocyte fraction of adipose tissue (Figure 3A). NINJ1 is a genetically regulated transcript, and may play an important role in controlling S_I and BMI by modulating expression of genes in macrophages. Thus, we tested the effect of experimental manipulation of NINJ1 expression on cellular pathways in THP1 cells, a human monocytic leukemia cell line widely used as a model for monocyte-macrophage differentiation (18). Exposure to phorbol esters ([phorbol 12-myristate 13-acetate (PMA/TPA)] ) induces differentiation of THP1 monocytes into macrophages (18). Differentiated THP1 macrophages showed a strong induction of NINJ1 expression (Figure 3B) compared with undifferentiated cells. Transduction of THP1 cells by NINJ1-specific lentiviral shRNA stably knocked down its expression at baseline and in the PMA/TPA-induced state (Figure 3C). Comparison of NINJ1 knockdown cells with control shRNA treated THP1 macrophage cells by RNA-Seq analysis identified genes modulated by NINJ1 (Supplemental Table 18). ShRNA mediated knockdown of NINJ1 induction in PMA/TPA-treated THP1 macrophages strongly abrogated induction of several genes including MMP9, COL6A2, FCRLA, and THBS1. Induction of genes including EDNRA and SEMA3D was enhanced in NINJ1 knockdown THP1 macrophages. These results were validated by independent experiments and by qRT-PCR (Figure 3, D–K). Bioinformatic analysis identified a significant enrichment of genes in chemokine signaling pathway, cytokine-cytokine receptor interaction, cell adhesion, and angiogenesis among genes modulated by NINJ1 (Supplemental Table 19). Thus, our in vitro studies identified target genes and pathways through which NINJ1 may modulate obesity and/or insulin sensitivity.

Figure 3.

Knockdown of NINJ1 expression in vitro in THP1 cells modulated expression of target genes. A, Relative expression of NINJ1 in pooled RNA samples from stromal vascular fractions compared with the adipocyte fraction of sc adipose tissue in AAs. B, Phorbol ester exposed (TPA/PMA, 10 ng/mL for 48 h) differentiated THP1 macrophages showed strong induction of NINJ1 expression compared with undifferentiated (control) cells. C, Transduction of THP1 cells by NINJ1 specific lentiviral shRNA stably knocked down its expression at baseline and in the TPA-induced differentiated state compared with control-shRNA lentiviral particle–treated cells. Stable knockdown efficiency of NINJ1 expression determined by qRT-PCR from two independent experiments (each with 2–3 biological replicates for each condition are shown. Expression of two other macrophage genes (CD68 and LCP1) was unchanged. D–K, ShRNA-mediated abrogation of NINJ1 induction in PMA/TPA treated THP1 macrophages significantly abrogated induction of several genes, while enhancing the expression of other genes. Validation of selected RNA-seq identified genes by qRT-PCR are shown. #, P < .01 lv-shRNA-control vs lv-shRNA-NINJ1 in basal or undifferentiated condition; *, P < .01 lv-shRNA-control vs lv-shRNA-NINJ1 in TPA treated or differentiated THP1 cells.

Discussion

T2D-related phenotypes are complex and heterogeneous; the pathophysiology of T2D likely involves altered gene expression in multiple tissues and pathways. The molecular processes and/or genetic variation that may contribute to population ancestry-specific pathophysiology of altered glucose homeostasis in AAs are poorly understood. The present study addresses this gap by combining quantitative intermediate traits related to glucose homeostasis, gene expression in tissues important to insulin action, and genome-wide genotype data in AAs. It successfully identified: 1) transcripts, networks of coexpressed transcripts, and biological pathways in adipose and muscle associated with S_I; and 2) genetic controls on the expression of transcripts involved in modulating insulin sensitivity. Discordant modulation of transcripts in some pathways suggests tissue-specific mechanisms of cellular glucose homeostasis. Thus, this study provides novel insights into the molecular mechanisms of insulin sensitivity in AAs. Adipose and muscle expression datasets from cohorts of similar phenotypes and comparable statistical power, regardless of ethnicity, are not available for direct comparison. Thus, ethnic-specific characteristics of S_I-modulating genes and pathways cannot be conclusively defined from this study.

S_I, a quantitative estimate of insulin sensitivity, reflects the net capacity for insulin to promote glucose disposal in insulin-responsive tissues and to inhibit endogenous glucose production. The insulin-modified FSIGT followed by MINMOD analysis (a dynamic test) as implemented in our study to evaluate S_I is comparable to steady-state measurement of insulin sensitivity by the euglycemic-hyperinsulinemic clamp (29, 30). However, MINMOD analysis merges the effects of insulin in promoting peripheral glucose utilization and suppressing hepatic (or endogenous) glucose production (31, 32). Thus, insulin sensitivity measured by this method is comparatively less reliable in individuals with impaired insulin secretion/and or significant insulin resistance (32), and some of the observations in this study may have been influenced by these inherent limitations of the FSIGT. Given that most insulin-mediated glucose disposal occurs in muscle, deranged expression of transcripts in skeletal muscle likely plays an important role in insulin resistance. However, a greater number of adipose tissue transcripts (relative to muscle) were associated with S_I in this AA cohort (2212 vs 145; FDR < 1%), despite identical statistical power. Limitations of assessing insulin sensitivity by FSIGT may have driven this observation. Nevertheless, this result is similar to a study by Sears et al (22), which used the hyperinsulinemic-euglycemic clamp to measure insulin sensitivity. Comparison of basal adipose and muscle tissue gene expression between IR (N = 53) and IS (N = 19) individuals in that study identified 524 differentially expressed transcripts (at FDR 5%) in adipose and only 81 differentially expressed transcripts in muscle (22). Thus, distinct tissue-specific molecular mechanisms in modulating insulin sensitivity seem to be a plausible explanation for this observation. Stronger associations of S_I were observed with sc adipose transcripts (maximum adjusted r² for a transcript = 0.21) compared with skeletal muscle (r² = 0.15), likely reflecting an important role of the adipose transcriptome in modulating insulin sensitivity (potentially via influence on glucose homeostasis in other tissues, including muscle). Recent studies revealed that adipose tissue has an important role (as a dominant regulator) in controlling whole-body glucose homeostasis in health and disease state (33). In adipocytes, insulin resistance and inflammation lead to production and release of free fatty acids and insulin resistance–provoking proinflammatory cytokines. These factors contribute to the accumulation of toxic lipid metabolites in muscle and liver, which impairs insulin signaling and activates pathways in those tissues, thereby modulating insulin sensitivity (34, 35). Many of the insulin resistance–associated molecular changes in these tissues are likely post-transcriptional and cannot be detected in this study that focused on transcriptional modulation.

Glucose homeostasis phenotypes are correlated with obesity; and we observed a significant overlap among transcripts associated with these traits. The magnitudes of associations between S_I and adipose and muscle transcript levels were reduced after adjusting for BMI. Thus, the associations between expression levels of many transcripts and S_I may reflect effects of BMI and are likely important in obesity-mediated insulin resistance.

It is thought that reduction in mitochondrial oxidative phosphorylation (OXPHOS) gene expression may be causally linked to impaired metabolic flexibility associated with insulin resistance and T2D. However, tissue-specific mechanisms and whether OXPHOS has a causal role in insulin resistance remain under debate (1, 36, 37). In this study, expression of PGC1α (PPARGC1A gene; probe ILMN_1750062), a key OXPHOS gene, was positively correlated with S_I in adipose (β = 2.23; P = 2.12 × 10⁻⁹; q-value = 2.61 × 10⁻⁷), but its expression in muscle showed no significant correlation (β = −0.059; P = .75) with S_I. Despite a relatively low mitochondrial mass compared with overall volume, the adipocyte interprets nutritional and hormonal cues in its microenvironment and determines appropriate mitochondrial responses (38). Expressions of many OXPHOS pathway genes are positively associated with S_I in adipose under fasting conditions (9, 22, 39); this, supports down-regulation of this pathway in insulin resistant AA AAGMEx participants. Several nuclear encoded mitochondrial genes appeared as hub genes in a network of 372 coexpressed transcripts (the “pink” module) in adipose tissue. This network module was enriched for OXPHOS pathway genes (KEGG: hsa00190), and was positively and negatively correlated with S_I and BMI, respectively. A recent study in mice showed that adipose-specific impairment of mitochondrial function may alter whole-body insulin sensitivity and produce insulin resistance and the metabolic syndrome (40). Thus, a compromised mitochondrial OXPHOS pathway in adipose may single-handedly, or collectively with other molecular defects, modulate insulin sensitivity.

Studies analyzing transcript expression in muscle have demonstrated reductions in expression of OXPHOS pathway genes in T2D (41, 42). However, this finding was not present in all studies, potentially reflecting differences in methodology, sample size, degree of insulin resistance, age, obesity, and ethnicity (1). No change in expression of OXPHOS pathway transcripts was seen in muscle of IR Europeans (43), whereas it was increased in insulin resistant Asian Indians (44). Furthermore, recent metabolomics and physiological studies in humans and animal models challenge the concept of muscle mitochondrial deficiency (including reduced fatty acid oxidation) as a common cause of insulin resistance (45, 46). Despite mild reductions in mitochondrial content in IR individuals, skeletal muscles have sufficient mitochondrial reserve to meet the excess demand of fatty acid oxidation (47, 48). Targeted metabolomic analysis reported excessive, not reduced, mitochondrial β-oxidation in initial stages of overnutrition and insulin resistance (45). Overload of the OXPHOS pathway may lead to incomplete oxidation of metabolic substrates, generation of reactive oxygen species, and trigger insulin resistance. Concordant with this, significant up-regulation of OXPHOS pathway genes was observed in muscle of IR AAs in AAGMEx. However, individual transcripts were only mildly modulated. The status of mitochondrial OXPHOS pathway genes identified in this cross-sectional study in resting individuals after an overnight fast suggests a unique tissue-specific mechanism of cellular energy homeostasis.

Most of the enriched biological pathways in S_I-associated adipose transcripts related to cellular metabolic activities were positively associated with S_I. Ingenuity pathway and gene set enrichment analyses also identified significant enrichment for canonical pathways not directly associated with metabolism, which may indirectly modulate S_I by altering the adipose microenvironment and adipocyte signaling. These pathways included leukocyte extravasation signaling, Tec kinase signaling, integrin signaling, Fcγ receptor-mediated phagocytosis in macrophages and monocytes, IL-8 signaling, remodeling of epithelial adherens junctions, and toll-like receptor signaling. Thirty genes in the integrin-signaling pathway were significantly negatively associated with S_I (Activation z score, −4.24), indicating up-regulation of this pathway in insulin resistance. A network of 416 coregulated transcripts (the “red” module) was enriched for leukocyte transendothelial migration pathway (KEGG: hsa04670); while a network of 704 transcripts (the “blue” module) was enriched for lysosomal genes (KEGG: hsa04142) in adipose. More than 30% of transcripts in these two modules were negatively correlated with S_I. Interaction of the hub genes in these modules, namely ARHGAP30 in the “red” module and CD163 in the “blue” module, were identified with other genes in these modules. Some intra-modular interactions are supported by the literature (IPA interaction network database), whereas others support the existence of novel transcriptional regulatory mechanisms that may be relevant in modulating S_I in AAs. The CD163 gene in macrophages is involved in endocytosis and lysosomal degradation of hemoglobin/haptoglobin complexes and may be involved in local and systemic insulin sensitivity by protecting tissues from free hemoglobin-mediated oxidative damage (49, 50). The ARHGAP30 gene was identified as a hub gene in an adipose coexpression module associated with circulating triglycerides in Finns and Mexican Americans (51). ARHGAP30 is near the USF1 gene in chromsome1q21–24, a region implicated in T2D susceptibility (52). Further statistical and functional studies are required to identify “key drivers” in these gene networks to achieve a mechanistic understanding of the processes modulating S_I.

Considering significant cis-eQTL results with transcript-S_I association data identified cis-eSNPs for 363 and 42 S_I-associated transcripts, respectively, in adipose and muscle. Further integrative analysis successfully identified genetic polymorphisms in NINJ1, AGA, and CLEC10A, three novel loci associated with S_I and expression of S_I-associated transcripts in adipose from AAs. Expression of NINJ1 is negatively (inversely) associated with S_I. A 3′UTR NINJ1 SNP (rs7033638) had the strongest association. The “C”-allele (freq = 0.36) was associated with higher S_I and lower adipose NINJ1 transcript expression. This variant was also a cis-eSNP in muscle eQTL data in our AAGMEx cohort. A coding missense variant rs2275848 (in linkage disequilibrium [LD] with rs7033638; r² = 0.81) was a cis-eSNP (P = 1.07 × 10⁻⁷) for NINJ1 in this study and, in the Multi Tissue Human Expression Resource adipose eQTL data (53) was associated with early-onset extreme obesity (54). The role of NINJ1 gene in human adipose is unknown; however, its expression is increased in adipose tissue of db/db mice (vs controls) (55) and adipose expression is induced by TNFα (56). NINJ1 is expressed by myeloid lineage cells in human blood, particularly CD14+ monocytes and CD68+ macrophages. NINJ1 is a cell-adhesion molecule; it mediates cell communication and enhances entry, migration, and activity of monocytes and macrophages in inflammation (57). In addition, NINJ1 regulates interactions between leukocytes and vascular endothelial cells and may play a role in vascular remodeling (57, 58). NINJ1 is a p53 target and may be involved in p53-dependent premature senescence, cell proliferation, and apoptosis (59). NINJ1 knockout (whole body) leads to postnatal lethality in mice (59). Analysis of transcripts in adipose tissue fractions in this study showed higher expression of NINJ1 in the stromal vascular fraction (including macrophages and vascular endothelial cells) compared with the adipocyte fraction. Its expression is increased during monocyte-to-macrophage differentiation. ShRNA-mediated abrogation of NINJ1 induction in phorbol ester–differentiated THP1 macrophages modulated expression of genes in chemokine signaling, cytokine-cytokine receptor interaction, cell adhesion, and angiogenesis pathways. Published genome-wide and gene-specific studies showed the association of these NINJ1-modulated pathways and genes, including matrix metalloproteinase, collagen VI, and thrombospondin with insulin sensitivity and obesity (22, 60–62). Thus, NINJ1 is a strong candidate gene that modulates inflammation/immune responses and adipose tissue angiogenesis; it may further modulate insulin sensitivity in adipose tissue.

In conclusion, this study presents a comprehensive characterization of adipose and muscle tissue transcripts, biological pathways, and gene-expression networks involved in the modulation of S_I in AAs. Despite limited overlap of S_I-associated transcripts between adipose and muscle at the individual transcript level, pathway-based, gene set enrichment analysis suggested modulation of the same pathways in both tissues. Transcript levels of genes involved in eIF2, eIF4-p70S6K, mTOR signaling pathways, and ribosome function were positively correlated with S_I, whereas the transcript level of genes in lysosome and proteasome pathways were negatively (inversely) correlated with S_I in both tissues. However, genes involved in leukocyte extravasation signaling showed adipose-specific regulation, and genes involved in oxidative phosphorylation had discordant regulation between tissues. Although multiple pathways are involved in the complex process of insulin resistance, discordant adipose- and muscle-specific regulation of oxidative phosphorylation seems to be important. Identification of genetic polymorphisms in NINJ1, AGA, and CLEC10A, three novel loci that modulate the expression of S_I-associated transcripts in adipose tissue as cis-eSNPs and are also associated with S_I, demonstrates the success of integrative multiomics approaches in mapping eQTLs of physiologic importance. Identification of SNPs associated with expression of a subset of S_I-associated transcripts suggests a genetic regulatory mechanism of insulin resistance and may contribute to the distinctive pathophysiology of T2D in AAs.