Genome-Wide Association Study Identifies Genetic Loci Associated With Fat Cell Number and Overlap With Genetic Risk Loci for Type 2 Diabetes

The number and size of fat cells determine body fat and are also risk factors for type 2 diabetes as previously reviewed (1). Early studies demonstrated that the number of fat cells is set in adolescence (2,3). However, more recent studies have challenged that view. There is a high rate of fat cell turnover in adult human adipose tissue, and it is increased recruitment of new fat cells, rather than decreased cell death, that explains increased fat cell number in obesity (4). Furthermore, increased body weight over time, as well as weight regain after the initial drop in body weight following bariatric surgery, is indeed associated with elevations in fat cell number (5–7). The plasticity of adipose cellularity has clinical consequences (8). Subjects with obesity have greater ability to increase fat cell size after overfeeding when they are insulin sensitive. Furthermore, size and number of fat cells may have impact on type 2 diabetes as previously discussed (1). Taken together, the studies suggesting that fat cells are generated over the entire life span and that this process has clinical consequences make genetic studies of the regulation of the cell number an important issue.

A number of genes and pathways controlling recruitment of new fat cells from precursor cells cultured in vitro have been described, but their importance for governing fat cell number and amount of body fat in humans in vivo is unclear (7,9). Family history is a strong risk factor for both overweight/obesity and type 2 diabetes, and recent genome-wide association studies (GWAS) have identified numerous common genetic variants associated with these traits (10,11). Interestingly, genetic predisposition to type 2 diabetes, but not obesity, is associated with impaired ability to produce adipocytes in the subcutaneous depot (12). This impairment is associated with enlarged fat cell size. Turnover studies suggest that a state of adipose hypertrophy is attributed to decreased recruitment of fat cells (13). Thus, variations in fat cell number may have impact on the development of type 2 diabetes.

In a recent GWAS, we identified a number of loci that are potentially important for fat cell volume (14). In the current study, we have used the same unique GENetics of Adipocyte Lipolysis (GENiAL) cohort to perform a GWAS of fat cell number. We also assessed whether genetic variants reported to be associated with BMI or type 2 diabetes displayed consistent association with fat cell number. Finally, genetic variants associated with fat cell number and displaying genotype-specific gene expression in adipose tissue were taken forward for functional evaluation.

The GENiAL cohort includes 273 men and 718 women and has previously been described (15) (also described in Supplementary Materials). HOMA of insulin resistance as a measure of insulin resistance was calculated from fasting levels of glucose and insulin as previously described (16). Subcutaneous adipose tissue (SAT) was obtained with a needle aspiration biopsy lateral to the umbilicus as previously described (15). Abdominal fat cell number was assessed in 953 subjects.

The study was approved by the local committee on research ethics at Huddinge hospital (Sweden) and explained in detail to each participant. Informed consent was obtained from all participants.

The method for determining fat cell number has previously been described in detail (17). Briefly, isolated fat cells were prepared from SAT by collagenase digestion and subjected to measures of cell diameter, which was used to calculate the average weight of fat cells. More than 95% of the weight of fat cells comprises lipids. The estimated abdominal SAT (ESAT) weight was calculated with a formula based on waist-to-hip ratio, sex, age, waist circumference, and body fat as previously described and validated with DEXA (17). Thereafter, fat cell number in the ESAT region was calculated as the weight of ESAT divided by the mean fat cell weight.

Genotyping in GENiAL has previously been described (15). After quality control, 896 samples and 9,714,326 single nucleotide polymorphism (SNPs) were available for phenotypic analysis. Fat cell number is normally distributed according to visual inspection of data in the cohort. We conducted a GWAS of abdominal fat cell number in PLINK (https://pngu.mgh.harvard.edu/purcell/plink/) (15) using linear regression, assuming an additive genetic model, and adjusting for population structure, age, and sex. Genome-wide significance was set at P < 5.00 × 10⁻⁸, and suggestive significance was set at P < 1.00 × 10⁻⁵. Results were visualized with FUMA (18) and LocusZoom (19). Conditional analyses, where the lead SNP was included as a covariate, were used to assess the number of signals in loci with multiple GWAS significant SNPs. Sensitivity analysis was conducted, whereby BMI was also included as a covariate.

Genetic variants associated with fat cell number were used in a Mendelian randomization (MR) analysis for examination of the possible causal effect of fat cell number on BMI, type 2 diabetes, and waist-to-hip ratio adjusted for BMI (WHRadjBMI) (see also Supplementary Materials and Supplementary Table 1). R package TwoSampleMR (20) was used to perform a bidirectional MR analysis. For the first analysis, fat cell number was used as an exposure to obtain the causal estimate of its association with outcomes BMI, type 2 diabetes, and WHRadjBMI. For the second analysis BMI, type 2 diabetes, and WHRadjBMI were used as exposures, while fat cell number was used as the outcome. Summary statistics were obtained with the IEU GWAS R package (21). MR analyses was conducted using the inverse variance weighting method to obtain the causal estimate of the fat cell number for the respective outcomes (for sensitivity analyses, see Supplementary Materials).

Data from GWAS Catalog (https://www.ebi.ac.uk/gwas/) were retrieved on 31 March 2020. Evidence of genotype-specific gene expression was retrieved from GTEx database (www.gtexportal.org) (22) in August 2020. The purpose of the analysis was to determine whether there was evidence that the SNPs have effects on gene expression levels in relevant tissues. All SNP-gene pairs listed in GTEx and reported herein reach false discovery rate <5%. There was no threshold for physical distance. Variant Effect Predictor (VEP) was used to assess predicted functional consequences of SNPs (https://www.ensembl.org/info/docs/tools/vep/index.html [accessed 2 August 2021]) (23). The functional annotation of the mammalian genome (FANTOM5) data set (24) was used to assess gene expression in human adipose-derived stem cells (hASCs) undergoing in vitro differentiation to adipocytes. Transcripts whose expression were >10 tags per million at some point during the time course were defined as detected in hASCs.

hASCs were isolated from SAT of a male donor (16 years old, BMI 24 kg/m²) and have previously been described in detail (25). One million proliferating hASCs suspended in 90 µL R buffer were transfected with 200 pmol ON-TARGETplus SMARTpool siRNAs targeting candidate genes or control siRNA no. 1 (Horizon Discovery, Waterbeach, U.K.) using Neon Transfection System (Invitrogen, Carlsbad, CA) according to the manufacturer’s protocol and as previously described (14). Subsequently, cells were plated in antibiotic-free medium. This was annotated as day −4. Medium was replaced 24 h posttransfection, and differentiation to adipocytes was started with medium replacement 96 h posttransfection (day 0) following medium replacement every third day. The RNA and medium were collected at proliferation stage 2 days posttransfection (day −2), at differentiation start (day 0), and at days 3 and 7 after the induction of differentiation. Product identification numbers for siRNA can be found in Table 1.

hASCs were electroporated at proliferation stage (day −4 before initiation of differentiation). Two days posttransfection (day −2), the cells were incubated with media containing 5 μmol/L 5-ethynyl-2′-deoxyuridine (EdU) for 24 h. EdU-positive cells and total cell number (nuclear staining) were assessed with the Click-iT Plus EdU Alexa Fluor 555 kit (C10352; Invitrogen) according to manufacturer protocols. Rate of proliferation (EdU-positive cells) was normalized to the number of nuclei representing number of cells in each well. Quantification of cells was performed with CellInsight CX5 High Content Screening (HCS) Platform (Thermo Fischer Scientific, Waltham, MA) with integrated “Object detection” protocol.

Lipids were stained and quantified at differentiation day 7, e.g., 11 days posttransfection, with BODIPY 493/503 (0.2 µg/mL; Molecular Probes, Thermo Fisher Scientific) as previously described (15).

NucleoSpin RNA II kit (MACHEREY-NAGEL, Düren, Germany) and the iScript cDNA synthesis kit (Bio-Rad Laboratories, Hercules, CA) were used to prepare RNA and synthesize cDNA. RNA concentration and quality were measured with Varioskan LUX Multimode Microplate Reader and µDrop Plate (Thermo Fisher Scientific). Quantitative RT-PCR was performed with 1 ng cDNA and SYBR Green primer pairs (Sigma-Aldrich) in 10 μL reaction on the QuantStudio Real-Time PCR machine (Thermo Fisher Scientific). Relative gene expression was calculated with the 2^−ΔCt method (26). Primer pairs are listed in Table 1.

Clinical data were analyzed with multiple regression in JMP 14. Data from in vitro experiments were analyzed with t test. All P values are two tailed. When specific hypotheses are tested, a nominal P < 0.05 is used as the significance threshold.

Summary statistics from the GWAS analysis and experimental data are available on request from I.A.D.

Characteristics of the GENiAL cohort have previously been reported (15). Obesity was positively associated with abdominal subcutaneous fat cell number in the cohort (R² = 0.40, P < 0.001), whereas insulin resistance as assessed by HOMA of insulin resistance was inversely correlated with fat cell number in multiple regression with adjustment for BMI of study participants (partial R² = 0.057, P < 0.001).

In a GWAS we identified 44 SNPs in nine genetic loci (as defined by FUMA [18]), which passed the threshold for GWAS significance for association with fat cell number (P < 5.00 × 10⁻⁸) (Fig. 1A and B, Supplementary Fig. 1, and Supplementary Table 2). Conditional analyses, using the lead or a proxy where FUMA lacks information on the lead SNP as a covariate, provided no evidence for additional signals in any loci (Supplementary Table 2). A further 337 SNPs demonstrated suggestive associations (P < 1 × 10⁻⁵) with fat cell number (Supplementary Table 3). Very few genetic signals for complex traits contain only one or two SNPs; instead, most contain a large number of SNPs, albeit at different levels of association due to being coinherited to varying degrees with the lead SNP. Therefore, to prioritize loci for functional follow-up, we focused on loci with more than three SNPs associated with fat cell number according to visual inspection in LocusZoom and, for SNPs, displaying suggestive association with fat cell number, minor allele frequency >2%. The analysis resulted in 17 genetic loci associated with fat cell number, of which 2 loci pass threshold for GWAS significance, on chromosomes 2 (lead SNP rs149660479-G, β = 0.183, SE 0.029, P = 8.71 × 10⁻¹⁰) (Fig. 1C and Supplementary Table 4) and 7 (rs147389390-deletion, β = 0.113, SE 0.018, P = 1.49 × 10⁻⁹) (Fig. 1D and E and Supplementary Table 4). The locus on chromosome 2 overlaps SPTBN1 and the locus on chromosome 7 C1GALT1 and LOC101927354. Fat cell number–associated SNPs in SPTBN1 and C1GALT1 are located within introns or 3′-untranslated regions of these genes. No SNP in Supplementary Table 4 has been reported to be associated with clinical traits according to GWAS Catalog.

Sensitivity analyses including BMI as an additional covariate provided similar results (Supplementary Fig. 2A); however, the quantile-quantile (Q-Q) plot suggests that adding BMI as a covariate introduces inflation of test statistics (Supplementary Fig. 2B). We believe this inflation is due to the skewed distribution of BMI in the cohort (Supplementary Fig. 2C). The SNPs with suggestive evidence of association in the original analyses demonstrated similar effect sizes in both analyses (Spearman rank correlation coefficient = 0.93) (Supplementary Fig. 2D).

Bidirectional MR analysis was conducted to assess direction of possible causal relationships between fat cell number and BMI, WHRadjBMI, or type 2 diabetes (Supplementary Materials). A total of nine instrumental variables (Supplementary Table 5) were used for fat cell number. Fat cell number showed no significant causal association with BMI with use of the inverse variance weighting method (odds ratio [OR] 0.98, 95% CI 0.93–1.03, P = 0.54). We obtained similar estimates while accounting for uncorrelated horizontal pleiotropy with MR-Egger (OR 0.92, 95% CI 0.73–1.151, P = 0.52) and accounting for correlated horizontal pleiotropy using MR weighted median (OR 0.97, 95% CI 0.92–1.03, P = 0.47) or weighted mode (OR 0.97, 95% CI 0.89–1.05, P = 0.49). To check whether BMI instead leads to a difference in fat cell number, we instead conducted MR using BMI as the exposure and fat cell number as the outcome. There was no significant association of the effect of BMI on fat cell number (Supplementary Table 6). Similarly, there was no significant casual association between fat cell number, WHRadjBMI, or type 2 diabetes (Supplementary Materials and Supplementary Tables 5 and 6). With MR Steiger we estimated the direction of causality to be fat cell number leading to a difference in BMI (Steiger P = 7.23 × 10⁻⁷⁴), WHRadjBMI (Steiger P = 1.21 × 10⁻⁸), and type 2 diabetes (Steiger P = 1.4 × 10⁻⁸).

We also explored the fat cell number–associated SNPs (P < 1.00 × 10⁻⁵) for evidence of genotype-specific gene expression patterns, and, where this was observed, we selected the protein-encoding genes for follow-up investigation (Table 2). Selected genes were in the same linkage disequilibrium block as the SNP according to LocusZoom. Of note, of the SNPs in Table 2, inclusion of BMI as a covariate had limited impact on the association effect size; maximum change was for rs141481897: β_original = 0.119, β_BMI = 0.095. The protein-encoding genes also had to be expressed in our hASCs. We identified five genes that were taken forward, i.e., SPATS2L, KCTD18, MAST4, FAH, and HEATR3. These genes were knocked down by siRNA in proliferating hASCs. For three genes, knockdown efficiency varied between 90 and 70%, whereas for HEATR3 knockdown efficiency of 50–70% was observed (Fig. 2A). We could not achieve satisfactory knockdown of MAST4, and this gene was therefore not taken forward for further functional studies (data not shown). Knockdown of two genes, SPATS2L and KCTD18, decreased number of proliferating cells. The other investigated genes had no impact on proliferation (Fig. 2B). The impact of SPATS2L and KCTD18 on proliferation was not associated with altered accumulation of lipids as measure of adipogenesis (Supplementary Fig. 3). Interestingly, among fat cell number–associated SNPs, rs565245989 in the KCTD18 locus is the only one that comprises a frameshift mutation and is predicted by VEP to have a deleterious effect on gene function (Supplementary Table 3).

We next assessed the importance of impaired fat cell number for genetic predisposition to type 2 diabetes by overlapping GWAS results for the two traits. SNPs associated with type 2 diabetes or related traits were retrieved from GWAS Catalog (n = 1,569) and the more recent article by Vujkovic et al. (n = 174) (27); 1,382 of these SNPs were assayed in our analysis of fat cell number. Forty-eight of the type 2 diabetes SNPs displayed nominal association with fat cell number, of which 30 SNPs showed consistent association; i.e., the type 2 diabetes risk allele was associated with lower fat cell number (Table 3). Eleven risk alleles for type 2 diabetes were associated with larger fat cell number, whereas for seven SNPs risk allele could not be determined in GWAS Catalog. Of the 30 consistent type 2 diabetes–fat cell number SNPs, 16 demonstrated expression quantitative trait loci (eQTLs) in adipose tissue according to GTEx (Table 3); 13 eQTL genes were expressed in our hASCs and were therefore taken forward for functional evaluation by siRNA-mediated knockdown in proliferating hASCs (SETD2, RPL8, ZNF34, ZNF251, ZNF7, COMMD5, HSD17B12, CYB5D2, ZZEF1, PEPD, MAP1LC3A, STOML1, and THOC5). The knockdown efficiency was satisfactory for most studied genes and varied between 90 and 50% (Fig. 3A). Knockdowns of ZNF34, ZNF7, and ZZEF1 were at ∼50% at day −2 or at differentiation start (day 0) but less efficient at the end of differentiation. We could not achieve satisfactory knockdown for ZNF251, and this gene was therefore not taken forward for further functional studies (data not shown). Knockdown of seven genes (RPL8, ZNF7, HSD17B12, ZZEF1, PEPD, MAP1LC3A, and THOC5) decreased number of proliferating cells (Fig. 3B). Results for RPL8, HSD17B12, and PEPD were consistent with genetic data; e.g., rs2294120-G is associated with higher RPL8 expression, more fat cells, and protection against type 2 diabetes, and RPL8 knockdown inhibited fat cell proliferation. The impact of HSD17B12 and PEPD on proliferation was not associated with altered accumulation of lipids as measure of adipogenesis (Supplementary Fig. 3). The few cells that survived knockdown of RPL8 and continue to differentiate developed very aberrant adipocyte and lipid droplets morphology. This precluded quantification of lipids (data are not shown).

For comparison we also retrieved SNPs associated with BMI (n = 3,254) from the GWAS Catalog. Of these SNPs, 2,987 were assayed in our analysis of fat cell number; 136 of the BMI SNPs displayed nominal association with fat cell number. Of these, 38 SNPs displayed consistent association; i.e., the BMI increasing allele was associated with higher fat cell number (Supplementary Table 7).

For the five genes displaying consistent effects according to genetic analysis and knockdown experiments (SPATS2L, KCT18, RPL8, HSD17B12, and PEPD), we next evaluated potential mechanisms by which these genes influenced fat cell number. We explored the impact of gene knockdown on genes encoding central stimulators of cell cycle progression in hASCs, i.e., cyclins CCND1, CCND3, and CCNG2 (28). We also measured TP53, which has ability to induce cell cycle arrest and apoptosis; to differentiate between the two cell fates, we measured the proapoptotic marker BAX in parallel. Knockdown of RPL8 upregulated expression of CCND1. In contrast, expression of CCND3 was first downregulated (day −2) and then upregulated at day 0 (Fig. 4A). Knockdown of PEPD temporally downregulated all three cell cycle markers (Fig. 4B). Knockdown of RPL8 upregulated expression of TP53 and BAX (Fig. 4C and D). PEPD knockdown reduced BAX at day −2 but otherwise had no impact on cell cycle arrest or apoptosis markers. Knockdown of SPATS2L, KCT18, and HSD17B12 did not show any effect on the expression of studied cell cycle markers (data not shown). Thus, available data do not support that impact on expression of cycle progression and apoptosis markers is a major mechanism of action of identified candidate genes for fat cell number.

The genetic factors that may be in control of fat cell size number and function have predominantly been investigated previously with use of a candidate gene approach (14). Only two articles with GWAS have been published (14,15) dealing with fat cell size and lipolysis. In this present study, we have, with a large-scale GWAS, demonstrated two genome-wide significant loci (SPTBN1, C1GALT1) associated with abdominal subcutaneous fat cell number. These loci have no such association with fat cell size in the presently investigated cohort (14). We also describe in the current study a region on chromosome 2 reaching the suggestive threshold for which eQTLs together with siRNA knockdown experiments support the notion that SPATS2L and KCTD18 control adipose cell number. Furthermore, observed enrichment for SNP alleles with opposing effects on fat cell number and type 2 diabetes risk is consistent with the notion that an inherited impaired ability to expand adipose cell number predisposes to diabetes. With further eQTL analysis together with knockdown experiments, we identified shared genetic loci for type 2 diabetes and fat cell number, highlight potential underlying candidate genes (RPL8, HSD17B12, and PEPD), and demonstrate their involvement in promoting proliferation of adipose precursor cells.

The GWAS significant locus on chromosome 2 harbors the SPTBN1 gene. SPTBN1 encodes β II spectrin, which is important for maintaining the cell structure and acts as scaffold for proteins within cells. SPTBN1 has been implicated in cancer development and cardiovascular disease (29). Furthermore, GWAS have implicated SPTBN1 in osteoporosis development, possibly through facilitating cell cycle progression (30). Interestingly, adipocytes and bone cells have the same stem cell origin (31). The GWAS significant locus on chromosome 7 harbors C1GALT1, which encodes T-synthase and regulates O-glycosylation of proteins. C1GALT1 and glycosylation are essential for normal development and have been implicated in inflammatory disease and oncogenesis (32). However, although SPTBN1 and C1GALT1 are candidate genes in the chromosome 2 and 7 loci, we could not functionally link the fat cell number–associated SNPs directly to the function of these genes, and we therefore did not take these genes forward for functional follow-up.

Among SNPs displaying suggestive association with fat cell number, their role as eQTLs together with siRNA knockdown results give support to the notion that SPATS2L and KCTD18 on chromosome 2 control adipose cell number. Consistent with our findings in hASCs, knockdown of SPATS2L in myoblast has been reported to induce growth arrest (33). Furthermore, one SNP in SPATS2L was associated with fasting plasma glucose in a Chinese GWAS (34). The function of KCTD18 has to our knowledge not been described (35).

A possible causal link between adipose morphology, i.e., the relationship between fat cell size and number, and type 2 diabetes has so far only been examined epidemiologically and for fat cell volume (36,37). By overlapping established risk alleles for type 2 diabetes with our GWAS of fat cell number, we revealed an overrepresentation of type 2 diabetes risk alleles that were associated with a lower number of fat cells. In these analyses, we used nominal association with fat cell number, since we analyzed predetermined hypotheses, i.e., SNP with established role in type 2 diabetes development. Bidirectional MR analyses to investigate potential causal relationships suggested that, while no significant associations were identified, fat cell number contributes to BMI, WHR, and type 2 diabetes, but there is no evidence for the opposite direction of effect.

Our analyses highlighted three genes controlling hASCs proliferation and comprising candidate genes for type 2 diabetes, RPL8, HSD17B12, and PEPD. RPL8 encodes a ribosomal protein; however, a peptide of RPL8 has previously been reported to stimulate T-cell proliferation (38), suggesting that the gene has functions beyond translation. mRNA measurements of cyclins suggested that RPL8 could control proliferation by controlling progression through the cell cycle but possibly also by influencing apoptosis. HSD17B12 encodes an enzyme that controls elongation of long-chain fatty acids. Knockdown of HSD17B12 has been reported to either stimulate or inhibit cell proliferation (39). PEPD encodes pepdidase C, which cleaves di- and tripeptides containing carboxyl-terminal proline and is involved in protein degradation. However, as a ligand, peptidase C can bind directly to the epidermal growth factor receptor and regulate cellular metabolism and cell proliferation through stimulation of PI3K/Akt (40). Our analysis of cell cycle markers suggested that PEPD knockdown inhibited cell cycle progression. Thus, all three herein described candidate GWAS candidate genes for fat cell number and type 2 diabetes have been linked to fat cell number by diverse mechanisms. The analyses presented here complement previous -omics analyses to identify type 2 diabetes GWAS signals with impact in adipose tissue (41,42).

There are some caveats with the present investigation. Adipocyte number is higher in obesity, which is strongly determined by central nervous system (CNS) pathways, although peripheral pathways are also important (4,43,44). One limitation of the current study is that we focused on local subcutaneous adipose regulation of fat number and ignored CNS effects. The GTEx database used for analysis of eQTLs has poor coverage of specific types of neurons, and the functional studies were focused on one adipose region. Therefore, we cannot say whether detected GWAS signals have impact on CNS pathways or whether similar results are true for other adipose depots—such as the visceral. Also, GTEx lacks data on adipose precursor cells. In the type 2 diabetes analyses, we focused functional follow-up studies on SNPs displaying opposing effects on cell number and diabetes risk; this does not exclude that alleles associated with more fat cells and increased diabetes risk could be important. We used nominal association as a threshold is analysis of type 2 diabetes SNPs, which increases the risk of false-positive findings. Another limitation is that siRNA experiments assess the function of genes and not the role of specific genetic variants. Although a large number of subjects were investigated, there is no independent confirmation of the findings.

To conclude, the findings presented herein identify SPATS2L, KCTD18, RPL8, HSD17B12, and PEPD to be of potential importance in controlling fat cell numbers (plasticity), the size of body fat, and diabetes risk.

Funding. R.J.S. is supported by the University of Glasgow Lord Kelvin/Adam Smith Fellowship. I.A.D. is supported by the Strategic Research Program in Diabetes at Karolinska Institutet (Genetic and long-term epigenetic studies of changes in adipose function), Swedish Research Council (2019-00997), Novo Nordisk Foundation (NNF200C0063582), and Swedish Diabetes Association (DIA2019-407).

The funders had no role in study design; in the collection, analysis, or interpretation of data; in the writing of the manuscript; or in the decision to submit the manuscript for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. I.A.D. designed the study, analyzed data, and wrote the manuscript. A.K. designed and performed experiments, analyzed data, and wrote the manuscript together with A.A., R.J.S., P.A., and I.A.D. analyzed data and reviewed and edited the manuscript. I.A.D. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.