Shared genetic loci for body fat storage and adipocyte lipolysis in humans

Total body fat and central fat distribution are heritable traits and well-established predictors of adverse metabolic outcomes. Lipolysis is the process responsible for the hydrolysis of triacylglycerols stored in adipocytes. To increase our understanding of the genetic regulation of body fat distribution and total body fat, we set out to determine if genetic variants associated with body mass index (BMI) or waist-hip-ratio adjusted for BMI (WHRadjBMI) in genome-wide association studies (GWAS) mediate their effect by influencing adipocyte lipolysis. We utilized data from the recent GWAS of spontaneous and isoprenaline-stimulated lipolysis in the unique GENetics of Adipocyte Lipolysis (GENiAL) cohort. GENiAL consists of 939 participants who have undergone abdominal subcutaneous adipose biopsy for the determination of spontaneous and isoprenaline-stimulated lipolysis in adipocytes. We report 11 BMI and 15 WHRadjBMI loci with SNPs displaying nominal association with lipolysis and allele-dependent gene expression in adipose tissue according to in silico analysis. Functional evaluation of candidate genes in these loci by small interfering RNAs (siRNA)-mediated knock-down in adipose-derived stem cells identified ZNF436 and NUP85 as intrinsic regulators of lipolysis consistent with the associations observed in the clinical cohorts. Furthermore, candidate genes in another BMI-locus (STX17) and two more WHRadjBMI loci (NID2, GGA3, GRB2) control lipolysis alone, or in conjunction with lipid storage, and may hereby be involved in genetic control of body fat. The findings expand our understanding of how genetic variants mediate their impact on the complex traits of fat storage and distribution.

Obesity, with abdominal fat deposition, is a risk factor for common metabolic diseases. Both overall obesity, as assessed by body mass index (BMI), and a central fat deposition, as assessed by waist-hip ratio adjusted for BMI (WHRadjBMI), display strong genetic predisposition 1,2 . By genome-wide association studies (GWAS) numerous genetic risk loci for BMI and body fat distribution have been identified 1,2 . In most cases, the underlying susceptibility genes in these loci and regulatory pathways remain to be defined. Although brain regulation of food intake is believed to be of central importance for BMI, local regulation of metabolism in peripheral organs may also be important 3 . In addition, expression of genes in GWAS loci associated with WHRadjBMI is enriched in adipose tissue, and some are differentially expressed between different adipose depots, giving support to the hypothesis that genetic susceptibility to high WHRadjBMI is mediated by effects on local metabolic pathways in adipose tissue 4 .
One mechanism in adipose tissue implicated in the control of fat mass is hydrolysis of triglycerides, i.e., lipolysis. High spontaneous and low catecholamine-stimulated lipolysis are associated with higher body weight in both cross-sectional and prospective cohorts 5,6 . Disturbances in adipocyte lipolysis have also been associated with central fat deposition 4,7 . We have recently reported the first GWAS of abdominal spontaneous and isoprenaline-stimulated adipose lipolysis in the unique GENiAL cohort 8 .
The present study aimed to determine whether genetic variants identified by GWAS to be associated with BMI 9 and/or central fat deposition (e.g. WHRadjBMI) 10 overlap with genetic variants reported to be associated with lipolysis 8 . To link genetic variants to specific genes, overlapping risk alleles for these traits were thereafter examined for allele-dependent gene expression through in silico analysis. Genes with allele-dependent expression patterns in adipose tissue were functionally evaluated in human adipose derived mesenchymal stem cells (hASCs). Data mining. In our analysis of BMI we used the SNPs listed in the file Meta-analysis_Locke_et_ al + UKBiobank_2018_top_941_from_COJO_analysis_UPDATED which was reached by a link in Yengo et al. 2 to https:// porta ls. broad insti tute. org/ colla borat ion/ giant/ index. php/ GIANT_ conso rtium_ data_ files and 2.2 BMI and Height GIANT and UK BioBank Meta-analysis Summary Statistics.
In our analysis of WHRadjBMI we used the SNPs listed in the file whradjbmi.giant-ukbb.meta.1.merged. indexSnps.combined.parsed which was reached by a link from Supplementary Table 1 in Pulit et al. 1 to https:// github. com/ lindg rengr oup/ fatdi stnGW AS.
The FANTOM 5 dataset 13 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 (TPM) at some point during the time course were defined as detected in hASCs.
Candidate genes were selected based on expression quantitative trait locus (eQTL) data in the GTEx 14 (www. gtexp ortal. org) database and were retrieved in December 2019 and August 2020. All SNP-gene pairs listed in GTEx and reported herein reach FDR < 5%. p-values for eQTL genes are nominal and taken from GTEx. There was no threshold for physical distance. Among SNP-gene pairs in Tables 1 and 2, four genes were located > 100 Kb from the SNP; only one gene was located at a distance > 500 Mb from the SNP, EBPL (~ 1 Mb distance).
Mendelian randomisation. Mendelian randomisation (MR) analysis was conducted using two sample MR as implemented by the online mrbase software (mrbase.org/pmid29846171). SNPs that were identified as being shared loci (GWAS-significant loci for BMI or WHRadjBMI and nominally significant (p < 0.05) for basal or stimulated lipolysis) were used as the exposure. For BMI, SNPs were also filtered for effect direction consistency. The largest datasets to date for BMI 2 and WHRadjBMI 10 were used as outcomes for this analysis. For BMI, the dataset used to select SNPs above was consistent with that used here 2 . For WHRadjBMI, the data set used above 1 was not available in MRbase, therefore the largest GWAS of WHRadjBMI available was used instead 10 . Settings chosen were: No check for LD between SNPs, Minimum LD Rsq for proxies = 0.8, MAF threshold for palindromes = 0.3, attempt to align strands for palindromic SNPs (all default settings), analyses using MR Egger, MR Egger (bootstrap), Weighted median, Inverse variance-weighted, weighted mode and weighted mode (NOME). The possibility of horizontal pleiotropy and fulfilment of the NOME assumption was tested using the I2 statistic, calculated as per Bowden et al. 15 .
Cell culture and transfection with small interfering RNA. Isolation, growth, differentiation, and characterization of hASCs isolated from subcutaneous adipose tissue of a male donor (16 years old, BMI 24 kg/ m 2 ) donor have been described in detail previously in Refs. 13,16 . hASCs were transfected 24 h before induction of adipogenesis using ON-TARGETplus SMARTpool small interfering RNAs (siRNAs) targeting candidate genes or control siRNA #1 (Dharmacon/Horizon Discovery Group, Lafayette, CO, US) and Dharmafect 3 transfection agent (Dharmacon/Horizon Discovery Group) as previously described 17 . Differentiation to adipocytes was started by medium replacement 24 h post-transfection and medium was subsequently replaced every third day. www.nature.com/scientificreports/ The RNA and medium were collected at days 2, 6 and 9 after the induction of differentiation and cells were fixed for lipid staining at day 9. For siNUP85 hASCs cells were also transfected 24 h postinduction of differentiation using a NEON electroporator (Invitrogen, Carlsbad, CA, US) according to the manufacturer's protocol. Briefly, for one transfection reaction 1 million hASCs suspended in 90 µl of R buffer was mixed with 200 pmol of siRNA targeting NUP85 or a control siRNA #1 and transfection was performed using 100 µl NEON electroporation tip. Electroporation conditions were 1600 V, 20 ms width, 1 pulse. Subsequently, cells were plated in antibiotic free medium at a density of 120,000 cells/well in 24-well plates or 10,000 cells/well in 96-well plates. Medium was replaced 24 h post-transfection and subsequently every third day. For this type of experiment, the cells were cultured until day 3, 6 and 10 of differentiation, at which time RNA and medium were collected or cells were fixed for lipid staining.
Product identification numbers for siRNA are listed in Supplemental Table 1.

Measurement of lipolysis and lipid accumulation in in vitro differentiated adipocytes.
For glycerol measurements, medium was collected at day 9 of differentiation of hASCs for all tested genes and additionally at day 10 for siNUP85 when transfected at day 1 of differentiation. Glycerol is a valid end-product of lipolysis as it is re-utilized by fat cells to a negligible extent and was measured in cell culture medium as described 18 . Amounts of glycerol were normalized to the number of cells (evaluated as stained nuclei and described below) in each well. Assessment of isoprenaline-stimulated lipolysis in siNUP85-transfected hASCs was performed as described previously 19 . In brief, cells were incubated for 3 h in DMEM/F12 medium supplemented with 20 mg/ml of BSA and with or without isoprenaline (1 µmol/l). Glycerol in media was measured as described above. On collection of medium, cells from the same wells were lysed in RIPA buffer, protein concentration was measured, and glycerol levels were then corrected for total protein content.
Lipids were stained and quantified at differentiation day 9 for all the genes and additionally on day 10 for siNUP85 as previously described 8 . Briefly, following fixation with paraformaldehyde neutral lipids were stained with Bodipy 493/503 (0.2 µg/ml; Molecular Probes, Thermo Fisher Scientific, Waltham, MA, US) and nuclei (DNA) were stained with Hoechst 33342 (2 µg/ml; Molecular Probes) for 20 min at room temperature. Accumulation of intracellular lipids (Bodipy) and cell number (Hoechst) were quantified with CellInsight CX5 High Content Screening (HCS) Platform (Thermo Fischer Scientific) with integrated protocols. Total Bodipy fluorescence (lipid droplets) was normalized to the amount of nuclei in each well.
RNA isolation and analysis of gene expression. NucleoSpin RNA II kit (Macherey-Nagel, Düren, Germany) was used to extract total RNA from adipocytes. RNA concentration and quality were measured using Varioskan LUX multimode microplate reader and µDrop plate (ThermoFisher Scientific). Subsequently, Microarray assay and analysis. Clariom S arrays (Thermo Fisher Scientific) were used for global transcriptome profiling of primary adipocyte cultures treated with siRNA against ZNF436 (n = 6), NUP85 (n = 6), or respective non-targeting siRNA control (n = 6) at day 6 of differentiation. The arrays were pre-processed in Transcriptome Analysis Console (Thermo Fisher Scientific) using SST-RMA, which includes normalization, summarization of probes to probesets, and background correction. We excluded 6,341 (ZNF436 samples) and 6,316 (NUP85 samples) probe sets with detection p > 0.05 in at least one sample i.e., not detected in at least one sample, leaving 15,107 and 15,132 probe sets, respectively, for downstream analysis. SAMR using block permutations was used to detect differentially expressed genes 21 . Webgestalt (www. webge stalt. org) was used to identify differentially expressed gene ontologies between siRNA treated and control samples. The KEGG (https:// www. genome. jp/ kegg/) database was used to identify genes in the gluconeogenesis 22,23 , and wikipathways (https:// www. wikip athwa ys. org) in the fatty acid biosynthesis pathway.   . When specific hypotheses are tested, a nominal p < 0.05 is used as the significance threshold. In explorative transcriptome analysis multiple testing is taken into account and FDR 10% used as threshold. Correlation between lipolysis and thermogenic gene expression was evaluated by linear regression in JMP 14.

Results
Shared genetic loci for body fat storage and adipocyte lipolysis. Data mining to identify shared genetic loci for BMI and adipocyte lipolysis is summarized in Fig. 1. A lookup of SNPs previously associated with BMI 2 was conducted in the results of the GWAS of lipolysis 8 . Among 941 SNPs reported to be associated with BMI 2 , 20 displayed nominal association (p < 0.05) with spontaneous lipolysis with consistent effect directions for BMI and spontaneous lipolysis (Supplementary Table 2). Five of these SNPs demonstrated expression quantitative trait loci (eQTLs) in SAT according to GTEx (Table 1). Furthermore, of 941 SNPs associated with BMI, 21 displayed nominal association with isoprenaline-stimulated lipolysis with expected discordant effect direction for BMI and stimulated lipolysis (Supplementary Table 3). Eight of the latter SNPs demonstrated eQTLs in SAT according to GTEx (Table 1). We next assessed genetic influence on central fat distribution. Among index-SNPs in 346 genetic loci associated with WHRadjBMI 1 , 12 displayed nominal associations (p < 0.05) with spontaneous lipolysis; eight of these SNPs demonstrated eQTLs in SAT according to GTEx (Table 2, Supplementary 4). 20 SNPs displayed associations with WHRadjBMI and isoprenaline-stimulated lipolysis (Supplementary Table 5) of which ten SNPs were eQTLs in SAT (Table 2). In the case of WHRadjBMI, effect directions were not considered because there is limited information regarding the relationship between lipolysis in different adipose depots and adipose distribution. None of the SNPs mentioned herein are among SNPs previously reported in the GENiAL cohort to display GWAS suggestive association with lipolysis 8 .
Functional evaluation of candidate genes in shared BMI-lipolysis loci. Three eQTL genes (ZNF436, WARS2, TBX15) in the shared BMI-spontaneous lipolysis loci and five eQTL genes (ZNF436, FLJ20021, INVS, STX17, ALDH2) in the shared BMI-stimulated lipolysis loci were expressed in hASCs (Table 1, Supplementary  Fig. 1a). Of these, five genes (ZNF436, WARS2, TBX15, INVS and STX17) were taken forward for functional siRNA screen in hASCs. FLJ20021 was excluded as, until recently, it has been annotated as an uncharacterized lncRNA 24 . ALDH2 was excluded because it has recently been established as a positive regulator of adipocyte differentiation 25 . We knocked-down the five candidate genes prior to induction of differentiation, followed by evaluation of glycerol levels in conditional medium as a marker of lipolysis. Knock-down of ZNF436, TBX15 and STX1, but not WARS2 and INVS, decreased glycerol levels relative to negative control siRNA (Fig. 2a). To assess whether the impact on lipolysis was direct or secondary to disturbances in adipogenesis, we stained the cells for  www.nature.com/scientificreports/ neutral lipid content. Knock-down of TBX15, but not the other investigated genes, decreased amount of lipids (Fig. 2b). Results for the BMI-panel genes ZNF436 and STX17 were of particular interest because knock-down of these two genes affected glycerol release without affecting lipid content supporting the hypothesis that the impact on lipolysis was direct, and not indirect via influence on adipogenesis.
In parallel, we evaluated the effect of knock-down efficiency of the studied genes at the mRNA level. In general, knock-down efficiency was good, i.e., varied between 90 and 70% with exception of TBX15 where knockdown efficiency reached only 50% at day 6 ( Fig. 3a, b and Supplementary Fig. 2a-c). Since for TBX15 no intrinsic effect on lipolysis could be ensured the limited knockdown was considered to be of less importance. For the genes affecting glycerol we evaluated whether gene knock-down effects could be linked to altered expression levels of key genes such as LIPE and PNPLA2 encoding lipases, PLIN1 encoding the lipid coating protein Perilipin 1, or expression of APIDOQ encoding the adipocyte marker adiponectin. Knock-down of ZNF436 caused temporary downregulation of PLIN1, PNPLA2 and ADIPOQ (Fig. 3a). Knock-down of STX17 upregulated LIPE and ADI-POQ at both day 6 and 9 of differentiation, while upregulation of PLIN1 and PNPLA2 was temporal (Fig. 3b). Knock-down of TBX15 had minor effects on studied gene expression ( Supplementary Fig. 2b).
Functional evaluation of candidate genes in shared WHRadjBMI-lipolysis loci. We next selected eQTL genes in the shared WHRadjBMI-lipolysis loci for functional evaluation in vitro. Four genes in the shared WHRadjBMI-spontaneous lipolysis loci were expressed in hASCs: TBX15, WARS2 (both included in analysis of BMI-loci above), ZNF263 and PCK1 (Table 2). Seven genes in the shared WHRajdBMI-stimulated lipolysis loci were expressed in hASCs: EBPL, NID2, GGA3, MRPS7, NUP85, GRB2, and PEMT ( Table 2, Supplementary Fig. 1b). Knock-down of PCK1, NID2, GGA3, NUP85 and GRB2 decreased levels of glycerol by 20-70% (Fig. 2c). Knock-down of EBPL, MRPS7 and PEMT had no effect (Fig. 2c). Among the genes controlling glycerol release, accumulation of lipids was decreased by knock-down of PCK1, NID2 and GRB2. (Fig. 2e). Knock down of ZNF263 showed no effect on glycerol or lipid levels most likely due to due to poor knockdown in the tested conditions (data not shown).
The inhibition of spontaneous lipolysis by PCK1 knock-down was not consistent with the genetic data, where the T allele is associated with high gene expression and low spontaneous lipolysis, and this gene was therefore irrelevant for further studies. NUP85 and GGA3 were the only two genes from the WHRadjBMI panel that affected glycerol release without affecting lipid content. However, knocking down GGA3 at different time points during differentiation decreased lipid accumulation (data not shown). Based on this and its effect on key adipocyte genes (Fig. 3c, also see below) we could not rule out an effect of adipogenesis. Thus, NUP85 was the only remaining gene from the WHRadjBMI panel that affected glycerol accumulation without affecting lipid content supporting that the impact on lipolysis was direct, and not indirect via influence on adipogenesis. Since the NUP85 SNP rs9909443 was associated with isoprenaline-stimulated lipolysis (Table 2), we also evaluated effects of NUP85 knock-down on isoprenaline-stimulated lipolysis, which was decreased compared to the control siRNA (Fig. 2d). In this experiment in basal condition (no isoprenaline) we did not detect lower glycerol release in siNUP85 samples compared to control siRNA. However, this is only a 3 h "window" of lipolysis and the accumulative effect seems to be important to obtain effects on glycerol release by NUP85.
As for BMI genes, we evaluated knock-down efficiency of WHRadjBMI genes and assessed whether expression of known lipolysis genes could be affected. Knock-down efficiency varied from 75 to 95% ( Supplementary  Fig. 2e-i) except for PCK1 and GGA3 ( Supplementary Fig. 2d and Fig. 3c). These genes were very low expressed in the beginning of differentiation and therefore it is possible that the siRNA-transfection conditions need to be further optimized. Furthermore, knock-down of WHRadjBMI-associated genes had less effect on adipocyte gene expression than BMI-associated genes. Knock-down of PCK1 upregulated LIPE and temporally downregulated PNPLA2 and ADIPOQ (Supplementary Fig. 2d). Knock-down of GGA3, NID2 and GRB2 temporary downregulated all studied genes ( Fig. 3c and Supplementary Fig. 2f, h) while knock-down of NUP85 had no effect (Fig. 3d).

ZNF436 and NUP85 target genes. The analysis of expression of central genes controlling lipolysis
(LIPE, PLIN1, and PNPLA2) did not reveal a functional link to lipolysis for the shared GWAS-candidate genes for BMI/WHRadjBMI and lipolysis. Therefore, two genes were taken forward for global transcriptome analysis: ZNF436 from the BMI and NUP85 from the WHRajdBMI analysis. They were selected because the genetic, Figure 2. Effects of RNA interference-mediated knock-down on lipolysis and lipids in human adipose-derived stem cells. hASCs were transfected with control siRNA oligonucleotide (siNegC) or indicated siRNAs targeting specific genes. (a,b) Expression of ZNF436, WARS2, TBX15, INVS and STX17 was knocked down using siRNA in hASCs 24 h prior induction of differentiation and followed until differentiation day 9, upon which accumulated glycerol in medium was measured to assess spontaneous lipolysis (a) and accumulation of neutral lipids was evaluated by lipid staining (b). (c,e) Expression of PCK1, EBPL, NID2, GGA3, MRPS7, NUP85, GRB2 and PEMT was knocked down using siRNA in hASCs 24 h prior induction of differentiation and followed until differentiation day 9, upon which glycerol in medium was measured to assess spontaneous lipolysis (c) and accumulation of neutral lipids was evaluated in the cells (e). Results are based on four biological/independent experiments. Results were analyzed using t-test and presented in fold change ± SD relative to negative control Neg C. ***p < 0.005, **p < 0.01, *p < 0.05. (d) Expression of NUP85 was knocked down using siRNA in hASCs 24 h before induction of adipogenenic differentiation of hASCs. At day 9 of differentiation cells were incubated with isoprenaline (ISO) and the effects of knock-down on lipolysis were evaluated. Results are based on four biological/independent experiments. Results were analyzed using t-test and are presented as fold change ± SD relative to 0 M isoprenaline (ISO) of respective condition (siNegC or siNUP85). ***p < 0.005, **p < 0.01. www.nature.com/scientificreports/ metabolic, and functional results for lipolysis were consistent, e.g., rs967605-C was associated with high ZNF436 expression, BMI and spontaneous lipolysis, and ZNF436 knock-down inhibited glycerol release. Furthermore, for both genes it was reasonable to assume that the impact on lipolysis was indirect via influence in gene expression; ZNF436 is a transcription factor 26 and NUP85 encodes a protein forming part of nuclear pores, which in turn are involved in transcriptional regulation [27][28][29] . Transcriptome analysis revealed 396 genes that were differentially expressed between ZNF436 knock-down and control cells with FDR < 10%, of which 203 were downregulated (Supplementary Table 6). Downregulated genes were enriched for the gene ontology (GO) "regulation of lipid metabolic process" Table 3). Expression of three genes (PNPLA3, PDE3B and MGLL) reported to be involved in lipolysis 30 were among ZNF436 regulated genes (Supplementary Table 6, Supplementary Fig. 3). There was no evidence of a compensatory impact on gluconeogenesis or fatty acid biosynthesis as only gene in each of these pathways were affected by ZNF436 knockdown, i.e. ENO1 (fold change 0.66) and ACSL4 (0.59).
336 genes were differentially expressed between NUP85 knockdown and control cells with FDR < 10%, of which only five were upregulated (Supplementary Table 7). Downregulated GO included "extracellular structure organization" and "regulation of lipid metabolic process" (Supplementary Table 8 Supplementary Fig. 3).

Correlation between expression of GWAS genes controlling lipolysis and thermogenesis genes.
Lipolysis is believed to be important, though not necessary, for brown and beige adipose thermogenesis 31 . To assess the relationship between novel lipolysis-regulated genes described herein and thermogenesis, we correlated the expression of the genes ZNF436, STX16, NUP85, NID2, GGA3, GRB2 with thermogenesis markers 32 in a separate cohort of obese women 33 . A variable number of thermogenesis genes were  . Effects of RNA interference-mediated knock-down of candidate genes in human human adiposederived stem cells. Human adipose-derived stem cells were transfected with control siRNA (siNegC) or indicated siRNAs targeting specific genes (ZNF436, STX17, GGA3 and NUP85) and followed until differentiation day 6 and 9, upon which the expression of target genes, LIPE, PLIN1, PNPLA2 and ADIPOQ was monitored. Relative gene expression was normalized to the reference gene 18s. Results were analyzed using t-test and presented as fold change ± SD relative to control siRNA (NegC) of corresponding time point (n = 6, for GGA3 n = 11). ***p < 0.005, **p < 0.01, *p < 0.05. www.nature.com/scientificreports/ correlated to the lipolysis regulating genes (Supplementary Table 9). However, for the majority of these genes there was an inverse correlation between lipolysis and thermogenesis gene expression suggesting that the novel lipolysis genes described herein do not have a consistent effect on thermogenesis.

Mendelian randomisation analysis. To assess relationship between BMI and WHRadjBMI and lipolysis
we used Mendelian Randomisation analysis. We provide evidence of causal effects of spontaneous and stimulated lipolysis on BMI but not WHRadjBMI (Supplementary Table 10, Supplementary Figs. 4-5). These results should be interpreted with caution, as the selection of SNPs for mendelian randomisation analyses was nonstandard, i.e., the SNPs were not GWAS significant. Whilst the results for spontaneous lipolysis are convincing, those for stimulated lipolysis are curious, as the MR Egger is null, despite the lack of horizontal pleiotropy and the NOME assumption being fulfilled (IGX2 = 1). In addition, it should be noted that the use of a smaller dataset for the WHRadjBMI MR analysis (GIANT consortium) 10 compared to the SNP selection analysis (GIANT consortium plus UK Biobank) 1 likely underestimates any causal effect of lipolysis on WHRadjBMI. We have also analysed directionally consistency for all BMI and WHRadjBMI GWAS SNPs for association with lipolysis in GENiAL. Results were significant for BMI but not WHRadjBMI (Supplementary Table 11).

Discussion
The findings reported here enhance our understanding of how genetic variants mediate their impact on the complex traits of fat storage and distribution. In the shared lipolysis-BMI or lipolysis-WHRadjBMI associated SNP panels we identified five and seven SNPs, respectively, displaying allele dependent gene expression for genes expressed in fat cells. Among these genes, we provided functional evidence supporting the hypothesis Table 3. Gene sets overrepresented among ZNF436 or NUP85 regulated genes.   www.nature.com/scientificreports/ that transcriptional regulation of ZNF436 and NUP85 have intrinsic effects on lipolysis consistent with the associations in the clinical cohort and propose that these genes hereby are involved in the genetic regulation of BMI and WHRadjBMI (Fig. 4). Furthermore, candidate genes in one more BMI-locus (STX17) and two more WHRadjBMI loci (NID2, GGA3 and GRB2) control lipolysis alone, or in conjunction with lipid storage, and may hereby be involved in body fat storage.
In the BMI-associated locus on chromosome 1 we showed that rs967605-C is associated with higher BMI, adipocyte spontaneous lipolysis, and ZNF436 expression. Consistent with these associations found in clinical cohorts knock-down of ZNF436 inhibited adipocyte lipolysis, whereas no impact on lipid accumulation was observed. ZNF436 is a member of the zinc finger transcription factor family and may act as a negative regulator on gene transcription mediated by MAPK signaling 26 . ZNF436 mRNA is enriched in adipocytes, and in heart and brain as compared to cells from other organs 13,26 . ZNF436 knock-down followed by transcriptome analysis revealed that ZNF436 acts as a regulator of many genes involved in lipid metabolic processes. However, ZNF436 did not control lipolysis via transcriptional regulation of hitherto described genes in the canonical human lipolysis process 30,34 . Importantly, the expression of a set of other genes reported to influence lipolysis or related processes, outside the canonical lipolysis pathway, was regulated by ZNF436 knock-down (Supplementary Table 6, Table 4). These are therefore described in more detail below: • QRFP encoding the neuropeptide 26RFa, which controls phosphorylation of HSL and PLIN1 and hereby inhibits lipolysis 35 . • PDE4A encoding one subtype of phosphodiesterase 4, which inhibits lipolysis 36 .
• ARHGAP26 which is involved in endocytosis and inhibits lipolysis 37 .
• TTC39B encoding a scaffolding protein that promotes ubiquitination and degradation of liver X receptor, a transcription factor that in turn stimulates lipolysis. In addition, TTC39B antisense treatment protects against obesity in mice 39 .
On chromosome 9 rs10118701-G was associated with higher BMI, STX17 expression, and lower stimulated lipolysis. Knock-down of STX17 inhibited spontaneous lipolysis. We did not follow up these findings to determine whether the gene had an opposing inhibitory effect on stimulated as compared to spontaneous lipolysis consistent with the opposing genetic association in the clinical cohort. One example of a gene with such opposing effects is PLIN1, which suppresses spontaneous and facilitates stimulated lipolysis 40 . STX17 encodes a key component of the autophagosome, which in turn has been functionally linked to lipolysis 41 .
In the WHRadjBMI-associated locus on chromosome 17 we showed that rs9909443-T is associated with higher stimulated lipolysis and NUP85 gene expression, as well as lower WHRadjBMI. NUP85 knock-down resulted in consistent inhibition of spontaneous and stimulated lipolysis without affecting lipids giving support to the hypothesis that NUP85, rather than other genes in this locus, is causally linked to lipolysis. Central fat accumulation has been associated with reduced stimulated lipolysis 42 , but the relationship between lipolysis and fat distribution is less straight forward for fat distribution than for BMI since there are depot specific differences in gene expression and fat cell metabolism between e.g. abdominal and femoral fat depots 43 . NUP85 encompasses an essential component of the nuclear pore complex important for nuclear transport but has also been reported to be involved in DNA replication and gene activation 44 . NUP85 has also been reported to be involved in chemotaxis of monocytes via membrane clustering of CCR2 45 . However, CCR2 is expressed at extremely low levels in human adipocytes 46 and this function of NUP85 can therefore not be justified in in adipocytes. The inhibitory effect of NUP85 knock-down on lipolysis was accompanied by downregulated expression of numerous genes. Potentially, downregulated expression of ADRB1 could contribute to inhibition of stimulated lipolysis. Other NUP85 regulated genes may mediate the effects of NUP85 deficiency on lipolysis. NUP85 regulated genes previously implicated in pathways of potential importance for lipolysis are listed below: • BMPR2 deficiency in adipocytes inhibits phosphorylation of the lipid-droplet-coating protein perilipin 47 and hereby impairs stimulated lipolysis. • EIF4EBP1. Simultaneous lack of EIF4EBP1 and EIF4EBP2 increases sensitivity to diet-induced obesity and insulin resistance in mice 48 and is involved in the regulation of the lipid-droplet-coating protein ATGL 49 . • DDIT4 is a fasting-regulated p53 target gene. Forced expression of DDIT4 augment lipolysis in adipocytes 50 .
In summary, the herein reported regulation of lipolysis by NUP85 describes a new role for this gene in adipose biology.
Further analysis of WHRadjBMI loci revealed that at the chromosomes 14 and 17 knock-down of NID2, GRB2, and GGA3, respectively, inhibited both glycerol release and lipid accumulation. The simultaneous effects of these genes on lipid accumulation and on glycerol release prevented us from drawing any conclusion as regards which metabolic processes link SNPs in these loci to body fat distribution. In the joint BMI-WHRadjBMI locus on chromosome 1 the inhibition of spontaneous lipolysis by TBX15 knock-down was not consistent with the genetic data where the T allele was associated with high TBX15 gene expression and low spontaneous lipolysis. It is therefore likely that other functions in adipose tissue linked to TBX15 may mediate the impact of this locus of body fat storage 51 .
One limitation of the study is that no adjustment for multiple testing was used when selecting SNPs associated with lipolysis for functional follow up. The approach was justified by strong prior biological knowledge of the relationships between obesity and the cellular phenotype in question. We have tried to balance the risk www.nature.com/scientificreports/ of false-positives and negatives results by using a p-value threshold and combined this with a strong and clear biological rationale, that is the importance of lipolysis in the function of adipose tissue. Importantly, a causal relationship between lipolysis and BMI was supported by both Mendelian randomization analysis and overall enrichment of SNPs displaying directionally consistent effects on BMI and lipolysis. The lack of such evidence for WHRadjBMI may be related to limited power of the study and a more complex association between lipolysis in different fat depots and adipose distribution. Another limitation is that we only looked for eQTL genes retrieved from the GTEx portal which only reports genetic effects on gene expression in tissue samples and not those limited to adipocytes. Thus, we may not have identified all BMI/WHRadjBMI-SNPs controlling gene function in adipocytes. Furthermore, among the genetic loci with genes controlling lipolysis in siRNA experiments, only rs146182298 (NID2) and rs1328757 (PCK1) were potential lead eQTL SNPs, i.e., they were among the SNPs most strongly associated with expression of the specific gene according to visual inspection in GTEx. A third limitation is that we only assessed lipolysis in abdominal SAT although other adipose depots might be more relevant 52 , but are even less accessible for sampling in adequate numbers. In addition, glycerol release was used to assess lipolysis. Glycerol represents an overall activation of the regulatory enzymes in the lipolysis cascade. Additional measurement of enzymatic activities and protein levels would be necessary to determine which steps in the lipolysis pathway is controlled by studied genes. Finally, gene knockdown may, due to strong effect on cell function, result in adaptive effects on metabolism not induced by specific functional genetic variants. To validate the effect of specific genes on lipolysis, knockdown experiments could be complemented with transient transection, or with e.g. single base substitution experiments. Such comprehensive experiments were, however, beyond the scope of this screening of genes in GWAS loci.
In summary, our study highlights that genetic susceptibility to central obesity might be associated with risk of impaired adipose lipolysis. We provided functional evidence supporting the notion that transcriptional regulation of ZNF436, and NUP85 have intrinsic effects on lipolysis consistent with the associations in the clinical cohort and propose that these genes hereby are involved in the genetic regulation of BMI and WHRadjBMI (Fig. 4).

Data availability
Summary statistics from the GWAS analysis of spontaneous and isoprenaline-stimulated lipolysis, experimental data and gene array data are available upon request (ingrid.dahlman@ki.se).