bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

“Multiple subtypes and mechanisms of their development.”

So Yun Min1,2, Anand Desai1, Zinger Yang2, Agastya Sharma1, Ryan M.J. Genga1,2,4, Alper Kucukural3, Lawrence Lifshitz1, René Maehr1,4, Manuel Garber1,3, Silvia Corvera1,*.

1Program in Molecular Medicine, University of Massachusetts Medical School, Worcester, MA 01655, USA

2Graduate School of Biomedical Sciences, University of Massachusetts Medical School, Worcester, MA 01655, USA

3Program in Bioinformatics, University of Massachusetts Medical School, Worcester, MA 01655, USA 4Department of Medicine, Center of Excellence, University of Massachusetts Medical School, Worcester, MA 01655, USA

*Corresponding author: [email protected]

KEY WORDS: ADSC, mesenchymal stem cells, pre-adipocyte, adipocyte differentiation, adipocyte development, leptin, adiponectin, , brown adipocyte, beige adipocyte, thermogenic , , diabetes

SUMMARY

Human depots perform numerous diverse physiological functions, and are differentially linked to metabolic disease risk, yet only two major human adipocyte subtypes have been described, white and “brown/brite/beige.” The diversity and lineages of adipocyte classes have been studied in mice using genetic methods that cannot be applied in . Here we circumvent this problem by studying the fate of single mesenchymal progenitor cells obtained from human adipose tissue. We report that a minimum of four human adipocyte subtypes can be distinguished by transcriptomic analysis, specialized for functionally distinct processes such as adipokine secretion and thermogenesis. Evidence for the presence of these subtypes in adult humans is evidenced by differential expression of key adipokines leptin and adiponectin in bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

isolated mature adipocytes. The human adipocytes most similar to the mouse “brite/beige” adipocytes are enriched in mechanisms that promote iron accumulation and protect from oxidative stress, and are derived from progenitors that express high levels of cytokines such as IL1B, IL8, IL11 and the IL6 family cytokine LIF, and low levels of the transcriptional ID1 and ID3. Our finding of this adipocyte repertoire and its developmental mechanisms provides a high-resolution framework to analyze human adipose tissue architecture and its role in systemic and metabolic disease.

INTRODUCTION

Adipose tissue plays critical roles in systemic metabolism, through multiple functions that include lipid storage and release, as well as secretion of cytokines such as leptin and adiponectin which control satiety and energy utilization [1]. In addition to its metabolic roles, adipose tissue plays a pivotal role in thermoregulation, both by providing an insulating layer under the skin and by the production of heat [2]. Other functions of adipose tissue include mechanical and immune- protective roles in multiple sites including the mesentery [3].

Currently, three functionally distinct adipocyte subtypes in mammals are recognized, which are best defined in the mouse [4]. White adipocytes are lipogenic and responsible for lipid storage and release; brown adipocytes form a defined depot restricted to the interscapular region and contain high basal levels of the mitochondrial uncoupler UCP1, required for their thermogenic function; and “brite/beige” adipocytes are interspersed within white adipose tissue and express UCP1 in response to stimulation. Lineage tracing and expression studies point to distinct developmental origins for these adipocyte subtypes [5, 6]. In adult humans, no specific depot is solely composed of UCP1-containing adipocytes, but UCP-1 positive cells can be found interspersed within supraclavicular, paravertebral, and perivascular adipose tissue [7], resembling the “brite/beige” phenotype. Thus, in humans, only two adipocyte types, UCP-1 positive (thermogenic) and UCP-1 negative (white) have been recognized to exist.

How two types can accomplish the numerous functions of human adipose tissue is unclear. Increased functionality of adipose tissue may be conferred by the different connections to vascular and neural networks present in distinct depots [4], which could play a role in diversifying adipocyte functions. For example, significant differences in lipolytic capacity and cytokine bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

secretion profiles are seen between depots that are composed of ostensibly similar adipocyte types [8-10]. Alternatively, additional yet uncharacterized developmentally and epigenetically distinct adipocyte subtypes that are not readily distinguished by their morphology may exist.

Recent advances in the ability to study single-cell transcriptomes can be leveraged to better understand the cellular composition of adipose tissue [11-13]. Single-cell transcriptomics of non- mature adipocytes in mouse fat has revealed specific populations of adipocyte stem cells primed to expand in response to adrenergic stimulation [13], stromal cells capable of inhibiting adipogenesis [11], and cells with enhanced adipogenic or fibro-inflammatory properties [14]. The use of single-cell transcriptomics to define mature adipocyte subtypes is complicated by the large size and high buoyancy of these cells, and the fact that the extensive proteolytic digestion required to isolate them from the adipose tissue compromises their viability and alters their phenotype [15]. While elegant solutions to these limitations have been developed [16], they are dependent on transgenic approaches and are thus less amenable to application for human studies.

As an alternative approach to studying human adipocyte subtypes and their development, we leveraged a recent finding by our laboratory that allows us to generate large numbers of mesenchymal progenitor cells from human adipose tissue with little loss of multipotency [17-19]. Many of these progenitor cells differentiate into large, long-lived unilocular adipocytes [18], and notably, some correspond to thermogenic adipocytes, as they respond to stimulation by inducing UCP1 and other brown adipose tissue markers [19]. To identify the distinguishing features of progenitor cells that give rise to adipocytes, and of the adipocyte subtypes generated from these progenitors, we obtained transcriptomic profiles before and after adipogenic induction, and before and after thermogenic stimulation of clonal cell populations derived from single mesenchymal progenitor cells.

Our results uncovered a minimum of four distinct adipocyte subtypes differing in profiles associated with distinct, recognizable adipocyte metabolic functions. Progenitor cells that give rise to each adipocyte subtype express distinct complements of transcriptional regulators and cytokines, many of which can explain existing results of the effects of genetic ablation or overexpression of these factors on adipose tissue function. Thus, our results provide a high- resolution perspective on the complexity of human adipose tissue, and propose tangible models bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

that can be further tested to better understand human adipose tissue function and systemic metabolism.

RESULTS

Morphological evidence for distinct adipocyte subtypes

To explore the mechanisms by which human adipocyte subtypes originate, we developed an approach summarized in Figure 1a and detailed in the Methods. Cells sprouting from human adipose tissue explants are plated as single cells, and proliferation allowed until near confluence, when they are split 1:3 into three wells of a 384 well multiwell plate. One well is maintained in a non-differentiated state (C), two wells are subjected to adipose differentiation (M), and one of the differentiate wells (F) is stimulated with Forskolin (Fsk) for the last 3 days of culture. Wells are then imaged and RNA extracted. Phase images of clones after differentiation and Fsk activation revealed three main subtypes. One consisted of cells that proliferated to confluence, but fail to accumulate lipid droplets in response to differentiation, and were therefore not considered further in this study (Figure 1b). A second subtype of cells accumulated lipid droplets, but were not visibly affected by Fsk (Figure 1c), suggesting a weak response to the lipolytic effects of this drug. A third subtype of cells accumulated lipid droplets and responded to Fsk with visible changes in droplet size and number (Figure 1d), suggesting a strong lipolytic response. To determine whether these differences were explained by expected stochastic variation, or whether they reflect the existence of true distinct adipocyte subtypes, we analyzed the frequency distributions of droplet size, measured after image processing as exemplified (Figure 1 b-d, lower panels). Aggregated values for lipid droplet size from all clones (Figure 1e), produced a monophasic histogram representative of the normal variation of adipocyte droplet sizes in these cells collectively. However, when we plotted the mean droplet size per clone, we obtained a non- monophasic distribution (Figure 1f), revealing the existence of adipocyte subtypes that differ in mean droplet size. We then quantified the responsiveness of each clone by comparing the size of lipid droplets before and after Fsk treatment (Figure 1g). Again, Fsk responsiveness was not represented by a monophasic distribution, as the histogram of the values (droplet size in MDI+Fsk as % of MDI only value) contained at least two peaks, a major peak centered around 100% (no response) and another at around 60% (decrease of lipid droplet size to 60% of non-stimulated bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

value) (Figure 1h). These results suggest that at least two adipocyte subtypes exist, based on their mean droplet size and differential responsiveness to Fsk.

Figure 1. Single mesenchymal progenitors give rise to morphologically different adipocyte subtypes. a. Mesenchymal progenitor cells from adipose tissue are induced to proliferate under pro-angiogenic conditions in a 3-dimensional hydrogel. Single cells are isolated, and proliferation allowed until near confluence, when they are split 1:3 into three wells of a 384 well multiwell plate. One well is maintained in a non-differentiated state (C), two wells are subjected to adipose differentiation (M), and one of the differentiate wells is stimulated with Forskolin (Fsk) for the last bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

3 days of culture (F). Wells are then imaged and RNA extracted. b. Example of a clone that failed to undergo adipose differentiation; c. Example of a clone that underwent adipocyte differentiation but was unresponsive to Fsk as assessed by lipid droplet size; d. Example of a clone that underwent adipocyte differentiation and responded to Fsk with decrease in lipid droplet size. e. Frequency distribution of all lipid droplets measured in all clones that underwent adipocyte differentiation. f. Frequency distribution of the mean droplet size per clone in clones that underwent adipocyte differentiation. g. Measurement of mean droplet sizes per clone comparing M and F conditions. h. Frequency distribution of responsiveness to Fsk, where mean lipid droplet size is expressed as percent of droplet size in M for each specific clone.

Identification of transcriptionally distinct adipocyte subtypes

We then asked whether the morphological evidence for distinct human adipocyte subtypes is supported by differences in gene expression. We extracted RNA from 156 samples comprising non-differentiated (C), differentiated (M) and differentiated plus Fsk-stimulated (F) conditions for each of the 52 clones obtained (Figure 1a). Principal component analysis revealed clear segregation of C, M and F conditions (Figure 2a), verifying that differentiation and Fsk stimulation had major effects on gene expression in these clones. Interestingly, a broader spread is seen in the first principal component of M and F compared to C conditions, indicating generation of divergent transcriptional landscapes following induction of differentiation and thermogenic stimulation.

We then analyzed the transcriptomes of clones in the M condition using hierarchical clustering. As a first approach, we sought to focus on that were associated with key properties of adipocytes. Thus, we clustered genes that correlated with lipid droplet size (as assessed in Figure 1) with a Pearson rank order correlation coefficient higher than 0.5 or lower than -0.5 (Supplementary Table 1). Unsupervised hierarchical clustering of these 447 genes resulted in 4 major clusters (Figure 2b) indicating 4 distinct adipocyte subtypes. To determine how these adipocyte subtypes related to classical adipose tissue traits, we looked at the expression levels of the canonical adipocyte genes ADIPOQ (Adiponectin), SLC2A4 (GLUT4) and LEP (Leptin) within these clusters. Clones in Clusters 1 and 2 expressed the highest levels of ADIPOQ and SLC2A4 (Figure 2 c,d), but surprisingly the highest expression of LEP was within cluster 4, which displayed lower expression levels of ADIPOQ and GLUT4 (Figure 2e). Mean lipid droplet size in Cluster 4 was bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

also significantly smaller compared to Cluster 1 (Figure 2f). These results reveal the existence of adipocyte subtypes with distinct transcriptomic states associated with key adipocyte functions.

To verify the robustness of these results, we performed an additional, unsupervised clustering analysis of the M samples using the top 1500 genes displaying the highest variance. This analysis produced 5 distinct clusters, overlapping with the clusters generated by hierarchical clustering of genes correlating with lipid droplet size (Supplementary Figure 2). Importantly, LEP and ADIPOQ- expressing adipocytes were again segregated in two different clusters.

To test whether the differences in ADIPOQ and LEP gene expression between adipocyte subtypes would be reflected by the levels of expression of their encoded , we performed immunofluorescence on adipocytes differentiated from unsorted progenitors obtained from two independent subjects. In both cases, cells expressing LEP and very little ADIPOQ, and vice versa, could be readily found (Figure 2g,h). Quantitative analysis of expression was done on 100 cells from each subject. Histograms of these values revealed very different frequency distributions of cells expressing ADIPOQ or LEP (Figure 2, i,j), consistent with these proteins being expressed by different adipocyte types. In addition, differences in the frequency distributions between the two subjects were observed (Figure 2, compare panel i with panel j). Notably, the inter-subject difference in distribution, i.e. the relative number of cells expressing each , correlated with the difference in overall expression of ADIPOQ and LEP mRNA between the two subjects (Figure 2k).

To explore whether the development of adipocytes preferentially expressing ADIPOQ or LEP mRNA occurs in vivo, we stained for LEP and ADIPOQ in isolated primary adipocytes obtained by collagenase digestion of human subcutaneous adipose tissue. We found that expression in mature unilocular adipocytes is heterogenous (Figure 2l) and that some adipocytes expressed high levels of only one of these proteins. Interestingly, staining of adipocytes from mouse epididymal adipose tissue has already been remarked as being similarly heterogeneous, with some cells expressing no detectable leptin [20]. While immunostaining alone cannot distinguish between stochastic variations in protein levels and the existence of true cell subtypes, the consistency between gene expression levels in clones and single-cell immunostaining of primary cultured and freshly isolated cells support the existence of distinct subtypes of white adipocytes specialized for production of specific adipokines. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

To enumerate the genes that define these specific clusters, and to delineate pathways enriched by these genes, we performed differential expression and pathway enrichment analysis between Clusters 1 and 4 as defined in Figure 2b. We find 1308 differentially expressed genes with a p- adjusted value < 0.05. Of these, 174 genes were more highly expressed by > 5-fold in Cluster 1, and 48 genes were more highly expressed > 5-fold in Cluster 4 (Figure 2m and Supplementary Table 2). Pathway enrichment analysis of genes enriched in Cluster 1 included PPARγ signaling and pathways, containing genes such as DGAT2, FABP4, PLIN1 and PLIN4 (Figure 2n and Supplementary table 3), and genes enriched in Cluster 4 were associated with extracellular matrix interaction and TGFβ signaling (Figure 2o and Supplementary table 3). Individual genes that were significantly enriched in Cluster 4 included SFRP2, an important regulator the Wnt signaling pathway which is central to adipogenesis [21].

bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Figure 2. Adipocyte progenitors give rise to distinct adipocyte subtypes with different functional properties. a. Principal component analysis of 52 clones analyzed under non- differentiated (C, black symbols), differentiated (M, green symbols) and differentiated plus (Fsk) stimulated (F, red symbols) conditions. b. Unsupervised hierarchical clustering of TPM values for 447 genes (Supplementary Table 1) correlated with lipid droplet size. Each row is independently scaled from blue (lower) to red (higher) values. c-e. TPM values for the genes indicated in the y axis for the clones in the x-axis Panels on the left are the mean and SEM of TPM values from each cluster defined by the heat map above. f. Mean droplet size for the clones in the x-axis. Panels on the left are the mean and SEM for each cluster as defined by the heat map above. Statistical comparison between all columns was done using the Kruskal-Wallis test corrected for multiple comparisons by controlling the false discovery rate using the Benjamini, Krieger and Yekutieli test as implemented in GraphPad Prism 7. *=p<0.05; **=p<0.01; ****=p<0.0001. g,h. Immunofluorescence staining of differentiated adipocytes from two individuals using to leptin and adiponectin, and Hoechst for nuclear staining. Arrows indicate adipocytes expressing mostly leptin (green), mostly adiponectin (red) or both (yellow). i,j. Frequency distribution of staining intensities per cell for leptin (green) or adiponectin (red) from 100 cells in fields exemplified in g and h. k. qRT-PCR of LEP and ADIPOQ in cultures exemplified in g and h. Values are expressed as fold difference relative to the lowest detectable value. l. Immunofluorescence staining of primary human adipocytes isolated by collagenase digestion. Green is leptin and red is adiponectin. Top panel is a single optical plane and bottom is the projection of all optical planes comprising the 3-dimensional volume of the cells. m. Volcano plot comparing cluster 1 and cluster 4. Red and green symbols are genes enriched in cluster 1 or cluster 4, respectively. n,o. KEGG pathway enrichment of enriched genes in cluster 1 or cluster 4, respectively.

Identification and characterization of the “brite/beige” adipocyte subtype

We then asked whether any of the clusters would correspond to adipocytes of the “brite/beige” phenotype, as defined by the ability to respond to Fsk stimulation with expression of canonical thermogenic genes. Adipocytes differentiated from heterogeneous progenitor cultures were highly responsive to Fsk stimulation, with strong induction of UCP1, DIO2 and CIDEA after acute (6-24h) and chronic (24h to 7 days) exposure (Figure 3a), and with high expression of UCP1 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

protein after chronic exposure (Figure 3b). Notably, immunofluorescence staining of UCP1 was heterogeneous, consistent with “brite/beige” adipocytes representing a subtype within the population (Figure 3b). We then analyzed the effect of Fsk on lipid droplet size and expression of thermogenic genes in each clone. Most clones responded to Fsk with a decrease in adipocyte droplet size (Figure 3c), as expected from stimulation of in response to the drug. In addition, Fsk stimulated ADIPOQ expression in Clusters 1 and 2, and inhibited LEP expression in Cluster 4 (Figure 3d,e), which are responses previously seen in this cell population [19]. The brown adipose tissue genes DIO2 and CIDEC were more strongly induced in Cluster 2 (Figure 3f,g), suggesting this cluster as corresponding to the “brite/beige” adipocyte subtype.

For unknown reasons, UCP1 reads were detected only in a few clones, despite strong RT-PCR signals (Figure 3a) and accumulation of UCP1 protein (Figure 3a,b). To further test whether Cluster 2 represents “brite/beige” adipocytes, we leveraged recent findings that the levels of the long non-coding RNA, LINC00473, highly correlated with UCP1 levels in “brite/beige” human adipose tissue and in adipocytes derived from this depot [22]. Indeed, we verified this correlation by analyzing the levels of LINC00473 prior to stimulation by Fsk to the levels of UCP1 following stimulation in adipocytes differentiated from non-cloned cells from 3 independent subjects (Figure 3h). LINC00473 was highly enriched in Cluster 2 (Figure 3 i,j), consistent with this cluster representing the “brite/beige” adipocyte subtype.

Figure 3. Cluster 2 corresponds to “brite/beige” adipocytes. a. RT-PCR of thermogenic genes in non-differentiated (C) and differentiated cells at different times of exposure to Fsk. Values are expressed as fold relative to the lowest value. Ct values bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

for UCP1 were 29-31 at t=0 and decreased to 20-22 at t=6h in response to Fsk. Symbols are means and bars SEM of two independent experiments using cells from two subjects assayed in duplicate. b. Immuofluoresence staining using antibodies to mitochondrial Hsp70 (red) and UCP1 (green) of differentiate cells without or with exposure to Fsk for 7 days. c-g. TPM values for the genes indicated in the y-axis for the clones in the x-axis, segregated into clusters as defined in Figure 2. Black symbols and traces are the values for M and red symbols and traces are values for F. Panels on the left are the mean and SEM of TPM values from each cluster. Statistical comparison between M and F in each cluster was done using ANOVA, with correction for multiple comparisons using the Holm-Sidak test as implemented in GraphPad Prism 7. *=p<0.05; **=p<0.01; ***=p<0.001; ****=p<0.0001. h. Levels of Linc00473 and UCP1 analyzed by RT-PCR in differentiated adipocytes with and without Fsk stimulation for 6h, and values expressed as fold relative to lowest value. Levels of Linc00473 in non-stimulated cells are plotted against levels of UCP1 after stimulation. Linear regression analysis was performed using GraphPad Prism 7. i,j. TPM values for Linc00473 for the clones in the x-axis, segregated into clusters as defined in Figure 2. Bars are the mean and SEM of TPM values from each cluster.

To define features that distinguish “brite/beige” adipocytes from other subtypes, we performed differential expression analysis between Cluster 2 and the combined Clusters 1, 3 and 4 (Figure 4a). We found 85 genes to be differentially expressed with a p-adjusted value < 0.05. Of these, 12 genes were increased by 1.5-fold or more, and 32 genes were decreased by -1.5-fold or more in Cluster 2 relative to all other clusters (Supplementary Table 4). Pathway enrichment analysis of genes increased in Cluster 2 with a p-adjusted value of 0.1 or lower (Supplementary Table 5) revealed pathways related to carbohydrate, carbon and nitrogen metabolism. It is notable that the gene family of carbonic anhydrases was highly enriched (p-adj = 4.505E-4) by the presence of CA3 and CA9 (Figure 4a, Supplementary Table 5). Genes decreased in Cluster 2 corresponded to the ferroptosis and mineral absorption pathways (Figure 4c), which are defined by the gene SLC40A1 (, FPN1) (Figure 4a, Supplementary Table 5), the only mechanism known to mediate iron egress from cells.

We then examined the effects of Fsk on the levels of CA3, CA9, and SLC40A1/FPN1 in all clones. Notably, Fsk stimulated expression of CA3 in Cluster 2, but not in other clusters (Figure 4d). Interestingly, CA3 is thought to be highly tissue-specific for , consistent with bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

previously observed similarities between thermogenic adipose tissue and muscle [6]. Expression of CA9 also was stimulated by Fsk, but to a lesser extent (Figure 4e). Fsk did not significantly affect expression of SLC40A1/FPN1, which remained under-represented in both Clusters 1 and 2 (Figure 4f). Taken together, these results indicate that Cluster 2 comprises “brite/beige” adipocytes, and these adipocytes are primed to increase metabolism, accumulate iron, and resist ensuing oxidative stress.

Figure 4. Differential gene expression analysis identifies genes that characterize human “brite/beige” adipocytes. a. Volcano plot of differential expression analysis results comparing Cluster 2 with all other clusters. b,c. KEGG pathway analysis of genes increased or decreased in Cluster 2 relative to all other clusters. d-f. TPM values for the genes indicated in the y-axis for the bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

clones in the x-axis, segregated into clusters as defined in Figure 2. Black symbols and traces are the values for M and red symbols and traces are values for F. Panels on the left are the mean and SEM of TPM values from each cluster. Statistical comparison between M and F in each cluster was done using ANOVA, with correction for multiple comparisons using the Holm-Sidak test as implemented in GraphPad Prism 7. ***=p<0.001.

Characteristics of progenitor cells that give rise to adipocyte subtypes.

The results of differential expression analysis indicate that adipocyte subtypes have different physiological functions; Cluster 1 represents a highly lipogenic subtype, Cluster 2 is a “brite/beige” thermogenic subtype and Cluster 4 is specialized for leptin secretion. To explore the mechanisms by which distinct adipocyte subtypes might be generated, we analyzed the features of the progenitor cells (C condition) that give rise to cells in each cluster. We first compared the progenitors that gave rise to Cluster 1 with all others to define what genes might predict the development of highly lipogenic adipocytes. 36 genes were differentially expressed by 2-fold with a p-adjusted value of 0.05 (Supplementary Table 6). Only 2 of these genes (EVI2A and CCDC69) were up-regulated in progenitors giving rise to Cluster 1 (Figure 5a). Pathway enrichment analysis of genes depleted in Cluster 1 progenitor cells revealed the KEGG pathway for IL17 signaling, which includes CXCL6, CXCL5, CXCL1 (Figure 5b and Supplementary Table 7).

To define what genes might predict the development of “brite/beige” adipocytes, we then compared the progenitors that gave rise to Cluster 2 with all others. 39 genes were differentially expressed by 2-fold with a p-adjusted value of 0.05 (Supplementary Table 8). Of these, 18 had higher expression and 21 had lower expression in comparison to all other clones (Figure 5c). Pathway enrichment analysis of genes enriched in Cluster 2 progenitors mapped to the KEGG pathway for IL17 signaling, including CSF3, IL1B, PTGS2 and CXCL8 (Figure 5d, Supplementary Table 9). Other genes significantly enriched in Cluster 2 include NR4A2, a cyclic AMP-responsive gene that has been implicated in enhancing UCP1 [23]. The 21 genes under- represented in Cluster 2 were not enriched in any currently specified pathway but mapped significantly to MAD box and basic helix-loop-helix gene families (MEF2C, ID1 and NPAS1) (Figure 5e, Supplementary Table 9). bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

To define what genes might predict the development of adipocytes expressing LEP, we compared the progenitors giving rise to Cluster 4 with all others. This comparison did not result in any significantly differentially expressed genes. As an alternative approach, we searched for genes expressed in the C condition that might correlate significantly with LEP levels in the M condition. We first defined the spurious level of correlation by calculating the Spearman rank order correlation coefficient with 10 sets of 52 random numbers (to match the number of analyzed clones, generated using the “RAND” function in Excel). The mean and 95% CI of the Spearman correlation coefficient between genes in the C condition and the random number sets was compared with their correlation with LEP in M (Figure 5f). The difference between the Spearman correlation coefficient between random numbers and LEP values decreased asymptotically (Figure 5f, right panel), and we further analyzed the genes displaying differences larger than 0.05 (Supplementary Tables 10 and 11). The gene most strongly positively correlated was ZNF395, which has been previously associated with adipogenesis [24, 25], and pathways most enriched included the pathway (genes ENO1 and ENO2) and pathways associated with pathogenic infection which are defined by cytoskeletal genes (Supplementary Table 11). The genes most negatively correlated included PAPPA (pappalysin), which regulates local bioavailability of IGF-1 and is differentially expressed in pre-adipocytes from different depots [26], and BSCL2, which is associated with lipid droplet formation and when mutated gives rise to congenital generalized lipodystrophy. Pathways most enriched in genes negatively correlated with LEP expression after differentiation included and terpenoid synthesis pathways (Supplementary Table 11). bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Figure 5. Features of progenitor cells that give rise to adipocyte subtypes. a. Volcano plot of differential expression analysis results comparing progenitors corresponding to Cluster 1 with all others. b. KEGG pathway analysis of genes under-represented in Cluster 1. c. Volcano plot of expression analysis results comparing progenitors corresponding to Cluster 2 with all others. d. KEGG pathway analysis of genes over-represented in Cluster 2. e. Gene family enrichment of genes under-represented in Cluster 2. f. Spearman rank order correlation between genes in C condition and the mean and 95% confidence interval of 10 sets of random numbers (RANDOM) or between genes in C condition and LEP levels in M condition. Right panel is the difference between the correlation coefficients seen in the left panel.

bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

DISCUSSION The major finding of this study is the existence of multiple distinct human adipocyte subtypes, differing in gene expression profiles related to key adipose tissue functions. These findings expand our understanding of adipocyte heterogeneity from white vs “brown/brite/beige” to subtypes within these categories that can collectively mediate the multiple diverse functions of adipose tissue. Our findings were made by analyzing the trancriptomes of cell populations derived from single human mesenchymal pluripotent cells after subjecting these populations to specific adipogenic and thermogenic stimuli. Among many advantages of this approach, one is that stochastic variations in degree and timing of gene expression displayed by single cell types, such as those seen in populations of 3T3- adipocytes [27-31] are averaged out in the population, and thus adipocyte subtype identity is determined from average population responsiveness to adipogenic and thermogenic stimulation. Our finding of distinct adipocyte phenotypes is consistent with prior observations of heterogeneity within adipocytes from human and mouse adipose tissue [32]. While these prior observations could be potentially attributed to stochastic variation, our results showing underlying gene expression architecture suggest that they reflect the existence of human adipocyte subtypes. Moreover, our findings trace these differences to progenitor cells that give rise to each subtype.

While we expected to find two adipocyte subtypes, corresponding to white and “brite/beige” adipocytes, we were surprised to find additional subtypes differing in key adipocyte functions, such as lipogenic capacity and LEP and ADIPOQ expression (Figure 2e). This finding may help understand the basis for the known differences in LEP expression associated with different depots [33], and the association between systemic LEP levels and degree of adiposity, which correlates but is not fully explained by adipocyte size [34], or lipid content. For example, LEP expression is not increased under conditions where adipocyte lipid accumulation is increased due to inhibition of lipolysis [35]. Indeed, mean lipid droplet size in adipocytes preferentially expressing LEP (Cluster 4) was significantly smaller compared to other clusters (Figure 2f). We also find that adipocytes specialized for LEP expression are highly responsive to Fsk (Figure 3e), suggesting that systemic LEP levels might be defined by the relative number and responsiveness of these cells. Indeed, levels of LEP expression in mixed adipocyte populations is explained by the number of adipocytes expressing high LEP levels, rather than the mean levels of expression per cell (Figure bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

2i-k). Interestingly, SFRP2, a potent antagonist of Wnt signaling, is amongst the most highly differentially expressed genes in these LEP-expressing adipocytes. Wnt suppresses adipocyte differentiation, and this effect is antagonized by SFRPs [36]. Moreover, levels of circulating SFRP2 are positively correlated with BMI in humans [37]. Together, these results suggest that adipocytes that express LEP can stimulate adipogenesis through regulation of Wnt signaling, and that this mechanism could contribute to the positive relationship between systemic LEP levels and increased adiposity.

Our results also distinguished an adipocyte subtype corresponding to the “brite/beige” phenotype, which is defined by increased mitochondrial biogenesis and uncoupling upon cAMP elevation . We find that “brite/beige” human adipocytes have features that enable them to accommodate increased mitochondrial biogenesis and oxidative stress following stimulation. Most salient are the decrease in SLC40A1/FPN1, the only known mechanisms for iron efflux [38], and the increased expression of carbonic anhydrases CA2 and CA3, which potently protect cells from oxidative stress [39-41]. Interestingly, CA3 is expressed at much higher levels in human skeletal muscle [41, 42] consistent with the notion that thermogenic adipose tissue and muscle share developmental features [6]. Iron metabolism is tightly associated with human adipose tissue function, as seen by increased levels of genes leading to iron efflux in response to overfeeding and their reversal with weight loss [43, 44]. Iron is a rate limiting and regulatory factor in adipose tissue browning and mitochondrial respiration [45, 46], and thereby is tightly associated with redox stress. The parallel down-regulation of SLC40A1/FPN1 to increase cellular iron, and up-regulation of CA2 and CA3 to protect from redox stress, provide a permissive environment for rapid mitochondrial biogenesis and enhanced respiratory flux. These are further enhanced in response to cAMP as evidenced by the elevation of CA3 expression in “brite/beige” adipocytes in response to Fsk (Figure 4d).

Our experimental paradigm allows us to explore the differential gene expression landscape of progenitors that give rise to adipocyte subtypes. “Brite/beige” adipocyte progenitors display high expression levels of cytokine genes, which could modulate the local concentrations of immune and stromal cells, which in turn could affect fate determination. Indeed, multiple reports exist of paracrine interactions with immune cells that enhance thermogenic activation [47-49]. In addition to cytokines, progenitors that give rise to “brite/beige” adipocytes display enhanced expression of PTGS2, the rate-limiting step in prostaglandin synthesis. PTGS2 has been directly associated with bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

formation of brown adipocytes in mice [50], and partial PTGS2 deficiency leads to obesity [51]. The results shown here suggest that the role of this may be conserved in humans, and that it may contribute to browning through effects on “brite/beige” adipocyte progenitors.

In addition to cytokines, genes that regulate gene expression are differentially expressed in progenitors for distinct adipocyte subtypes. For example, progenitors of Cluster 1, which represents adipocytes with high lipogenic capacity, are highly enriched in CCDC69, a coiled-coil domain protein implicated in central spindle assembly [52], which is enriched in human adipose tissue (www.proteinatlas.org/ENSG00000198624-CCDC69/tissue). The 21 genes decreased in Cluster 2 relative to other clones mapped significantly to MAD box and basic helix-loop-helix gene families (MEF2C, ID1 and NPAS1). Id1 depletion results in increased generation of thermogenic adipose tissue in mice, consistent with its lower levels in human progenitors fated for the “brite/beige” phenotype [53, 54].

In summary, our results provide evidence for multiple human adipocyte subtypes specialized for key functions associated with adipose tissue, arising from progenitors that can be distinguished by their complement of expressed cytokines and transcriptional regulatory factors. Many of our results are consistent with published observations of the role of these regulatory factors, and many additional results provide the basis for exploration of additional currently unknown mechanisms of adipocyte differentiation and adipose tissue function. The concept that adipose tissue function results from the ensemble of multiple subtypes of cells, rather than from modulation of function of a single cell type will help deconvolve the complex relationships between adipose tissue biology and metabolic disease. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 2. Alternative approach to hierarchical clustering. Unsupervised hierarchical clustering of TPM values for the 1500 genes displaying the highest variance in the M condition. K-means clustering was used to define gene clusters across the vertical axis. Clone identifiers are color coded to identify the cluster each clone associated with using the strategy bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

depicted in Figure 2. Bottom panel shows TPM values for LEP and ADIPOQ across the defined clusters.

ACKNOWLEDGEMENTS

This work was supported by grant DK089101-04 to SC. We acknowledge the use of services from the UMASS Bioinformatics Core, supported by NIH CTSA grant UL1 TR000161-05.

AUTHOR CONTRIBUTIONS

SC supervised this work. SYM, AD, ZY, AS, RMJG, AK, LL, RM, MG: hypothesis generation, conceptual design, data analysis, and manuscript preparation. SYM, AD, ZY, AS, LL conducted experiments and data analysis.

DECLARATION OF INTERESTS

The authors declare no competing interests.

MATERIALS AND METHODS Clonal isolation and expansion of cells Cells were obtained as outlined in a previously published method [19]. Briefly, explants from human subcutaneous adipose tissue were embedded in Matrigel (BD Biosciences Cat. No. 356231) and cultured in EBM-2 medium supplemented with endothelial growth factors (EGM-2MV) (Lonza) for 14 days. Sprouting cells were collected using dispase, plated on 15-cm tissue culture plates and grown to confluency. Media from these plates (25 mL per plate) was collected, spun at 2000 x g for 10 min, filtered through 0.2 μM filter and mixed with FBS (65% Media and 35% FBS). This FBS enriched medium was then mixed with fresh EGM-2MV media in a 1:1 ratio and used for collecting and maintaining sorted single cells. For single cell sorting, a vial of frozen cells was revived and cells sorted as singlets into 384-well plates using a BSL3 BD FACSAria Cell Sorter (BD Biosciences). Clones that grew to confluency were further split 1:3. Briefly, media was removed, and cells were washed with 1X PBS (100 μL). Trypsin (1X) + 0.05% collagenase (50 μL total) was added and incubated at 37ᵒC until the cells were detached. Conditioned media (45 μL) was added and suspensions divided into 3 wells. Additional 50 μL of conditioned media was added to each well and cells were grown to confluence. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Cells in control group (C condition) were maintained in DMEM + 10% FBS. Adipogenic differentiation of two of the three wells was induced by providing confluent cells with DMEM + 10% FBS supplemented with 0.5 mM 3-isobutyl-1-methylxanthine, 1 μM dexamethasone and 1 μg/ml (MDI). Half of MDI media was removed daily and replaced with media with 2x MDI. After 72h the MDI media was replaced with DMEM + 10% FBS. 50% of the media was replaced with fresh media every 48h until day 14. At that point, one of the two wells was stimulated by adding 10 μM Forskolin (Fsk), every 24h for 72 h. RNA extraction Total RNA extraction was performed using NucleoSpin RNA XS kit (Clontech Laboratories Inc. Cat. No. 740902). Briefly, cells were detached as mentioned in previous section, snap frozen in liquid nitrogen and stored at -80ᵒC overnight. Next day, RNA was prepared from a total of 168 samples across three conditions (C, M and F) according to the manufacturer’s instructions. RNA quality was assessed using High Sensitivity Fragment Analyzer. cDNA library preparation cDNA library was prepared using SMART-Seq v4 3’ DE kit according to the manufacturer’s instructions (Clontech Laboratories Inc. Cat. No. 635040). For each clone, 1 ng of total RNA was used as input to synthesize cDNA which was subsequently amplified. Control RNA was provided with the kit. All samples were processed concurrently during library preparation. Prior to purification, the amplified cDNA’s were grouped in fourteen pools of twelve samples each with a unique barcode (Clonetech Oligo dT in-line indexes or IL 1-12) ligated to the 3’end during cDNA synthesis. Pooled and purified libraries were fragmented using Nextera XT DNA Sample Preparation Kit (Illumina, Cat. Nos. FC-131-1024, FC-131-1096). Illumina sequencing indexes (provided with the SMART-Seq v4 3’ DE kit) were added to the fragmented libraries and indexed libraries were amplified. For intermittent purification steps, Agencourt AMPure XP PCR purification kit (5 ml Beckman Coulter Part No. A63880; 60 ml Beckman Coulter Part No. A63881) was used. Purified samples were analyzed on a High Sensitivity Fragment Analyzer (Agilent Technologies). The pooling strategy and the combinations of Illumina indices employed in this experiment are given in Table 1.

Table 1: Pooling and indexing strategy for 168 samples grouped in fourteen pools. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Pooling strategy Indexing strategy Pool no. Clonetech in-line index No. of clones Pool no. Illumina Reverse Index (Index 1) Illumina Forward Index (Index 2) 1 IL-1 to IL-12 12 1 1 1 2 IL-1 to IL-12 12 2 2 2 3 IL-1 to IL-12 12 3 3 3 4 IL-1 to IL-12 12 4 4 4 5 IL-1 to IL-12 12 5 5 5 6 IL-1 to IL-12 12 6 6 6 7 IL-1 to IL-12 12 7 7 7 8 IL-1 to IL-12 12 8 8 8 9 IL-1 to IL-12 12 9 9 1 10 IL-1 to IL-12 12 10 10 2 11 IL-1 to IL-12 12 11 11 3 12 IL-1 to IL-12 12 12 12 4 13 IL-1 to IL-12 12 13 1 5 14 IL-1 to IL-12 12 14 2 6 (A) (B)

Denaturation and dilution of purified libraries Purified libraries were denatured and diluted as outlined in the “Denature and Dilute Libraries Guide” designed for the NextSeq sequencing platform by Illumina. Libraries were sequenced across three different sequencing runs with final loading concentration equal to 1.8 pM. Un- indexed control library PhiX (NextSeq™ PhiX Control Kit FC-110-3002 by Illumina) was denatured and diluted to 1.8 pM according to the “Denature and Dilute Libraries Guide” and used as a spike- in to provide improved nucleotide diversity. Sequencing The choice of sequencing kits and the decision on number of pools to be sequenced in a single run were based on the ENCODE guidelines for RNA-seq experiments. The libraries were sequenced on NextSeq 500 platform using NextSeq® 500/550 High Output Kits v2 (75 cycles; Cat. No. FC-404- 2005) for paired-end sequencing. In this, 35 read-1 cycles and 40 read-2 cycles were performed with adapter trimming feature turned off. Transcript/gene sequence information was exclusively present in read 1 whereas most of read 2 included sequence of the matched IL’s. Data acquisition and analysis: The FASTQ files obtained were de-multiplexed using SMART-Seq DE3 Demultiplexing Software by Clonetech, which organized reads into individual libraries by IL’s. Bowtie was used to align the reads to the human assembly hg19 [55] . The aligned reads were then processed by RSEM to generate un-normalized, TPM, and FPKM counts [56]. Sequencing run quality was analyzed and bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

clones that did not have high-fidelity sequence data or that had low Sanger quality scores were eliminated from further analysis. Further review and inspection of the data set revealed no significant variation between the sequencing runs. Furthermore, upon distribution analysis, no skew was detected between runs. Clones that did not have sufficient count results in all three conditions, Control (C), MDI (M), and Forskolin (F), were disregarded from this portion of the study. In total, 12 out of the 168 sequenced samples were excluded from this portion of the study, leaving 52 clones under 3 conditions (C, M and F). Clustering A matrix containing TPM counts of genes in “M” clones, and values for mean droplet sizes was analyzed using the Broad Institute “Morpheus” software (https://software.broadinstitute.org/morpheus/). Genes with a correlation coefficient greater than 0.5 and less than -0.5 (to account for negative monotonic trends) to lipid droplet size were then used for hierarchical clustering, using one minus the Spearman’s coefficient of correlation to calculate distances between clusters and then used the average of complete and single linkage to generate the dendrogram shown in Figure 2b.

Pathway Analysis Differential expression between clusters defined in Figure 2b was calculated using DESeq [57]. Lists of differentially expressed genes were uploaded to ToppGene [58] via the ToppFun uploader. ToppFun default parameters were used to generate KEGG and GO Term lists. Pathways were considered to be significantly enriched if the p-adj value was less than 0.05. Immunofluorescence: Antibodies against human adiponectin and leptin were purchased from Thermo Fisher (Adiponectin Monoclonal (19F1); Cat. No. MA1-054, Leptin Polyclonal Antibody; Cat. No. PA1-051). Cells were rinsed twice with 1X PBS (pH 7.4) and fixed with 4% PFA for 15 minutes followed by three washes with 1X PBS. Fixed cells were permeabilized using freshly prepared permeabilization buffer (1% FBS and 0.5% Triton X-100 in 1X PBS) for 30 minutes and incubated with a cocktail of anti-adiponectin (1:100) and anti-leptin (1:100) primary antibodies for 2 hrs. Cells were washed three times with permeabilization buffer followed by incubation with a cocktail of anti-mouse (Alexa Fluor® 594 Goat anti-mouse, Thermo Fisher A11005; 1:1000), anti-rabbit (Alexa Fluor® 488 Donkey anti-rabbit, Thermo Fisher A21206; 1:1000) and DAPI (Hoechst bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

33342, trihydrochloride trihydrate; Thermo Fisher H3570; 1:2000) for 30 minutes. After three washes with permeabilization buffer, cells were mounted using ProLong™ Gold Antifade Mountant (Thermo Fisher P36930). Antibody dilutions were prepared fresh in permeabilization buffer. All incubation and wash steps were performed at room temperature on a . Image acquisition and processing: Cells were imaged with a Zeiss Axiovert 200M Fluorescence microscope. ImageJ (FIJI) was used for image processing and normalization. Mean fluorescence intensity was calculated on individual cells and the values were used to plot frequency distribution using GraphPad Prism v.7. Adipocyte droplet size was quantified using Fiji software from brightfield images of individual wells of 384- well multiwells taken with a 20x objective. Image contrast was enhanced (background subtraction 50, enhance contrast 0.5), and images were converted to 8-bit greyscale images. The images were subsequently binarized, and size in pixels2was recorded for all objects in the field with a circularity between 0.5 and 1. Real Time-PCR Total RNA was extracted and reverse-transcribed according to the manufacturer’s protocol (iScript™ cDNA Synthesis Kit, BIO-RAD). cDNA was loaded in duplicate and qPCRs were performed using the BIO-RAD CFX System (BIO-RAD, Hercules, CA, USA) and iQ™ SYBR Green™ Supermix (BIO-RAD cat. No. 1708882). Primers were designed through the Primer3 software platform. SYBR Primer specificity was confirmed via melt curve analysis.

QUANTIFICATION AND STATISTICAL ANALYSIS.

GraphPad Prism 7.0 was used for all analyses. Parametric or non-parametric test were chosen based on results from the D’Agostino-Pearson omnibus normality test, and are described in each figure. Heatmaps were plotted using Morpheus (Broad Institute).

REFERENCES 1. Kusminski, C.M., P.E. Bickel, and P.E. Scherer, Targeting adipose tissue in the treatment of obesity-associated diabetes. Nat Rev Drug Discov, 2016. 15(9): p. 639-60. 2. Nedergaard, J. and B. Cannon, Brown adipose tissue as a heat-producing thermoeffector. Handb Clin Neurol, 2018. 156: p. 137-152. 3. Zwick, R.K., et al., Anatomical, Physiological, and Functional Diversity of Adipose Tissue. Cell Metab, 2018. 27(1): p. 68-83. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

4. Cinti, S., The adipose organ: morphological perspectives of adipose tissues. Proc Nutr Soc, 2001. 60(3): p. 319-28. 5. Sanchez-Gurmaches, J. and D.A. Guertin, Adipocytes arise from multiple lineages that are heterogeneously and dynamically distributed. Nat Commun, 2014. 5: p. 4099. 6. Wang, W. and P. Seale, Control of brown and beige fat development. Nat Rev Mol Cell Biol, 2016. 17(11): p. 691-702. 7. Shinoda, K., et al., Genetic and functional characterization of clonally derived adult human brown adipocytes. Nat Med, 2015. 21(4): p. 389-94. 8. Lee, M.J., Y. Wu, and S.K. Fried, Adipose tissue heterogeneity: implication of depot differences in adipose tissue for obesity complications. Mol Aspects Med, 2013. 34(1): p. 1-11. 9. Newby, F.D., M.N. Sykes, and M. DiGirolamo, Regional differences in adipocyte lactate production from . Am J Physiol, 1988. 255(5 Pt 1): p. E716-22. 10. Jensen, M.D., Regulation of forearm lipolysis in different types of obesity. In vivo evidence for adipocyte heterogeneity. J Clin Invest, 1991. 87(1): p. 187-93. 11. Schwalie, P.C., et al., A stromal cell population that inhibits adipogenesis in mammalian fat depots. Nature, 2018. 559(7712): p. 103-108. 12. Hepler, C., et al., Identification of functionally distinct fibro-inflammatory and adipogenic stromal subpopulations in visceral adipose tissue of adult mice. Elife, 2018. 7. 13. Burl, R.B., et al., Deconstructing Adipogenesis Induced by beta3-Adrenergic Activation with Single-Cell Expression Profiling. Cell Metab, 2018. 28(2): p. 300-309 e4. 14. Sempere, J.M., et al., Single cell-derived clones from human adipose stem cells present different immunomodulatory properties. Clin Exp Immunol, 2014. 176(2): p. 255-65. 15. Viswanadha, S. and C. Londos, Optimized conditions for measuring lipolysis in murine primary adipocytes. J Lipid Res, 2006. 47(8): p. 1859-64. 16. Roh, H.C., et al., Simultaneous Transcriptional and Epigenomic Profiling from Specific Cell Types within Heterogeneous Tissues In Vivo. Cell Rep, 2017. 18(4): p. 1048-1061. 17. Tran, K.V., et al., Distinct adipocyte progenitor cells are associated with regional phenotypes of perivascular aortic fat in mice. Mol Metab, 2018. 9: p. 199-206. 18. Rojas-Rodriguez, R., et al., Generation of Functional Human Adipose Tissue in Mice from Primed Progenitor Cells. Tissue Eng Part A, 2018. 19. Min, S.Y., et al., Human 'brite/beige' adipocytes develop from capillary networks, and their implantation improves metabolic in mice. Nat Med, 2016. 22(3): p. 312-8. 20. Barr, V.A., et al., Insulin stimulates both leptin secretion and production by rat white adipose tissue. Endocrinology, 1997. 138(10): p. 4463-72. 21. Christodoulides, C., et al., Adipogenesis and WNT signalling. Trends Endocrinol Metab, 2009. 20(1): p. 16-24. 22. Tran, K.-V., et al., A long-non-coding RNA, LINC00473, confers the human adipose tissue thermogenic phenotype through enhanced cAMP responsiveness. bioRxiv, 2018. 23. Kumar, N., et al., Orphan NOR-1 enhances 3',5'-cyclic adenosine 5'- monophosphate-dependent -1 gene transcription. Mol Endocrinol, 2008. 22(5): p. 1057-64. 24. Wei, S., et al., Emerging roles of finger proteins in regulating adipogenesis. Cell Mol Life Sci, 2013. 70(23): p. 4569-84. 25. Hasegawa, R., et al., Identification of ZNF395 as a novel modulator of adipogenesis. Exp Cell Res, 2013. 319(3): p. 68-76. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

26. Davidge-Pitts, C., C.J. Escande, and C.A. Conover, Preferential expression of PAPPA in human preadipocytes from omental fat. J Endocrinol, 2014. 222(1): p. 87-97. 27. Katz, L.S., E. Geras-Raaka, and M.C. Gershengorn, Heritability of fat accumulation in white adipocytes. Am J Physiol Endocrinol Metab, 2014. 307(3): p. E335-44. 28. Loo, L.H., et al., Heterogeneity in the physiological states and pharmacological responses of differentiating 3T3-L1 preadipocytes. J Cell Biol, 2009. 187(3): p. 375-84. 29. Le, T.T. and J.X. Cheng, Single-cell profiling reveals the origin of phenotypic variability in adipogenesis. PLoS One, 2009. 4(4): p. e5189. 30. Nagayama, M., T. Uchida, and K. Gohara, Temporal and spatial variations of lipid droplets during adipocyte division and differentiation. J Lipid Res, 2007. 48(1): p. 9-18. 31. Shigematsu, S., S.L. Miller, and J.E. Pessin, Differentiated 3T3L1 adipocytes are composed of heterogenous cell populations with distinct receptor kinase signaling properties. J Biol Chem, 2001. 276(18): p. 15292-7. 32. Varlamov, O., et al., Cell-autonomous heterogeneity of nutrient uptake in white adipose tissue of rhesus macaques. Endocrinology, 2015. 156(1): p. 80-9. 33. Montague, C.T., et al., Depot- and sex-specific differences in human leptin mRNA expression: implications for the control of regional fat distribution. Diabetes, 1997. 46(3): p. 342-7. 34. Guo, K.Y., et al., Effects of obesity on the relationship of leptin mRNA expression and adipocyte size in anatomically distinct fat depots in mice. Am J Physiol Regul Integr Comp Physiol, 2004. 287(1): p. R112-9. 35. Wu, J.W., et al., Fasting energy homeostasis in mice with adipose deficiency of desnutrin/adipose . Endocrinology, 2012. 153(5): p. 2198-207. 36. Bodine, P.V., et al., The Wnt antagonist secreted frizzled-related protein-1 is a negative regulator of trabecular bone formation in adult mice. Mol Endocrinol, 2004. 18(5): p. 1222- 37. 37. Crowley, R.K., et al., SFRP2 Is Associated with Increased Adiposity and VEGF Expression. PLoS One, 2016. 11(9): p. e0163777. 38. Donovan, A., et al., The iron exporter ferroportin/Slc40a1 is essential for iron homeostasis. Cell Metab, 2005. 1(3): p. 191-200. 39. Silagi, E.S., et al., Expression of Carbonic Anhydrase III, a Nucleus Pulposus Phenotypic Marker, is Hypoxia-responsive and Confers Protection from Oxidative Stress-induced Cell Death. Sci Rep, 2018. 8(1): p. 4856. 40. Shi, C., et al., Carbonic anhydrase III protects osteocytes from oxidative stress. FASEB J, 2018. 32(1): p. 440-452. 41. Wade, R., et al., Nucleotide sequence, tissue-specific expression, and location of human carbonic anhydrase III: the human CAIII gene is located on the same chromosome as the closely linked CAI and CAII genes. Proc Natl Acad Sci U S A, 1986. 83(24): p. 9571-5. 42. Zimmerman, U.J., et al., Anti-oxidative response of carbonic anhydrase III in skeletal muscle. IUBMB Life, 2004. 56(6): p. 343-7. 43. Segrestin, B., et al., Adipose tissue expansion by overfeeding healthy men alters iron gene expression. J Clin Endocrinol Metab, 2018. 44. Moreno-Navarrete, J.M., et al., modulates iron-related proteins in adipose tissue. Diabetes Care, 2014. 37(4): p. 1092-100. 45. Kusminski, C.M., et al., MitoNEET-driven alterations in adipocyte mitochondrial activity reveal a crucial adaptive process that preserves insulin sensitivity in obesity. Nat Med, 2012. 18(10): p. 1539-49. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

46. Kusminski, C.M., J. Park, and P.E. Scherer, MitoNEET-mediated effects on browning of white adipose tissue. Nat Commun, 2014. 5: p. 3962. 47. Jun, H., et al., An immune-beige adipocyte communication via nicotinic acetylcholine receptor signaling. Nat Med, 2018. 24(6): p. 814-822. 48. Finlin, B.S., et al., Mast Cells Promote Seasonal White Adipose Beiging in Humans. Diabetes, 2017. 66(5): p. 1237-1246. 49. Brestoff, J.R., et al., Group 2 innate lymphoid cells promote beiging of white adipose tissue and limit obesity. Nature, 2015. 519(7542): p. 242-6. 50. Vegiopoulos, A., et al., -2 controls energy homeostasis in mice by de novo recruitment of brown adipocytes. Science, 2010. 328(5982): p. 1158-61. 51. Fain, J.N., L.R. Ballou, and S.W. Bahouth, Obesity is induced in mice heterozygous for cyclooxygenase-2. Prostaglandins Other Lipid Mediat, 2001. 65(4): p. 199-209. 52. Pal, D., et al., Role of a novel coiled-coil domain-containing protein CCDC69 in regulating central spindle assembly. , 2010. 9(20): p. 4117-29. 53. Patil, M., et al., Id1 Promotes Obesity by Suppressing Brown Adipose Thermogenesis and White Adipose Browning. Diabetes, 2017. 66(6): p. 1611-1625. 54. Patil, M., B.K. Sharma, and A. Satyanarayana, Id transcriptional regulators in adipogenesis and adipose tissue metabolism. Front Biosci (Landmark Ed), 2014. 19: p. 1386-97. 55. Langmead, B., et al., Ultrafast and memory-efficient alignment of short DNA sequences to the . Genome Biol, 2009. 10(3): p. R25. 56. Li, B. and C.N. Dewey, RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics, 2011. 12: p. 323. 57. Love, M.I., W. Huber, and S. Anders, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol, 2014. 15(12): p. 550. 58. Chen, J., et al., ToppGene Suite for gene list enrichment analysis and candidate gene prioritization. Nucleic Acids Res, 2009. 37(Web Server issue): p. W305-11. bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 1. Genes correlated with lipid droplet size

Pearson correlation Spearman rank order gene coefficient correlation coefficient Droplet Size 1 1 C4orf3 0.78 0.78 AOC3 0.77 0.82 G0S2 0.76 0.82 VKORC1L1 0.75 0.67 MLXIPL 0.75 0.73 MCAM 0.75 0.77 GPAM 0.75 0.77 ACOT2 0.74 0.76 SCD 0.73 0.81 C14orf180 0.73 0.79 CIDEC 0.73 0.77 FIBCD1 0.72 0.56 DGAT2 0.72 0.77 SORBS1 0.72 0.69 PRKAR2B 0.72 0.73 DBI 0.72 0.67 FAM213A 0.71 0.71 SHMT1 0.71 0.73 LPL 0.71 0.81 PLIN4 0.71 0.81 HCAR2 0.71 0.68 PPP1R1A 0.71 0.65 DLAT 0.71 0.59 FABP4 0.71 0.78 LINC01140 0.71 0.68 KCNK3 0.71 0.76 PTPRF 0.71 0.61 PLIN1 0.71 0.78 SORT1 0.7 0.74 TUSC5 0.7 0.74 ACACB 0.7 0.72 RAB20 0.7 0.73 KLB 0.7 0.73 GPD1 0.7 0.8 GPATCH11 0.69 0.66 AQP7 0.69 0.8 HADH 0.69 0.67 ECH1 0.69 0.71 ACSL1 0.69 0.78 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

CDKN2C 0.69 0.81 LPCAT3 0.68 0.57 PDE3B 0.68 0.71 ZNF117 0.68 0.68 BCL2L2-PABPN1 0.68 0.64 CSPG4 0.68 0.58 AIFM2 0.68 0.67 LAMB3 0.68 0.75 PNMA1 0.67 0.67 GPI 0.67 0.66 SAT1 0.67 0.56 THEMIS2 0.67 0.6 ENY2 0.67 0.58 ACLY 0.67 0.62 CKMT2 0.67 0.63 MGLL 0.67 0.56 FABP5 0.67 0.72 ELOVL5 0.67 0.69 MPC1 0.67 0.59 TMEM170B 0.66 0.64 ANKRD9 0.66 0.61 TM7SF2 0.66 0.61 PNPLA3 0.66 0.63 TMEM135 0.66 0.57 ENO1 0.66 0.66 ACSS2 0.66 0.69 DTX1 0.66 0.65 PPARG 0.66 0.62 ALG9 0.66 0.59 NCOA4 0.66 0.57 GHR 0.66 0.67 APBB1IP 0.66 0.63 ALDH1L1 0.66 0.78 BTG1 0.66 0.68 AGPAT2 0.66 0.68 MESP1 0.66 0.68 ACSF2 0.65 0.67 ISOC1 0.65 0.66 CDR1 0.65 0.63 ITGA7 0.65 0.57 PFKFB3 0.65 0.64 SLC16A7 0.65 0.63 CYB5A 0.65 0.66 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

FZD4 0.65 0.63 EBP 0.64 0.66 CKS1B 0.64 0.58 SREBF1 0.64 0.68 FADS1 0.64 0.66 PDK1 0.64 0.64 PNPLA2 0.64 0.77 PCK1 0.64 0.63 LAMA4 0.64 0.65 IMMT 0.64 0.58 CLMN 0.64 0.6 ACER3 0.64 0.76 ITIH5 0.64 0.7 ECHDC1 0.64 0.57 PC 0.64 0.67 DECR1 0.64 0.61 ABCG1 0.64 0.59 PLA2G16 0.64 0.71 CD36 0.64 0.63 TMEM37 0.64 0.61 FAM107B 0.64 0.58 APOE 0.64 0.59 ADRBK2 0.63 0.56 TSPAN13 0.63 0.54 MRPS16 0.63 0.57 MPST 0.63 0.56 GLUL 0.63 0.55 MVD 0.63 0.58 DIXDC1 0.63 0.59 HK2 0.62 0.58 SLC25A51 0.62 0.54 SCRN2 0.62 0.61 ELOVL6 0.62 0.62 CS 0.62 0.58 AQP7P1 0.62 0.69 FDPS 0.62 0.57 BCKDHA 0.62 0.65 FASN 0.62 0.65 MMP28 0.62 0.71 TACC3 0.62 0.6 AOC2 0.62 0.69 COX17 0.62 0.55 ETFDH 0.62 0.65 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

HRASLS5 0.62 0.76 PLEKHG6 0.62 0.56 HILPDA 0.62 0.65 FAM219B 0.61 0.64 CALB2 0.61 0.59 ZNF462 0.61 0.58 CAV2 0.61 0.47 DLD 0.61 0.53 CITED4 0.61 0.54 KCNIP2 0.61 0.65 KLHL25 0.61 0.49 MARCH1 0.61 0.38 STAT5A 0.61 0.55 DUSP6 0.6 0.47 BOK 0.6 0.56 SLC25A5 0.6 0.53 ACACA 0.6 0.59 PYGL 0.6 0.61 NDUFA6 0.6 0.57 UQCRFS1 0.6 0.57 CAMK1 0.6 0.52 SIRT2 0.6 0.5 LGALS12 0.6 0.74 NR1H3 0.6 0.5 LIMS1 0.6 0.54 IQCH 0.6 0.47 GYG2 0.6 0.68 GLRX5 0.6 0.61 CHRDL1 0.6 0.67 CYCS 0.6 0.46 PTGER3 0.6 0.63 ERV3-1 0.6 0.62 UCP2 0.6 0.78 NDUFB5 0.59 0.55 ZNF664 0.59 0.61 NSDHL 0.59 0.53 PPP2R1B 0.59 0.69 COL4A2 0.59 0.63 ACAA2 0.59 0.61 BBOX1 0.59 0.61 CCDC88A 0.59 0.48 BCKDK 0.59 0.62 APOC1 0.59 0.62 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

RXRA 0.59 0.6 CSAD 0.59 0.58 LBR 0.59 0.6 CDC42EP4 0.59 0.58 CCDC107 0.59 0.58 TSPO 0.59 0.51 CAT 0.59 0.46 OSBPL11 0.59 0.51 DHRS11 0.59 0.52 PLXND1 0.59 0.63 AZGP1 0.59 0.72 ARL6IP1 0.59 0.52 OIP5-AS1 0.59 0.52 AAMDC 0.59 0.58 ACADL 0.59 0.58 THRSP 0.58 0.77 MIR1244-1 0.58 0.57 ZPR1 0.58 0.55 SLC25A11 0.58 0.58 TECR 0.58 0.54 ITIH1 0.58 0.7 LIPE 0.58 0.7 COL4A1 0.58 0.57 GRPEL1 0.58 0.48 SUN2 0.58 0.46 LOC101929371 0.58 0.76 ADIRF 0.58 0.59 KIAA0408 0.58 0.54 UGP2 0.58 0.52 CYC1 0.58 0.51 ARRB1 0.58 0.46 HINT2 0.58 0.52 TMEM56 0.57 0.59 PDHA1 0.57 0.63 TALDO1 0.57 0.5 ANXA6 0.57 0.55 MYL6 0.57 0.54 NET1 0.57 0.43 CHI3L2 0.57 0.68 PEX19 0.57 0.46 PFKP 0.57 0.54 HP 0.57 0.67 ACADVL 0.57 0.49 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

ACADS 0.57 0.61 PECR 0.57 0.52 FAM162A 0.57 0.59 KLHDC8B 0.57 0.51 FOXP4-AS1 0.56 0.56 PCSK9 0.56 0.5 PEX13 0.56 0.47 CHP1 0.56 0.46 SEMA3G 0.56 0.71 PDHB 0.56 0.51 COX20 0.56 0.51 ACADM 0.56 0.53 LPIN1 0.56 0.64 TEAD2 0.56 0.55 LDHA 0.56 0.54 COX6C 0.56 0.46 MMP8 0.56 0.72 MYZAP 0.56 0.48 SCARB1 0.56 0.52 RNF125 0.56 0.57 PANK3 0.56 0.49 H3F3A 0.56 0.52 TF 0.56 0.63 ALKBH5 0.55 0.49 ANP32A 0.55 0.53 SOS1 0.55 0.49 INAFM1 0.55 0.52 ADIPOQ 0.55 0.75 RETSAT 0.55 0.46 TSPAN15 0.55 0.63 MTHFD1 0.55 0.47 AIFM1 0.55 0.53 MGAT2 0.55 0.42 CEBPA 0.55 0.64 MTFP1 0.55 0.56 GRB2 0.55 0.47 PPP2R5A 0.55 0.46 AK1 0.55 0.55 BLCAP 0.55 0.49 AK2 0.55 0.53 TXNIP 0.55 0.55 GPT2 0.55 0.5 FOXRED1 0.55 0.53 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

TMC6 0.54 0.68 NR3C1 0.54 0.43 ATP5O 0.54 0.48 PDZD2 0.54 0.57 PGRMC1 0.54 0.42 PTMA 0.54 0.58 LINC01119 0.54 0.47 CTC1 0.54 0.48 AQPEP 0.54 0.54 GCHFR 0.54 0.47 RBM8A 0.54 0.56 MRAP 0.54 0.56 EHMT1 0.54 0.56 UBE2K 0.54 0.47 AK4 0.54 0.53 PALM2 0.54 0.45 MRPL27 0.53 0.46 CXorf56 0.53 0.51 RHOT2 0.53 0.51 NOL3 0.53 0.61 ACAT2 0.53 0.44 PGK1 0.53 0.6 SYNPO 0.53 0.51 ZEB2 0.53 0.46 ALDH9A1 0.53 0.41 LPHN2 0.53 0.45 COL5A3 0.53 0.51 VPREB3 0.53 0.46 FNDC4 0.53 0.59 UQCRC2 0.53 0.54 MLYCD 0.53 0.55 SLC7A6 0.53 0.47 HES6 0.53 0.48 ME3 0.53 0.5 TMBIM6 0.53 0.48 LSS 0.53 0.43 PDP2 0.53 0.42 PEMT 0.53 0.51 KIAA0040 0.53 0.5 BCAP31 0.53 0.47 COX7B 0.52 0.49 LRIG1 0.52 0.34 RNASE4 0.52 0.51 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

LINC00632 0.52 0.42 CTDSP1 0.52 0.51 ADCY5 0.52 0.5 BMPER 0.52 0.45 TOMM40 0.52 0.43 TROVE2 0.52 0.51 CA9 0.52 0.45 HN1L 0.52 0.52 VDAC1 0.52 0.46 GSTCD 0.52 0.43 TPI1 0.52 0.53 MRPS14 0.52 0.49 ACVR1C 0.52 0.66 NDUFB8 0.52 0.51 HAS2 0.52 0.36 MRPL15 0.52 0.48 MGST3 0.52 0.44 MCM7 0.52 0.49 FKBP3 0.52 0.51 SCAMP5 0.52 0.47 KIF21A 0.52 0.56 SLC25A3 0.51 0.43 PRRT4 0.51 0.47 MTCH2 0.51 0.41 ACHE 0.51 0.47 UHMK1 0.51 0.38 CCND3 0.51 0.36 SIK2 0.51 0.53 DUS1L 0.51 0.43 HEIH 0.51 0.41 CLCA2 0.51 0.37 ATP5G3 0.51 0.45 RBP4 0.51 0.66 UNC93B1 0.51 0.53 PPP6C 0.51 0.52 ACVR2B 0.51 0.41 CRYAB 0.51 0.46 COX5A 0.51 0.41 H2AFZ 0.51 0.43 RASSF4 0.51 0.5 USP2 0.51 0.47 BUD13 0.51 0.55 COX7A2 0.51 0.45 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

NDUFC2 0.51 0.44 WHAMM 0.51 0.5 MDFIC 0.51 0.46 AP3B2 0.5 0.43 SERPINB1 0.5 0.52 GPD1L 0.5 0.42 DTYMK 0.5 0.43 CCNH 0.5 0.49 BTBD6 0.5 0.49 NPR1 0.5 0.53 VCP 0.5 0.35 PEX11A 0.5 0.56 POLD2 0.5 0.47 COMMD10 0.5 0.42 PCOLCE2 0.5 0.5 NDUFV1 0.5 0.43 ENO2 0.5 0.55 CDKN2AIPNL 0.5 0.37 OAS1 0.5 0.52 LOC439933 0.5 0.41 ACSL5 0.5 0.52 RTN3 0.5 0.45 MIR22HG 0.5 0.45 NUTM2B-AS1 0.5 0.4 PMVK 0.5 0.49 BCAR3 0.5 0.41 NFU1 0.5 0.37 FERMT2 0.5 0.5 TRHDE-AS1 0.5 0.45 HADHA 0.5 0.4 CHL1 0.5 0.55 SLC2A4 0.5 0.65 FAM101B 0.5 0.44 CAMK2N1 -0.5 -0.49 CYBRD1 -0.5 -0.52 SDAD1 -0.5 -0.53 PAMR1 -0.5 -0.51 RPS19 -0.5 -0.46 SAMHD1 -0.5 -0.49 TGIF1 -0.5 -0.47 ANXA4 -0.5 -0.52 C1orf85 -0.5 -0.48 MAP1LC3B -0.5 -0.43 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

AHCY -0.5 -0.54 FBLN1 -0.5 -0.52 41896 -0.5 -0.47 ABCA9 -0.5 -0.48 ARHGAP29 -0.51 -0.51 THSD1 -0.51 -0.56 ZNF516 -0.51 -0.49 RPL19 -0.51 -0.43 SMPD1 -0.51 -0.55 SLC35F5 -0.51 -0.51 ADAM33 -0.51 -0.56 FABP3 -0.51 -0.51 SLC2A10 -0.51 -0.44 GATA6 -0.51 -0.43 SLC2A12 -0.51 -0.51 CLIC2 -0.52 -0.5 PERP -0.52 -0.46 DCTD -0.52 -0.52 CLIP3 -0.52 -0.49 CPT1A -0.52 -0.56 GDF15 -0.52 -0.51 TNRC6C -0.52 -0.55 TPP1 -0.53 -0.46 WTAP -0.53 -0.57 GPNMB -0.53 -0.54 CHPF -0.53 -0.5 TFPT -0.54 -0.53 PPP1R12C -0.54 -0.5 IL6ST -0.54 -0.53 HEXB -0.54 -0.5 AHNAK2 -0.54 -0.53 SELM -0.54 -0.5 DAP -0.54 -0.51 GRN -0.54 -0.48 SLC15A3 -0.55 -0.57 IDUA -0.55 -0.48 UBE2L6 -0.55 -0.48 SRXN1 -0.55 -0.58 SLC40A1 -0.55 -0.56 CRTAP -0.55 -0.55 ERP44 -0.55 -0.56 IFT20 -0.56 -0.56 TIMP2 -0.56 -0.56 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

USP53 -0.57 -0.57 FTL -0.57 -0.52 UNC50 -0.57 -0.54 DPP4 -0.57 -0.59 JAM3 -0.57 -0.56 SQSTM1 -0.57 -0.59 PPIB -0.57 -0.56 NENF -0.59 -0.56 IL15RA -0.59 -0.57 MAP1A -0.6 -0.62 FAM3C -0.6 -0.6 PRNP -0.6 -0.68 DCLK1 -0.61 -0.66 FTH1 -0.62 -0.58 DRAM1 -0.62 -0.66 PLXNC1 -0.63 -0.65 MOSPD1 -0.63 -0.63 ABR -0.63 -0.62 SCPEP1 -0.64 -0.65 NPDC1 -0.65 -0.69 LAMP1 -0.66 -0.61 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 2. Genes differentially expressed between Cluster 1 and Cluster 4

Enriched in Cluster 1 Enriched in cluster 4

Fold change Fold change Gene p-adjusted value Gene p-adjusted value (linear) (linear) C15orf38-AP3S2 (C15orf38- LGALS12 (galectin 12) 3.83E-12 284.39 0.037151639 43.25 AP3S2 readthrough) OLR1 (oxidized low density AQP7 ( 7) 5.46E-40 184.52 0.00459878 38.62 lipoprotein receptor 1) C14orf180 ( open 1.11E-07 182.39 KIAA1324L (KIAA1324 like) 0.025501441 21.40 reading frame 180) GAPLINC (gastric GPD1 (glycerol-3- adenocarcinoma associated, 6.29E-12 163.07 0.001306205 21.31 dehydrogenase 1) positive CD44 regulator, long intergenic non-coding RNA) CACNB4 ( voltage-gated HP (haptoglobin) 1.75E-07 159.81 0.011331814 21.01 channel auxiliary subunit beta 4) KCTD16 ( AZGP1 (alpha-2- 1, 3.49E-25 156.25 tetramerization domain 0.006547112 20.70 zinc-binding) containing 16) TF () 1.93E-15 154.61 HAS3 (hyaluronan synthase 3) 0.020916873 20.39 ITIH1 (inter-alpha-trypsin inhibitor RASL11B (RAS like family 11 4.65E-06 145.83 0.019710153 19.09 heavy chain 1) member B) PCK1 (phosphoenolpyruvate CEMIP (cell migration inducing 1.52E-15 144.67 3.01E-17 19.07 carboxykinase 1) hyaluronidase 1) FABP4 ( binding protein PRKG2 ( cGMP- 3.14E-18 144.28 0.000717436 18.10 4) dependent 2) PCDHGA12 ( CHIA (chitinase, acidic) 2.04E-04 140.33 0.016822205 17.32 gamma subfamily A, 12) THRSP ( hormone 1.77E-09 133.56 HOXB5 ( B5) 0.03652482 15.76 responsive) LTF (lactotransferrin) 7.67E-04 129.91 ZNF283 ( protein 283) 0.007544988 15.35

PLIN1 (perilipin 1) 3.25E-15 129.09 LOC101928650 0.009173736 15.13 TMC6 (transmembrane channel AARD (alanine and rich 6.31E-09 113.54 0.018792948 14.84 like 6) domain containing protein) SFRP2 (secreted frizzled related PLIN4 (perilipin 4) 1.55E-21 101.05 8.14E-10 13.88 protein 2) TRARG1 (trafficking regulator of RASSF9 (Ras association 3.30E-13 98.70 0.013270877 13.69 GLUT4 (SLC2A4) 1) domain family member 9) HCAR2 (hydroxycarboxylic acid PUS10 (pseudouridine synthase 8.82E-26 97.10 0.023550016 12.01 receptor 2) 10) REPS2 (RALBP1 associated LPL () 8.00E-11 91.56 0.030938901 11.74 Eps domain containing 2) CKMT1B (creatine kinase, 2.53E-02 87.66 DPT (dermatopontin) 0.006009189 11.35 mitochondrial 1B) HRASLS5 (HRAS like suppressor SLC35E2B ( 1.34E-09 86.62 0.000169024 11.31 family member 5) 35 member E2B) KLB (klotho beta) 1.26E-11 83.93 PRR15 ( rich 15) 0.002280082 10.98 PLXDC2 (plexin domain CHI3L2 (chitinase 3 like 2) 8.89E-22 82.36 0.010601626 10.78 containing 2) LOC101926960 (uncharacterized 3.69E-06 79.67 CERKL (ceramide kinase like) 0.026171322 10.50 LOC101926960) ADIPOQ (adiponectin, C1Q and TNFRSF10C (TNF receptor 6.81E-15 74.47 0.037387153 10.21 collagen domain containing) superfamily member 10c) SORBS1 (sorbin and SH3 domain 9.55E-13 73.61 IL32 (interleukin 32) 0.046389922 9.88 containing 1) ALDH1L1 (aldehyde ARNT2 (aryl hydrocarbon dehydrogenase 1 family member 1.54E-07 73.30 2.86E-06 9.44 receptor nuclear translocator 2) L1) KCNB1 (potassium voltage-gated 6.04E-05 71.24 HDAC9 ( deacetylase 9) 0.035103613 9.44 channel subfamily B member 1)

KCNIP2 (potassium voltage-gated CHRNE (cholinergic receptor 3.34E-08 67.01 0.028170118 9.24 channel interacting protein 2) nicotinic epsilon subunit) bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

AOC2 (amine oxidase, 3.53E-22 65.56 ASPN (asporin) 1.62E-05 9.06 containing 2) DTX1 (deltex E3 ITGA11 ( subunit alpha 8.46E-07 65.39 0.000703074 7.75 1) 11) C21orf58 ( open SEMA3G (semaphorin 3G) 9.67E-09 64.58 0.022306312 7.52 reading frame 58) THSD1 (thrombospondin type 1 UCP2 (uncoupling protein 2) 6.27E-28 63.21 0.00459878 7.34 domain containing 1) CIDEC (cell death inducing DFFA MAMDC2 (MAM domain 5.37E-25 63.06 0.007354357 7.28 like effector c) containing 2) LIPE (lipase E, hormone sensitive MMP11 (matrix 5.68E-18 62.54 0.001089517 6.93 type) metallopeptidase 11) LOC101929371 3.55E-06 60.77 LOXL4 (lysyl oxidase like 4) 0.002958287 6.76 MEGF6 (multiple EGF like CA2 (carbonic anhydrase 2) 3.47E-02 60.16 0.016389033 6.54 domains 6) PMS2P5 (PMS1 homolog 2, TSPAN15 ( 15) 1.70E-05 55.41 mismatch repair system 0.020677449 6.13 component pseudogene 5) STAC (SH3 and rich CA3 (carbonic anhydrase 3) 2.59E-02 54.23 0.000297827 5.92 domain) S100B (S100 calcium binding CHST10 (carbohydrate 1.17E-02 50.98 0.0011656 5.77 protein B) sulfotransferase 10) MLXIPL (MLX interacting protein 3.79E-11 48.99 CTSK ( K) 7.17E-06 5.73 like) AOC3 (amine oxidase, copper ABLIM1 ( binding LIM 3.75E-21 48.78 0.044242384 5.67 containing 3) protein 1) BBOX1 (gamma-butyrobetaine IL15RA (interleukin 15 receptor 7.56E-09 47.66 0.010451444 5.49 hydroxylase 1) subunit alpha) FABP5 (fatty acid binding protein 4.33E-38 46.72 TBX1 (T-box 1) 0.035589777 5.32 5) ZDHHC14 (zinc finger DHHC- CHRDL1 (chordin like 1) 1.36E-05 46.71 0.044235358 5.29 type containing 14) ICAM1 (intercellular adhesion APOC1 (apolipoprotein C1) 8.68E-17 44.68 4.29E-06 5.28 molecule 1) FGR (FGR proto-oncogene, Src CHMP1B (charged 4.65E-03 42.80 0.023425364 5.05 family ) multivesicular body protein 1B) ABCD2 (ATP binding cassette 1.08E-06 42.52 PHC1 (polyhomeotic homolog 1) 0.006996637 5.04 subfamily D member 2) KIAA0408 (KIAA0408) 1.25E-04 39.63 MESP1 (mesoderm posterior 7.43E-13 38.13 bHLH 1) PPP1R1A (protein 1 1.97E-03 36.85 regulatory inhibitor subunit 1A) TM7SF2 (transmembrane 7 2.44E-29 36.24 superfamily member 2) TMEM132C (transmembrane 2.77E-03 34.73 protein 132C) ACVR1C (activin A receptor type 3.99E-06 34.11 1C) MARCO ( receptor 1.16E-03 33.75 with collagenous structure) ICA1 (islet cell autoantigen 1) 4.51E-03 32.51 CPB1 (carboxypeptidase B1) 3.24E-08 32.00 ULK4P1 (ULK4 pseudogene 1) 2.86E-03 30.21 PRXL2A (peroxiredoxin like 2A) 1.20E-15 29.71 TDO2 (tryptophan 2,3- 6.34E-03 29.15 ) RBPMS2 (RNA binding protein, 1.89E-02 28.48 mRNA processing factor 2) AP3B2 (adaptor related protein 8.79E-04 27.56 complex 3 subunit beta 2) SOX18 (SRY-box 18) 1.78E-03 27.00 PRKAR2B (protein kinase cAMP- dependent type II regulatory 3.11E-12 26.86 subunit beta) ADIRF (adipogenesis regulatory 1.00E-14 25.16 factor) bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

CDKN2C (cyclin dependent 9.96E-16 24.92 kinase inhibitor 2C) MCAM (melanoma 6.33E-06 24.80 molecule) DGAT2 (diacylglycerol O- 1.20E-63 24.52 acyltransferase 2) G0S2 (G0/G1 switch 2) 1.10E-15 23.43 TMEM37 ( 5.16E-06 22.17 37) SCD (stearoyl-CoA desaturase) 4.33E-38 22.15 RBP4 (retinol binding protein 4) 1.13E-17 21.98 AQP7P1 (aquaporin 7 1.63E-03 21.88 pseudogene 1) LVRN (laeverin) 1.70E-04 21.53 MMP28 (matrix metallopeptidase 1.16E-02 20.74 28) SLC2A4 (solute carrier family 2 2.37E-06 20.67 member 4) RTN4RL1 (reticulon 4 receptor 1.60E-04 20.47 like 1) GRB14 ( receptor 3.77E-02 20.11 bound protein 14) TMEM170B (transmembrane 1.55E-12 18.90 protein 170B) ITIH5 (inter-alpha-trypsin inhibitor 1.09E-07 18.88 heavy chain family member 5) KCNK3 (potassium two pore domain channel subfamily K 4.57E-13 18.13 member 3) ABCG1 (ATP binding cassette 9.62E-05 18.00 subfamily G member 1) PLEKHG6 (pleckstrin and RhoGEF domain containing 8.24E-04 18.00 G6) CEBPA (CCAAT enhancer 1.30E-09 17.53 binding protein alpha) ACHE ( 2.52E-02 17.32 (Cartwright blood group)) CYP4F12 (cytochrome P450 4.01E-02 17.23 family 4 subfamily F member 12) LDHD (lactate dehydrogenase D) 1.01E-06 16.97 FAM47E (family with sequence 4.56E-02 16.54 similarity 47 member E) PLIN5 (perilipin 5) 1.62E-05 16.52 MRAP (melanocortin 2 receptor 3.58E-10 15.11 accessory protein) GYG2 (glycogenin 2) 8.91E-22 14.89 MAOB (monoamine oxidase B) 1.94E-03 14.87 GPAM (glycerol-3-phosphate 4.76E-25 14.23 acyltransferase, mitochondrial) LAIR1 (leukocyte associated 9.73E-04 13.91 immunoglobulin like receptor 1) CKMT2 (creatine kinase, 9.85E-04 13.70 mitochondrial 2) PTGER3 (prostaglandin E 1.31E-05 12.99 receptor 3) CD36 (CD36 molecule) 4.53E-05 12.83 PDE3B ( 3B) 1.14E-02 12.71 LINC01140 (long intergenic non- 1.11E-08 12.49 protein coding RNA 1140) AQP3 ( (Gill blood 1.47E-06 12.39 group)) PLA2G2A ( A2 1.07E-03 11.57 group IIA) bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

SHMT1 (serine 1.08E-23 11.44 hydroxymethyltransferase 1) ACSL5 (acyl-CoA synthetase long 3.81E-13 11.40 chain family member 5) ACSS2 (acyl-CoA synthetase 1.06E-28 10.98 short chain family member 2) ACADL (acyl-CoA dehydrogenase 1.27E-02 10.94 long chain) LBP (lipopolysaccharide binding 1.60E-02 10.70 protein) MYZAP (myocardial zonula 2.05E-02 10.35 adherens protein) LAMB3 (laminin subunit beta 3) 6.53E-13 10.07 CALB2 ( 2) 2.98E-04 9.78 LRRN3 ( rich repeat 4.17E-03 9.69 neuronal 3) ADH4 (alcohol dehydrogenase 4 7.50E-03 9.65 (class II), pi polypeptide) LOC100129518 1.42E-02 9.64 PRL (prolactin) 3.76E-03 9.63 S1PR5 (sphingosine-1-phosphate 2.61E-02 9.62 receptor 5) CLIC6 (chloride intracellular 5.99E-03 9.49 channel 6) ACACB (acetyl-CoA carboxylase 3.69E-06 9.20 beta) PDK4 ( 3.67E-02 9.16 kinase 4) CITED4 (Cbp/p300 interacting transactivator with Glu/Asp rich 1.03E-07 8.85 carboxy-terminal domain 4) SNX10 ( 10) 2.03E-02 8.77 ACSF2 (acyl-CoA synthetase 1.74E-20 8.74 family member 2) OAS1 (2'-5'-oligoadenylate 3.15E-03 8.63 synthetase 1) GHR (growth ) 1.08E-23 8.50 PNPLA2 (patatin like phospholipase domain containing 2.55E-55 8.47 2) FZD5 (frizzled class receptor 5) 4.67E-02 8.34 TENM4 ( transmembrane 9.20E-04 8.34 protein 4) PALMD (palmdelphin) 2.33E-02 8.12 ACSL1 (acyl-CoA synthetase long 7.85E-52 7.81 chain family member 1) PCAT5 (prostate cancer 3.28E-02 7.80 associated transcript 5) KIF14 ( family member 14) 4.45E-03 7.75

CLMN (calmin) 5.81E-06 7.61 SNTN (sentan, cilia apical 2.19E-02 7.57 structure protein) APBB1IP (amyloid beta precursor protein binding family B member 9.36E-11 7.43 1 interacting protein) TMEM56 (transmembrane protein 4.01E-04 7.36 56) IL13RA2 (interleukin 13 receptor 4.18E-05 7.26 subunit alpha 2) FAT3 (FAT atypical 3) 2.32E-03 7.25 AGPAT2 (1-acylglycerol-3- 5.52E-66 7.19 phosphate O-acyltransferase 2) C12orf77 ( open 4.47E-02 7.07 reading frame 77) ZNF117 (zinc finger protein 117) 2.07E-10 7.03 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

CLCA2 ( 2.33E-02 6.81 accessory 2) GRK3 (G protein-coupled 2.65E-08 6.63 receptor kinase 3) APOC1P1 (apolipoprotein C1 2.81E-02 6.56 pseudogene 1) MPP7 (membrane palmitoylated 4.27E-02 6.52 protein 7) LRIG1 (leucine rich repeats and 7.99E-10 6.51 immunoglobulin like domains 1) CSPG4 (chondroitin sulfate 9.49E-03 6.48 proteoglycan 4) CHCHD10 (coiled-coil-helix-coiled- 7.25E-13 6.44 coil-helix domain containing 10)

PDE8B (phosphodiesterase 8B) 2.29E-03 6.42 SNCG (synuclein gamma) 1.41E-02 6.33 DIO2 ( 1.34E-02 6.28 2) PLA2G16 ( 5.50E-06 6.20 group XVI) PPP2R1B ( 2 2.59E-29 6.19 scaffold subunit Abeta) LOC100506368 (uncharacterized 3.85E-02 6.19 LOC100506368) HSPB2-C11orf52 (HSPB2- C11orf52 readthrough (NMD 1.20E-06 5.96 candidate)) ERV3-1 ( 3.10E-04 5.92 group 3 member 1, envelope) SLC29A4 (solute carrier family 29 3.79E-06 5.84 member 4) ACOT4 (acyl-CoA 4) 2.32E-03 5.79

C6 (complement C6) 4.13E-02 5.68 APOE (apolipoprotein E) 2.34E-05 5.64 PCSK9 (proprotein convertase 1.82E-04 5.64 subtilisin/kexin type 9) DLAT (dihydrolipoamide S- 1.88E-27 5.58 acetyltransferase) ITGA7 (integrin subunit alpha 7) 7.48E-10 5.55 CDC42EP4 (CDC42 effector 1.56E-14 5.37 protein 4) FAM107B (family with sequence 3.43E-07 5.36 similarity 107 member B) ACADS (acyl-CoA 1.34E-09 5.31 dehydrogenase short chain) FZD4 (frizzled class receptor 4) 3.53E-17 5.31 DBI (diazepam binding inhibitor, 2.35E-84 5.23 acyl-CoA binding protein) FASN (fatty acid synthase) 6.43E-12 5.23 CYB5A ( type A) 1.84E-35 5.15 F11R (F11 receptor) 1.05E-05 5.10 SORT1 () 2.83E-04 5.10 DUSP4 (dual specificity 4.50E-04 5.07 phosphatase 4) ABCC6 (ATP binding cassette 1.50E-02 5.07 subfamily C member 6) KLHL25 (kelch like family 8.79E-04 5.06 member 25) SLC25A10 (solute carrier family 1.42E-04 5.06 25 member 10) HES6 (hes family bHLH 1.86E-03 5.04 transcription factor 6) bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 3. Pathway enrichement analysis of genes differentially expressed between Cluster 1 and Cluster 4. KEGG Hit in Query List q-value Hit Count in Hit Count in Pathway p-value Bonferroni Query List Genome Name Pathways for ACADL,ACADM,RXRA,FABP4,FABP5,ACSL PPAR genes 1,ADIPOQ,SCD,PLIN5,PLIN1,PLIN4,SORBS signaling 6.86E-17 1.17E-13 21 72 enriched in 1,PPARG,AQP7,ACSL5,DBI,NR1H3,EHHAD pathway Cluster 1 H,CD36,LPL,PCK1 CDO1,HK2,ACACA,ACACB,MAOB,ACADL,A CADM,ACADS,ACAT2,PDHA1,ACLY,ACO2,G LYCTK,MGLL,ACOT4,TST,ADH4,ACSL1,CK MT1B,CKMT2,FASN,SAT1,PEMT,FDPS,CSA D,PLA2G2A,AK1,ACAA2,UQCRFS1,ALDH3A 2,ALDOC,MOCS1,ALPL,AOC2,AGPAT2,AOX 1,SHMT1,GPAM,CS,MTHFD1,EBP,ASPA,AO Metabolic 3.35E-16 5.69E-13 84 1272 C3,AKR1C3,PNPLA3,MVD,MVK,ACSL5,ATP pathways 5MC3,CHIA,AUH,BCKDHA,MCCC1,ACSS2, BDH1,ME3,NSDHL,LPIN1,SUOX,DGAT2,DH CR24,GLUL,DLAT,PCYT2,DLD,GYG2,ACSM 5,ACOT2,CKMT1A,PNPLA2,PANK3,COX17, OGDH,ALG10B,PLA2G16,NAMPT,EHHADH, ALG9,TM7SF2,LSS,HADH,PC,PCCB,PCK1

ACACA,ACADL,ACADM,ACADS,ACAT2,ELO Fatty acid 3.41E-14 5.80E-11 16 48 VL5,ACSL1,FASN,SCD,ACAA2,ELOVL6,AC metabolism SL5,PECR,FADS1,EHHADH,HADH ACADM,ACADS,ACAT2,ACAA2,ALDH3A2,A Valine, leucine OX1,AACS,AUH,BCKDHA,MCCC1,DLD,OXC and isoleucine 6.48E-13 1.10E-09 15 48 T1,EHHADH,HADH,PCCB degradation HK2,ACADM,ACADS,ACAT2,PDHA1,ACO2, Carbon 1.05E-10 1.78E-07 19 114 GLYCTK,ALDOC,SHMT1,CS,ACSS2,ME3,DL metabolism AT,DLD,CAT,OGDH,EHHADH,PC,PCCB Pyruvate ACACA,ACACB,ACAT2,PDHA1,ALDH3A2,A 1.70E-10 2.89E-07 12 39 metabolism CSS2,ME3,LDHD,DLAT,DLD,PC,PCK1 Propanoate ACACA,ACACB,ACADM,ACAT2,ECHDC1,B 4.96E-09 8.43E-06 10 32 metabolism CKDHA,ACSS2,DLD,EHHADH,PCCB ACADL,ACADM,ACADS,ACAT2,ADH4,ACSL Fatty acid 1.15E-08 1.96E-05 11 44 1,ACAA2,ALDH3A2,ACSL5,EHHADH,HADH degradation ACAT2,ACO2,GLYCTK,SHMT1,CS,GLUL,DL Glyoxylate and D,CAT,PCCB dicarboxylate 2.25E-08 3.82E-05 9 28 metabolism Citrate cycle PDHA1,ACLY,ACO2,CS,DLAT,DLD,OGDH,P 4.45E-08 7.56E-05 9 30 (TCA cycle) C,PCK1 ACACA,ACACB,PFKFB3,FASN,ADIPOQ,SC AMPK D,PPARG,SLC2A4,PPP2R1B,PPP2R5A,IGF signaling 9.64E-08 1.64E-04 16 121 1,SREBF1,PPARGC1A,LIPE,CD36,PCK1 pathway Butanoate ACADS,ACAT2,AACS,BDH1,ACSM5,OXCT1, 3.90E-07 6.63E-04 8 28 metabolism EHHADH,HADH Regulation of PDE3B,MGLL,FABP4,PLIN1,AQP7,PTGER3, lipolysis in 1.11E-06 1.88E-03 10 54 PNPLA2,NPR1,PLA2G16,LIPE adipocytes ELOVL5,ACOT4,SCD,ELOVL6,PECR,ACOT Biosynthesis 2,FADS1 of unsaturated 1.33E-06 2.25E-03 7 23 fatty acids Fatty acid ELOVL5,ACOT4,ACAA2,ELOVL6,ELOVL3,A 2.48E-06 4.22E-03 7 25 elongation COT2,HADH Glycerolipid GLYCTK,MGLL,ALDH3A2,AGPAT2,GPAM,P 2.58E-06 4.39E-03 10 59 metabolism NPLA3,LPIN1,DGAT2,PNPLA2,LPL Tryptophan MAOB,ACAT2,ALDH3A2,AOX1,CAT,OGDH,E 7.34E-06 1.25E-02 8 40 metabolism HHADH,HADH Cholesterol EBP,NSDHL,DHCR24,TM7SF2,LSS biosynthesis, squalene 2,3- 4.77E-06 8.11E-03 5 11 epoxide => cholesterol bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Pathways for genes enriched in LAMA1,COL1A1,ITGA11,LAMA2,THBS1,THB ECM-receptor Cluster 4 4.73E-04 3.93E-01 6 82 S3 interaction TGF-beta ACVR2A,DCN,BMP2,ID1,TGIF1,THBS1 signaling 5.38E-04 4.47E-01 6 84 pathway Focal PDGFA,LAMA1,COL1A1,ITGA11,LAMA2,BIR 7.11E-04 5.91E-01 9 199 adhesion C3,CCND2,THBS1,THBS3 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 4. Genes differentially expressed between Cluster 2 and all others.

Higher in Cluster 2 Lower in Cluster 2

Fold Fold change p-adjusted change p-adjusted Gene (linear) value Gene (linear) value

CA3 (carbonic anhydrase 3) 12.24 0.01382 PKIB (cAMP-dependent protein kinase inhibitor beta) -7.12 0.01724 PRL (prolactin) 7.89 0.00103 LOXL4 (lysyl oxidase like 4) -4.18 0.00103 CA9 (carbonic anhydrase 9) 7.14 0.01642 CEMIP (cell migration inducing hyaluronidase 1) -3.66 0.00155 LINC01119 (long intergenic non-protein coding RNA 1119) 2.50 0.00155 MYOZ2 (myozenin 2) -3.59 0.03180 HIST1H1E (histone cluster 1 H1 family member e) 2.08 0.02994 SFRP4 (secreted frizzled related protein 4) -3.43 0.00206 TGFBI (transforming growth factor beta induced) 1.80 0.00199 SLC40A1 (solute carrier family 40 member 1) -2.98 0.00103 PFKFB3 (6-phosphofructo-2-kinase/fructose- 2,6-biphosphatase 3) 1.71 0.04477 CTSK (cathepsin K) -2.44 0.00860 COL4A1 (collagen type IV alpha 1 chain) 1.69 0.01744 PODN (podocan) -2.42 0.02186 TXNIP (thioredoxin interacting protein) 1.68 0.04375 DPP4 (dipeptidyl peptidase 4) -2.40 0.00155 COL4A2 (collagen type IV alpha 2 chain) 1.64 0.00513 FMOD (fibromodulin) -2.31 0.00919 FAM241A (family with sequence similarity 241 member A) 1.63 0.01966 FBLN1 (fibulin 1) -2.20 0.01724 H1FX (H1 histone family member X) 1.51 0.00608 DCLK1 (doublecortin like kinase 1) -2.19 0.02186 SOD3 ( 3) -2.08 0.04195 SVEP1 (sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1) -2.03 0.04195 MAN1C1 (mannosidase alpha class 1C member 1) -2.01 0.04195 MBP ( basic protein) -2.00 0.01783

DRAM1 (DNA damage regulated autophagy modulator 1) -1.90 0.00315 CAMK2N1 (calcium/calmodulin dependent protein kinase II inhibitor 1) -1.90 0.01382 SLIT3 (slit guidance ligand 3) -1.79 0.02902 BHLHE41 (basic helix-loop-helix family member e41) -1.78 0.01389 MDK (midkine) -1.77 0.04726 SOD2 (superoxide dismutase 2) -1.72 0.02940 EML1 (echinoderm associated protein like 1) -1.72 0.03797 USP53 (ubiquitin specific peptidase 53) -1.70 0.04195 IFIT3 ( induced protein with tetratricopeptide repeats 3) -1.66 0.04195 NUMBL (NUMB like, endocytic adaptor protein) -1.66 0.04938 GNPTAB (N-acetylglucosamine-1-phosphate subunits alpha and beta) -1.64 0.02661 SLC20A1 (solute carrier family 20 member 1) -1.59 0.03797 GPNMB (glycoprotein nmb) -1.58 0.01966 SLC20A2 (solute carrier family 20 member 2) -1.57 0.04195 FTH1P3 ( heavy chain 1 pseudogene 3) -1.52 0.00155 ABR (ABR, RhoGEF and GTPase activating protein) -1.51 0.03797 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 5. Pathway enrichment analysis of genes differentially expressed between Cluster 2 and all others.

Hit Hit Count Count q-value KEGG Pathway Name p-value in Query in Hit in Query List Bonferroni List Genom e Pathways for Glycolysis (Embden- GPI,GAPDH,PKM,PFKL,PFKP,ENO1,E genes enriched Meyerhof pathway), 6.24E-16 2.55E-13 9 25 NO2,PGK1,TPI1 in Cluster 2 glucose => pyruvate Glycolysis / GPI,GAPDH,PKM,PFKL,PFKP,LDHA,E 2.45E-13 1.00E-10 10 67 NO1,ENO2,PGK1,TPI1 Glycolysis, core module involving three-carbon 5.08E-12 2.08E-09 6 12 GAPDH,PKM,ENO1,ENO2,PGK1,TPI1 compounds Biosynthesis of amino GAPDH,PKM,PFKL,PFKP,ENO1,ENO2, 1.15E-09 4.69E-07 8 75 acids PGK1,TPI1 GPI,GAPDH,PKM,PFKL,PFKP,ENO1,E Carbon metabolism 1.47E-09 6.01E-07 9 114 NO2,PGK1,TPI1 Gluconeogenesis, oxaloacetate => fructose- 8.30E-09 3.40E-06 5 17 GAPDH,ENO1,ENO2,PGK1,TPI1 6P GAPDH,PFKFB3,PFKL,LDHA,ENO1,EN HIF-1 signaling pathway 2.77E-07 1.13E-04 7 101 O2,PGK1 COL4A1,COL4A2,COL6A1,COL6A2,LA ECM-receptor interaction 1.52E-06 6.22E-04 6 82 MA4,ITGA5 Central carbon metabolism 9.44E-06 3.86E-03 5 65 ,PKM,PFKL,PFKP,LDHA in cancer Fructose and mannose 1.27E-05 5.19E-03 4 33 PFKFB3,PFKL,PFKP,TPI1 metabolism Protein digestion and COL3A1,COL4A1,COL4A2,COL6A1,CO 4.62E-05 1.89E-02 5 90 absorption L6A2 MYC,COL4A1,COL4A2,COL6A1,COL6 PI3K-Akt signaling pathway 1.18E-04 4.82E-02 8 342 A2,LAMA4,ITGA5,PRL COL4A1,COL4A2,COL6A1,COL6A2,LA Focal adhesion 2.34E-04 9.56E-02 6 199 MA4,ITGA5 Pentose phosphate 3.04E-04 1.25E-01 3 30 GPI,PFKL,PFKP pathway RNA degradation 3.63E-04 1.49E-01 4 77 PFKL,PFKP,ENO1,ENO2 Small cell cancer 5.06E-04 2.07E-01 4 84 MYC,COL4A1,COL4A2,LAMA4 Amoebiasis 8.38E-04 3.43E-01 4 96 COL3A1,COL4A1,COL4A2,LAMA4 Nitrogen metabolism 2.50E-03 1.00E+00 2 17 CA9,CA3 GPI,P4HA2,MTMR6,GAPDH,PKM,NMR Metabolic pathways 3.02E-03 1.00E+00 13 1272 K1,PFKL,PFKP,LDHA,ENO1,ENO2,PG K1,TPI1 Galactose metabolism 8.21E-03 1.00E+00 2 31 PFKL,PFKP AGE-RAGE signaling pathway in diabetic 9.49E-03 1.00E+00 3 99 COL3A1,COL4A1,COL4A2 complications Pathways for GCLM,FTH1,FTL,MAP1LC3B,SLC40A1 Ferroptosis 5.86E-08 2.77E-05 6 40 genes decreased ,PRNP in Cluster 2 Mineral absorption 1.49E-04 7.05E-02 4 51 CYBRD1,FTH1,FTL,SLC40A1 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 6. Genes differentially expressed between progenitors for Cluster 1 and all others.

Enriched in progenitors giving rise to Cluster 1 Decreased in progenitors giving rise to Cluster 1

Fold Change Fold Change Gene p adjusted value Gene p adjusted value (Linear) (Linear) EVI2A (ecotropic viral 3.81 0.033427884 NEFM ( medium) -174.53 0.022725442 integration site 2A) CCDC69 (coiled-coil domain 2.53 0.035515468 PLSCR3 ( 3) -43.61 0.00014831 containing 69) SLC22A3 (solute carrier family 22 -21.10 0.016915217 member 3) CXCL1 (C-X-C motif chemokine -20.64 0.030480859 ligand 1) CXCL6 (C-X-C motif chemokine -20.30 0.020206449 ligand 6) NBL1 (NBL1, DAN family BMP -16.66 0.042373531 antagonist) MMP1 (matrix metallopeptidase 1) -15.30 0.017375974 CXCL5 (C-X-C motif chemokine -14.62 0.032115504 ligand 5) STC1 (stanniocalcin 1) -8.13 0.030480859 INHBA (inhibin subunit beta A) -8.09 0.016915217 ID1 (inhibitor of DNA binding 1, HLH -7.12 0.047758672 protein) DACT1 (dishevelled binding -6.94 0.000397007 antagonist of beta 1) ATOH8 (atonal bHLH transcription -6.30 0.035515468 factor 8) JAM2 (junctional adhesion molecule -6.08 0.038224756 2) SCHIP1 (schwannomin interacting -5.95 0.00014831 protein 1) ENC1 (ectodermal-neural cortex 1) -4.92 0.018319432 ANXA2P3 ( pseudogene -4.32 0.016377587 3) UBASH3B (ubiquitin associated and -4.05 0.024168171 SH3 domain containing B) PPIAL4G (peptidylprolyl A -3.30 0.018319432 like 4G) C1QTNF1 (C1q and TNF related 1) -3.11 0.029872169 SLC2A1 (solute carrier family 2 -3.11 0.016377587 member 1) SLC16A3 (solute carrier family 16 -2.87 0.016377587 member 3) MIR210HG (MIR210 host gene) -2.79 0.007867149 RORA (RAR related orphan receptor -2.77 0.03479213 A) TPM1 ( 1) -2.70 0.039134788 SNHG9 (small nucleolar RNA host -2.46 0.03479213 gene 9) TNS1 (tensin 1) -2.38 0.031506188 CCDC124 (coiled-coil domain -2.36 0.042373531 containing 124) FTH1P3 (ferritin heavy chain 1 -2.30 0.033486735 pseudogene 3) CDKN2A (cyclin dependent kinase -2.24 0.045705177 inhibitor 2A) CLEC11A (C-type lectin domain -2.19 0.033317835 containing 11A) MICAL2 (microtubule associated , calponin and LIM -2.18 0.018319432 domain containing 2) TNFRSF12A (TNF receptor -2.15 0.018319432 superfamily member 12A) FGF2 (fibroblast growth factor 2) -2.00 0.049391853 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 7. Pathway enrichment analysis of genes differentially expressed in progenitors for Cluster 1.

q-value Hit Count in Hit Count Category KEGG Pathway Name p-value Hit in Query List Bonferroni Query List in Genome

TRAF4,MMP1,MMP3,CXCL6,CXCL5,CXCL IL-17 signaling pathway 1.40E-11 6.77E-09 9 93 1,CXCL3,IL6,CXCL8 Cytokine-cytokine receptor INHBA,CXCL6,CXCL5,CXCL1,CXCL3,IL6,C 2.50E-05 1.20E-02 7 270 interaction XCL8 Pathways TNF signaling pathway 2.62E-05 1.26E-02 5 108 MMP3,CXCL5,CXCL1,CXCL3,IL6 enriched in Chemokine signaling 3.09E-04 1.49E-01 5 182 CXCL6,CXCL5,CXCL1,CXCL3,CXCL8 genes pathway decreased in Signaling pathways Cluster 1 regulating pluripotency of 1.09E-03 5.27E-01 4 139 INHBA,WNT5A,FGF2,MEIS1 progenitors stem cells EGFR tyrosine kinase 2.20E-03 1.00E+00 3 79 GAS6,FGF2,IL6 inhibitor resistance NOD-like receptor signaling 2.29E-03 1.00E+00 4 170 CXCL1,CXCL3,IL6,CXCL8 pathway bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 8. Genes differentially expressed between progenitors for Cluster 2 and all others.

Enriched in progenitors giving rise to Cluster 2 Decreased in progenitors giving rise to Cluster 2

Fold Change Fold Change Gene p adjusted Gene p adjusted (Linear) (Linear)

CSF3 (colony stimulating factor 3) 23.48 0.01673 ACTC1 (actin, alpha, 1) -11.39 0.03611 PPP1R14C ( IL11 (interleukin 11) 12.32 0.02760 regulatory inhibitor subunit 14C) -10.80 0.02929 CXCL8 (C-X-C motif chemokine 11.25 0.00010 SPIN2A (spindlin family member 2A) ligand 8) -5.89 0.02614 OAS1 (2'-5'-oligoadenylate 10.57 0.00448 DLX2 (distal-less homeobox 2) synthetase 1) -5.54 0.02760 NR4A2 (nuclear receptor 8.15 0.00011 CLCA2 (chloride channel accessory 2) subfamily 4 group A member 2) -5.26 0.00878 VAMP8 (vesicle associated membrane IL1B (interleukin 1 beta) 8.08 0.02028 protein 8) -5.01 0.02760 IFI6 (interferon alpha inducible ID1 (inhibitor of DNA binding 1, HLH 5.27 0.00000 protein 6) protein) -4.39 0.00878 LIF (LIF, family 4.09 0.02760 RELN (reelin) cytokine) -3.91 0.01012 PTGS2 (prostaglandin- CLEC3B (C-type lectin domain family 3 3.99 0.00290 endoperoxide synthase 2) member B) -3.68 0.03078

CASC15 (cancer susceptibility 15) 3.29 0.01337 THBD (thrombomodulin) -3.59 0.01012 IFIT1 (interferon induced protein MRAP2 (melanocortin 2 receptor 2.62 0.00912 with tetratricopeptide repeats 1) accessory protein 2) -3.13 0.04151 IRF7 (interferon regulatory factor HSPB2-C11orf52 (HSPB2-C11orf52 2.42 0.04085 7) readthrough (NMD candidate)) -3.04 0.01817 TMEM132A (transmembrane 2.42 0.03984 NFASC (neurofascin) protein 132A) -2.93 0.01337 PID1 (phosphotyrosine interaction HSPB7 (heat shock B 2.30 0.04306 domain containing 1) (small) member 7) -2.57 0.04135 THEMIS2 (thymocyte selection GREM2 (gremlin 2, DAN family BMP 2.29 0.01660 associated family member 2) antagonist) -2.52 0.00010 IFI44 (interferon induced protein 2.24 0.01944 TUBB3 ( beta 3 class III) 44) -2.47 0.00004 ZC3H12A (zinc finger CCCH-type 2.12 0.02614 MEF2C (myocyte enhancer factor 2C) containing 12A) -2.32 0.03682 TNFAIP6 (TNF alpha induced 2.03 0.01012 CTSK (cathepsin K) protein 6) -2.22 0.04135

NPAS1 (neuronal PAS domain protein 1) -2.08 0.04135 TP53I11 (tumor protein inducible protein 11) -2.04 0.00878 FMN2 (formin 2) -2.02 0.04135 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 9. Pathway and Gene family enrichment analysis of genes differentially expressed in progenitors for Cluster 2.

q-value Hit Count in Hit Count in Name p-value Hit in Query List Bonferroni Query List Genome IL-17 signaling pathway 1.38E-06 2.68E-04 4 93 CSF3,IL1B,PTGS2,CXCL8 Cytokine-cytokine receptor 3.23E-06 6.27E-04 5 270 CSF3,IL1B,IL11,LIF,CXCL8 KEGG Pathways interaction enriched in NOD-like receptor signaling 1.53E-05 2.96E-03 4 170 IL1B,OAS1,IRF7,CXCL8 genes increased pathway in Cluster 2 progenitors NF-kappa B signaling pathway 9.01E-05 1.75E-02 3 95 IL1B,PTGS2,CXCL8

Chloride channel accessory 3.51E-03 6.68E-02 1 4 CLCA2 Myocyte enhancer factor 2 3.51E-03 6.68E-02 1 4 MEF2C proteins|MADS box family Basic helix-loop-helix proteins 4.11E-03 7.81E-02 2 110 ID1,NPAS1 Tudor domain 4.39E-03 8.34E-02 1 5 SPIN2A Gene families containing|Spindlin family (genenames.org) 5.27E-03 1.00E-01 1 6 ACTC1 enriched in genes decreased DAN family 6.14E-03 1.17E-01 1 7 GREM2 in Cluster 2 progenitors Vesicle associated membrane 6.14E-03 1.17E-01 1 7 VAMP8 proteins Small heat shock proteins 9.63E-03 1.83E-01 1 11 HSPB7

Cathepsins 1.31E-02 2.49E-01 1 15 CTSK

Tubulins 2.26E-02 4.30E-01 1 26 TUBB3 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 10. Genes in progenitors correlated with LEP levels in M condition

Positively Spearman Negatively Spearman correlated correlation correlated correlation Gene (Spearman coefficient Gene (Spearman coefficient correlation (LEP in M - correlation (LEP in M - coefficient) Random) coefficient) Random) ZNF395 (zinc finger protein 395) 0.62 0.123 PAPPA (pappalysin 1) -0.61 -0.14 PAPPA-AS1 (PAPPA antisense ENO2 (enolase 2) 0.55 0.119 RNA 1) -0.54 -0.117 BSCL2 (BSCL2, seipin lipid ENO1 (enolase 1) 0.53 0.116 droplet biogenesis associated) -0.55 -0.112 CHSY3 (chondroitin sulfate PTX3 (pentraxin 3) 0.56 0.113 synthase 3) -0.56 -0.105 NDRG1 (N-myc downstream regulated DHCR7 (7-dehydrocholesterol 1) 0.53 0.107 reductase) -0.51 -0.101 TMEM147 (transmembrane GLTP (glycolipid transfer protein) 0.57 0.104 protein 147) -0.48 -0.097 TECR (trans-2,3-enoyl-CoA KRT34 ( 34) 0.51 0.101 reductase) -0.48 -0.094 LSG1 (large 60S subunit nuclear GPI (glucose-6-phosphate isomerase) 0.5 0.094 export GTPase 1) -0.51 -0.093 HNRNPA1 (heterogeneous nuclear ATP6V0E1 (ATPase H+ ribonucleoprotein A1) 0.49 0.094 transporting V0 subunit e1) -0.47 -0.093 SDE2 (SDE2 maintenance ATP1B3 (ATPase Na+/K+ homolog) 0.48 0.091 transporting subunit beta 3) -0.49 -0.092 ATP6V1F (ATPase H+ transporting KRT19 () 0.48 0.09 V1 subunit F) -0.47 -0.089 LGALS8 (galectin 8) 0.49 0.089 ITGAE (integrin subunit alpha E) -0.46 -0.088 TRA2B (transformer 2 beta homolog) 0.47 0.087 SC5D (sterol-C5-desaturase) -0.46 -0.088 TUBA1B (tubulin alpha 1b) 0.47 0.086 FBLN5 (fibulin 5) -0.49 -0.087 COX14 (cytochrome c oxidase DARS (aspartyl-tRNA synthetase) 0.47 0.083 assembly factor COX14) -0.48 -0.086 CDCP1 (CUB domain containing protein DSTN (, actin 1) 0.46 0.081 depolymerizing factor) -0.45 -0.083 SNORD50A (small nucleolar RNA, C/D TMEM59 (transmembrane protein box 50A) 0.45 0.081 59) -0.45 -0.08 PLPP1 (phospholipid phosphatase BNIP3 (BCL2 interacting protein 3) 0.45 0.079 1) -0.4 -0.079 NFATC2IP (nuclear factor of activated T GDE1 (glycerophosphodiester cells 2 interacting protein) 0.45 0.077 phosphodiesterase 1) -0.43 -0.078 DIPK2A (divergent protein kinase G6PD (glucose-6-phosphate domain 2A) 0.44 0.076 dehydrogenase) -0.44 -0.078 CCDC58 (coiled-coil domain containing NUDT9 (nudix 9) 58) 0.45 0.074 -0.38 -0.077 RNF126 (ring finger protein 126) 0.44 0.074 MIR6723 (microRNA 6723) -0.38 -0.077 CLIC1 (chloride intracellular channel 1) 0.42 0.074 TP53TG1 (TP53 target 1) -0.43 -0.077 HLA-C (major histocompatibility RGS2 (regulator of G protein complex, class I, C) 0.44 0.073 signaling 2) -0.43 -0.077 BRD9 (bromodomain containing 9) 0.43 0.073 SERINC3 (serine incorporator 3) -0.44 -0.077 LRRFIP1 (LRR binding FLII interacting ISCU (iron-sulfur cluster assembly protein 1) 0.42 0.073 enzyme) -0.38 -0.076 ADAT1 (adenosine deaminase, TXNIP (thioredoxin interacting protein) 0.43 0.072 tRNA specific 1) -0.39 -0.076 ICMT (isoprenylcysteine carboxyl TUBB6 (tubulin beta 6 class V) 0.43 0.071 ) -0.39 -0.076 HSD17B12 (hydroxysteroid 17- LMNB2 ( B2) 0.41 0.071 beta dehydrogenase 12) -0.4 -0.076 HNRNPK (heterogeneous nuclear GAMT (guanidinoacetate N- ribonucleoprotein K) 0.42 0.07 methyltransferase) -0.38 -0.075 FAM118A (family with sequence SERPINB8 (serpin family B member 8) 0.43 0.069 similarity 118 member A) -0.39 -0.075 UAP1 (UDP-N-acetylglucosamine UNC50 (unc-50 inner nuclear pyrophosphorylase 1) 0.42 0.069 membrane RNA binding protein) -0.4 -0.075 SLC2A3 (solute carrier family 2 member GULP1 (GULP PTB domain 3) 0.39 0.069 containing engulfment adaptor 1) -0.4 -0.075 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

ARF6 (ADP ribosylation factor 6) 0.39 0.069 TCTN3 (tectonic family member 3) -0.42 -0.075 CFB (complement factor B) 0.42 0.068 CETN2 (centrin 2) -0.38 -0.074 PPP6C (protein phosphatase 6 QSOX1 (quiescin sulfhydryl oxidase 1) 0.41 0.068 catalytic subunit) -0.38 -0.074 CRAMP1 (cramped regulator HIST1H4K (histone cluster 1 H4 homolog 1) 0.41 0.068 family member k) -0.38 -0.074 ADGRL2 (adhesion G protein-coupled ZNHIT1 (zinc finger HIT-type receptor L2) 0.4 0.067 containing 1) -0.43 -0.074 TMEM98 (transmembrane protein NGF (nerve growth factor) 0.39 0.067 98) -0.44 -0.074 FIBP (FGF1 intracellular binding MSH6 (mutS homolog 6) 0.39 0.067 protein) -0.39 -0.073 ALG3 (ALG3, alpha-1,3- P4HA2-AS1 (P4HA2 antisense RNA 1) 0.35 0.067 mannosyltransferase) -0.39 -0.073 ANTXR1 (ANTXR cell adhesion MYLK ( light chain kinase) 0.42 0.066 molecule 1) -0.39 -0.073 PSMD13 ( 26S subunit, non- MDH1 (malate dehydrogenase 1) ATPase 13) 0.4 0.066 -0.39 -0.073 SKP1 (S-phase kinase associated PPAN (peter pan homolog) 0.39 0.066 protein 1) -0.4 -0.073 CCN4 (cellular communication network BRAT1 (BRCA1 associated ATM factor 4) 0.38 0.066 activator 1) -0.39 -0.072 PSMD6 (proteasome 26S subunit, MFAP3 (microfibril associated protein 3) 0.37 0.066 non-ATPase 6) -0.4 -0.072 MAPKAPK3 (mitogen-activated protein GSTZ1 (glutathione S-transferase kinase-activated protein kinase 3) 0.35 0.066 zeta 1) -0.4 -0.072 HSPA5 (heat shock protein family A FUNDC1 (FUN14 domain (Hsp70) member 5) 0.35 0.066 containing 1) -0.42 -0.072 EGLN1 (egl-9 family hypoxia inducible TWSG1 (twisted BMP factor 1) 0.35 0.066 signaling modulator 1) -0.43 -0.072 PPP1R18 (protein phosphatase 1 ZNF226 (zinc finger protein 226) regulatory subunit 18) 0.41 0.065 -0.41 -0.072 BCKDHB (branched chain keto RPL13 (ribosomal protein L13) acid dehydrogenase E1 subunit 0.41 0.065 beta) -0.36 -0.071 DLGAP4 (DLG associated protein 4) 0.4 0.065 DOLK (dolichol kinase) -0.38 -0.071 ZCCHC7 (zinc finger CCHC-type PTMA (prothymosin alpha) 0.38 0.065 containing 7) -0.38 -0.071 KCTD9 (potassium channel CYP51A1 (cytochrome P450 tetramerization domain containing 9) 0.37 0.065 family 51 subfamily A member 1) -0.38 -0.071 NIPA2 (NIPA magnesium RAI14 (retinoic acid induced 14) 0.35 0.065 transporter 2) -0.4 -0.071 RAD23B (RAD23 homolog B, TXNDC5 (thioredoxin domain nucleotide excision repair protein) 0.4 0.064 containing 5) -0.4 -0.071 PPA2 (pyrophosphatase AK4 (adenylate kinase 4) 0.4 0.064 (inorganic) 2) -0.4 -0.071 SDF2 (stromal cell derived factor AES (amino-terminal enhancer of split) 0.39 0.064 2) -0.42 -0.071 FUBP1 (far upstream element binding YIPF2 (Yip1 domain family protein 1) 0.39 0.064 member 2) -0.36 -0.07 MME (membrane SLC17A5 (solute carrier family 17 metalloendopeptidase) 0.39 0.064 member 5) -0.36 -0.07 ZDHHC16 (zinc finger DHHC-type BTG3 (BTG anti-proliferation factor 3) 0.38 0.064 containing 16) -0.36 -0.07 NCOA4 (nuclear receptor RBM3 (RNA binding motif protein 3) 0.36 0.064 coactivator 4) -0.36 -0.07 MEAK7 (MTOR associated protein, eak- MBD4 (methyl-CpG binding 7 homolog) 0.35 0.064 domain 4, DNA glycosylase) -0.37 -0.07 ADGRE2 (adhesion G protein-coupled MICA (MHC class I polypeptide- receptor E2) 0.35 0.064 related sequence A) -0.37 -0.07 CLCN3 (chloride voltage-gated HMGA1 (high mobility group AT-hook 1) 0.4 0.063 channel 3) -0.37 -0.07 HLA-A (major histocompatibility MARCKSL1 (MARCKS like 1) complex, class I, A) 0.37 0.063 -0.38 -0.07 SLC16A3 (solute carrier family 16 CBR4 (carbonyl reductase 4) member 3) 0.37 0.063 -0.38 -0.07 PFKP (phosphofructokinase, platelet) 0.36 0.063 TFDP2 (transcription factor Dp-2) -0.39 -0.07 SLC38A6 (solute carrier family 38 KRT18 () 0.36 0.063 member 6) -0.39 -0.07 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

CDCA4 (cell division cycle associated FOCAD (focadhesin) 4) 0.34 0.063 -0.39 -0.07 PEBP1 (phosphatidylethanolamine TYMP (thymidine phosphorylase) 0.34 0.063 binding protein 1) -0.41 -0.07 HDHD5 (haloacid dehalogenase like AKR1C1 (aldo-keto reductase hydrolase domain containing 5) 0.34 0.063 family 1 member C1) -0.41 -0.07 DCUN1D3 (defective in cullin CNPY2 (canopy FGF signaling neddylation 1 domain containing 3) 0.34 0.063 regulator 2) -0.41 -0.07 L3HYPDH (trans-L-3- CTSH (cathepsin H) 0.34 0.063 dehydratase) -0.36 -0.069 CCAR2 (cell cycle and RABEPK (Rab9 effector protein regulator 2) 0.34 0.063 with kelch motifs) -0.36 -0.069 NCKIPSD (NCK interacting protein with MTRNR2L1 (MT-RNR2 like 1) SH3 domain) 0.4 0.062 -0.36 -0.069 MTRR (5-methyltetrahydrofolate- TM9SF2 (transmembrane 9 methyltransferase superfamily member 2) reductase) 0.38 0.062 -0.37 -0.069 DYNC1H1 ( cytoplasmic 1 heavy HAGHL (hydroxyacylglutathione chain 1) 0.37 0.062 hydrolase like) -0.39 -0.069 GNPDA2 (glucosamine-6- MIR210HG (MIR210 host gene) 0.37 0.062 phosphate deaminase 2) -0.39 -0.069 PIK3IP1 (phosphoinositide-3- TUBA1C (tubulin alpha 1c) 0.37 0.062 kinase interacting protein 1) -0.39 -0.069 ERG28 (ergosterol biosynthesis SIRPA (signal regulatory protein alpha) 0.37 0.062 28 homolog) -0.4 -0.069 FBXL3 (F-box and leucine rich repeat SELENBP1 (selenium binding protein 3) 0.36 0.062 protein 1) -0.42 -0.069 SLC25A24 (solute carrier family 25 HSPH1 (heat shock protein family member 24) 0.36 0.062 H (Hsp110) member 1) -0.42 -0.069 VAMP4 (vesicle associated LYSMD3 (LysM domain containing 3) 0.36 0.062 4) -0.41 -0.069 TMEM176B (transmembrane protein ZFAND5 (zinc finger AN1-type 176B) 0.35 0.062 containing 5) -0.35 -0.068 NAGPA (N-acetylglucosamine-1- DDX41 (DEAD-box 41) phosphodiester alpha-N- 0.35 0.062 acetylglucosaminidase) -0.35 -0.068 MEMO1 (Methylation modifier for class I MGMT (O-6-methylguanine-DNA HLA) 0.35 0.062 methyltransferase) -0.35 -0.068 PITPNB (phosphatidylinositol transfer ZMAT3 (zinc finger matrin-type 3) protein beta) 0.35 0.062 -0.35 -0.068 GBE1 (1,4-alpha-glucan branching CCDC92 (coiled-coil domain enzyme 1) 0.34 0.062 containing 92) -0.36 -0.068 RPLP0P2 (ribosomal protein lateral TCEAL9 (transcription elongation stalk subunit P0 pseudogene 2) 0.39 0.061 factor A like 9) -0.36 -0.068 TPCN1 (two pore segment MOK (MOK protein kinase) 0.39 0.061 channel 1) -0.36 -0.068 VPS4B ( 4 WIPI1 (WD repeat domain, homolog B) 0.39 0.061 phosphoinositide interacting 1) -0.36 -0.068 ARL8A (ADP ribosylation factor like TMEM106B (transmembrane GTPase 8A) 0.38 0.061 protein 106B) -0.37 -0.068 TMEM38B (transmembrane AJUBA (ajuba LIM protein) 0.38 0.061 protein 38B) -0.37 -0.068 CHD1 (chromodomain helicase DNA SCAMP3 (secretory carrier binding protein 1) 0.38 0.061 membrane protein 3) -0.37 -0.068 HIGD2A (HIG1 hypoxia inducible DBNL (drebrin like) 0.37 0.061 domain family member 2A) -0.37 -0.068 TUBB (tubulin beta class I) 0.35 0.061 MTRNR2L8 (MT-RNR2 like 8) -0.37 -0.068 TIMMDC1 ( of inner NGLY1 (N-glycanase 1) mitochondrial membrane domain 0.34 0.061 containing 1) -0.38 -0.068

SECISBP2 (SECIS binding protein 2) 0.34 0.061 KIF3B (kinesin family member 3B) -0.38 -0.068 VPS33A (VPS33A, PCGF6 (polycomb group ring finger 6) 0.34 0.061 CORVET/HOPS core subunit) -0.38 -0.068 WSB1 (WD repeat and SOCS box ANAPC13 (anaphase promoting containing 1) 0.34 0.061 complex subunit 13) -0.38 -0.068 ARFRP1 (ADP ribosylation factor NAE1 (NEDD8 activating enzyme related protein 1) 0.33 0.061 E1 subunit 1) -0.4 -0.068 COMMD3-BMI1 (COMMD3-BMI1 LOXL4 (lysyl oxidase like 4) 0.33 0.061 readthrough) -0.41 -0.068 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

LMAN2 (lectin, mannose binding DDX50 (DExD-box helicase 50) 0.39 0.06 2) -0.35 -0.067 ADI1 (acireductone dioxygenase SERPINE1 (serpin family E member 1) 0.38 0.06 1) -0.35 -0.067 SERBP1 (SERPINE1 mRNA binding TSPAN6 (tetraspanin 6) protein 1) 0.37 0.06 -0.35 -0.067 VEGFA (vascular endothelial growth SYNGR1 (synaptogyrin 1) factor A) 0.37 0.06 -0.35 -0.067 BBIP1 (BBSome interacting SH2B3 (SH2B adaptor protein 3) 0.36 0.06 protein 1) -0.35 -0.067 CCDC107 (coiled-coil domain CD59 (CD59 molecule (CD59 containing 107) 0.36 0.06 blood group)) -0.36 -0.067 PIGP (phosphatidylinositol glycan SERPINB2 (serpin family B member 2) 0.36 0.06 anchor biosynthesis class P) -0.4 -0.067 FLNB ( B) 0.35 0.06 UBB (ubiquitin B) -0.4 -0.067 NPDC1 (neural proliferation, LOXL1 (lysyl oxidase like 1) 0.35 0.06 differentiation and control 1) -0.35 -0.066 MSANTD4 (Myb/SANT DNA PGK1 (phosphoglycerate kinase 1) binding domain containing 4 with 0.35 0.06 coiled-coils) -0.35 -0.066 SLC8B1 (solute carrier family 8 member MRPL32 (mitochondrial ribosomal B1) 0.34 0.06 protein L32) -0.36 -0.066 MAGT1 ( TUSC1 (tumor suppressor candidate 1) 0.34 0.06 1) -0.36 -0.066 ZEB2 (zinc finger E-box binding TKT (transketolase) homeobox 2) 0.33 0.06 -0.36 -0.066 NRAS (NRAS proto-oncogene, LHFPL6 (LHFPL tetraspan GTPase) 0.33 0.06 subfamily member 6) -0.35 -0.065 TMEM70 (transmembrane protein PFN1 ( 1) 0.33 0.06 70) -0.36 -0.065 RPL36A (ribosomal protein L36a) 0.33 0.06 SYNPO2 (synaptopodin 2) -0.41 -0.065 MTFP1 (mitochondrial fission process ALAD (aminolevulinate 1) 0.33 0.06 dehydratase) -0.35 -0.064 MROH1 (maestro heat like repeat CORO1C (coronin 1C) 0.37 0.059 family member 1) -0.36 -0.064 GLT8D2 (glycosyltransferase 8 PLP2 (proteolipid protein 2) 0.37 0.059 domain containing 2) -0.4 -0.064 CCDC112 (coiled-coil domain TMEM179B (transmembrane containing 112) 0.36 0.059 protein 179B) -0.4 -0.064 ATP6V1E1 (ATPase H+ MT1E (metallothionein 1E) 0.36 0.059 transporting V1 subunit E1) -0.34 -0.063 RPS6KB1 (ribosomal protein S6 GTPBP2 (GTP binding protein 2) 0.36 0.059 kinase B1) -0.34 -0.063 HNRNPD (heterogeneous nuclear MTRNR2L2 (MT-RNR2 like 2) ribonucleoprotein D) 0.36 0.059 -0.35 -0.063 GPR137B (G protein-coupled RFC4 ( subunit 4) 0.36 0.059 receptor 137B) -0.35 -0.063 LINC01119 (long intergenic non-protein PDGFRL (platelet derived growth coding RNA 1119) 0.36 0.059 factor receptor like) -0.35 -0.063 GPT2 (glutamic--pyruvic transaminase NPC2 (NPC intracellular 2) 0.35 0.059 cholesterol transporter 2) -0.36 -0.063 MCFD2 (multiple coagulation WDR45B (WD repeat domain 45B) 0.35 0.059 factor deficiency 2) -0.36 -0.063 TMEM258 (transmembrane ANXA6 (annexin A6) 0.35 0.059 protein 258) -0.34 -0.062 ALDH2 (aldehyde dehydrogenase NRG1 (neuregulin 1) 0.35 0.059 2 family member) -0.35 -0.062 ALKBH5 (alkB homolog 5, RNA DBI (diazepam binding inhibitor, ) 0.35 0.059 acyl-CoA binding protein) -0.35 -0.062 CITED2 (Cbp/p300 interacting PTOV1 (prostate tumor transactivator with Glu/Asp rich carboxy- overexpressed 1) terminal domain 2) 0.35 0.059 -0.34 -0.061 FAHD1 (fumarylacetoacetate hydrolase SHQ1 (SHQ1, H/ACA domain containing 1) 0.34 0.059 ribonucleoprotein assembly factor) -0.34 -0.061 CSE1L (chromosome segregation 1 FAM98A (family with sequence like) 0.34 0.059 similarity 98 member A) -0.34 -0.061 MXI1 (MAX interactor 1, dimerization ATP6V1A (ATPase H+ protein) 0.34 0.059 transporting V1 subunit A) -0.34 -0.061 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

NOC2L (NOC2 like nucleolar associated ZUP1 (zinc finger containing transcriptional ) 0.34 0.059 ubiquitin peptidase 1) -0.34 -0.061 ATP6V0B (ATPase H+ RRAS2 (RAS related 2) 0.34 0.059 transporting V0 subunit b) -0.34 -0.061 FANK1 ( type III and NOL7 (nucleolar protein 7) 0.34 0.059 repeat domains 1) -0.33 -0.06 SMURF2 (SMAD specific E3 ubiquitin PTGR1 (prostaglandin reductase protein ligase 2) 0.34 0.059 1) -0.36 -0.06 PIM3 (Pim-3 proto-oncogene, DYNC2LI1 (dynein cytoplasmic 2 serine/threonine kinase) 0.33 0.059 light intermediate chain 1) -0.36 -0.06 EDEM2 (ER degradation PDLIM1 (PDZ and LIM domain 1) enhancing alpha-mannosidase like 0.32 0.059 protein 2) -0.36 -0.06 AGBL5 (ATP/GTP binding protein KLF10 (Kruppel like factor 10) 0.34 0.058 like 5) -0.32 -0.059 IMP3 (IMP3, U3 small nucleolar PRXL2B (peroxiredoxin like 2B) ribonucleoprotein) 0.37 0.058 -0.32 -0.059 C4orf3 (chromosome 4 open reading SETMAR (SET domain and frame 3) 0.37 0.058 mariner transposase fusion gene) -0.32 -0.059 NDUFB10 (NADH:ubiquinone SNX15 (sorting nexin 15) 0.36 0.058 subunit B10) -0.32 -0.059 IDI1 (isopentenyl-diphosphate AXL (AXL receptor tyrosine kinase) 0.33 0.058 delta isomerase 1) -0.32 -0.059 TICAM2 (toll like receptor adaptor AKR1C2 (aldo-keto reductase molecule 2) 0.33 0.058 family 1 member C2) -0.32 -0.059 JPT1 (Jupiter microtubule associated ME1 (malic enzyme 1) homolog 1) 0.33 0.058 -0.32 -0.059 SLC39A14 (solute carrier family 39 PPP1R7 (protein phosphatase 1 member 14) 0.33 0.058 regulatory subunit 7) -0.33 -0.059 ING3 (inhibitor of growth family member GTF2B (general transcription 3) 0.32 0.058 factor IIB) -0.34 -0.058 IL32 (interleukin 32) 0.31 0.058 PRR13 (proline rich 13) -0.34 -0.058

PDGFC (platelet derived growth factor GALNT7 (polypeptide N- C) 0.3 0.058 acetylgalactosaminyltransferase 7) -0.34 -0.058 PPP6R3 (protein phosphatase 6 ZDHHC24 (zinc finger DHHC-type regulatory subunit 3) 0.3 0.058 containing 24) -0.34 -0.058 NRIP1 (nuclear receptor interacting WBP2 (WW domain binding protein 1) 0.3 0.058 protein 2) -0.34 -0.058 RELB (RELB proto-oncogene, NF-kB NFIC ( C) subunit) 0.34 0.057 -0.3 -0.058 PFDN6 (prefoldin subunit 6) 0.35 0.057 PEPD (peptidase D) -0.32 -0.058

PCMTD2 (protein-L-isoaspartate AHNAK (AHNAK ) (D-aspartate) O-methyltransferase 0.33 0.057 domain containing 2) -0.32 -0.058 AHCYL1 (adenosylhomocysteinas EXT1 (exostosin glycosyltransferase 1) 0.33 0.057 e like 1) -0.32 -0.058 ARV1 (ARV1 homolog, fatty acid RPS25 (ribosomal protein S25) 0.33 0.057 homeostasis modulator) -0.33 -0.058 HILPDA (hypoxia inducible lipid droplet IRF2BP2 (interferon regulatory associated) 0.33 0.057 factor 2 binding protein 2) -0.33 -0.058 ATP5MG (ATP synthase FOXD1 () 0.33 0.057 membrane subunit g) -0.33 -0.058 GGA1 (golgi associated, gamma RAB34 (RAB34, member RAS adaptin ear containing, ARF binding oncogene family) protein 1) 0.32 0.057 -0.33 -0.058 RIN1 (Ras and interactor 1) 0.32 0.057 MGP (matrix Gla protein) -0.33 -0.058 IGFBP5 (insulin like growth factor FOS (Fos proto-oncogene, AP-1 binding protein 5) 0.32 0.057 transcription factor subunit) -0.33 -0.058 S100A16 (S100 calcium binding protein TTC3 (tetratricopeptide repeat A16) 0.31 0.057 domain 3) -0.33 -0.058 TMEM14B (transmembrane protein NBR2 (neighbor of BRCA1 14B) 0.37 0.056 lncRNA 2) -0.33 -0.058

TMEM123 (transmembrane protein 123) 0.36 0.056 ITSN1 (intersectin 1) -0.33 -0.058 TMEM8A (transmembrane protein CPE (carboxypeptidase E) 0.35 0.056 8A) -0.33 -0.058 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

TOX2 (TOX high mobility group box C16orf87 ( open family member 2) 0.35 0.056 reading frame 87) -0.33 -0.058 LINC00847 (long intergenic non- KLF11 (Kruppel like factor 11) 0.33 0.056 protein coding RNA 847) -0.33 -0.058 SMIM4 (small integral membrane RHOC (ras homolog family member C) 0.33 0.056 protein 4) -0.32 -0.057 SMYD3 (SET and MYND domain SESN1 (sestrin 1) containing 3) 0.32 0.056 -0.32 -0.057 DNAJB5 (DnaJ heat shock protein RAD50 (RAD50 double strand family (Hsp40) member B5) 0.32 0.056 break repair protein) -0.32 -0.057 HLA-B (major histocompatibility LOC100506100 (uncharacterized complex, class I, B) 0.32 0.056 LOC100506100) -0.32 -0.057 ILK (integrin linked kinase) 0.32 0.056 EBPL (EBP like) -0.32 -0.057 PDK1 (pyruvate dehydrogenase kinase RGL1 (ral guanine nucleotide 1) 0.32 0.056 dissociation stimulator like 1) -0.32 -0.057 GPR176 (G protein-coupled receptor KPNA3 ( subunit alpha 176) 0.32 0.056 3) -0.33 -0.057 LUZP1 ( protein 1) 0.32 0.056 PRDX3 (peroxiredoxin 3) -0.33 -0.057 NUMB (NUMB, endocytic adaptor REEP5 (receptor accessory protein) 0.31 0.056 protein 5) -0.3 -0.056 EIF4A1 (eukaryotic translation initiation PSMD10 (proteasome 26S factor 4A1) 0.31 0.056 subunit, non-ATPase 10) -0.3 -0.056 AGPAT5 (1-acylglycerol-3-phosphate O- ZFYVE27 (zinc finger FYVE-type acyltransferase 5) 0.31 0.056 containing 27) -0.3 -0.056 LRPPRC (leucine rich SRSF2 (serine and arginine rich splicing pentatricopeptide repeat factor 2) 0.31 0.056 containing) -0.3 -0.056 NDUFAF4 (NADH:ubiquinone TNFAIP3 (TNF alpha induced protein 3) oxidoreductase complex assembly 0.3 0.056 factor 4) -0.3 -0.056 BAZ1A (bromodomain adjacent to zinc ABCA8 (ATP binding cassette finger domain 1A) 0.3 0.056 subfamily A member 8) -0.3 -0.056 WDFY1 (WD repeat and FYVE domain UBP1 (upstream binding protein 1) containing 1) 0.3 0.056 -0.3 -0.056 MRPL9 (mitochondrial ribosomal NUP50 ( 50) 0.35 0.055 protein L9) -0.3 -0.056 ANKRD36B ( domain SNX9 (sorting nexin 9) 36B) 0.35 0.055 -0.3 -0.056 COMMD4 (COMM domain GRPEL1 (GrpE like 1, mitochondrial) 0.32 0.055 containing 4) -0.3 -0.056 SPAG4 (sperm associated antigen 4) 0.32 0.055 RNASEK ( K) -0.31 -0.056 ARHGDIA (Rho GDP dissociation SIGMAR1 (sigma non-opioid inhibitor alpha) 0.32 0.055 1) -0.31 -0.056 FASTKD5 (FAST kinase domains DPP7 (dipeptidyl peptidase 7) 0.3 0.055 5) -0.31 -0.056 RAB35 (RAB35, member RAS BDH2 (3-hydroxybutyrate oncogene family) 0.3 0.055 dehydrogenase 2) -0.31 -0.056 HSPB2 (heat shock protein family RPL37 (ribosomal protein L37) 0.29 0.055 B (small) member 2) -0.31 -0.056 SLC44A2 (solute carrier family 44 BCL3 (B cell CLL/lymphoma 3) 0.29 0.055 member 2) -0.32 -0.056 FKBP10 (FK506 binding protein 10) 0.29 0.055 FKBP7 (FK506 binding protein 7) -0.32 -0.056 TIMP3 (TIMP metallopeptidase inhibitor FAM3C (family with sequence 3) 0.29 0.055 similarity 3 member C) -0.32 -0.056 SRSF8 (serine and arginine rich INKA2 (inka box actin regulator 2) 0.33 0.054 splicing factor 8) -0.32 -0.056 PRKD1 (protein kinase D1) 0.32 0.054 MTRNR2L9 (MT-RNR2 like 9) -0.32 -0.056 SENP2 (SUMO specific peptidase ZNF350 (zinc finger protein 350) 0.32 0.054 2) -0.32 -0.056 OPTN (optineurin) 0.32 0.054 CASC4 (cancer susceptibility 4) -0.33 -0.056 GTDC1 (glycosyltransferase like YIPF3 (Yip1 domain family domain containing 1) 0.31 0.054 member 3) -0.3 -0.055 APMAP (adipocyte plasma FLNC (filamin C) 0.31 0.054 membrane associated protein) -0.31 -0.055 MCL1 (MCL1, BCL2 family apoptosis PENK (proenkephalin) regulator) 0.31 0.054 -0.33 -0.055 PNN (, associated CUTA (cutA divalent cation protein) 0.31 0.054 tolerance homolog) -0.31 -0.054 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

RELA (RELA proto-oncogene, NF-kB GNAI1 (G protein subunit alpha i1) subunit) 0.31 0.054 -0.31 -0.054 CLIC4 (chloride intracellular channel 4) 0.31 0.054 CD63 (CD63 molecule) -0.32 -0.054 TPST1 (tyrosylprotein HPCAL1 ( like 1) 0.31 0.054 sulfotransferase 1) -0.33 -0.054 FDPS (farnesyl diphosphate ARRDC4 ( domain containing 4) 0.31 0.054 synthase) -0.33 -0.054 MVD (mevalonate diphosphate IRX3 (iroquois homeobox 3) 0.31 0.054 decarboxylase) -0.33 -0.054 RHOT2 (ras homolog family member RBP4 (retinol binding protein 4) T2) 0.29 0.054 -0.33 -0.054 EIF5A (eukaryotic translation initiation PRAF2 (PRA1 domain family factor 5A) 0.29 0.054 member 2) -0.33 -0.054 ADH5 (alcohol dehydrogenase 5 HRAS (HRas proto-oncogene, GTPase) 0.29 0.054 (class III), chi polypeptide) -0.31 -0.053 FLNA (filamin A) 0.29 0.054 FKBP9 (FK506 binding protein 9) -0.31 -0.053 UFC1 (ubiquitin-fold modifier PLEC () 0.29 0.054 conjugating enzyme 1) -0.31 -0.053 SERTAD2 (SERTA domain containing TTC8 (tetratricopeptide repeat 2) 0.32 0.053 domain 8) -0.32 -0.053 UPRT (uracil SMS (spermine synthase) phosphoribosyltransferase 0.32 0.053 homolog) -0.32 -0.053 LINC01278 (long intergenic non- GYS1 ( synthase 1) 0.31 0.053 protein coding RNA 1278) -0.29 -0.052 MGST1 (microsomal glutathione S- TUBB4B (tubulin beta 4B class IVb) 0.31 0.053 transferase 1) -0.29 -0.052 LSMEM2 (leucine rich single-pass LSM10 (LSM10, U7 small nuclear membrane protein 2) 0.3 0.053 RNA associated) -0.29 -0.052 NELFE (negative elongation factor SORBS2 (sorbin and SH3 domain complex member E) 0.3 0.053 containing 2) -0.29 -0.052 EIF5A2 (eukaryotic translation initiation MINOS1-NBL1 (MINOS1-NBL1 factor 5A2) 0.3 0.053 readthrough) -0.29 -0.052 HSPB8 (heat shock protein family ITGA5 (integrin subunit alpha 5) 0.3 0.053 B (small) member 8) -0.3 -0.052 BZW1 (basic leucine zipper and W2 AK3 (adenylate kinase 3) domains 1) 0.3 0.053 -0.3 -0.052 LPP (LIM domain containing preferred TMEM30A (transmembrane translocation partner in lipoma) 0.3 0.053 protein 30A) -0.3 -0.052

IRF7 (interferon regulatory factor 7) 0.3 0.053 ACVR1 (activin A receptor type 1) -0.3 -0.052 MBNL2 (muscleblind like splicing SDC4 (syndecan 4) 0.3 0.053 regulator 2) -0.3 -0.052 GFER (growth factor, augmenter of C7orf25 ( open regeneration) 0.29 0.053 reading frame 25) -0.3 -0.052 BMPER (BMP binding endothelial FBXO22-AS1 (FBXO22 antisense regulator) 0.29 0.053 RNA 1) -0.3 -0.052 LRRC59 (leucine rich repeat containing NUP133 () 59) 0.32 0.052 -0.3 -0.052 WDR41 (WD repeat domain 41) 0.3 0.052 MKLN1 (muskelin 1) -0.3 -0.052 MARK3 (microtubule affinity regulating NINJ2 (ninjurin 2) kinase 3) 0.3 0.052 -0.3 -0.052 ARL4A (ADP ribosylation factor like SLC2A10 (solute carrier family 2 GTPase 4A) 0.29 0.052 member 10) -0.3 -0.052 SRSF9 (serine and arginine rich splicing NDUFV2 (NADH:ubiquinone factor 9) 0.29 0.052 oxidoreductase core subunit V2) -0.3 -0.052 KCTD5 (potassium channel NBDY (negative regulator of P- tetramerization domain containing 5) 0.28 0.051 body association) -0.3 -0.052 SMIM11A (small integral STMN2 ( 2) 0.32 0.051 membrane protein 11A) -0.3 -0.052 FAAP24 (FA core complex RFLNB (refilin B) 0.32 0.051 associated protein 24) -0.3 -0.052 AKAP9 (A-kinase anchoring RPL21 (ribosomal protein L21) 0.3 0.051 protein 9) -0.3 -0.052 SNRPA (small nuclear ribonucleoprotein PRR16 (proline rich 16) polypeptide A) 0.3 0.051 -0.31 -0.052 SLC39A10 (solute carrier family EAF1 (ELL associated factor 1) 0.3 0.051 39 member 10) -0.31 -0.052 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

SLC2A5 (solute carrier family 2 member DEPTOR (DEP domain containing 5) 0.3 0.051 MTOR interacting protein) -0.31 -0.052 ZSWIM7 (zinc finger SWIM-type EZR (ezrin) 0.3 0.051 containing 7) -0.31 -0.052 TPMT (thiopurine S-methyltransferase) 0.3 0.051 INSIG1 (insulin induced gene 1) -0.31 -0.052 SENP5 (SUMO specific peptidase DTD1 (D-tyrosyl-tRNA deacylase 1) 0.3 0.051 5) -0.31 -0.052 RCOR3 (REST corepressor 3) 0.3 0.051 ANXA10 (annexin A10) -0.31 -0.052 SRSF1 (serine and arginine rich splicing POLR2F (RNA II factor 1) 0.3 0.051 subunit F) -0.31 -0.052 PRELID1 (PRELI domain containing 1) 0.29 0.051 TBX18 (T-box 18) -0.31 -0.052 SLC38A5 (solute carrier family 38 NSDHL (NAD(P) dependent member 5) 0.28 0.05 steroid dehydrogenase-like) -0.31 -0.052 SEC24B (SEC24 homolog B, COPII AKR1C3 (aldo-keto reductase coat complex component) 0.28 0.05 family 1 member C3) -0.31 -0.052 IQGAP1 (IQ motif containing GTPase PCOLCE2 (procollagen C- activating protein 1) 0.28 0.05 endopeptidase enhancer 2) -0.31 -0.052 SNORD81 (small nucleolar RNA, C/D ITM2B (integral membrane protein box 81) 0.28 0.05 2B) -0.31 -0.052 RPSAP52 (ribosomal protein SA GSR (glutathione-disulfide pseudogene 52) 0.28 0.05 reductase) -0.32 -0.052 SSR4 (signal sequence receptor VGLL3 (vestigial like family member 3) 0.31 0.05 subunit 4) -0.32 -0.052 TPI1 (triosephosphate isomerase 1) 0.31 0.05 NIT2 (nitrilase family member 2) -0.32 -0.052 DDX19A (DEAD-box helicase XPO6 (exportin 6) 0.31 0.05 19A) -0.28 -0.051 ARHGEF2 (Rho/Rac guanine FUCA2 (alpha-L-fucosidase 2) nucleotide exchange factor 2) 0.31 0.05 -0.28 -0.051 PHLDB2 (pleckstrin homology like MFSD3 (major facilitator domain family B member 2) 0.31 0.05 superfamily domain containing 3) -0.28 -0.051 IMMP1L (inner mitochondrial membrane DUSP23 (dual specificity peptidase subunit 1) 0.31 0.05 phosphatase 23) -0.28 -0.051 SLC39A8 (solute carrier family 39 DAP (death associated protein) member 8) 0.31 0.05 -0.28 -0.051 RAB29 (RAB29, member RAS MTX2 (metaxin 2) 0.31 0.05 oncogene family) -0.29 -0.051 LMBR1 (limb development EMP1 (epithelial membrane protein 1) 0.31 0.05 membrane protein 1) -0.3 -0.051 MUS81 (MUS81 structure-specific B4GALT1-AS1 (B4GALT1 subunit) 0.31 0.05 antisense RNA 1) -0.3 -0.051 ARL6IP5 (ADP ribosylation factor RAPH1 (Ras association (RalGDS/AF- like GTPase 6 interacting protein 6) and pleckstrin homology domains 1) 0.31 0.05 5) -0.31 -0.051 LOC100129917 (uncharacterized PSMA5 (proteasome subunit LOC100129917) 0.3 0.05 alpha 5) -0.31 -0.051 ANPEP (alanyl aminopeptidase, OAT (ornithine aminotransferase) membrane) 0.29 0.05 -0.31 -0.051 WASL (Wiskott-Aldrich syndrome UCK2 (uridine-cytidine kinase 2) 0.29 0.05 like) -0.31 -0.051 HNRNPU (heterogeneous nuclear VAMP3 (vesicle associated ribonucleoprotein U) 0.29 0.05 membrane protein 3) -0.32 -0.051 ARPC5L (actin related protein 2/3 TMED3 (transmembrane p24 complex subunit 5 like) 0.29 0.05 trafficking protein 3) -0.32 -0.051 FJX1 (four-jointed box kinase 1) 0.29 0.05 TRIB1 (tribbles pseudokinase 1) -0.32 -0.051 CMTM6 (CKLF like MARVEL MRPL46 (mitochondrial ribosomal transmembrane domain containing protein L46) 0.29 0.05 6) -0.27 -0.05 PSMG3 (proteasome assembly RPL6 (ribosomal protein L6) 0.29 0.05 chaperone 3) -0.27 -0.05 FDFT1 (farnesyl-diphosphate TMEM217 (transmembrane protein 217) 0.29 0.05 farnesyltransferase 1) -0.27 -0.05 GNPNAT1 (glucosamine-phosphate N- STYXL1 (serine/threonine/tyrosine acetyltransferase 1) 0.29 0.05 interacting like 1) -0.27 -0.05 SYNCRIP ( binding COQ10A (coenzyme Q10A) cytoplasmic RNA interacting protein) 0.29 0.05 -0.27 -0.05 HSD17B1 (hydroxysteroid 17-beta BBS1 (Bardet-Biedl syndrome 1) dehydrogenase 1) 0.29 0.05 -0.27 -0.05 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

ZCRB1 (zinc finger CCHC-type FGF1 (fibroblast growth factor 1) and RNA binding motif containing 0.29 0.05 1) -0.28 -0.05 JPT2 (Jupiter microtubule associated SRP14 (signal recognition particle homolog 2) 0.29 0.05 14) -0.28 -0.05 ADGRE5 (adhesion G protein-coupled TRUB2 (TruB pseudouridine receptor E5) 0.29 0.05 synthase family member 2) -0.28 -0.05 UNKL (unkempt family like zinc MTRNR2L1 (MT-RNR2 like 1) 0.29 0.05 finger) -0.28 -0.05 COX7A1 (cytochrome c oxidase NOTCH1 (notch 1) 0.29 0.05 subunit 7A1) -0.28 -0.05

BEX3 ( expressed X-linked 3) -0.28 -0.05 TMEM206 (transmembrane protein 206) -0.29 -0.05 CSTA (cystatin A) -0.29 -0.05 MRPS11 (mitochondrial ribosomal protein S11) -0.29 -0.05 ITFG1 (integrin alpha FG-GAP repeat containing 1) -0.29 -0.05 NUPL2 (nucleoporin like 2) -0.29 -0.05 ASAH1 (N-acylsphingosine amidohydrolase 1) -0.29 -0.05 GALNT1 (polypeptide N- acetylgalactosaminyltransferase 1) -0.29 -0.05 RNASEH1 (ribonuclease H1) -0.29 -0.05 GTF3C6 (general transcription factor IIIC subunit 6) -0.3 -0.05 HECTD1 (HECT domain E3 ubiquitin protein ligase 1) -0.3 -0.05 PGD (phosphogluconate dehydrogenase) -0.3 -0.05 LETMD1 (LETM1 domain containing 1) -0.3 -0.05 POMZP3 (POM121 and ZP3 fusion) -0.3 -0.05 C1GALT1 (core 1 synthase, glycoprotein-N- acetylgalactosamine 3-beta- galactosyltransferase 1) -0.3 -0.05 MEPCE (methylphosphate capping enzyme) -0.3 -0.05 TBC1D7 (TBC1 domain family member 7) -0.31 -0.05 TMEM50A (transmembrane protein 50A) -0.31 -0.05 MTRNR2L10 (MT-RNR2 like 10) -0.32 -0.05 bioRxiv preprint doi: https://doi.org/10.1101/537464; this version posted January 31, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Table 11. Pathway enrichment analysis of genes in progenitors correlated with LEP levels in M condition .

Hit Hit Count q-value Count in KEGG Pathway Name p-value in Hit in Query List Bonferroni Query Genome List

TUBA1B,TUBB6,TUBB4B,EZR,KRT18,TU Pathogenic Escherichia coli infection 1.54E-07 2.09E-04 9 55 BA1C,TUBB,ARPC5L,ARHGEF2 Glycolysis (Embden-Meyerhof pathway), 1.83E-06 2.48E-03 6 25 PFKP,PGK1,GPI,ENO1,ENO2,TPI1 glucose => pyruvate CITED2,BNIP3,HRAS,RRAS2,NRAS,RELA Mitophagy - animal 7.10E-06 9.62E-03 8 65 ,RHOT2,OPTN Glycolysis, core module involving three- 2.57E-05 3.48E-02 4 12 PGK1,ENO1,ENO2,TPI1 carbon compounds TUBA1B,TUBB6,TUBB4B,HRAS,PDGFC, 6.67E-05 9.03E-02 8 88 NRAS,TUBA1C,TUBB

ITGA5,IQGAP1,HRAS,SDC4,VEGFA,EZR, Proteoglycans in cancer 7.99E-05 1.08E-01 12 203 FLNA,FLNB,FLNC,RRAS2,NRAS,TIMP3

Gluconeogenesis, oxaloacetate => fructose- 1.16E-04 1.57E-01 4 17 PGK1,ENO1,ENO2,TPI1 6P Pathways for genes HLA-A,HLA-B,HLA- positively Phagosome 1.47E-04 2.00E-01 10 154 C,ITGA5,TUBA1B,TUBB6,TUBB4B,DYN correlatedwith LEP C1H1,TUBA1C,TUBB levels in M condition PDK1,PGK1,VEGFA,EGLN1,RELA,SERPI HIF-1 signaling pathway 1.76E-04 2.39E-01 8 101 NE1,ENO1,ENO2 MCL1,TUBA1B,HRAS,NGF,NRAS,TUBA Apoptosis 3.06E-04 4.15E-01 9 138 1C,RELA,LMNB2,CTSH PFN1,DYNC1H1,FLNA,FLNB,FLNC,RELA Salmonella infection 3.81E-04 5.16E-01 7 86 ,ARPC5L MYLK,ITGA5,PFN1,IQGAP1,HRAS,FGF1 Regulation of actin 4.88E-04 6.61E-01 11 212 ,PDGFC,EZR,RRAS2,NRAS,ARPC5L Glycolysis / Gluconeogenesis 6.01E-04 8.14E-01 6 67 PFKP,PGK1,GPI,ENO1,ENO2,TPI1 Biosynthesis of amino acids 1.09E-03 1.00E+00 6 75 PFKP,PGK1,GPT2,ENO1,ENO2,TPI1 HNRNPA1,HNRNPK,HNRNPU,SRSF1,SR Spliceosome 1.18E-03 1.00E+00 8 134 SF2,TRA2B,SNRPA,SRSF9 EGFR tyrosine kinase inhibitor resistance 1.43E-03 1.00E+00 6 79 NRG1,AXL,HRAS,PDGFC,VEGFA,NRAS

Carbon metabolism 2.02E-03 1.00E+00 7 114 PFKP,PGK1,GPT2,GPI,ENO1,ENO2,TPI1 ATP6V1A,AHCYL1,ATP6V1E1,ATP6V0B, Metabolic pathways 2.44E-08 3.57E-05 46 1272 PIGP,GALNT1,GAMT,CYP51A1,BDH2,A Steroid biosynthesis 1.06E-05 1.56E-02 5 20 CYP51A1,NSDHL,SC5D,FDFT1,DHCR7 Cholesterol biosynthesis, squalene 2,3- 1.66E-05 2.43E-02 4 11 CYP51A1,NSDHL,SC5D,DHCR7 epoxide => cholesterol ATP6V1A,ATP6V1E1,ATP6V0B,ATP6V1 Oxidative 3.93E-05 5.75E-02 10 133 F,NDUFB10,NDUFV2,ATP6V0E1,COX7 Pathways for genes A1,ATP5MG,PPA2 negatively Terpenoid backbone biosynthesis 3.23E-04 4.72E-01 4 22 FDPS,IDI1,ICMT,MVD correlatedwith LEP Pentose phosphate pathway (Pentose levels in M condition 7.17E-04 1.00E+00 3 12 PGD,TKT,G6PD phosphate cycle) ATP6V1A,ATP6V1E1,ATP6V1F,ATP6V0 Collecting duct acid secretion 7.29E-04 1.00E+00 4 27 E1 ATP6V1A,ATP6V1E1,ATP6V0B,ATP6V1 Vibrio cholerae infection 1.09E-03 1.00E+00 5 51 F,ATP6V0E1 Pentose phosphate pathway, oxidative 1.39E-03 1.00E+00 2 4 PGD,G6PD phase, glucose 6P => ribulose 5P