A spatiotemporal proteomic map of human adipogenesis

White adipocytes function as major energy reservoirs in humans by storing substantial amounts of triglycerides, and their dysfunction is associated with metabolic disorders; however, the mechanisms underlying cellular specialization during adipogenesis remain unknown. Here, we generate a spatiotemporal proteomic atlas of human adipogenesis, which elucidates cellular remodelling as well as the spatial reorganization of metabolic pathways to optimize cells for lipid accumulation and highlights the coordinated regulation of protein localization and abundance during adipocyte formation. We identify compartment-specific regulation of protein levels and localization changes of metabolic enzymes to reprogramme branched-chain amino acids and one-carbon metabolism to provide building blocks and reduction equivalents. Additionally, we identify C19orf12 as a differentiation-induced adipocyte lipid droplet protein that interacts with the translocase of the outer membrane complex of lipid droplet-associated mitochondria and regulates adipocyte lipid storage by determining the capacity of mitochondria to metabolize fatty acids. Overall, our study provides a comprehensive resource for understanding human adipogenesis and for future discoveries in the field.

• C19orf12 is an adipocyte specific LD protein and regulator of lipid storage SUMMARY White adipocytes function as the major energy reservoir in humans by storing substantial amounts of triglycerides and their dysfunction is associated with metabolic disorders. However, the mechanisms underlying cellular specialization during adipogenesis remain unknown. Here, using a high-sensitivity-high throughput workflow, we generated a spatiotemporal proteomic atlas of human adipogenesis encompassing information for ~8.000 proteins. Our systematic approach gives insights into cellular remodeling and the spatial reorganization of metabolic pathways to optimize cells for lipid accumulation and highlights the coordinated regulation of protein localization and abundance during adipogenesis. More specifically, we identified a compartmentspecific regulation of protein levels and localization changes of metabolic enzymes to reprogram branched chain amino acid and one-carbon metabolism to provide building blocks and reduction equivalents for lipid synthesis. Additionally, we identified C19orf12 as a differentiation induced and adipocyte-specific lipid droplet (LD) protein, which interacts with the translocase of the outer membrane (TOM) complex of LD associated mitochondria and regulates adipocyte lipid storage.
Overall, our study provides a comprehensive resource for understanding human adipogenesis and for future discoveries in the field.

INTRODUCTION
Living organisms have evolved the capacity to store energy in the form of fat in lipid droplets (LDs).
The core of these storage organelles contain neutral lipids, such as triglycerides and an average healthy adult stores 10-25 kg of fat primarily in white adipose tissue (WAT), with each kg equivalent to 9,000 kcal 1 .
WAT characterized by few, but large adipocytes (hypertrophy) is associated with insulin resistance and the secretion of pro-inflammatory cytokines, whereas WAT displaying a higher number of small adipocytes (hyperplasia) is linked to a metabolically healthy phenotype [8][9][10] . The fact that the balance between hyperplasia and hypertrophy associates strongly with the risk of developing metabolic complications of obesity, underscores the importance to understand how adipocytes acquire their remarkable capacity for lipid storage and mobilization during differentiation, and to identify cellular processes that underly healthy adipogenesis and lipid dynamics.
Recent technological advances in transcriptomics have significantly improved our understanding of adipocyte heterogeneity and transcriptional networks underlying adipogenesis [11][12][13][14][15] . However, it has become increasingly clear that post-transcriptional processes are also critical for regulating protein levels and activity during adipogenesis [16][17][18][19] . These dynamic processes, along with the remodeling of organelles can be more thoroughly understood at the proteome level, which remains poorly characterized in the context of adipogenesis. To date, proteomic analyses have provided static snapshots of the protein landscape, but not yet covered the spatiotemporal aspects of adipogenesis [20][21][22][23] . As a result, our current understanding of proteomic remodeling and the subcellular organization during adipocyte differentiation remains incomplete.
To address this gap, we have generated a comprehensive spatiotemporal proteomic map across four different human adipocyte models. Our deep and quantitative temporal profiling of protein level changes allowed us to determine proteomic evolution across adipogenesis, which we then compared to primary human white adipocytes and WAT to identify conserved proteomic changes in human adipogenesis. Using a machine-learning-based organelle proteomics approach, we mapped multiple changes in protein localization during adipogenesis. We revealed the coordinated remodeling of metabolic pathways at the level of protein abundance and localization to support de novo lipogenesis, and identified a yet unknown LD protein, C19orf12, as a novel regulator of adipocyte function. We found that C19orf12 localizes to LD-mitochondrial contact regions, and its depletion leads to impaired lipolysis, and increased lipid droplet accumulation. In a human patient cohort, we find C19orf12 expression in WAT inversely correlated with obesity associated clinical parameters, underlying the key function of C19orf12 in human adipocyte lipid storage. Overall, our study offers a comprehensive resource for understanding the temporally resolved core proteome changes in human adipogenesis, as well as the reorganization of organelles and metabolic pathways that drive human adipogenesis.

The temporally resolved core proteome of human adipogenesis
To define the core proteome trajectory during human adipogenesis, we performed liquid chromatography-mass spectrometry (LC-MS) proteomics over the time course of differentiation across different human adipocyte models (Fig.1A). All models are derived from human adipocyte precursor cells (hAPCs) isolated from the stromal-vascular fraction (SVF) of WAT and have the capacity to differentiate into adipocytes upon treatment with pro-adipogenic cocktails. Two cell types were non-immortalized (SGBS 2 and hAPC 3,4 ), whereas one was immortalized (hWA 5 ). To control for the potential effects of the immortalization process, we generated an additional cell model in which hAPCs were immortalized by introducing telomerase reverse transcriptase into the AAVS1 safe harbor locus using CRISPR/Cas9 engineering (herein termed ThAPC). Thus, a total of four human model systems were included in this study to map the conserved trajectory of adipogenesis independent of cell type-specific effects. Information regarding the origins, immortalization procedures, and differentiation protocols is summarized in Fig.S1A and Table1.
For comparison, we also included paired primary samples of subcutaneous abdominal mature adipocytes (pACs), SVF (which contains immature adipocyte precursors), and intact WAT from seven donors (see Methods, Table1). These served as reference points for in vivo adipogenesis.
First, we assessed the adipogenic capacity of all four models. By measuring lipid accumulation by BODIPY staining followed by fluorescence microscopy, and mRNA levels of well-established adipogenic marker genes, we found all models to display high differentiation efficiencies ( Fig.S1B and Fig.S1C). For proteomic analyses, we sampled the undifferentiated state and at least six time points after induction of differentiation in three biological replicates for each model. Tryptic peptides were analyzed in 1h single shots in data-independent acquisition (DIA) mode (Materials and Methods, Fig.1A). This approach enables higher identification rates over a larger dynamic range and fewer missing values compared to data-dependent acquisition (DDA). Spectronaut analysis quantified 5,979-7,061 protein groups in the four cell models, 3,638-6,403 in the primary samples (pAC/SVF/WAT), and 8,268 in the complete dataset (Fig.S1D). The MS signals spanned an abundance range of five orders of magnitude (Fig.1B) and were highly reproducible among replicates, with an average Pearson's correlation coefficient of 0.97. We found that a considerable proportion (56-65%) of the proteome underwent remodeling during the differentiation process (ANOVA, FDR 0.01) in all four models. About 6-14% of the quantified proteins showed a more than 10-fold change compared to the undifferentiated state or were exclusive to either the undifferentiated or mature state (Fig.S1E). In total, 3,934 proteins were significantly regulated in at least three of the four models (Fig.1E, F), of which 1,979 proteins displayed a conserved temporal trajectory during adipogenesis (see Methods and Fig.1G). When comparing the samples, the Pearson correlation coefficients were high within the same time points between the systems (Fig.S1F), indicating that there are universal proteomic features of adipogenesis that can be recapitulated in multiple in vitro models. This enabled us to identify the universal events of human adipogenesis, which are not dependent on donors and are not affected by immortalization.
Principal component analysis (PCA) showed the temporal trajectory of differentiation states in vitro (the four cell models) and in vivo (SVF vs. pACs) in components 1 and 2, where the adipogenic process of the cell lines projected towards mature adipocytes (Fig.1C). An increase in LD and lipid biosynthesis protein levels was a key factor driving the shift in component 1 (Fig.1D) and we observed a significant increase in protein levels of canonical adipogenesis markers in all four models during differentiation (Fig.S1G). However, our PCA analysis showed that hWA cells reached a less mature state than the other models. This finding is consistent with the reduced expression of classical adipocyte marker genes (Fig.S1C) and lower Pearson correlation coefficients between the proteomes of fully differentiated cell models and pACs (Fig.S1H).
We next performed hierarchical clustering analysis on the conserved temporal profiles. Our results identified distinct clusters that represented early, intermediate, and late responses during adipogenesis (Fig.1H). The early phase was characterized by the downregulation of proteins involved in cell cycle progression, cytoskeletal and extracellular matrix remodeling, secreted factors, and the chaperone system. The transient cluster of temporarily upregulated proteins was characterized by glycosaminoglycan degradation and lysosomal pathways. In addition, the intermediate phase was also defined by a significant increase in enzymes involved in fatty acid and triglyceride metabolism as well as in many mitochondria-related functions that maintain the increased energy demands for triglyceride synthesis, such as the TCA cycle, respiratory chain, and ATP synthesis. Simultaneously, the levels of proteins involved in DNA replication, telomere maintenance, ciliary proteins, and WNT signaling (a pathway that inhibits adipogenesis [6][7][8] decreased. In the late phase of adipogenesis, we observed downregulation of spliceosomal and mRNA-processing proteins, and upregulation of cholesterol biosynthesis proteins. Overall, these findings suggest comprehensive remodeling of cellular processes during adipogenesis, with distinct temporal regulation of specific functional pathways.

A spatial map of adipogenesis
To add a spatial dimension to our proteome of adipogenesis, we utilized protein correlation profiling (PCP), a technique that allows for the analysis of organellar protein localization based on relative abundance profiles. Briefly, for PCP, cells are mechanically lysed, and the organelles are separated by density-gradient centrifugation. Next, proteins are quantified across gradient fractions by LC-MS, and the generated abundance profiles, which are characteristic of the residual cellular compartments, are used to predict protein localization by machine learning 9 .
We applied PCP to mature adipocytes and pre-adipocytes using the SGBS model to identify proteins that displayed different locations during adipogenesis ( Fig.2A). 1h LC-MS DIA single shot analyses led to 3,500-5,600 quantified proteins per fraction (Fig.S2A) resulting in cellular maps with higher proteomic coverage and identification rates, less MS runtime, and increased reproducibility compared with traditional DDA-based approaches 10 . A comparison of the median profiles of marker proteins from different cellular compartments indicated clear separation of organelles in our gradients (Fig.2B). Supervised hierarchical clustering of the median profiles from the biological replicates showed distinct clusters for cellular compartments (Fig.2C). By employing support vector machine (SVM)-based supervised learning, we were able to predict primary and potential secondary protein localization using the generated abundance profiles, allowing for the comprehensive annotation of protein distribution. Organellar cluster boundaries were determined using established marker proteins selected based on GO annotations (Methods). Our maps of adipocytes and preadipocytes provide localization information for a total of 7,619 and 8,250 proteins, respectively. Out of these, 4,918 proteins in adipocytes and 4,635 proteins in preadipocytes were confidently assigned to specific organelle clusters using SVMs (see Methods and Fig.2D and 2E). A mean prediction accuracy of 91% was achieved for marker proteins. More than half of the proteins were classified as derived from at least two organelle distributions ( Fig.S2B and S2C), consistent with previous studies 11 .

Protein localization changes in adipogenesis
SVMs assigned a total of 4,426 proteins with high confidence in both the undifferentiated progenitor cells and the adipocytes, and 898 of them exhibited divergent organelle assignments ( Fig.3A). We leveraged these spatial proteomics data and the time-resolved core proteome of adipogenesis to comprehensively characterize organelle remodeling during adipogenesis. By integrating information from both datasets, we were able to predict the proportion of each organelle in the total proteome. Our findings showed that, during adipogenesis, there was an increase in the percentage of mitochondrial, endoplasmic reticulum, endosomal, and LD proteins, whereas the proportion of cytosolic, nuclear, and ribosomal proteins decreased. These changes in organelle composition reflected an overall increase in the total protein mass of all compartments involved in lipid metabolism and secretory functions, which ultimately led to a state that closely resembled the proportional organelle distribution in pACs (Fig.3B, S3A).
Our analysis revealed that certain compartments exchanged proteins more frequently than others ( Fig. 3C). Notably, we observed a reorganization of the translational machinery during adipogenesis. In mature adipocytes, the N-terminal acetyltransferase C (NATC) complex was associated with ribosomes, while in preadipocytes, the subunits of the same complex exhibited a diffuse distribution across all fractions ( Figure 3D and 3E). Indeed, NATs can bind to ribosomes where they perform N-terminal acetylation in a co-translational manner to regulate protein degradation rates and interactions 12 . Strikingly, among NAT complexes, NATC is particularly important to modify mitochondrial proteins, which are strongly induced in adipogenesis 13 . As another example, we mapped the translocation of numerous mitochondrial proteins, including deoxythymidylate kinase DTYMK, which is involved in pyrimidine biosynthesis. During adipogenesis, DTYMK displayed increased mitochondrial targeting, with a concomitant decrease in the cytosol (Fig.3F).
Additionally, we observed an alternative mechanism contributing to changes in protein expression during adipogenesis, which involved the specific expression of isoforms with distinct localizations.
This phenomenon was observed for SLC25A10, the mitochondrial dicarboxylate carrier responsible for succinate transport and predominantly expressed in white adipose tissue 14 . During adipogenesis, isoform 2 with a nuclear profile was downregulated, while isoform 1 with mitochondrial localization significantly increased ( Figure 3F, G and H). We further confirmed the isoform switch-driven relocalization through immunofluorescence staining of SLC25A10 using an antibody recognizing both isoforms. The staining showed SLC25A10 localization in the nucleus of preadipocytes and in mitochondria of adipocytes ( Figure 3I). In summary, our findings demonstrate that approximately 20% of all mapped proteins undergo localization changes, suggesting that the regulation of protein localization contributes to cellular differentiation processes during adipogenesis."

Cooperative protein localization and abundance changes drive metabolic reprogramming
To gain a better understanding of how organelles respond during adipogenesis, we conducted cluster analysis of temporal protein profiles assigned to specific organelles as exemplified here for mitochondria. Our analysis indicated that the notable increase in the total amount of mitochondrial protein during adipogenesis (Fig.3C) was accompanied by the upregulation of various mitochondrial pathways, including the TCA cycle, respiratory chain complexes, and branched chain amino acid (BCAA) catabolism (Fig.4A), consistent with previous findings that degradation of the amino acids valine, leucine, and isoleucine provides an essential pool of acetyl-CoA for de novo lipogenesis in adipocytes 15 . Our spatiotemporal data integration revealed compartmentspecific regulation of both the levels and localization of BCAA catabolism enzymes (Fig.4B).
Specifically, we found that during the proliferative phase of adipocyte precursor cells, BCAT1 and BCAT2, the first enzymes in the BCAA degradation pathway, are in the cytosol. This localization enables the degradation of BCAA to produce glutamine, a key component required for de novo nucleotide biosynthesis, which is critical for cell division. However, during differentiation, BCAT1 was downregulated, whereas BCAT2 was upregulated and translocated to the mitochondria ( Fig.4C). As all mitochondrial BCAA enzymes increase their levels during differentiation, we hypothesize that the upregulation of these enzymes, coupled with BCAT2 translocation to the mitochondria, shifts the pathway from cytosol to mitochondrial BCAA degradation, leading to the production of acetyl-CoA via the TCA cycle, which is essential for de novo lipogenesis in adipogenesis.
Unexpectedly, the increase in mitochondrial protein mass during adipogenesis was accompanied by a decline in the levels of mitochondrial enzymes involved in one-carbon metabolism, which activates and transfers one-carbon units for biosynthetic processes. Given the counter-regulation of this set of enzymes to the mitochondrial protein pool, together with the observation that enzymes for 10-formyltetrahydrofolate catabolism were enriched among the protein with localization changes (Fig.S3B), we investigated the reorganization of one-carbon metabolism during adipogenesis. While levels of all mitochondrial enzymes of the pathway decreased, the cytosolic branch of the pathway was upregulated (Fig.4D). Meanwhile, both isoforms of 10formyltetrahydrofolate dehydrogenase, ALDH1L1 and ALDH1L2, which catalyze the last reaction of the pathway and promote NADPH release, changed localization from the mitochondria to cytosolic protein complexes, as indicated by their protein profiles, as well as confirmed by coimmunostaining with the mitochondrial marker TOM20 (Fig.4E, F, G and H). Cytosolic enzymes for purine and methionine synthesis catalyzing reactions consuming one carbon metabolism intermediates and cytosolic NADPH decreased. Notably, in proliferating cells, the electrochemical potential difference between mitochondrial NADH and cytosolic NADPH is responsible for driving the serine cycle in the direction that catabolizes serine in the mitochondria and synthesizes it in the cytosol, as shown in previous studies 16 . However, when the mitochondrial component of the serine cycle is lost, the direction of the cytosolic part of the cycle is reversed, leading to cytosolic serine degradation and NADPH 17 production. Given this regulatory mechanism, compartmentspecific adjustments of enzyme levels during adipogenesis may also lead to an increase in cytosolic NADPH synthesis, thereby supporting de novo lipogenesis by providing the necessary reduction equivalents. Compartment-specific regulation of one-carbon metabolism and BCAA enzyme levels was also present in pACs compared to SVF (FigS3C and S3D), indicating that metabolic reprogramming also occurs in vivo during adipogenesis. Together, our findings highlight the coordinated control of protein localization and levels to reprogram metabolic pathways to provide building blocks and reduction equivalents for fatty acid synthesis in adipogenesis.

Spatial organization of lipid metabolism in human white adipocytes
The defining feature of white adipocytes is their specialization for lipid storage and the formation of large LDs. Therefore, we used our spatial-temporal atlas to investigate the organization of lipid metabolism and to define the adipocyte LD proteome. Hierarchical clustering of the protein profiles had revealed that proteins organized into protein complexes were clearly separated from cytosolic proteins using the PCP approach (Fig.2C). Unexpectedly, annotation enrichment analysis identified not only the partitioning of several prominent complexes, including mTOR, proteasome or chaperonin complexes, into this protein complex cluster, but also an enrichment for fatty acid biosynthesis (Fig.5A). Notably, the enzymes ACLY, FASN, ACACA, and ACACB, which are involved in all steps of fatty acid biosynthesis, from citrate via acetyl-CoA and malonyl-CoA to the final product palmitate, were also sorted into this cluster. Their protein profiles were nearly identical ( Fig.5B), suggesting potential condensate formation or a special arrangement of these enzymes within the cytosol in adipocytes.
Although adipocytes are the major cell type for lipid storage, a high confident adipocyte LD proteome is lacking so far. To establish the human white adipocyte LD proteome and to exclude contaminants from the set of proteins with LD classifications, we filtered for proteins that were enriched in the LD fraction compared to the total proteome, as all known LD marker proteins showed this behavior (Fig.5C). Clustering analysis of significantly altered LD proteins revealed a well-coordinated and time-dependent regulation of the LD proteome during adipogenesis, which exhibited high consistency across all models examined (Fig.5D). Following adipogenesis induction, a rapid surge in seipin (BSCL2) levels was observed, highlighting its crucial role in the budding of LDs from the endoplasmic reticulum [18][19][20] . In contrast, the proteins involved in lipid mobilization displayed an upregulation pattern in the later stages. Among the canonical lipolysisassociated proteins, PNPLA2 and its essential cofactor ABHD5 exhibited coordinated upregulation and were assigned to a cluster showing intermediate upregulation. Conversely, LIPE and MGLL were upregulated during the later stages of adipogenesis. Furthermore, alterations in the expression levels of perilipins, a family of structural proteins responsible for coating LDs and regulating lipase access 21 , were also observed. Initially, the ubiquitously expressed member of the perilipin family, PLIN2, was downregulated, whereas the adipocyte-specific member PLIN4 was upregulated late in the adipogenesis process.

C19orf12 is an adipocyte specific LD protein regulating lipid storage
To identify adipocyte specific LD proteins as potential candidates to promote the exceptional characteristics of adipocytes for lipid storage and dynamics, we overlaid the LD proteome from SGBS cells with the LD proteome from hAPCs, which we also generated by SVM-based protein correlation profiling and subsequent filtering for LD enrichment versus the total proteome (Fig.S4A,   S4B). In the adipocyte LD proteomes, we detected most known LD proteins with most of these proteins involved in triglyceride and sterol metabolism (Fig.S4C). We identified 48 LD proteins that were common to both white adipocyte models, out of which 23 were specific to adipocytes To delve deeper into their functional significance, we integrated the core proteome data of adipogenesis with the adipocyte-specific LD proteome. Our analysis revealed 10 LD proteins that exhibited significant regulation during adipogenesis. Among these candidates, LGALS12 and C19orf12 particularly captured our attention due to their highly conserved temporal trajectory across all four human adipogenesis models ( Fig.6B and C) and their substantial increases in expression exceeded those of other candidate proteins in both cell lines and pACs when compared to the stromal vascular fractions SVFs, suggesting a potential functional role in adipocyte lipid storage ( Fig.6D and Fig.S4D).
LGALS12, an exclusively adipose tissue-expressed protein, has previously been associated with visceral adipose tissue mass 23 and demonstrated to influence lipolytic signaling on LDs. Our analysis revealed two isoforms of LGALS12, of which only one isoform (isoform2) being induced during adipogenesis and localizing to LDs (Fig.6E).
C19orf12, is associated with neurodegenerative diseases 24 , but has a genetic association with body mass index (Fig.S4E) and shows cell type enriched expression for adipocytes 25 . However, its localization and function in adipocytes remain unknown. To address this question, we conducted a series of experiments to investigate its localization and potential role in lipid storage.
Firstly, we confirmed the PCP predicted localization of C19orf12 at LDs ( Figure 6F and S4F) by immunofluorescence staining. We observed characteristic ring-like structures formed by C19orf12 around LDs stained by BODIPY (Fig.6G). Next, we performed co-immunoprecipitation coupled with proteomics to map its interactome in both preadipocytes and mature adipocytes. Notably, we identified an interaction between C19orf12 and the TOM complex, which is responsible for mitochondrial protein import (Fig.6H, S4G). This interaction suggests that C19orf12 may function at LDs-mitochondrial interaction sites. Therefore, to examine the subcellular localization of C19orf12 in relation to mitochondria, we conducted co-immunostainings using for C19orf12 and the mitochondrial marker TOM20. These experiments revealed a colocalization of C19orf12 with mitochondria in close proximity to LDs at the contact regions between these organelles ( Fig.6I and J).
Next, we sought to explore the functional significance of C19orf12 for adipocyte function and lipid storage at the cellular level. We performed knockdowns by reverse transfection with siRNA in preadipocytes prior to their differentiation, using an optimized protocol 26 . The knockdown resulted in a sustained decrease in C19orf12 protein levels throughout the differentiation process

DISCUSSION
Here, we address a major gap in our current understanding of human adipogenesis, shedding light onto the cellular and metabolic changes that occur during the formation of fat cells. By analyzing temporal proteomics data from multiple models of human adipogenesis and incorporating spatial proteomics, we have uncovered a highly coordinated process of cellular remodeling. Our findings offer a high-resolution view of the sequential changes in protein isoforms, abundance, and organelle organization, shedding light on so far unseen spatial organization of metabolic processes.
One particularly significant finding is the identification of the spatial organization of lipid metabolism enzymes within larger assemblies that exhibit distinct fractionation characteristics compared to soluble cytosolic proteins. Notably, all enzymes involved in de novo lipogenesis display nearly identical profiles in the proteomic co-fractionation, suggesting the condensation of diffusely localized enzymes into discrete foci. Indeed, phase separation is a frequently employed mechanism for controlling the biochemical activity of these enzymes and maintaining metabolic homeostasis. Recent studies have described the structures of human ACACA filaments, revealing that ACACA-citrate dimers polymerize into a polymer with high enzymatic activity in the presence of citrate 28 . However, our work suggests the formation of even larger condensates of all enzymes throughout de novo lipogenesis, which might enable efficient substrate channeling and serve as a means to regulate substrate flux through the pathway.
An important aspect highlighted by our study is the role of protein localization changes in cellular differentiation. We have discovered that a striking 20% of all proteins translocate during adipogenesis, thereby implying that protein translocation may play a more significant and general role in cellular differentiation processes. Several of these events may contribute to cellular specialization for lipid storage. Notably, we have observed the assembly of the NATC complex, responsible for modifying mitochondrial proteins through N-terminal acetylation on ribosomes.
Previous research has linked NATC to the modification of mitochondrial proteins, which are highly upregulated during adipogenesis 13 . This suggests that the assembly of the NATC complex may play a supportive role in facilitating adipogenesis by ensuring the proper modification of mitochondrial proteins.
By addressing the previously unexplored spatial aspects of adipogenesis, our study has revealed a sophisticated remodeling of protein localization and abundance in pathways that contribute to de novo lipogenesis. Notably, we observed a diversion of BCAA catabolism away from cytosolic nucleotide biosynthesis towards mitochondrial degradation for citrate and acetyl-CoA synthesis.
The translocation of BCAT2 from the cytosol to mitochondria, where it assembles with downstream enzymes into a metabolome complex, may enhance the flux through the mitochondrial part of the pathway 29 , which is highly upregulated during adipogenesis. Similarly, we have identified a counter regulation and localization changes of enzymes involved in mitochondrial and cytosolic onecarbon metabolism. This coordinated regulation may result in the redirection of the cytosolic part of the cycle towards serine degradation and NADPH production, similar as it has been reported for the ablation of mitochondrial one-carbon metabolism enzymes 17 and thereby contributing to cellular NADPH pools necessary for providing reduction equivalents for fatty acid synthesis Our spatiotemporal atlas of adipogenesis has revealed novel factors that are crucial for adipocyte biology. We uncovered that C19orf12 localizes specifically to the contact sites between lipid droplets (LDs) and mitochondria, where it interacts with the mitochondrial protein import machinery. These LD-mitochondrial contact sites have been recognized for their importance in facilitating the transfer of fatty acids released from LDs into mitochondria for beta oxidation or reciprocally, providing mitochondrial-generated ATP for triglyceride storage. Moreover, In summary, we have generated a comprehensive cellular map of human adipocytes, encompassing annotations for 8,000 proteins and their changes during adipogenesis. This resource offers researchers in the fields of lipid droplets and adipogenesis a platform to analyze protein expression, metabolic pathways, and organelle composition throughout adipogenesis.

DATA AVAILABILITY
Source data for each of the figures are provided with this paper. Proteomic raw data and Spectronaut search tables are available via ProteomeXchange under the identifier PXD043338. Username: reviewer_pxd043338@ebi.ac.uk Password: 0ISnZgLL

Human sample acquisition
Samples from subcutaneous abdominal WAT was obtained by needle aspiration under local anesthesia (as described 31 ) from five women and two men (mean standard deviation for age 60.73.5 years and body mass index 28.76.6 kg/m 2 ). Mature fat cells and SVF were isolated by collagenase digestion as previously described 4 . All studies were approved by the Regional Board of Ethics in Stockholm and all participants provided informed written consent.

Cell immortalization
To immortalize hAPCs, we stably integrated the EF-1 alpha promoter driving the expression of  mM chloroacetamide (CAA)) in the same conditions but in the dark. The combined supernatants were then digested overnight at 25°C, 1000rpm. The next day peptides were acidified using 1µl trifluoroacetic acid and purified on C18 stage tips as described before 33 .

Human genetic association analysis
Human genetic association analysis was performed using MAGMA (Multi-marker Analysis of GenoMic Annotation) 34 scores in the LDKP 22 . MAGMA scores for genes of proteins with lipiddroplet localization during adipogenesis were plotted for adiposity-related phenotypes as indicated in the figures. A significance threshold of p < 2.5 x 10 -6 is generally considered significant for MAGMA. A more stringent significance threshold of p < 3.125 x 10 -7 was derived by Bonferroni correction to account for the 8 phenotypes tested. Manhattan plots were generated in R using ggplot2.

C19orf12 Clinical Correlations
Expression level of C19orf12 was retrieved from a previously published WAT microarray study compromising 56 women with or without obesity 27 . This microarray data is publicly available in the NCBI Gene Expression Omnibus repository under the accession numbers GSE25401. In brief, C19orf12 expression was correlated to clinical parameters using the cor function in R (method = ''spearman'').

Proteome Analysis
For proteome analysis quantified proteins were filtered for at least two valid values among three

Protein Correlation Profiling Analysis
LFQ intensities for each protein among the organellar fractions were scaled from 0-1. To the fraction with the maximum intensity, the value of 1 was assigned, whereas fractions in which the protein was not quantified were set to 0. For the generation of median protein, the median values from the biological replicates for each fraction were calculated and a second 0-1 scaling step was performed. Pearson correlations between profiles of biological replicates were calculated for each protein and proteins with a correlation <0 in any of the comparisons were discarded.

SVM-Based Organelle prediction
Marker proteins were selected based on their documented GO-annotations and robust quantification in our dataset for main cellular compartments and protein complexes. The set was used for parameter optimization and training of the SVM based supervised learning approach implemented in Perseus software 35 . Parameters were set to Sigma=0.2 and C=8. For SGBS mature adipocytes, four and for the preadipocytes three and hAPC cells one biological replicate were used for organelle classifications using the second organelle assignment option in the Perseus software Version 1.5.6.2. Organelle predictions were filtered for positive assignment to at least one organelle. The correlation value between the experimentally determined protein profile and the assigned in silico generated combination profile is given as measure for the quality of organelle assignment. The alpha value (0-1) is a quantitative measure for the second organelle contribution. Proteins with alpha=0 were defined as single organelle localizing proteins.    In the overlay, DAPI is shown in blue, BODIPY in gray, SLC25A10 in green and TOM20 in magenta. Scale bar = 25µm. for ALDH1L1 and ALDH1L2 in preadipocytes and adipocytes, respectively. In the overlay, BODIPY is shown in grey, ALDH1L1 and ALDH1A2 in green and TOM20 in magenta. Scale bar = 25µm and 10µm in inlay.   LGALS12 Iso2