PPARG governs adipogenic differentiation and cell state plasticity in well-differentiated and dedifferentiated liposarcoma

(153/160 words) 26 Well-differentiated and dedifferentiated liposarcoma (WD/DD LPS) represent a pathological 27 continuum, often coexisting within the same tumor. While the dedifferentiated component is 28 clinically aggressive, marked by rapid growth and metastatic potential, the evolutionary 29 relationship between WD and DD LPS remains unknown. To investigate this, we performed 30 single-nucleus RNA sequencing on matched WD and DD tumor regions. Both compartments 31 shared a predominant population of undifferentiated mesenchymal cells, but only WD regions 32 contained cells expressing adipocytic differentiation markers and PPARG target genes. Given the 33 central role of PPARG in coordinating lipid metabolism and mitochondrial biogenesis during 34 adipogenesis, these findings suggest that loss of this program may underlie the poorly 35 differentiated, proliferative phenotype of DD LPS. Functional studies confirmed that PPARG 36 activation in DD LPS cells induces lipid accumulation, reduces proliferation, and impairs tumor 37 growth in vivo. These support a model in which impaired adipogenic differentiation underlies DD 38 LPS pathology and identify PPARG as a potential therapeutic target to promote differentiation 39 and suppress tumor progression. 40 41 Teaser (92/130 characters) 42 43 PPARG reprograms DD liposarcoma toward adipogenesis, reducing proliferation and tumor 44 growth 45 46 MAIN TEXT (/15,000 words) 47 48

49 50 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 2 of 17 Liposarcoma (LPS) is the most common soft tissue sarcoma subtype in adults, and the 51 incidence is rising(1, 2). Understqanding of its biology lags behind other malignancies and 52 treatment options remain limited(3-5). Surgical resection is the primary treatment, but recurrence 53 occurs in more than 50% of cases and there are currently no curative options for recurrent disease 54 (6-8). These challenges underscore the urgent need for novel therapeutic strategies. 55 The World Health Organization (WHO) classifies LPS into five major subtypes: well-56 differentiated (WD), dedifferentiated (DD), myxoid, pleomorphic, and myxoid-pleomorphic. 57 Among these, WD and DD LPS are the most prevalent and are increasingly recognized as a 58 spectrum of the same disease—collectively termed WD/DD LPS—due to shared defining 59 genomic feature: amplification of the 12q13-15 locus, including MDM2(9, 10). 60 Clinically, WD/DD LPS manifests in various forms: as a homogeneous WD or DD LPS 61 tumor, or as a tumor with distinct components of both WD and DD LPS (11, 12). Although WD 62 LPS is generally indolent, some cases progress rapidly to DD LPS(13). Conversely, DD LPS—63 while generally associated with a high metastatic potential and a six-fold increased risk of 64 death—can occasionally follow a more protracted course(14-18). The observation that tumors can 65 recur with a histology different from the original subtype, including instances of DD LPS 66 recurring as WD, suggests that tumor differentiation exists along a dynamic continuum rather 67 than as a fixed binary state(12). These clinical scenarios suggest a dynamic, bidirectional 68 differentiated-dedifferentiated state, but the underlying molecular mechanisms remain poorly 69 understood. This clinical heterogeneity also indicates that histologic classification alone is 70 insufficient for the prognostication that drives treatment decisions. Molecular markers such as 71 MDM2 amplification level and IGF2BP3 expression have shown improved prognostic value, 72 emphasizing the need for a deeper understanding of tumor evolution in WD/DD LPS (19, 20). 73 Histologically, WD LPS is composed of mature adipocytes and lipoblasts with nuclear 74 atypia, while DD LPS consists of undifferentiated spindle cells with high mitotic rate and reduced 75 expression of adipogenic markers (7, 13, 21). Both subtypes are thought to originate from 76 mesenchymal precursors such as adipocyte progenitor cells or preadipocytes. Genomic analyses 77 have revealed similar somatic mutation profiles between WD and DD LPS, yet DD tumors exhibit 78 greater genomic complexity, including additional copy number alterations beyond 12q13–15—79 such as amplifications at 5p and 14q and deletions at 11q23–24, 19q13, 3q29, 9p22–24, or 80 17q21—suggesting secondary genomic events may drive dedifferentiation(22-26). 81 Viewing WD/DD LPS as a dynamic spectrum provides a framework to improve diagnosis, 82 prognostication, and therapeutic development. Although bulk genetic profiling has yielded 83 valuable insights, it has yet to elucidate the molecular mechanisms that govern differentiation 84 state and tumor progression. Single-cell and spatial omics approaches offer a powerful means to 85 dissect intra-tumoral heterogeneity and trace tumor evolution. A recent study by Gruel et al. used 86 single-cell transcriptomics to demonstrate that both WD and DD components originate from a 87 shared adipocyte stem cell and that differentiation in DD LPS is suppressed by TGF-β 88 signaling(27). 89 Building on this work, we performed single-nucleus RNA sequencing (snRNA-seq) on 90 paired WD and DD tumor components from patients with WD/DD LPS. Our data revealed 91 significant inter-patient heterogeneity, but a shared core architecture: both WD and DD 92 compartments were predominantly composed of undifferentiated mesenchymal cells, with only 93 WD regions containing cells expressing markers of preadipocytes and mature adipocytes. 94 Pseudotime trajectory analysis revealed a clear transcriptional path toward adipogenic 95 differentiation in WD tumors, associated with activation of PPARG target genes. 96 These findings led us to test whether PPARG activation could reprogram DD LPS cells 97 toward a more differentiated, less aggressive state. We found that the splice variant PPARG2 98 promoted lipid accumulation, reduced proliferation in vitro, and suppressed tumor growth in vivo. 99 Moreover, pharmacologic activation of PPARG with rosiglitazone impaired the growth of DD 100 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 3 of 17 LPS xenografts. Together, these results suggest that loss of PPARG-driven differentiation 101 underlies dedifferentiation in LPS and that restoring this program may represent a promising 102 therapeutic strategy. 103 104

105 106 snRNA-seq identifies molecular heterogeneity in WD/DD LPS 107 108 The molecular basis of WD/DD LPS transition remains poorly understood, despite their 109 frequent co -occurrence within the same tumor. This spatial juxtaposition provides a unique 110 opportunity to investigate dedifferentiation in situ —offering direct insight into tumor evolution, 111 plasticity, and progression. We hypothesized that distinct transcriptional programs and 112 differentiation states underlie the histological differences between WD and DD components, and 113 that single-nucleus RNA sequencing (snRNA-seq) could resolve these programs at high resolution. 114 This approach is uniquely suited for LPS because it enables detection of mature adipocytes, which 115 are typically lost in single -cell RNA-seq using droplet -based platforms(28). This is particularly 116 important in LPS, where mature adipocytes and lipid-laden lipoblasts define WD histology and may 117 represent a differentiation endpoint or lineage anchor. Capturing these populations allows for more 118 accurate reconstruction of adipogenesis and identification of dedifferentiation events. 119 To identify tumors with clear histologic demarcation between WD and DD components, we 120 screened primary, untreated WD/DD LPS cases based on CT imaging and confirmed subtype 121 identity using histology and DNA -FISH ( Fig. 1A –C). Three tumors were selected for analysis 122 (Table 1 ). The tumors analyzed varied widely in size (8 –46 cm) and clinical behavior. This 123 heterogeneity reflects the clinical challenges posed by WD/DD LPS and underscores the need for 124 a deeper molecular understanding. WD components showed abundant mature adipocytes and 125 lipoblasts, while DD regions were composed of densely cellular, non -lipogenic spindle cells (Fig. 126 1B, Fig. S1A–B). MDM2 amplification was confirmed in both regions , validating their malignant 127 origin and enabling distinction between tumor and non-tumor nuclei in downstream analyses. 128 Following tissue dissociation and nuclei extraction, cDNA libraries were constructed and 129 sequenced. After read alignment and quality filtering using Cell Ranger and CellBender, we 130 integrated datasets by patient and subtype using Seurat (29-31). Dimensionality reduction and 131 clustering revealed 22 distinct transcriptional populations (Fig. 2A). Of these, 10 clusters expressed 132 canonical LPS oncogenes ( MDM2, CDK4, HMGA2) (Fig. 2B, Fig. S2), suggesting a tumor cell 133 identity. To validate this classification, we performed copy number inference, which revealed 134 characteristic gains on chromosome 12q13 –15—encompassing the MDM2 locus—specifically in 135 these clusters ( Fig. 2C). Inferred CNAs provided orthogonal confirmation of tumor identity and 136 helped distinguish malignant from stromal and immune populations. 137 To focus our analysis on tumor -intrinsic heterogeneity, we re -clustered MDM2-amplified 138 nuclei and identified nine transcriptionally distinct tumor cell states (Fig. 2D) (32). These included 139 proliferative clusters, metabolically active populations, and cells with mesenchymal, immune-like, 140 or adipogenic features (Fig. S3A -D). These data reveal extensive intra -tumoral diversity and 141 support the existence of discrete tumor cell states that may influence growth, recurrence, and 142 therapeutic response. 143 144 WD tumors uniquely harbor differentiated adipocyte-like cells 145 146 We next asked whether specific tumor cell populations were preferentially enriched in either 147 the WD or DD regions. Among the nine tumor -intrinsic clusters, two (clusters 5 and 9) stood out 148 for their high expression of adipocyte lineage markers including PPARG, FABP4, and ADIPOQ, 149 consistent with a differentiated adipogenic phenotype ( Fig. 3A -B). Notably, these cells lacked 150 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 4 of 17 expression of adipocyte progenitor markers such as COL6A1, COL6A2, and FKBP10, reinforcing 151 the idea that they are committed to adipocyte fate rather than representing an intermediate precursor 152 state (Fig. 3C). These adipocyte-like clusters accounted for approximately 9% of all tumor cells 153 and were detected exclusively within the WD component, suggesting they may represent a 154 terminally differentiated cell population (Fig. 3D, Fig. S3A). We refer to these as “WD-specific 155 clusters.” 156 To determine whether other tumor cell populations were similarly restricted to one 157 component or shared between both, we examined the distribution of all clusters across matched 158 WD and DD samples. Most clusters were present in both components, including proliferative, 159 mesenchymal-like, and metabolically active populations , indicating that despite histological 160 divergence, the majority of tumor cells share a conserved transcriptional architecture (Fig. 3E). We 161 refer to these as common clusters, as they exhibited similar abundance and transcriptional profiles 162 in WD and DD regions ( Fig. 3F). The presence of shared transcriptional states across spatially 163 distinct histologies suggests that WD and DD tumors retain a core transcriptional architecture, and 164 that differentiation states are not driven by global reprogramming but instead by selective 165 enrichment or loss of specific states. The key distinction between components was the exclusive 166 presence of the adipocyte -like, WD -specific clusters —a feature entirely absent from the DD 167 regions. 168 To investigate the lineage relationship between tumor cell populations, we performed copy 169 number inference on WD and DD tumor cells. In one representative tumor (patient #069), both WD 170 and DD components shared canonical gains on 12q and 7q, consistent with a common clonal origin. 171 However, only the WD -specific clusters harbored additional CNAs —including gains on 6q and 172 14q—suggesting these cells diverged from the shared progenitor through a distinct genetic 173 trajectory ( Fig. 3G ). This supports a model in which both WD - specific and common clusters 174 originate from a shared population of undifferentiated tumor cells, with differentiation governed by 175 acquisition of CNAs and/or activation of lineage-specific transcriptional programs. 176 It is unclear whether WD and DD represent distinct, parallel tumor lineages or reflect a 177 continuum in which one subtype gives rise to the other. Shared transcriptional architecture and copy 178 number alterations support a common clonal origin . However, the presence of additional CNAs 179 unique to the WD component —absent from DD —suggests the WD compartment may arise via 180 differentiation from a less differentiated DD -like progenitor (Fig. 3H). In this model, WD tumor 181 cells acquire new genetic features or engage specific transcriptional programs that enable 182 adipogenic differentiation and give rise to the histologically well -differentiated phenotype. The 183 complete absence of adipocyte-like cells in DD regions further supports the idea that differentiation, 184 rather than dedifferentiation, explains the emergence of WD histology in some cases. This concept 185 is supported by r ecent studies that show that both WD and DD components arise from a shared 186 adipocyte stem cell –like progenitor(27). Understanding the mechanisms that constrain or permit 187 this differentiation could provide new opportunities to manipulate tumor cell state with novel 188 systemic therapies. 189 190 Pseudotime analysis implicates PPARG in adipogenic differentiation 191 192 We next sought to identify the regulatory programs that govern the transition from 193 undifferentiated tumor cells to adipocyte -like states. Given the restricted presence of adipogenic 194 clusters in WD tumors, we hypothesized that WD/DD subtype divergence reflects altered 195 transcriptional trajectories rather than the presence of entirely distinct lineages. To test this, we 196 performed pseudotime trajectory analysis using Monocle 3, anchoring the trajectory at the WD -197 specific adipocyte-like clusters as a terminal node(33). 198 Trajectory reconstruction revealed a complex network of potential paths originating within 199 the common clusters, which comprised transcriptionally heterogeneous populations with the 200 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 5 of 17 capacity to adopt multiple fates. Within this network, we identified a continuous trajectory linking 201 the common clusters to the WD-specific adipocyte-like clusters, suggesting a directed progression 202 toward adipogenic differentiation ( Fig. 4A). Along this path, we observed a gradual decrease in 203 MDM2 expression and a corresponding increase in HMGA2, two markers historically associated 204 with DD and WD histology, respectively ( Fig. 4B). These findings support a model in which the 205 common clusters serve as a transitional hub, giving rise to either differentiated adipocyte -like 206 cells—represented by the WD-specific clusters—or to alternative, non-lipogenic tumor states. 207 To gain insight into the transcriptional programs driving this transition, we identified genes 208 with significant dynamic expression across pseudotime and organized them into co -expression 209 modules (Fig. 4C). One module —designated group 3 —was characterized by low expression in 210 early pseudotime and progressive upregulation toward the terminal WD -specific clusters. This 211 pattern is consistent with a differentiation trajectory from common clusters to a mature adipocyte -212 like state and likely reflects genes activated during the acquisition of adipogenic identity. Notably, 213 group 3 was enriched for known transcriptional targets of PPARG and other key regulators of 214 adipogenesis. These findings suggest that adipogenic differentiation in WD tumors is orchestrated 215 by a coordinated transcriptional program and that loss or failure to activate this program may 216 underlie the dedifferentiated state in DD tumors. Given that PPARG is both necessary and sufficient 217 for adipocyte differentiation in normal and malignant contexts, these data raise the possibility that 218 reactivating PPARG signaling could restore differentiation capacity in DD LPS cells, offering a 219 potential therapeutic strategy(34). 220 221 PPARG2 promotes adipogenic differentiation and suppresses tumor growth in DD LPS 222 223 To explore whether PPARG activation might drive adipogenic differentiation in 224 dedifferentiated liposarcoma (DD LPS), we first generated a PPARG gene score based on canonical 225 PPARG transcriptional targets. This score was significantly enriched in the WD-specific adipocyte-226 like clusters identified in our single -nucleus RNA-seq dataset ( Fig. 5A), consistent with the idea 227 that PPARG activity marks a terminally differentiated state. These findings suggested that PPARG 228 is a key regulator of differentiation in LPS and prompted us to test whether activating this pathway 229 could reprogram DD tumor cells. 230 To functionally test whether PPARG activation is sufficient to induce adipogenic 231 differentiation in DD LPS, we overexpressed its two major isoforms —PPARG1, a ligand -232 dependent isoform, and PPARG2, a constitutively active variant with enhanced adipogenic 233 activity—in two DD LPS cell lines (LPS1 and LPS2) (Fig. 5B). Both isoforms were expressed from 234 the same lentiviral backbone, but PPARG2 protein levels were higher, consistent with prior reports 235 showing that PPARG2 is more stable than PPARG1 in mammalian cells (35, 36). Functionally, 236 PPARG2 induced strong expression of FABP4, a canonical PPARG target gene and adipogenic 237 marker, and promoted lipid droplet accumulation (Fig. 5C–D) (35). By contrast, PPARG1 had only 238 modest effects on differentiation marker expression and lipid accumulation, suggesting that 239 endogenous ligand availability may be insufficient to activate this isoform under baseline 240 conditions. In addition to driving differentiation, PPARG2 expression significantly suppressed 241 proliferation in both cell lines (Fig. 5E), consistent with induction of a terminal differentiation state. 242 These findings demonstrate that PPARG2 is sufficient to both activate adipogenic gene programs 243 and impair tumor cell growth in vitro, reinforcing its role as a central regulator of lineage fate in 244 LPS. 245 To assess the therapeutic relevance of PPARG activation in vivo, we generated 246 doxycycline-inducible LPS2 cells expressing either PPARG1 or PPARG2 and implanted them 247 subcutaneously into immunocompromised mice (Fig. 6A). Tumor growth was monitored by caliper 248 measurements, and doxycycline -containing chow was introduced once tumors reached 100 –200 249 mm³. Induction of PPARG2 led to a significant reduction in tumor growth, while PPARG1 250 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 6 of 17 induction had no measurable effect ( Fig. 6B ). Histological analysis of the xenografts revealed 251 heterogeneous tumor architecture, including the presence of lipid -rich regions reminiscent of the 252 WD components seen in human WD/DD LPS tumors (Fig. S5A). Although we did not observe 253 significant differences in the overall percentage of lipid -rich area between groups, this is likely 254 confounded by the growth-suppressive effects of PPARG2, which limited overall tumor expansion 255 and may have reduced the opportunity for detectable differentiation to accumulate at the tissue level 256 (Fig. S5B). 257 To test whether the limited activity of PPARG1 in vivo reflected insufficient ligand 258 availability, we treated mice bearing parental LPS2 xenografts with rosiglitazone, a synthetic 259 thiazolidinedione-class PPARG agonist. In adipose tissue, ligand -activated PPARG1 not only 260 drives adipogenic gene expression but also promotes a feed -forward loop by initiating PPARG2 261 transcription through a conserved PPARG response element(37). Consistent with this mechanism, 262 daily rosiglitazone treatment significantly slowed tumor growth ( Fig. S6A-B), indicating that 263 pharmacologic activation of PPARG1 can partially recapitulate the effects of PPARG2 in vivo. 264 These findings suggest that DD LPS cells retain a latent capacity to engage the PPARG 265 differentiation program, but fail to do so due to insufficient endogenous ligand activity. 266 Pharmacologic activation may therefore bypass this block and re-engage differentiation pathways. 267 While the effect size was more modest than that observed with direct PPARG2 expression, these 268

provide proof -of-concept that PPARG agonists can restrain tumor growth by shifting cell 269 state, and suggest a therapeutic opportunity to restore adipogenic differentiation using clinically 270 accessible compounds. 271 Finally, we examined expression of PPARG isoforms in primary human tumors. While 272 PPARG1 levels were comparable across WD and DD samples, PPARG2 expression was 273 significantly enriched in WD tumors and in the WD components of mixed WD/DD cases (Fig. 6C–274 D, Table 2). Together, these results support a model in which PPARG2 serves as a key driver of 275 adipogenic differentiation in liposarcoma and suggest that restoring PPARG activity may offer a 276 strategy to suppress proliferation and promote differentiation in DD tumors. 277 278

279 280 Well-differentiated and dedifferentiated liposarcomas (WD/DD LPS) are defined by 281 common amplification of 12q13 –15 oncogenes MDM2 and CDK4, and can co -exist in the same 282 patient, in the same tumor, with disparate behavior along this continuum of disease, both lipoma -283 like indolent and high -grade rapidly fatal components. The coexistence of WD and DD lineages 284 within a single tumor and the capacity for tumors to shift between these states over time suggests a 285 dynamic process of differentiation and dedifferentiation. Yet, the molecular mechanisms governing 286 these transitions remain poorly understood. 287 Using snRNA-seq, we defined the transcriptional landscape of spatially distinct WD and 288 DD components from patient tumors. While most tumor cells in both compartments shared a 289 common mesenchymal -like architecture, a rare population of adipocyte -like tumor cells was 290 identified exclusively in WD regions. These cells expressed canonical adipogenic markers and 291 formed a distinct transcriptional cluster, consistent with a terminal differentiation state. Pseudotime 292 trajectory analysis revealed a unidirectional path from shared progenitor -like clusters to this 293 adipocyte-like population, supporting a model in which the WD component may emerge from a 294 less differentiated tumor state. 295 We identified PPARG as a key regulator of this transition. A PPARG gene score was 296 enriched in the WD -specific clusters, and pseudotime -regulated genes were enriched for PPARG 297 targets and adipogenic modules. Among the PPARG isoforms, PPARG2 —a constitutively active 298 splice variant—was sufficient to induce lipid accumulation and suppress proliferation in DD LPS 299 cells. In vivo , inducible expression of PPARG2 impaired tumor growth, and pharmacologic 300 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 7 of 17 activation of PPARG1 with rosiglitazone partially recapitulated this effect, potentially through 301 feed-forward activation of PPARG2 expression. These findings suggest that PPARG signaling 302 governs lineage commitment in LPS and that loss of this program contributes to the dedifferentiated 303 state. 304 Importantly, our study suggests that subtype transitions in WD/DD LPS are not solely 305 dictated by genomic divergence. Although some copy number alterations were unique to WD 306 regions, many gene expression differences—particularly those related to differentiation—occurred 307 independently of CNAs. These results align with recent work demonstrating that both WD and DD 308 components arise from a common adipocyte stem cell –like progenitor and that DD tumor cells 309 retain latent adipogenic potential (27). In that study, adipogenic differentiation of DD cells was 310 actively suppressed by a TGF-β–rich microenvironment, suggesting that extrinsic cues can enforce 311 a dedifferentiated state. Complementary work has shown that PPARG2 and its downstream targets 312 are epigenetically silenced in DD LPS through hypermethylation of adipogenic super -313 enhancers(38). Treatment with the DNA demethylating agent 5 -aza-2’-deoxycytidine and the 314 PPARG agonist rosiglitazone restored PPARG2 expression and induced adipogenic differentiation 315 in DD tumor cells. Together, these findings reinforce the idea that both intrinsic transcriptional 316 programs and extrinsic environmental signals constrain differentiation in LPS and that therapeutic 317 reactivation of PPARG2 may overcome these blocks to restore a more differentiated, less 318 aggressive phenotype. 319 Together, these data support a model in which PPARG2 serves as a lineage -defining 320 regulator in LPS, and its loss —through transcriptional suppression and/or epigenetic silencing —321 permits or promotes the transition to a dedifferentiated state. This model aligns with clinical 322 behavior: WD and DD can co -exist within the same tumor or interconvert during disease 323 progression or recurrence. Importantly, our findings suggest that subtype transitions in WD/DD 324 LPS are not solely dictated by genomic divergence. Although some CNAs distinguish WD and DD 325 components, many gene expression differences —particularly those related to differentiation —326 occur independently of copy number alterations. This further underscores the role of epigenetic and 327 transcriptional regulation in shaping tumor cell identity. 328 Strategies that restore PPARG2 expression —through either demethylating agents, nuclear 329 receptor agonists, or chromatin remodeling —may provide a means to enforce differentiation and 330 limit aggressiveness in DD LPS. By converting a poorly differentiated, proliferative tumor into a 331 more indolent, adipocyte-like state, it may be possible to reduce recurrence, delay progression, and 332 improve survival. This approach mirrors successful differentiation therapies in other cancers, such 333 as ATRA in acute promyelocytic leukemia. 334 More broadly, this study highlights the power of single -cell and epigenomic profiling to 335 uncover lineage hierarchies and therapeutic vulnerabilities in mesenchymal tumors. WD/DD LPS 336 has long presented a clinical paradox, with tumors able to recur as more or less differentiated 337 subtypes. Our findings provide a mechanistic framework to explain these transitions and nominate 338 PPARG2 as a therapeutic entry point for differentiation -based strategies. Understanding how 339 PPARG activity is regulated —and how it can be restored —will be essential for translating this 340 approach into clinical benefit. 341 342

343 344 Experimental Design 345 346 The objective of this study was to define cellular heterogeneity, differentiation states, and 347 subtype transitions in well-differentiated and dedifferentiated liposarcoma (WD/DD LPS) using 348 single-nucleus RNA sequencing (snRNA-seq). We aimed to uncover lineage relationships 349 between WD and DD tumor components, identify molecular features of dedifferentiation, and 350 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 8 of 17 reveal potential therapeutic vulnerabilities. To achieve this, we collected freshly resected tumor 351 specimens from patients with histologically confirmed WD/DD LPS. Tumors with both WD and 352 DD regions were macrodissected and processed independently to preserve distinct transcriptional 353 programs. Nuclei were isolated and profiled using the 10x Genomics Chromium platform. 354 Sequencing data were analyzed using a standardized computational pipeline for quality control, 355 dimensionality reduction, clustering, and differential gene expression. 356 357 Patients 358 359 This study was conducted in accordance with the principles expressed in the Declaration 360 of Helsinki. Informed consent was provided by all patients or guardians granting access to tumor 361 tissue, serum and medical records. Databases including patient identifiers were maintained 362 according to our local institutional guidelines. The research was conducted under protocol #10-363 001857, approved by the UCLA Institutional Review Board (IRB). 364 Patients with primary, untreated WD/DD LPS undergoing initial surgical resection were 365 selected for inclusion. Preoperative imaging was used to assist with patient selection. Patients 366 with history of systemic chemotherapy, radiation or prior resection were excluded. 367 Clinicopathologic data was collected and stored in an encrypted database and maintained by the 368 study authors (KK, BK). 369 370 Tumor harvest, storage and validation 371 372 Fresh tissue was harvested at the time of surgery from both WD and DD components and 373 stored immediately in liquid nitrogen. Additional tissue from both components was placed in 374 formalin for permanent fixation and then embedded in paraffin. Tumor diagnosis and subtype was 375 validated by expert histological review and DNA FISH (Empire Genomics, SKU MDM2-376 CHR12-20-GROR) was used to confirm the presence of MDM2 amplification (authors SD, MN). 377

have been previously described(39). Tumors without MDM2 amplification were 378 excluded. 379 380 Immunohistochemistry 381 382 Formalin-fixed sections embedded in paraffin were cut at 4 μm thickness. Samples were 383 submerged in xylene for paraffin removal and then rehydrated using graded ethanol washes. 384 Sections were counterstained with hematoxylin and eosin. Brightfield slides were digitally 385 scanned on a ScanScope AT2 (Leica Biosystems, Vista, CA, USA) and analyzed using QuPath 386 version 0.2.3. 387 388 Nuclei isolation and cDNA library generation 389 390 After tumor diagnosis and subtype were confirmed, samples were thawed in RNALaterTM-391 ICE (ThermoFisher) overnight at -20°C. Nuclei were then isolated using the Chromium Nuclei 392 Isolation with RNase Inhibitor Kit (10X Genomics). For DD components, 50mg of tissue was 393 used and for WD components, 250mg of tissue was used from each sample. Nuclei concentration 394 and cell viability was determined using a Countess II FL Automated Cell Counter 395 (ThermoFisher). Cell viability <5% was used as a threshold to ensure high quality nuclei 396 isolation. Samples were also assessed by brightfield microscopy for the presence of significant 397 debris. Single nuclei suspensions generated from each sample were then used to construct 3’GEX 398 cDNA libraries (10x Genomics). 399 400 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 9 of 17 Single nuclei sequencing and processing 401 402 Single nuclei suspensions generated from each sample were used to construct 3’GEX 403 cDNA libraries (10x Genomics) followed by next-generation sequencing via NovaSeq 6000 404 (Illumina). Demultiplexed sequencing results were aligned to the reference genome and ambient 405 RNA was removed using CellBender. After filtering, 27,333 nuclei were included in the 406 computational analysis pipeline. The R package Seurat was applied to integrate samples, cluster 407 cells, and identify differentially expressed genes. LPS clusters were analyzed by pseudo-time 408 analysis (Monocle3) to establish trajectory inferences between WD and DD subtypes. 409 410 Cell lines 411 412 LPS1 and LPS2 are derived from DD LPS patient-derived xenografts, which have been 413 previously validated(40). Both lines were cultured in DMEM with 10% FBS and 414 penicillin/streptomycin, and maintained in a 37°C humidified, normoxic chamber supplemented 415 with 5% CO2. Cells were monitored regularly for the presence of mycoplasma. 416 417 Plasmids 418 419 TFORF3549 (pLX317-EGFP) was a gift from Feng Zhang (Addgene plasmid # 145025 ; 420 http://n2t.net/addgene:145025 ; RRID:Addgene_145025). TFORF3550 (pLX317-mCherry) was a 421 gift from Feng Zhang (Addgene plasmid # 145026 ; http://n2t.net/addgene:145026 ; 422 RRID:Addgene_145026). TFORF3138 (pLX317-PPARG1) was a gift from Feng Zhang 423 (Addgene plasmid # 144614 ; http://n2t.net/addgene:144614 ; RRID:Addgene_144614). 424 TFORF3139 (pLX317-PPARG2) was a gift from Feng Zhang (Addgene plasmid # 144615 ; 425 http://n2t.net/addgene:144615 ; RRID:Addgene_144615). pRSV-Rev was a gift from Didier 426 Trono (Addgene plasmid #12253; http://n2t.net/addgene:12253; RRID:Addgene_12253). 427 pMDLg/pRRE was a gift from Didier Trono (Addgene plasmid #12251; 428 http://n2t.net/addgene:12251; RRID:Addgene_12251). pCMV-VSV-G was a gift from Bob 429 Weinberg (Addgene plasmid #8454; http://n2t.net/addgene:8454; RRID:Addgene_8454). 430 Plasmids for inducible expression of mCherry, PPARG1, and PPARG2 were generated 431 using standard molecular cloning techniques (pCW57.1-mCherry, pCW57.1-PPARG1, and 432 pCW57.1-PPARG2). 433 434 Lentivirus Production and Transduction 435 436 Lentivirus was produced in HEK293T cells infected with pLX317 transfer plasmid, 437 pRSV-Rev, pMDLg/pRRE, and pCMV-VSV-G. At 2 days post-transfection, supernatant was 438 collected and filtered at 0.45 µm, distributed into 1 mL aliquots, then froze at -80C for future use. 439 For transduction, aliquots of virus were thawed and polybrene was added to 4 µg/mL. The 440 virus/polybrene mixture was added to PBS-rinsed cells and incubated for 16 hours. At that point 441 the virus/polybrene was replaced with fresh media. After 24 hours, selection was started by 442 treating cells with 2 µg/mL puromycin. 443 444 RT-qPCR 445 446 For cell lines, total RNA was harvested from approximately 300,000 cells. Cells were 447 washed with ice cold PBS then washed with RNAlater (Invitrogen). After aspirating the 448 RNAlater, cells were frozen at -80C for processing at a later point. Cells were thawed and total 449 RNA was collected using a Direct-zol RNA Miniprep kit (Zymo). 450 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 10 of 17 For patient tumors, approximately 10 mg was incubated in RNAlater-ICE (Invitrogen) 451 overnight at -80 °C. The tumor was removed from the RNAlater-ICE, and homogenized in 1 mL 452 trizol. Total RNA was isolated from the homogenate using RNeasy lipid tissue kit (Qiagen). 453 For both cell lines and patient tumors 1 µg RNA was reverse transcribed using iScriptTM 454 Reverse Transcription Supermix (BioRad). qPCR reactions were up using Power SYBR Green 455 PCR Master Mix (Applied Biosystems). For PPARG1 and PPARG2 measurements, unique 456 forward and a common reverse primers were used – PPARG1-fwd: 5’-457 GCCATTTTCTCAAACGAGAGTCAGCC-3’; PPARG2-fwd: 5’-458 TGACCCAGAAAGCGATTCCTTCA-3’; PPARG1/2-rev: 5’-459 ACGGAGAGATCCACGGAGCTGA-3’. For FABP4, the following primers were used – 460 FABP4-fwd: 5’-ACGAGAGGATGATAAACTGGTGG-3’; FABP4-rev: 5’-461 GCGAACTTCAGTCCAGGTCAAC-3’. 462 463 Lipid content quantification 464 465 Live cells were stained with 1 µM Bodipy 493/503 (Invitrogen) and 0.2 µg/mL Hoechst-466 33342 (Invitrogen). Cells were incubated in media containing dyes at 37°C for 10 min, then 467 washed well three times with PBS. Cells were imaged using an Evos Cell Imaging System. 468 QuPath was used to analyze images for fluorescent intensity. In short, 5 cells from each 469 sample and 3 samples from each condition were analyzed for a total of 15 cells per condition. 470 Mean Bodipy fluorescent intensity was determined from each region of interest. 471 472 Proliferation assays 473 474 Cells were seeded at a density of 20,000 cells/well in 6-well dishes. After 24 hours cells 475 were counted and used as day 0 measurements. At day 3 cells were counted from separate wells 476 and the data was fitted to an exponential growth model. The fitted growth rates were presented as 477 the proliferation rates with units of divisions/day. 478 479 Mouse xenografts 480 481 LPS2 cells were used to establish tumor xenografts. A total of 1,000,000 cells in a 1:1 482 mixture of Matrigel and PBS were injected into the one side of 7-week-old NSG mice. Tumors 483 were monitored for growth by calipering. Tumor volumes were calculated by the equation (length 484 x width2)/2. When tumors reach 100-200 mm3, treatment was initiated. Rosiglitazone was 485 administered via oral gavage daily at a concentration of 20 mg/kg. Mice were weighed daily and 486 exhibited no significant weight loss due to rosiglitazone treatment. Mice were euthanized once 487 tumors reached 2000 mm3. 488 LPS2 cells expressing doxycycline-inducible mCherry, PPARG1, or PPARG2 were used 489 to establish tumor xenografts. A total of 1,000,000 cells in a 1:1 mixture of Matrigel and PBS 490 were injected into the one side of 7-week-old NSG mice. Once tumors reached 100-200 mm3 all 491 mice were fed chow containing 625 mg/kg doxycycline hyclate (Envigo). Mice were euthanized 492 once tumors reached 2000 mm3. 493 494 Statistical Analysis 495 496 Statistical significance was established using a P value threshold < 0.05 with 95% 497 confidence intervals. Continuous, normally distributed data were evaluated using the two-tailed 498 T-tests for pairwise comparisons and ANOVA for comparisons involving multiple groups. The 499 normality of the data was confirmed using quantile-quantile (Q-Q) plots. The Bonferroni test was 500 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 11 of 17 employed for post-hoc analysis to identify specific group differences. Data lacking normal 501 distribution was assessed using the Mann-Whitney U test. All experiments were carried out in 502 duplicate or triplicate to ensure reliability. Statistical analyses were performed using Graphpad 503 Prism software, version 9.3.1, on a MacOS platform. Unless otherwise indicated, data are 504 reported as mean ± standard deviation (SD). 505 506

507 508 1. A. P. D. Tos, Liposarcoma: New entities and evolving concepts. Annals of Diagnostic 509 Pathology 4, (2000/08/01). 510 2. S. Bock, D. G. Hoffmann, Y. Jiang, H. Chen, D. Il’Yasova, Increasing Incidence of 511 Liposarcoma: A Population-Based Study of National Surveillance Databases, 2001–2016. 512 International Journal of Environmental Research and Public Health 17, 2710 (2020). 513 3. I. Judson et al., Doxorubicin alone versus intensified doxorubicin plus ifosfamide for first-514 line treatment of advanced or metastatic soft-tissue sarcoma: a randomised controlled 515 phase 3 trial. The Lancet Oncology 15, 415-423 (2014). 516 4. E. C. Borden et al., Randomized comparison of three adriamycin regimens for metastatic 517 soft tissue sarcomas. Journal of Clinical Oncology 5, (1987-Jun). 518 5. V. Bramwell, D. Anderson, M. Charette, Doxorubicin-based chemotherapy for the 519 palliative treatment of adult patients with locally advanced or metastatic soft tissue 520 sarcoma. Cochrane Database of Systematic Reviews 2019, (2003). 521 6. K. Ishii et al., Characteristics of primary and repeated recurrent retroperitoneal 522 liposarcoma: outcomes after aggressive surgeries at a single institution. Japanese Journal 523 of Clinical Oncology 50, (2020/12/16). 524 7. T. K, Well-differentiated liposarcoma and dedifferentiated liposarcoma: An updated 525 review - PubMed. Seminars in diagnostic pathology 36, (2019 Mar). 526 8. J. Chen, Y. Hang, Q. Gao, X. Huang, Surgical Diagnosis and Treatment of Primary 527 Retroperitoneal Liposarcoma. Frontiers in Surgery 8, (2021 Jun 4). 528 9. J. H. Choi, J. Y. Ro, The 2020 WHO Classification of Tumors of Soft Tissue: Selected 529 Changes and New Entities. Advances in Anatomic Pathology 28, (January 2021). 530 10. R. Conyers, S. Young, D. M. Thomas, Liposarcoma: Molecular Genetics and 531 Therapeutics. Sarcoma 2011, 1-13 (2011). 532 11. M. AF et al., Atypical lipomatous tumors/well-differentiated liposarcomas: clinical 533 outcome of 67 patients - PubMed. Orthopedics 34, (12/06/2011). 534 12. C. Nessim et al., Analysis of Differentiation Changes and Outcomes at Time of First 535 Recurrence of Retroperitoneal Liposarcoma by Transatlantic Australasian Retroperitoneal 536 Sarcoma Working Group (TARPSWG). Annals of Surgical Oncology 28, 7854-7863 537 (2021). 538 13. G. DS et al., Does "Low-Grade" Dedifferentiated Liposarcoma Exist? The Role of Mitotic 539 Index in Separating Dedifferentiated Liposarcoma From Cellular Well-differentiated 540 Liposarcoma - PubMed. The American journal of surgical pathology 47, (06/01/2023). 541 14. D. KM, K. MW, A. CR, B. MF, S. S, Subtype specific prognostic nomogram for patients 542 with primary liposarcoma of the retroperitoneum, extremity, or trunk - PubMed. Annals of 543 surgery 244, (2006 Sep). 544 15. E. HL, Atypical lipomatous tumor, its variants, and its combined forms: a study of 61 545 cases, with a minimum follow-up of 10 years - PubMed. The American journal of surgical 546 pathology 31, (2007 Jan). 547 16. M. C et al., The prognostic impact of dedifferentiation in retroperitoneal liposarcoma: a 548 series of surgically treated patients at a single institution - PubMed. Cancer 113, 549 (10/01/2008). 550 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 12 of 17 17. S. S, A. CR, R. E, B. MF, Histologic subtype and margin of resection predict pattern of 551 recurrence and survival for retroperitoneal liposarcoma - PubMed. Annals of surgery 238, 552 (2003 Sep). 553 18. E. MA et al., Lifelong Imaging Surveillance is Indicated for Patients with Primary 554 Retroperitoneal Liposarcoma - PubMed. Annals of surgical oncology 30, (2023 May). 555 19. K. D. Klingbeil et al., IGF2BP3 as a Prognostic Biomarker in Well-556 Differentiated/Dedifferentiated Liposarcoma. Cancers 15, 4489 (2023). 557 20. K. L. J. Bill et al., Degree of MDM2 Amplification Affects Clinical Outcomes in 558 Dedifferentiated Liposarcoma. Oncologist 24, 989-996 (2019). 559 21. A. Serguienko, P. Braadland, L. A. Meza-Zepeda, B. Bjerkehagen, O. Myklebost, 560 Accurate 3-gene-signature for early diagnosis of liposarcoma progression. Clinical 561 Sarcoma Research 10, (2020). 562 22. M. H. Jagosky et al., Genomic alterations and clinical outcomes in patients with 563 dedifferentiated liposarcoma. Cancer Med 12, 7029-7038 (2023). 564 23. A. M. Crago et al., Copy Number Losses Define Subgroups of Dedifferentiated 565 Liposarcoma with Poor Prognosis and Genomic Instability. Clinical Cancer Research 18, 566 (2012/03/01). 567 24. J. Barretina et al., Subtype-specific genomic alterations define new targets for soft-tissue 568 sarcoma therapy. Nature Genetics 2010 42:8 42, (2010-07-04). 569 25. E. L. Snyder et al., c‐Jun amplification and overexpression are oncogenic in liposarcoma 570 but not always sufficient to inhibit the adipocytic differentiation programme. The Journal 571 of Pathology: A Journal of the Pathological Society of Great Britain and Ireland 218, 572 292-300 (2009). 573 26. H. C. Beird et al., Genomic profiling of dedifferentiated liposarcoma compared to 574 matched well-differentiated liposarcoma reveals higher genomic complexity and a 575 common origin. Molecular Case Studies 4, a002386 (2018). 576 27. N. Gruel et al., Cellular origin and clonal evolution of human dedifferentiated 577 liposarcoma. Nature Communications 15, (2024). 578 28. W. Sun et al., snRNA-seq reveals a subpopulation of adipocytes that regulates 579 thermogenesis. Nature 587, 98-102 (2020). 580 29. G. X. Y. Zheng et al., Massively parallel digital transcriptional profiling of single cells. 581 Nature Communications 8, 14049 (2017). 582 30. S. J. Fleming et al., Unsupervised removal of systematic background noise from droplet-583 based single-cell experiments using CellBender. Nature Methods 20, 1323-1335 (2023). 584 31. Y. Hao et al., Dictionary learning for integrative, multimodal and scalable single-cell 585 analysis. Nature Biotechnology 42, 293-304 (2024). 586 32. T. I. Tickle T, Georgescu C, Brown M, Haas B, inferCNV of the Trinity CTAT Project. 587 https://github.com/broadinstitute/inferCNV, (2019). 588 33. C. Trapnell et al., The dynamics and regulators of cell fate decisions are revealed by 589 pseudotemporal ordering of single cells. Nature Biotechnology 2014 32:4 32, (2014-03-590 23). 591 34. M. Hernandez-Quiles, M. F. Broekema, E. Kalkhoven, Frontiers | PPARgamma in 592 Metabolism, Immunity, and Cancer: Unified and Diverse Mechanisms of Action. 593 Frontiers in Endocrinology 12, (2021/02/26). 594 35. R. Brunmeir, F. Xu, Functional Regulation of PPARs through Post-Translational 595 Modifications. International Journal of Molecular Sciences 19, 1738 (2018). 596 36. O. Van Beekum, V. Fleskens, E. Kalkhoven, Posttranslational Modifications of PPAR‐γ: 597 Fine‐tuning the Metabolic Master Regulator. Obesity 17, 213-219 (2009). 598 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 13 of 17 37. L. H et al., Activation of PPARγ2 by PPARγ1 through a functional PPRE in 599 transdifferentiation of myoblasts to adipocytes induced by EPA - PubMed. Cell cycle 600 (Georgetown, Tex.) 14, (2015). 601 38. N. Hattori et al., Frontiers | Epigenetic disruption of adipogenic gene enhancers in 602 dedifferentiated liposarcomas and its therapeutic value. Frontiers in Oncology 15, 603 (2025/04/30). 604 39. K. D. Klingbeil et al., Targeting Asparagine Metabolism in Well-605 Differentiated/Dedifferentiated Liposarcoma. Cancers 16, 3031 (2024). 606 40. D. Braas et al., Metabolomics Strategy Reveals Subpopulation of Liposarcomas Sensitive 607 to Gemcitabine Treatment. Cancer Discovery 2, (2012/12/01). 608 609 Acknowledgments 610 We thank the members of the Christofk and Kadera labs for their discussion and 611 constructive feedback. 612 613 Funding: 614 UCLA Jonsson Comprehensive Cancer Center (JCCC) seed grant (BEK and KDK) 615 UCLA Department of Surgery grant (BEK and KDK) 616 NIH R01 CA215185 and R01 CA215185 and R01 AR070245 (HRC) 617 American Cancer Society 133839-PF-19-203-01-CCG (BRW) 618 619 Author contributions 620 Conceptualization: BRW, KDK, BK, HRC 621 Methodology: BRW, KDK, BK, HRC 622 Investigation: BRW, KDK, FD, CD, CF 623 Supervision: BK, HRC 624 Writing—original draft: BRW, KDK 625 Writing—review & editing: BRW, KDK, BK, HRC 626 627 Competing interests: Authors declare that they have no competing interests. 628 629 Data and materials availability: All data, code, and materials used in the analyses must 630 be available in some form to any researcher for purposes of reproducing or extending the 631 analyses. Include a note explaining any restrictions on materials, such as materials transfer 632 agreements (MTAs). Include accession numbers to any data relevant to the paper and 633 deposited in a public database; include a brief description of the dataset or model with the 634 number. The DMA statement should include the following: “All data are available in the 635 main text or the supplementary materials.” 636 637 Figure and Table Legends 638 639 Fig. 1. Study overview and histologic characterization of WD/DD liposarcoma for 640 single-nucleus RNA sequencing. (A) Schematic of experimental workflow. Computed 641 tomography (CT) scan from a patient with untreated retroperitoneal WD/DD LPS (sagittal 642 view), demonstrating differential contrast enhancement consistent with spatially distinct 643 WD and DD tumor regions. Tumor samples were collected at the time of surgical 644 resection. Both WD and DD components were independently macrodissected from each 645 tumor, analyzed for histological and molecular features, and subjected separately to 646 snRNA-seq. (B) H&E staining of matched WD and DD components from the same tumor. 647 The WD region is characterized by abundant lipoblasts and adipocytic differentiation, 648 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 14 of 17 while the DD component shows increased cellularity and nuclear atypia. (C) DNA-FISH 649 for MDM2 reveals gene amplification (red), with chromosome 12 centromere probe 650 (green) as control. 651 652 Fig. 2. Single-nucleus transcriptomic profiling identifies malignant clusters and 653 reveals transcriptional heterogeneity in WD/DD LPS. (A) UMAP embedding of all 654 nuclei from WD and DD components across three patient tumors reveals 21 655 transcriptionally distinct clusters. (B) Expression of known LPS-associated genes was 656 used to annotate and predict clusters enriched for malignant tumor cells. (C) Inferred copy 657 number alterations (CNAs) show characteristic gains on chromosome 12q—encompassing 658 the MDM2 locus—specifically in the predicted tumor cell clusters. (D) Re-clustering of 659 tumor-enriched nuclei reveals nine malignant subclusters, each defined by distinct gene 660 expression signatures and predicted functional states. 661 662 Figure 3. Adipocyte-like tumor cells are restricted to the WD component. UMAPs 663 show expression of (A) mature adipocyte markers (ADIPOQ, LIPE, PLIN1), (B) 664 preadipocyte markers (CD36, FABP4, PPARG), and (C) adipocyte progenitor markers 665 (COL6A1, COL6A2, FKBP10), highlighting a distinct population of differentiated cells 666 exclusive to the WD component. (D) UMAPs split by WD and DD components confirm 667 the spatial restriction of this population. (E) Quantification of cluster composition by 668 subtype shows that adipocyte-like clusters are present only in WD tumors. (F) UMAP 669 colored by shared versus WD-specific cluster identity. (G) Inferred copy number profiles 670 from patient 069 demonstrate shared CNAs (12q, 7q) across both components and 671 additional gains (6q, 14q) restricted to the WD-specific cluster. (H) Model summarizing 672 the differentiation trajectory from shared progenitor states to the WD-specific adipocyte-673 like cell population. 674 675 Figure 4. Pseudotime analysis reveals transcriptional trajectories toward adipogenic 676 differentiation. (A) Monocle 3 trajectory analysis reconstructs a continuous path from 677 common tumor clusters to the WD-specific adipocyte-like clusters, suggesting progressive 678 differentiation from a shared progenitor state. (B) Along this trajectory, cells show 679 decreasing expression of MDM2 and increasing expression of HMGA2, markers 680 associated with DD and WD histology, respectively. (C) Genes dynamically regulated 681 across pseudotime were clustered into co-expression modules, revealing distinct 682 transcriptional programs associated with cell state transitions. One module—designated 683 group 3—was characterized by low expression early in pseudotime and progressive 684 upregulation toward the terminal WD-specific clusters. (D) Groups were analyzed for 685 overrepresentation of transcription factor target genes. 686 687 Figure 5. PPARG2 promotes adipogenic differentiation and suppresses proliferation 688 in DD LPS cells. (A) A PPARG gene score based on known transcriptional targets is 689 enriched in WD-specific clusters. (B) Western blot showing PPARG levels in DD LPS 690 cell lines with overexpression of PPARG1 and PPARG2. (C) qPCR analysis shows that 691 PPARG2 induces robust expression of FABP4, while PPARG1 has modest effects. (D) 692 Bodipy staining and quantification reveal lipid droplet accumulation in PPARG2-693 expressing cells, consistent with adipocyte-like differentiation. (E) PPARG2 significantly 694 reduces proliferation of DD LPS cell lines (LPS1 and LPS2), consistent with induction of 695 a terminal differentiation program. 696 697 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 15 of 17 Figure 6. Activation of PPARG impairs tumor growth and is associated with 698 adipogenic features in DD LPS. (A) Schematic of the in vivo experimental approach 699 using doxycycline-inducible expression of PPARG1 or PPARG2 in LPS2 xenografts. (B) 700 Tumor growth curves demonstrate that induction of PPARG2 significantly suppresses 701 tumor growth, while PPARG1 has no measurable effect. (C) qPCR analysis of primary 702 human tumors reveals that PPARG2, but not PPARG1, is significantly enriched in WD 703 tumors. (D) Analysis of publicly available RNA-sequencing data from matched WD and 704 DD components of WD/DD tumors confirms higher PPARG2 expression in WD regions, 705 further supporting its role as a marker of differentiation (GSE221494)(25). 706 707 708 709 710 711 712 713 714 Table 1. Clinicopathologic variables of WD/DD LPS tumor cohort. Tumor size is the 715 greatest dimension in centimeters. CT, chemotherapy; DOD, dead of disease; NA, 716 neoadjuvant; NED, no evidence of disease; RP, retroperitoneum; RT, radiation therapy. 717 Median follow-up is 2.2 years. 718 719 Identifier Age Gender Tumor Location Tumor Size NA CT NA RT Recurrence Survival LPS-062 52 male RP 15.2 No No No NED LPS-064 69 male RP 46 No No Yes DOD LPS-069 62 male Chest Wall 8 No No No NED .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 16 of 17 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 Table 2. Clinicopathologic variables of WD and DD LPS tumor. Age, tumor size and 764 follow-up are listed as median (range). Remaining variables are listed as frequency (n). 765 Patients were diagnosed between June 2019-Aug 2024, censor date May 30 2025. 766 767 Fig. S1. Histologic characterization of WD and DD tumor components. (A-B) H&E 768 staining of two WD/DD LPS tumors used in snRNAseq shows consistent histological 769 WD (n=13) DD (n=15) Age At Diagnosis (years) 51 (27-80) 66 (52-90) Gender Male 8 9 Race White 13 11 Asian 0 3 Black 0 1 Ethnicity Hispanic 0 3 Non-Hispanic 13 12 Disease State Primary 13 15 Recurrence 0 0 Tumor Location Retroperitoneum 7 13 Extremity 6 1 Chest wall 0 1 Tumor Size Greatest Dimension (cm) 20.5 (11.5-63) 0 25 (7.1-47) Neoadjuvant Chemotherapy Yes 0 0 No 13 15 Neoadjuvant Radiation Therapy Yes 0 0 No 13 15 Recurrence Yes 6 6 No 7 9 Follow-Up Years 2.1 (0-6.1) 3.4 (0.3-4.7) Survival Outcome No Evidence of Disease 7 10 Alive With Disease 6 3 Dead of Disease 0 2 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint Page 17 of 17 differences between WD and DD regions, with WD components displaying abundant 770 mature adipocytes and lipoblasts, and DD regions exhibiting dense cellularity and spindle 771 morphology. 772 773 Fig. S2. Identification of tumor clusters based on expression of LPS oncogenes and 774 copy number alterations. (A) UMAP feature plots for MDM2, CDK4, 775 and HMGA2 across all nuclei. (B) Violin plots further confirm selective expression of 776 these oncogenes as well as adipocyte, preadipocyte, and adipocyte progenitor markers. 777 778 Fig. S3. Inferred copy number gains in MDM2 are used to identify tumor cells. 779 UMAP profiles across all clusters highlight (A) amplification of MDM2 and (B) 780 annotation of tumor cell populations. 781 782 Fig S4. Functional annotation of re-clustered tumor cell states. (A) Pie chart shows 783 relative number of nuclei across the nine malignant clusters, including corresponding to 784 proliferative, metabolic, mesenchymal-like, immune-like, and adipocyte-like states. (B) 785 UMAP plots colored by predicted cell cycle state of each nuclei. (C) Number of nuclei 786 contributing to each cluster, split by patient and component. (D) The relative number of 787 nuclei per patient for each cluster. 788 789 Fig S5. Histological analysis of LPS2 xenografts with induction of PPARG isoforms. 790 (A) Representative H&E staining of LPS2 xenograft tumors following doxycycline-791 induced expression of PPARG1 or PPARG2 reveal focal lipid-rich regions. (B) 792 Quantification of lipid-positive area shows no differences, likely due to limited tumor size 793 following PPARG-mediated growth suppression. 794 795 Fig S6. PPARG activation decreases DD LPS tumor growth in vivo. (A) Rosiglitazone 796 activates PPARG1 activity, which drives expression of PPARG2. (B) Rosiglitazone 797 treatment of mice bearing parental LPS2 xenografts reduces tumor growth, supporting the 798 idea that spharmacologic activation of PPARG1 can partially recapitulate the effects of 799 PPARG2 in vivo. 800 801 802 803 .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint .CC-BY-NC-ND 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 21, 2025. ; https://doi.org/10.1101/2025.07.16.665211doi: bioRxiv preprint