Abstract
Dedifferentiated liposarcoma (DDLPS) is the most frequent high-grade soft tissue sarcoma subtype. It is characterized by a component of undifferentiated tumor cells coexisting with a component of well-differentiated adipocytic tumor cells. Both dedifferentiated (DD) and well-differentiated (WD) components exhibit MDM2 amplification, however their cellular origin remains elusive. Using single-cell RNA sequencing, DNA sequencing, in situ multiplex immunofluorescence and functional assays in paired WD and DD components from primary DDLPS tumors, we characterize the cellular heterogeneity of DDLPS tumor and micro-environment. We identify a population of tumor adipocyte stem cells (ASC) showing striking similarities with adipocyte stromal progenitors found in white adipose tissue. We show that tumor ASC harbor the ancestral genomic alterations of WD and DD components, suggesting that both derive from these progenitors following clonal evolution. Last, we show that DD tumor cells keep important biological properties of ASC including pluripotency and that their adipogenic properties are inhibited by a TGF-β-high immunosuppressive tumor micro-environment.
Similar content being viewed by others
Introduction
Liposarcomas are rare tumors of mesenchymal origin affecting adults and showing adipocytic differentiation. They encompass a heterogenous group of malignancies characterized by distinct clinical, histological, and molecular features, including well-differentiated liposarcoma (WDLPS)/atypical lipomatous tumor (ALT), dedifferentiated liposarcoma (DDLPS), myxoid liposarcoma, and pleomorphic liposarcoma1,2. DDLPS is the most frequent high-grade liposarcoma subtype with an estimated annual incidence of 0.2/100,0003. DDLPS generally develops from the retroperitoneum, the limbs, or the trunk, and typically coexists with WDLPS within the same tumors4. The gold-standard therapeutic approach in localized tumors consists of large, “en-bloc” surgical resection performed in reference care centers5,6, however local and/or metastatic relapses happen in up to 30% of DDLPS patients7,8. Advanced diseases are treated with standard cytotoxic chemotherapy regimens, including doxorubicin, ifosfamide, trabectedin, and eribulin, but DDLPS is typically chemoresistant with a response rate and overall survival below 20% and 18 months, respectively9. Immune checkpoint inhibitors have poor activity, except in rare patients harboring tertiary lymphoid structures10,11.
Microscopically, DDLPS present both lipogenic and undifferentiated components that are often distinguishable macroscopically and typically well separated. The extent of the dedifferentiated component is variable. Most frequently, it is composed of intermediate- to high-grade, non-lipogenic undifferentiated and spindle cells, assembled in different ways and with mixed inflammatory infiltrates. On the contrary, the lipogenic component is composed of well-differentiated, low-grade cells resembling mature adipose tissue12. Whereas local relapse can be associated with both components, distant metastases are always composed of dedifferentiated tumor cells13. WDLPS and DDLPS are characterized by recurrent 12q13-15 amplifications, resulting from supernumerary rings or giant chromosomes14 containing MDM2 in 100% of cases and CDK4 in 90% of cases15. This 12q13-15 amplification is the driving oncogenic alteration and is present in all tumor cells of WDLPS and DDLPS16. MDM2 is a functional antagonist of TP53, and its overexpression leads to the inactivation of TP53 transactivation17, but also to its binding to the chromatin to promote nucleotide synthesis that fuels tumor growth18. Apart from MDM2 amplification, DDLPS tumors are transcriptomically and genomically heterogeneous, with frequent copy number alterations and few recurrent molecular alterations19,20.
The molecular mechanisms governing the coexistence of both well-differentiated (WD) and dedifferentiated (DD) components in DDLPS, as well as their specificities in the tumor microenvironment, are poorly understood. Previous studies have implicated additional copy number alterations, such as gains in JUN and ASK1, in the emergence of the DD component and the transition from WDLPS to DDLPS21,22. On the contrary, other studies claimed that WD and DD components rather diverge early from a common precursor23. DDLPS are commonly associated with immune infiltrates and display macrophage and CD8+ T cell infiltration scores among the highest across all sarcoma subtypes20. However, recent advances in the characterization of soft tissue sarcoma micro-environment have shown that similarly to other subtypes, the composition of the micro-environment in DDLPS was highly heterogeneous, ranging from “cold” tumors with no significant infiltration by myeloid and lymphoid cells to “hot” tumors characterized by massive immune infiltrates, high expression of immune-related signatures and high levels of immune checkpoints10.
Over the last decade, single-cell RNA sequencing (scRNA-seq) has become a transformative technique to address intratumor heterogeneity and characterize the tumor microenvironment. It has been successfully applied to the characterization of several sarcoma subtypes24,25,26,27,28.
In this study, paired samples of WD and DD components from 11 freshly resected DDLPS are analyzed by scRNA-seq together with a cohort of other adipocytic tumors and normal adipose tissues. We show that WD and DD components are characterized by the existence of specific immune infiltrates and tumor cell populations. We identify a population of tumoral progenitor cells related to multipotent adipocytic progenitors found in normal adipose tissue from which WD and DD components show early genomic divergence. We use patient-derived xenograft models to show that DD tumor cells keep the biological properties of their progenitors, including pluripotency and regulation of their adipocytic differentiation properties by the TGF-β pathway. This study provides insights into DDLPS cellular composition and genomic evolution and illuminates opportunities for drug development.
Results
Cellular composition of dedifferentiated liposarcomas
scRNA-seq was performed on tumor samples from 11 primary untreated DDLPS patients undergoing surgery in our institution (3 males and 8 females, aged 42–72). Primary tumors were located in the retroperitoneum (N = 10), and the spermatic cord (N = 1). For each tumor, suspected paired WD and DD components were histologically distinguished by pathologists during the macroscopic examination of the fresh surgical specimen just after surgery and simultaneously submitted to sequencing. The diagnosis of DDLPS had been established in all cases on pre-operative biopsies and later confirmed on the surgical specimen. All samples were integrated using Harmony29 together with additional adipocytic tumors (WDLPS, N = 5 and benign lipoma, N = 1) and normal adipose tissues (N = 5) (Fig. 1A, Supplementary Data 1 and Supplementary Data 2). After quality control and doublets removal, the transcriptomes of 102,753 single cells (SC) were obtained, including 36,326 SC from 10 DDLPS-WD components and 31,017 SC from 9 DDLPS-DD components (Supplementary Tables 1 and 2). Unbiased clustering on principal component analysis (PCA)-reduced data using Seurat identified 42 main clusters. Gene expression profiles and expression of canonical markers enabled grouping these clusters into 10 main cellular subfamilies, including tumor cells, lymphoid cells, myeloid cells, cycling cells, adipocytes, mesothelial cells, mast cells, endothelial cells, pericytes, and red blood cells (RBC) (Fig. 1B, C, Supplementary Fig. 1, and Supplementary Data 3).
A Graphical view of the study roadmap. Fresh samples were collected from patients undergoing surgery for primary untreated DDLPS (N = 11) and other adipocytic tumors (N = 11) and processed for single-cell suspension. For DDLPS, paired samples in the well-differentiated area (WD, white star) and in the dedifferentiated area (DD, black star) were identified on pre-operative computed tomography, defined macroscopically on the surgical specimen, and later confirmed at the microscopic level by expert pathologists. All samples were analyzed by scRNA-seq using the 10X Genomics platform and integrated into a single adipocytic tumor atlas. A total of 102,753 single cells were recovered, clustering within 42 clusters according to Seurat, and used for subsequent analyses. B Uniform Manifold Approximation and Projection (UMAP) of the 10 main cell types identified according to expression of marker genes. C Dot plots showing the expression of 15 marker genes across the 10 cellular clusters. The size of the dot represents the proportion of cells expressing the marker and the spectrum of color indicates the average expression level of the marker (log1p transformed) scaled over cell clusters. D Relative proportion of each cell cluster according to the histological subtypes (Lipoma, N = 1; WDLPS, N = 5; DDLPS, N = 19; DDLPS-WD, N = 10; DDLPS-DD, N = 9). E: Relative proportion of each cell cluster according to Lipoma (N = 1), WDLPS (N = 5), and DDLPS (N = 11) patients. F: Relative proportion of each cell cluster between paired WD and DD samples from DDLPS patients (N = 8 paired samples). Color scales are similar for panels (B, D, E, and F). Source data are provided as a Source Data file.
Nontumoral cells from the tumor microenvironment (TME) were highly represented, and constituted more than 50% of all SC from the various histological tumor subtypes, with myeloid and lymphoid cells being the most abundant cellular subtypes in the TME of these tumors (Fig. 1D and Supplementary Table 3). However, in DDLPS samples, the level of tumor infiltration by non-malignant cells was highly heterogeneous according to patients, with some tumors showing up to 80% of stromal cells (P1, P10, P14) and others showing less than 40% of cells from the TME (P2, P15) (Fig. 1E and Supplementary Fig. 2). A similar heterogeneity was observed between paired samples from the same DDLPS tumors, with the DD component showing in most cases a higher stromal infiltration than its WD counterpart (Fig. 1F and Supplementary Table 4). The proportion of tumor cells and stromal populations in this dataset was not significantly associated with any clinical or pathological feature, including patient age, gender, tumor size, or grade (Supplementary Data 4). These results were confirmed by inference of TME composition by MCP counter on bulk mRNAseq of the same samples and on a larger cohort of 53 samples showing a range of immune infiltration, including “cold” samples with absent to rare immune cell infiltrates to “hot” tumors highly infiltrated by myeloid and lymphoid cells (Supplementary Figs. 3A, B, Supplementary Data 1 and Supplementary Data 2).
Diversity of myeloid and lymphoid cells in adipocytic tumors
Myeloid cells, including CD163+ M2-like macrophages, have previously been shown to be the most prevalent immune cell populations in soft tissue sarcomas and are thought to be major contributors to the immunosuppressive TME in these tumors30,31. Ten clusters of distinct myeloid and related cell subtypes were identified based on gene expression profiles and expression of canonical markers (Fig. 2A, B, Supplementary Figs. 4 and 5, and Supplementary Table 5 and Supplementary Data 5). They included: 1. Three clusters of dendritic cells (DC) (classical DC2 (cDC2) with high expression of CD1C, classical DC1 (cDC1) with high expression of CLEC9A and ID2, and plasmacytoid DC (pDC) with high expression of IRF8 and TCF4); 2. three clusters of monocytes (classical monocytes with high expression of CD14 and low expression of FCGR3A, intermediate monocytes with high expression of CD14, CD163 and intermediate expression of FCGR3A and CXCL8, and non-classical monocytes with low expression of CD14 and high expression of FCGR3A); 3. one cluster of myeloid-derived suppressor cells (MDSC) with expression of CD14, CD163, CD33, and OLR1; 4. one cluster of tumor-associated neutrophils (TAN) with high expression of CXCL8, S100A8 and G0S2; and 5. two clusters of tumor-associated macrophages (TAM), the first one with expression of CD163, FABP4, MSR1, and TREM2 (immunosuppressive or M2-like TAM) and the second one with high expression of CXCL10 and IFIT1 (anti-tumorigenic or M1-like TAM). Of note, intermediate and non-classical monocytes, as well as TAM2-like macrophages, showed expression of HMOX1, previously reported as a promoter of an immunosuppressive program and CD8+ T cell inhibition32,33. The relative proportions of DC, MDSC, and TAN were enriched within DDLPS tumors compared to lipoma and WDLPS, contrary to monocytes and macrophages that were more abundant in lipoma and WDLPS (Supplementary Fig. 5 and Supplementary Table 6). Moreover, within DDLPS, the immunosuppressive cell populations of M2-like TAM, MDSC, and TAN seemed enriched in the DD component compared to the WD component of the same tumors (Fig. 2C and Supplementary Table 7). This was also confirmed by immunohistochemistry (IHC) on an independent cohort of 15-paired WD and DD components of DDLPS, showing a significantly higher infiltration of the DD component by CD163-positive macrophages (Fig. 2D). In situ multiplex immunofluorescence in a series of 7 paired WD and DD DDLPS components confirmed the presence of immunosuppressive myeloid populations within DDLPS including CD163+ monocytes and macrophages, CD68+ CD163+ M2-like macrophages, S1008+ TAN and CD33+ MDSC. These myeloid cells were interspersed between tumor cells in both WD and DD components, with no specific spatial distribution (Fig. 2E). Quantification of the proportions of these different myeloid cell populations identified by multiplex IF confirmed the major heterogeneity of immune infiltrates across patients (Supplementary Data 6).
A UMAP of the 10 main myeloid cell types identified according to the expression of marker genes. B Dot plots showing the expression of 22 marker genes across the clusters (n = 26,557 single cells). C Proportion of the myeloid populations among all myeloid cells according to histological subtypes (Lipoma, N = 1; WDLPS, N = 5; DDLPS, N = 19; DDLPS-WD, N = 10; DDLPS-DD, N = 9). D Proportion of CD163-positive myeloid cells by immunohistochemistry (IHC) in the independent cohort of paired WD and DD DDLPS (N = 15). The mean value is indicated in bold ( ± SD). p-val = 2.6 × 10−6 (two-sided t test). E In situ multiplex immunofluorescence (mIF) staining of tumor cells (MDM2+), monocytes/macrophages (CD163+), M2-like macrophages (TAM2-like, CD163+ and CD68+), myeloid-derived suppressor cells (MDSC, CD33+) and tumor-associated neutrophils (TAN, S100A8+) (representative pictures from one paired tumor sample, n = 7 paired samples analyzed); scale bar: 50 µm. F UMAP of the 8 main lymphoid cell types. G Dot plots showing the expression of 10 marker genes across the clusters (n = 19,978 single cells). H Proportions of the lymphoid populations among all lymphoid cells according to histological subtypes (Lipoma, N = 1; WDLPS, N = 5; DDLPS, N = 19; DDLPS-WD, N = 10; DDLPS-DD, N = 9). I Proportion of CD3-positive (left panel, N = 21 DDLPS-WD and N = 19 DDLPS-DD) and FOXP3-positive (right panel, N = 15 DDLPS-WD and N = 10 DDLPS-DD) T cells by IHC in an independent cohort of WD and DD DDLPS components; the number of CD3 and FOXP3-positive cells was normalized on the same number of total cells in both components. The mean value is indicated in bold (± SD). p-val = 1 × 10−3 (CD3) and p-val = 4 × 10−4 (FOXP3) (two-sided t test). J mIF staining of tumor cells (MDM2+), T lymphocytes (CD4+ or CD8+), CD4 Treg cells (CD4+, FOXP3+), and exhausted CD8 T cells (CD8+, PD1+) (representative pictures from one paired tumor sample, n = 7 paired samples analyzed); scale bar: 50 µm. Abbreviations: c: classical; i: intermediate; nc: non-classical. ***: p-val ≤ 0.001. Source data are provided as a Source Data file.
Lymphoid cells, including T and B lymphocytes, have previously been shown to play a major role in soft tissue sarcoma TME and in the response to immune checkpoint inhibitors10,11. Eight distinct clusters of lymphoid cells were identified based on gene expression profiles and expression of canonical markers (Fig. 2F, G, Supplementary Figs. 6 and 7, Supplementary Table 8 and Supplementary Data 7). They included: 1. one cluster of plasma cells with high expression of MZB1; 2. one cluster of B lymphocytes with high expression of MS4A1 (also known as CD20); 3. three clusters of T lymphocytes, including CD4+ T cells with high expression of CD3D and CD4, CD4+ T regulatory (CD4+ Treg) cells with high expression of CD3D, CD4, FOXP3 and CTLA4, and CD8+ T cells with high expression of CD3D and CD8A; 4. one cluster of innate-like, FCGR3A+ CD8+ cells with co-expression of T cell (CD3D, CD8A) and NK cell (KLRD1 and FCGR3A) markers (NKT cells) and 5. two clusters of NK cells, including NKDIM-like cells with high expression of KLRD1 and FCGR3A and low expression of NCAM1 (also known as CD56), and NKBRIGHT-like cells with higher expression of NCAM1, strong expression of KLRD1 and low expression of FCGR3A. The proportion of B lymphocytes was higher in DDLPS compared to lipoma or WDLPS, whereas the proportion of NK cells was lower in DDLPS compared to the other histotypes (Supplementary Fig. 7 and Supplementary Table 9). Within DDLPS, NK and NKT cells were enriched in the WD component, whereas CD4+ Treg and CD8+ T cells were more abundant in the DD component (Fig. 2H and Supplementary Table 10). Moreover, CD8+ T cells in the DD component expressed higher levels of exhaustion markers, including PDCD1, TIGIT, LAG3, and CD27, as well as higher levels of apoptosis markers, including CASP4 and FASLG (Supplementary Fig. 8 and Supplementary Table 11), reflecting their impaired cytotoxic activity. Immunohistochemistry confirmed the higher abundance of CD3+ T lymphocytes and FOXP3+ Treg cells in the DD component (Fig. 2I). Last, in situ multiplex immunofluorescence reinforced these findings, with significantly higher proportions of CD4+ FOXP3+ Treg cells among all CD4+ T cells (5.9 % vs 3.2%, p = 2.9 × 10−28) and of CD8+ PD1+ exhausted T cells among all CD8+ T cells (71.8% vs 60.8%, p = 1.39 × 10−83) in the DD part compared to the WD part (Fig. 2J and Supplementary Data 8). The spatial distribution of T cells varied across patients and tumors; in some cases, T cells were scattered between tumor cells, whereas in others, they were grouped in clusters within tumor septa (Supplementary Fig. 9).
Altogether, these results show that despite major inter-tumor heterogeneity, the composition of DDLPS TME differs between WD and DD components with an enrichment of pro-tumoral, immunosuppressive cell subpopulations in the DD component, including TAM2-like, MDSC, TAN, CD4+ Treg and exhausted CD8+ T cells (Supplementary Table 12).
Tumor cell heterogeneity in DDLPS
To address whether the differences in the TME composition of DDLPS-WD and DD components were associated to coexistence of transcriptomically-distinct tumor cell populations, we next focused on tumor cells from DDLPS samples. 24,000 cells distributed into 6 main clusters were confidently labeled as DDLPS tumor cells based on overexpression of MDM2 and CDK4 and identification of chromosome 12q amplification using InferCNV (Fig. 3A and Supplementary Figs. 10, 11). Gene expression signatures and pathway analysis enabled us to gather these clusters into 6 main pathways: 1. adipocytic differentiation; 2. stemness; 3. extracellular matrix remodeling; 4. hypoxia; 5. angiogenesis, and 6. invasion and cell migration (Fig. 3B and Supplementary Data 9). As expected, the most abundant cluster in DDLPS-WD components was the adipocytic differentiation cluster, which gathered 38% of all tumor cells from WD components, whereas cells from DDLPS-DD components were mostly found in the invasion and angiogenesis clusters, which gathered 45% and 15% of all cells from the DD components, respectively (Fig. 3C). Surprisingly, the stemness cluster, characterized by overexpression of genes related to stem cells including CD44, CD55 and CD34, was almost exclusively composed of tumor cells from DDLPS-WD components (11% of WD tumor cells versus 3% of DD tumor cells, p-val ≤ 0.05) (Fig. 3C). Moreover, when looking at the entire scRNA-seq dataset, the same stemness cluster was also identified in other well-differentiated samples including WDLPS, lipoma and non-tumoral adipocytic tissues, suggesting that this cluster could highlight a population of adipocytic progenitors (Fig. 3D and Supplementary Table 13). Differential gene expression analysis between all tumor cells from WD and DD components and functional annotation using Toppfun showed enrichment of pathways involved in ECM remodeling, invasion, and angiogenesis in the DD component, and enrichment of pathways involved in lipid metabolism, TGF-β signaling and adipose stem cells in the WD component (Fig. 3E, Supplementary Data 10 and Supplementary Data 11). Unsupervised clustering using the 800 most variant genes on the cohort of adipocytic tumors analyzed by bulk RNA-seq (N = 53 samples) confirmed the overexpression of genes related to adipocytic differentiation, and stemness in most well-differentiated samples (including DDLP-WD, WDLPS, lipomas and adipose tissue). Besides, undifferentiated samples were characterized by overexpression of genes related to invasion, angiogenesis, and immune response (Fig. 3F and Supplementary Data 12), validating the signatures identified by scRNA-seq in these samples.
A UMAP of the 6 main tumor cell populations identified within DDLPS-WD (N = 10) and DDLPS-DD (N = 9) samples (N = 24,000 single cells) (left panel). All cells are characterized by strong expression of MDM2 and CDK4 (middle and right panel). B Dot plots showing the expression of 20 marker genes across the 6 identified cellular clusters. C Proportion of DDLPS-WD and DDLPS-DD tumor cells within each cell cluster. Adipo diff, p-val = 0.0 × 100; stemness, p-val = 3.3 × 10−113; ECM remodeling, p-val = 4.1 × 10−2; hypoxia p-val = 4.1 × 10−2; angiogenesis, p-val = 0.0 × 100; invasion: p-val = 2.8 × 10−13 (Chi2 test). D Distribution of cells within each of the 6 clusters in all samples analyzed by scRNA-seq and within each tissue type (Adipose tissue, N = 5; Lipoma, N = 1; WDLPS, N = 5; DDLPS, N = 19; DDLPS-WD, N = 10; DDLPS-DD, N = 9). E Barplots showing Gene Ontologies (GO), pathways, and coexpression identified with Toppfun analysis and found differentially and specifically (p-val_adj ≤ 0.05) overexpressed in DDLPS-WD (left panel) or DDLPS-DD (right panel) tumor cells after FDR correction for multiple testings. F Hierarchical clustering and scaled expression matrix using the 800 most variant genes across the larger cohort of adipocytic samples analyzed by bulk RNA-seq. Abbreviations: Adipo diff: adipocytic differentiation. ECM: extracellular matrix. UR: upregulated. *: p-val ≤ 0.05; ***: p-val ≤ 0.001; ns: p-val > 0.05. Source data are provided as a Source Data file.
DDLPS originate from adipocyte stromal progenitor cells
To better characterize the stemness population of tumor cells and determine if these cells could be DDLPS progenitors, all tumor cells from DDLPS-WD samples were selected and re-integrated together in a single dataset. Tumor cells showed consistent MDM2 and CDK4 expression and specific expression of the previously identified signatures: adipocytic differentiation (clusters 1, 2, 3, 4, 7, and 8), stemness (clusters 5, 6, 9), ECM remodeling and hypoxia (cluster 0), invasion (clusters 1, 8 and 11), and, in a very low proportion, angiogenesis (cluster 10) (Fig. 4A and Supplementary Data 13). We next compared this stemness signature with previously reported signatures of mammalian adipocyte stromal progenitor cells (ASPC) found in normal white adipose tissue and characterized by scRNA-seq and found that the stemness cells in DDLPS were highly similar to early multipotent adipocytic progenitors described in previous studies (G1, Merrick et al.34; hASPC2, Emont et al.35; P1, Schwalie et al.36) (Fig. 4B and Supplementary Table 14). In particular, these cells were characterized by overexpression of adipocyte stem cells canonical markers including CD55, CD34, CD44, ALDH1A3, MFAP5, PI16 and DPP4, whereas other tumor cell clusters expressed markers of preadipocytes or adipocytes in line with their adipocytic differentiation (Fig. 4C). Immunohistochemical staining using MDM2 and CD55 antibodies showed the presence of MDM2+CD55+ cells within the WD component of DDLPS tumors. These cells were present in the fibrous septa interspersed between tumor adipocytes and showed marked atypia (Fig. 4D). Last, in situ multiplex immunofluorescence in a cohort of 7 WD components of DDLPS confirmed the presence of cells coexpressing MDM2 together with multiple ASPC markers including CD55, CD44, CD34 and ALDH1 (Fig. 4E). Quantification of the proportion of these cells detected by multiplex immunofluorescence revealed that they represented between 0.5% and 1.2% of all MDM2+ tumor cells and between 0.2% and 0.6% of the total cells found in tumor samples (Supplementary Data 14).
A UMAP of the 12 main tumor cell clusters found in DDLPS-WD samples (N = 10), expression of MDM2, and application of the 6 main gene expression signatures identified in the whole dataset. B UMAP showing the expression of three previously reported adipocyte stem cell signatures in DDLPS-WD tumor cells. C Dot plots showing the expression of 10 ASPC marker genes and 7 pre-adipocytes/adipocytes marker genes across the stemness and adipocytic differentiation clusters. D MDM2 (pink) and CD55 (brown) immunohistochemical stainings of a human DDLPS-WD component showing the presence of double positive, undifferentiated tumor cells interspersed between tumor adipocytes (arrowheads) (Scale bars: 100 µm (upper panel) and 50 µm (lower panel)); n = 9 tumor samples analyzed. E In situ multiplex immunofluorescence staining of tumor adipocyte stem cells with MDM2, CD55, CD44, and ALDH1 antibodies; n = 7 tumor samples analyzed. Scale bar: 50 µm. F Representative example of the genomic evolution of DDLPS-WD and -DD components from tumor ASPC (Patient P15). Left panel: UMAP of the tumor clusters identified in WD and DD components (top) and of the expression of the stemness signature highlighting one cluster (WD-4) in the WD component of the tumor (bottom). Middle panel: inference of Copy Number Variations (CNVs) using InferCNV in all tumor cell clusters of patient P15. Right panel: reconstitution of P15 genomic clonal and subclonal evolution based on CNVs found in each cell cluster.
To investigate whether these tumor ASPC found in the WD component of DDLPS could be the progenitors of DDLPS, we compared the Copy Number Variations (CNVs) of tumor ASPC to the ones of other tumor cells in matched WD and DD components from 4 DDLPS patients. Strikingly, tumor ASPC were characterized by very few CNVs except for chromosome 12q amplification. Moreover, CNVs present in ASPC were conserved in both WD and DD components of the same tumor, which showed additional private clonal and sub-clonal genomic alterations (Fig. 4F and Supplementary Fig. 12). These results suggest that WD and DD components of DDLPS diverge early from a population of tumor adipocyte stem cells related to ASPC and later acquire additional genomic events private to each component, indicating also that the DD part does not come from the dedifferentiation of the WD part.
Genomic evolution of DDLPS
To confirm the scRNAseq findings at the genomic level, tumor DNA from 10 paired WD and DD DDLPS samples were extracted and submitted to shallow Whole Genome Sequencing for analysis of CNVs and to targeted Next Generation Sequencing using the SureSelect CD Curie CGP panel (Agilent) for analysis of Single Nucleotide Variations (SNVs) in 571 genes involved in oncogenesis (Fig. 5, Supplementary Data 15 and Supplementary Data 16).
A Pangenomic profiles assessed by shallow Whole Genome Sequencing (shWGS) of P14 (upper panel), P29 (middle panel), and P48 (lower panel) paired DDLPS samples showing major Copy Number Variations (CNVs) private to each WD (blue) and DD (red) component. B Genomic profiles of the chromosome 12q region containing MDM2 and CDK4 showing the shared amplicons between WD (red) and DD (blue) pairs for patients P14 (left panel), P29 (middle panel), and P48 (right panel). C Oncoprint summarizing the main CNVs found in WD and DD pairs analyzed by shWGS, either private to one component or shared by both components. The genomic profiles obtained after shWGS initially contained 52,725 segments that were combined into 296 final chromosomal regions containing the largest and most discriminant alterations spanning the 22 chromosomes, and represented on the Y-axis of the oncoprint. D Oncoprint summarizing the single nucleotide variations found by targeted DNA Sequencing using the SureSelect CD Curie CGP panel (Agilent) in paired WD and DD DDLPS samples.
The median number of CNVs detected across 296 segmented chromosomal regions distributed along the 22 chromosomes was 14 (range 7–73) in DDLPS-WD samples, and 26 (range 7–76) in DDLPS-DD samples (Supplementary Data 15). Analysis of CNVs in each paired WD and DD DDLPS components showed that the median number of shared CNVs between pairs was 11 (range 6–57), the median number of CNVs private to the WD component was 2 (range 0–16), and the median number of CNVs private to the DD component was 12 (range 1–40). Some pairs showed major differences in the CNVs profiles of both components, illustrating their early genomic divergence and independent clonal evolution (Fig. 5A). The only recurrent shared alteration in all pairs was the 12q amplification containing MDM2 and CDK4 (Fig. 5B, C).
The median number of SNVs detected in the 571 genes covered by the DNA panel in DDLPS-WD samples was 1 (range 0–4), and the median number of SNVs detected in DDLPS-DD samples was 1 (range 0–6) (Supplementary Data 16). Analysis of SNVs between paired WD and DD samples showed that 100% of SNVs were private either to the WD or to the DD component of each tumor (Fig. 5D).
Thus, except for the oncogenic MDM2 amplification on chromosome 12q, paired WD and DD components of DDLPS harbor private CNVs and SNVs. This validates the scRNAseq findings and demonstrates that the DD component does not develop from the dedifferentiation of WD tumor cells but that, on the contrary, WD and DD show genomic divergence from their common MDM2-amplified ancestor and later acquire genomic features private to each component following clonal evolution.
DD tumor cells keep the biological properties of their progenitors
Adipocyte stem cells found in white adipose tissue have previously been shown to be mesenchymal stem cells that commit to adipogenesis when exposed to a complete adipogenic cocktail but also show pluripotent properties, including the ability to differentiate into the osteoblastic lineage34. Moreover, TGF-β is a known anti-adipogenic factor that specifically regulates the identity and function of DPP4+ ASPC by increasing their proliferative capacity and inhibiting adipogenesis34.
To define whether DDLPS tumor cells kept the biological properties of their progenitors, we used four patient-derived xenograft (PDX) models established from the DD component of four different DDLPS tumors (Supplementary Table 15). The pathological review showed that all models recapitulated the phenotypic aspect of the human DD component of DDLPS, with a proliferation of high-grade, undifferentiated tumor cells expressing MDM2 (Supplementary Fig. 13). Next Generation Sequencing of PDX tumor DNA confirmed genomic profiles in accordance with DDLPS, with 12q amplification containing MDM2 and CDK4 in all models (Supplementary Fig. 14). The morphological, immunohistochemical and genomic profiles of PDX remained stables over tumor passages (Supplementary Figs. 15 and 16). Single-cell suspensions of the PDX were obtained, and cells were subsequently cultivated ex vivo in the presence of adipocytic differentiation cocktail (ADC) or osteogenic differentiation cocktail (ODC) for 14 or 21 days, respectively, before RT-qPCR analysis and specific staining (Fig. 6A). In all four PDX models, tumor cells showed striking pluripotent properties and were able to differentiate into both lineages as shown by the induction of expression of genes indicative of differentiation, and positive staining with Bodipy or Alizarin red dyes, indicative of adipocytic and osteogenic differentiation, respectively (Fig. 6B and Supplementary Figs. 17, 18). Of note, using RT-qPCR primer probes specific to human or murine genes, only human genes were concerned by adipocytic differentiation, demonstrating that mouse cell contamination was negligible (Supplementary Fig. 18 and Supplementary Table 16).
A Graphical view of the study roadmap. DDLPS tumors from four different PDX models were resected from nude mice and processed for single-cell suspension before being cultured in the presence of adipocytic differentiation cocktail (ADC) for 14 days or osteogenic differentiation cocktail (ODC) for 21 days and then later processed for characterization of differentiation using RT-qPCR and phenotypic staining. B Bodipy staining (green) after culture with ADC, and Alizarin red staining (red) after culture with ODC showing the multipotent properties of PDX SIL10AS tumor cells (Gx10). C Western Blot showing the expression of phospho-Smad2 (P-Smad2), total Smad2, and GAPDH in PDX SIL10AS cells after treatment with TGF-β and TGF-βR inhibitors SB431542 and galunisertib and quantification of P-Smad2/total Smad2 expression in comparison to TGF-β condition. D Bodipy staining of SIL10AS PDX cells cultured for 14 days in ADC, in the presence or absence of TGF-β and TGF-βR inhibitors SB431542 and galunisertib (left panel, Gx10) and quantification of the activity of TGF-β (p-val = 5.3 × 10−12) and TGF-βR inhibitors SB431542 (p-val = 3.7 × 10−6) and galunisertib (p-val = 3.3 × 10−9) on tumor cells adipocytic differentiation properties evaluated by Bodipy staining (N = 10 independent experiments, median ± SD, one-sided ANOVA test) (right panel). E Relative expression of FABP4 by RT-qPCR in PDX SIL10AS cells in the presence of TGF-β (p-val = ns) and TGF-βR inhibitors SB431542 (p-val = 1.7 × 10−3) and galunisertib (p-val = 1.1 × 10−4) (n = 3 independent experiments, mean ± SD, Chi2 test). F Western Blot showing the expression of FABP4 in PDX SIL10AS cells in the presence of TGF-β and TGF-βR inhibitors (n = 3 independent experiments). G In situ multiplex immunofluorescence staining of tumor cells (MDM2+), monocytes/macrophages (CD14+), and T lymphocytes (CD3+) together with TGFβ1; n = 7 tumor samples analyzed. Scale bar: 50 µm. H Quantification of intratumoral TGF-β concentration by ELISA in WD (N = 5) and DD (N = 4) samples from human DDLPS (mean ± SD). p-val = 1.4 × 10−2 (two-sided t test). Abbreviations: *: p-val ≤ 0.05; **: p-val ≤ 0.01; ***: p-val ≤ 0.001. Source data are provided as a Source Data file.
To evaluate the effect of the TGF-β pathway on DDLPS cells, PDX cells were cultured in the presence of TGF-β and TGF-β receptor inhibitors SB431542 or galunisertib. As expected, TGF-β treatment was associated with an induction of TGF-β pathway reflected by the induction of phospho-Smad2 expression at the protein level, whereas treatment with SB431542 and galunisertib inhibited TGF-β signaling (Fig. 6C). Treatment of PDX cells with TGF-β significantly inhibited the adipocytic differentiation properties of tumor cells, whereas TGF-β receptor inhibitors SB431542 and galunisertib restored adipocytic differentiation in all four PDX models (Fig. 6D–F and Supplementary Fig. 19). Moreover, TGF-β increased the proliferative properties of SIL10AS tumor cells whereas SB431542 and galunisertib decreased cell proliferation (Supplementary Fig. 20).
To address the likely origin of TGF-β within DDLPS samples, we looked at the expression of TGF-β and TGF-β receptor isoforms in DDLPS tumors analyzed by scRNAseq. We found that TGFB1 was expressed by myeloid and lymphoid cells in these tumors and, interestingly, by more macrophages, MDSC, and DC from the DD component compared to the WD one (Supplementary Fig. 21A, B). TGFB1, TGFB2, and TGFB3 were also expressed by a minority of tumor cells, predominantly in DDLPS-DD samples (Supplementary Fig. 21C). Moreover, all the three TGF-β receptor isoforms TGFBR1, TGFBR2, and TGFBR3 were expressed by tumor cells (Supplementary Fig. 21D–F). These findings were confirmed by in situ multiplex immunofluorescence in a cohort of seven paired DDLPS samples, showing that the proportion of TGF-β1-positive cells among all cells was significantly higher in DDLPS-DD samples compared to DDLPS-WD samples (27.9% vs 8.3%, p-val = 0 × 100) (Supplementary Data 17). Costaining with MDM2, CD14, and CD3 antibodies showed that the main sources of TGF-β1 were CD14+ myeloid cells and MDM2+ tumor cells and that the proportions of MDM2+ tumor cells and CD14+ myeloid cells expressing TGF-β1 were significantly higher in DDLPS-DD samples compared to DDLPS-WD samples (23.2% vs 12.9%, p-val = 2.6.10−235, and 50.1% vs 23.8%, p-val = 0 × 100) (Fig. 6G and Supplementary Data 17). Last, we quantified TGF-β concentration in paired WD and DD components from 4 DDLPS tumors and confirmed that TGF-β was significantly increased in the DD component compared to the WD component (Fig. 6H). This suggests that TGF-β likely plays a major role in the emergence of the DD component and in its immunosuppressive TME composition.
Discussion
Soft tissue sarcomas are a heterogeneous group of malignancies whose diversity is thought to result both from a variety of molecular alterations and from the lineage and stage of differentiation of their presumed mesenchymal cell of origin37. Among them, DDLPS is the most frequent high-grade STS subtype in adults and represents the evolution of indolent WDLPS into an aggressive and metastasizing tumor38. Although the molecular spectrum of WDLPS and DDLPS has been the subject of numerous studies20,39,40,41,42, the presumed cell of origin of both subtypes remains uncertain.
In this study, we performed scRNA-seq profiling, genomic profiling, and in situ multiplex immunofluorescence staining on a cohort of paired WD and DD components of primary human DDLPS tumors. This enabled us to identify a population of progenitor tumor cells showing transcriptomic and phenotypic similarities with ASPC recently described as multipotent progenitors of white adipocytic tissue, from which WD and DD components derive early following clonal evolution and acquire distinct TME (Fig. 7).
In the context of low-TGF-β exposition, DPP4+CD55+ liposarcoma adipocyte stem cells (LPS ASC) harboring MDM2 amplification can differentiate into the adipocytic lineage and give rise to the development of WDLPS and DDLPS-WD component, characterized by expression of adipocytic-related genes and anti-tumoral tumor micro-environment containing Natural Killer (NK) cells, Dendritic cells (DC) and monocytes. Following a maturation arrest signal during LPS ASC early adipocytic differentiation and acquisition of copy number events and genomic instability, clones of undifferentiated tumor cells arise and develop within a TGF-β-high microenvironment. TGF-β prevents the differentiation of tumor cells into the adipocytic lineage and favors cell proliferation, metastases, and development of a DDLPS-DD immunosuppressive TME containing myeloid-derived suppressor cells (MDSC), tumor-associated neutrophils (TAN), immunosuppressive tumor-associated macrophages (TAM-2 like), CD4+ Treg cells and exhausted CD8+ T cells.
ASPC are complex and heterogeneous cell populations of the stromal vascular fraction (SVF) of white adipose tissue and are defined as the SVF depleted of immune, endothelial, and red blood cells43,44. Over the last decade, scRNA-seq profiling has enabled the dissection of the cellular hierarchy and functions of mammalian ASPC, with the description of three main subpopulations: early adipose stem cells (ASC), pre-adipocytes, and adipogenesis regulators45. ASC have been widely studied and isolated in mice and human white adipose tissue34,35,36,46,47. These cells are characterized by the expression of stem-related genes, including Cd55, Dpp4, Ly6a, and Srfp4, and are highly proliferative in vitro34,35. When stimulated with a complete adipogenic differentiation cocktail, these cells can differentiate into pre-adipocytes and adipocytes34,36; however, they also show the capacity to differentiate into other mesenchymal lineages, including osteoblasts34. In our study, we identified a subpopulation of tumor cells present in the fibrous septa and along blood vessel walls of the DDLPS-WD component and whose transcriptomic and immunohistochemical profiles are highly similar to ASC. We showed that these tumoral ASC present the ancestral genomic alterations conserved both in the WD and DD components of these tumors, including the oncogenic chromosome 12q amplification containing MDM2. Inference of genomic alterations in paired WD and DD samples suggests that both components derive early from this ancestral progenitor and later acquire additional private clonal and sub-clonal CNVs, which we later confirmed at the genomic level using DNA sequencing. This early genomic divergence of WD and DD components from a common ancestor is in agreement with previous reports showing that both components share very few genomic alterations23,48.
Whether chromosome 12q amplification in an ASC is sufficient to promote the development of liposarcoma will require further investigation, however several data support that ASC are at the origin of both WDLPS and DDLPS. ASC have been described both in the subcutaneous and visceral white adipose tissue, which matches WDLPS and DDLPS preferred tumor locations, including retroperitoneum, trunk, and limbs35. Moreover, the DDLPS-DD component sometimes shows heterologous differentiation into the osteogenic or rhabdomyoblastic lineages, highlighting the multipotent properties of the tumor cells in vivo49,50,51. Besides, the transcriptomic landscape of DDLPS is related to one of the mesenchymal stem cells in the early steps of adipocytic differentiation52. Last, CD55, a major ASC marker, has recently been shown to be specifically overexpressed in DDLPS in comparison to a cohort of other sarcomas53.
We also characterized the TME of DDLPS components and revealed that the DD component is characterized by a TGF-β-high TME with increased immunosuppressive myeloid cell populations as well as CD4+ Treg cells and exhausted CD8+ T lymphocytes, compared to its WD counterpart. Of note, TGF-β has been shown to exert an anti-adipogenic effect on ASC by increasing ASC proliferative capacity and inhibiting adipocyte formation even in the presence of a complete adipogenic cocktail34,54. Using DD tumor cells from four DDLPS PDX models, we showed that the adipocytic differentiation properties of DD tumor cells are abrogated under TGF-β treatment, whereas TGF-β inhibitors SB431542 or galunisertib restore adipogenesis, even in the presence of additional TGF-β. Of note, the knockdown of SMAD4, a central regulator of TGF-β signaling, has been shown to increase adipogenesis in DDLPS cell lines55. Moreover, we showed using multiplex immunofluorescence that the main sources of TGF-β in human DDLPS are myeloid cells but also DD tumor cells, shaping a complex interplay between tumor and immune cells to maintain a protumoral TME. These results demonstrate that DDLPS tumor cells keep the biological properties of ASC and suggest that TFG-β inhibition could represent a new therapeutic opportunity to prevent the emergence of the DD component or favor its differentiation back into a low-grade adipocytic tumor.
This study has several limitations. First, despite the fact that all human tumor samples were processed in the same conditions, we can not exclude sample-specific batch effects, as fresh tumor samples were processed immediately following patients’ surgery. Second, the rarity of tumor ASC in DDLPS tumor samples was not compatible with their sorting for genome-wide CNV assessment and validation of scRNAseq data at the genomic level. Last, the potential activity of TGF-β inhibitors on DDLPS cells will need to be validated in vivo, keeping in mind that immunodeficient PDX models are not fully relevant in this context and that immunocompetent models of DDLPS are lacking.
Altogether, we provide the characterization of human DDLPS at the single-cell level and deliver insights on the oncogenesis of this sarcoma subtype. We propose a model of tumorigenesis based on the evolution from a tumor ASC to the development of two distinct tumor components, reconnecting tumor cell differentiation and TME.
Methods
The study complied with all relevant ethical regulations and was approved by Institut Curie institutional review board (IRB protocol DATA190160).
Human samples procurement and approvals
Anonymized tumoral and non-tumoral samples were collected from patients undergoing surgery at Institut Curie (Paris, France). All patients provided written informed consent. Consent was obtained to publish information on age, sex, medical center, and diagnosis. Age and gender were not considered in the study design, and the gender of participants was determined based on self-reporting.
Patient derived Xenografts (PDX)
PDX SIL10AS was established in Institut Curie, and PDX 15lp0542, 15lp2584, and 15lp4728 were established in Institut Régional du Cancer de Montpellier. The care and animal housing were in accordance with institutional guidelines and with the recommendations of the French Ethics Committee (Comité Ethique en matière d’ expérimentation animale de l’Institut Curie, National registration number: #118, authorization no.: 02163.02). Models were obtained from consented DDLPS patients after IRB approval. Tumors were engrafted in 7 to 8-week-old female nude mice, monitored weekly, and collected before they reached 10,000 mm3, the maximal tumor size permitted by the French Ethics Committee. Mice were euthanized by cervical dislocation. Mice were purchased from Charles River Laboratories and maintained under specific pathogen-free conditions. The housing facility was kept at 22 °C (± 2 °C) with a relative humidity of 30–70%. The light/dark cycle was 12 h light/12 h dark.
Sample preparation, cell isolation for scRNA-seq, library preparation and scRNA-seq
Tissues were processed within 1 h after tumor resection, and sorted cells were loaded in a 10x Chromium instrument within 3 h. Fresh tumor samples were cut into small pieces and then dissociated for 30 min at 37 C in a CO2-independent medium (Gibco) containing 150 mg/ml of Liberase TL (Roche, ref 05401020001) and DNase (Sigma Aldrich, ref DN25-10MG). Dissociated cells were then filtered with a 100 µm cell strainer and washed in PBS. Debris were removed by centrifugation (3000 × g for 10 min at 4 C) using the debris removal solution (Miltenyi Biotech, ref 130-109-398). Cells were then resuspended in PBS, counted, and adjusted at 106 cells/ml in PBS. 6000 cells were loaded on a 10X Chromium (10X Genomics), and libraries were prepared using a Single Cell 30 Reagent Kit (NextGem kit, 10X Genomics), according to the manufacturer’s protocol, targeting 3000 recovered cells per cell type and sample. Libraries were sequenced on an Illumina NovaSeq 6000 sequencing platform.
Pre-processing of patient scRNA-seq data
Reads were processed using the standard CellRanger pipeline (versions varied depending on the run from 3.0.0 to 3.0.1, 3.1.0, and more recently, 6.0.0) using the hg38 reference genome. The first analysis step was performed on each sample individually using the Seurat package in R and CellRanger’s filtered count tables as well as using the same initial filters on genes and cells (minimum 200 genes per cell, minimum 3 cells expressing each gene, no more than 20% UMIs mapping MT genes and restrictions for outliers on UMI and gene counts using 0.01 and 0.99 percentiles). In addition, doublets were predicted with DoubletFinder (version 2.0.3), scDblFinder (version 1.4.0), and scds (hybrid mode, version 1.6.0) with default parameters and cells predicted as doublet in at least one of these methods were removed. Ambient RNA contamination was accounted for using the SoupX package (version 1.5.2) and the previously filtered matrices, as well as CellRanger’s unfiltered count matrices for better ambient RNA composition estimation. As manual ambient contamination estimations using key marker genes such as HBB (red blood cells), IGKC (B cells), or CD14 (myeloid cells) gave variable results within the same sample, contamination was set to 5% for all samples (representing 90–95% of the ambient RNA removed according to SoupX authors). Raw counts were then adjusted and rounded to integers.
Dimensionality reduction and clustering of individual patient scRNA-seq data
Sample counts were normalized using Seurat’s SCTransform function with centering and scaling. PCA was performed on the 3000 most variable genes identified by SCTransform. Finally, UMAP dimensions were calculated on the first 50 PCA components from the 100 calculated. Shared nearest-neighbors and Louvain clustering were also performed on those components using the FindNeighbors and FindClusters functions from Seurat at several resolutions.
Integration, dimensionality reduction and clustering of patient scRNA-seq data
Directly after the pre-processing step, samples from the same patient were first merged and normalized together using Seurat’s SCTransform function with centering and scaling. To integrate all samples across patients and remove the batch effect, all data was merged and PCA was performed on 3000 common most variable genes determined by SelectIntegrationFeatures from Seurat. Finally, Harmony 0.1.0 was run with default parameters on the first 50 PCA components from the 100 calculated, with patients as a grouping variable. Finally, UMAP dimensions were calculated on the first 50 harmony dimensions. Shared nearest-neighbors and Louvain clustering were performed on the whole integration using the FindNeighbors and FindClusters functions from Seurat on the first 50 harmony dimensions at several resolutions. These steps were performed for both the full integration and the DDLPS-WD-specific subintegration with the corresponding samples.
Differential gene expression analysis and cluster annotation
Raw gene expressions were log-normalized with Seurat’s NormalizeData function and used in further downstream analysis. Differential gene expression analysis was performed with Seurat’s FindAllMarkers function using a Wilcoxon rank sum test with Bonferroni correction on genes expressed in at least 25% of cells (min.pct = 0.25). Genes with a log2-fold change ≥ 0.4 and an adjusted p-value ≤ 0.05 were considered as significantly over-expressed for a given cluster. Clusters co-expressing gene markers specific for two distinct cell types were removed (this was the case for 2 clusters from the full patient integration), as well as clusters with lower UMI and gene count expressing no obvious cell type markers and often significantly over-expressing nuclear lincRNA MALAT1 (4 clusters from the full patient integration). This left 102,753 cells in the full patient integration for further analysis. Clusters were labeled based on marker gene expression and differential gene expression analysis results. Clusters in PDX samples were labeled using the label transfer pipeline described in Seurat tutorials with FindTransferAnchors on log-normalized data using the first 50 PCA dimensions, the top 3000 variable features calculated by FindVariableFeatures on log-normalized counts, and parameters n.trees = 20, mapping.score.k = 100, recompute.residuals = F and k.filter = NA. TransferData was applied with k.weight = 20 and MappingScore with nn.method = “annoy” and k = 50.
Pathway enrichment analysis
Pathway enrichment analysis was performed using the Toppfun tool based on functional annotations and protein interactions network (https://toppgene.cchmc.org/enrichment.jsp).
Single-cell copy-number variation (CNV) and clonality analysis
CNVs were predicted for all cells clustering together in the “Tumor cells” group of the single-cell patient integration using the “Endothelial cells” group as the reference cell type and split according to patient and tissue type. The prediction was performed using the inferCNV package (version 1.11.2) and the GRCh38 annotated human genome in which HLA genes in chromosome 6p were removed to avoid bias in prediction due to frequent high correlations between these genes. inferCNV was run on cells with ≥ 1000 total UMI count only. Default options were used with a cutoff set to 0.1 (cutoff = 0.1), the denoise option activated (denoise = TRUE), and conservation of query and reference clusters (cluster_by_groups = TRUE and cluster_references = TRUE).
Endothelial cells within each sample were used as internal references and when too little or no endothelial cells were present, the immune cells were used instead.
Bulk RNA sequencing
RNAs were extracted from frozen tumors by mechanical crushing followed by TRIzol® reagent (Invitrogen, ref 15596018,) and purified with the NucleoSpin RNA kit (Macherey-Nagel, ref 740955.50). RNA quality was assessed with a Bioanalyzer instrument, and RNAs with an RNA Integrity Number above 7 were processed for sequencing. RNA sequencing libraries were prepared from 1 μg of total RNA using the Stranded mRNA prep Ligation-Illumina kit according to the manufacturer's recommendation. 100 bp paired-end sequencing was performed with the Illumina NovaSeq 6000 instrument. Reads were aligned to the human reference genome hg38/ GRCh38 using STAR 2.6.1a_08-27 (RRID:SCR_015899, https://github.com/alexdobin/STAR) with the following options: outFilterMismatchNoverLmax 0.04, alignIntronMin 20, alignIntronMax 1000000, outFilterMultimapNmax 20. Gene expression values (FPKM = fragments per kilobase per million reads) were computed by Cufflinks v2.2.1 (RRID:SCR_014597, http://cole-trapnell-lab.github.io/ cufflinks/), and further normalization between samples was done using quantile normalization (R/LIMMA v3.30.7 (RRID:SCR_010943)).
Genomic analyses
DNA was extracted from frozen tumor samples after 2 hours of incubation with proteinase K and 1 h with RNAse A, by a phenol/chloroform and Phase Lock Gel Light (Eppendorf, Hamburg, Germany) procedure. Quantity and quality were estimated in 2 steps. First, a nanodrop spectrophotometer was used to assess gDNA purity. For an optimal labeling yield, samples should have an A260/A280 ≥ 1.8 and A260/A230 ≥ 1.9. Then a qubit dsDNA BR Assay Kit was used to measure the double-stranded DNA concentration.
The design of a custom next-generation sequencing (NGS) panel called DRAGON (for the Detection of Relevant Alterations in Genes involved in Oncogenetics by NGS) and marketed by Agilent under the name of SureSelect CD Curie CGP has been developed specifically for the genomic analysis of tumors. It is composed of 571 genes of interest in oncology from diagnostic, prognostic, and molecular therapy points of view. The nucleotide sequence (variant calling is performed using Varscan2) as well as the number of copies (deletion and focal amplification) were explored. 50 ng of DNA input was used to prepare the library with the Agilent SureSelect XT-HS Target Enrichment System for Illumina Paired-End Multiplex Sequencing kit in accordance with the manufacturer’s protocol. The design uses 571 genes and an additional backbone of probes across the whole genome with an average resolution of one probe every 200 Kb. Additional use of the library allows the perform a complete CNV profile by shallow whole genome sequencing before the capture. This allowed us to determine a genomic profile spanning every chromosome. The copy number profile for each case was estimated using a combination of homemade R scripts and the facets package (v0.6.0) with a sex-specific unmatched-germline control, previously sequenced using the same panel for normalization. Libraries were sequenced over 2*100pb, including 10pb containing UMI (Unique Molecular Identifier) to reach an average depth of 1500X and a minimum depth of 300X on the region of interest.
Read alignment was performed with BWA mem (v0.7.15) on the Human reference genome (hg19 assembly). Concomitantly, paired-end and UMI reads were first converted from FASTQ format to unmapped BAM file using fbio FastqToBam (v0.7.0), and then the query name was sorted using PicardTools SortSam. UMI-tagged unmapped reads were then merged with previously aligned reads using PicardTools MergeBamAlignment. Using fgbio tools, aligned reads sharing the same molecular barcode were grouped together (GroupReadsByUmi, adjacency strategy, without mapping quality filter and allowing one mismatch between UMI sequences) and a filtered consensus read was produced per UMI (CallMolecularConsensusReads, using a phread-scaled error rate of 30 and a minimum number of reads to produce a consensus base set to 1; FilterConsensusReads with following parameters: --reverse-per-base-tags=true –min-reads=1 -E 0.05 -N 40 -e 0.1 -n 0.1). Information regarding UMI consensus was added to the alignment files using PicardTools MergeBamAlignment.
Average read depth on the targeted design and the number of reads per UMI consensus were produced as quality control.
Copy number alterations were called using the combination of homemade R scripts (v3.2.0, R Core Team, 2015) and facets package v0.6.0 with a minimum mapping quality of 0 and a minimum base quality of 15. Focal ( < 10 Mb) amplification and homozygous deletion were extracted from the hyper-segmented profile and aggregated with the normal profile. The resulting segments were then annotated with the targeted genes of the panel. The number of breakpoints was computed per sample in the case of normal segmentation. The genomic profiles obtained after shWGS contained 52,725 segments that were combined into 296 final chromosomal regions containing the largest and most discriminant alterations spanning the 22 chromosomes.
Variant calling of both single nucleotide variations (SNVs) and small insertion/deletions (indels) was then performed on the processed alignment files using samtools and varscan2 (v2.4.3), considering a minimum mapping quality of 0 and base quality of 15 and a maximum read depth of 1 million.
Variants were reported if the number of reads supporting the alternative allele was superior or equal to 2 at a locus covered by at least 10 reads and if the allelic ratio of this variant was superior or equal to 1%. When available, Varscan2 somatic mode was used to call variants using both tumor and matched-germline samples with similar parameters. Since VarScan2 mpileup2cns prevents SNVs from presenting multi-allelic positions to be called, mpileup2snp was therefore used to catch these alterations that were then added to the variants call using an in-house R script.
A dedicated strategy using transIndel (v0.1) was applied to call intermediate-size indels, with the following parameters: minimum mapping quality of 0, read depth above 10X, length between 10 bp and 1000 bp, covered by at least 30X and presenting an allelic ratio above 1%.
Quality controls were performed using FastQC (v0.11.5) and MultiQC (v1.7), and coverage quality control per sample was performed using a combination of awk, Samtools mpileup, and bedtools intersect.
Histopathology and immunohistochemistry
Fresh surgical specimens were analyzed by a pathologist after surgery in the Institut Curie Pathology Department. Suspected WD and DD components and normal adipose tissue were placed in RPMI and sent for RNAseq analyses. Each component had a mirror sample, which was fixed in formalin, processed, and embedded in paraffin blocks. All these mirror samples and all tumors were reviewed by pathologist experts in the field of sarcoma (RRePS network) to confirm the pathological subtype and the WD and DD components. Serial 5 µm sections were incubated with specific antibodies recognizing tumoral cells (MDM2, SantaCruz, ref NCL-L-CD163, clone 10D6, pH6, dil 1:100), macrophages (CD163, Novocastra, ref NCL-L-CD163, clone 10D6, pH6, dil 1:100), lymphocytes (CD3, Dako Agilent, ref A0452, pH9, dil 1:100, FOXP3, Master Diagnostica, ref MAD-000536-QD, clone SP97, pH9, ready-to-use), and DDLPS progenitors (CD55, CST, ref 31759, clone E7G2U, dil 1 :600, pH 9). Immunohistochemistry was processed in a Bond RX automated (Leica) with Bond Polymer refine detection kit (Leica, ref DS9800), associated with Bond Polymer Refine Red Detection (Leica, ref DS9390) in case of double staining. Antigen retrieval was performed in BOND Epitope Retrieval Solution 1 pH6 (Leica, ref AR9961) or Solution 2 pH9 (Leica, ref AR9640). Primary antibodies were incubated for 60 min at room temperature. Positive and negative controls were included in each slide run. Slides were counterstained with hematoxylin before mounting with resin. All pathological analyses were performed after scanning slides with Nanozoomer HT2.0 (Hamamatsu) at × 20 magnification. IHC-stained cells were quantified by taking images of multiple fields and normalizing them on the same number of total cells in WD and DD samples.
In situ multiplex immunofluorescence
Paraffin-embedded tissue blocks were cut with a microtome into fine slivers of 3 microns. Immunostaining was processed in a Bond RX automated (Leica) with Opal™ 7-Color IHC Kits (Akoya Biosciences, ref NEL821001KT) according to the manufacturer’s instructions. The multiplex panels consisted of the following antibodies: lymphoid panel: CD4 (Cell signaling, ref 48274, dil 1: 100), FOXP3 (Diagomics, ref MAD-000536QD, ready-to-use (RTU)), CD8 (Agilent, ref M710301, dil 1: 200), PD1 (Abcam, ref ab137132, dil 1: 500), MDM2 (Diagomics, ref MAD-000682QD, RTU); myeloid panel: CD68 (Agilent, ref M081401, dil 1: 800), CD163 (Leica, ref NCLCD163, dil 1: 200), S100A8 (Abcam, ref ab92331, dil 1: 100), CD33 (Abcam, ref ab270942, ref 1: 500), MDM2 (Diagomics, ref MAD-000682QD, RTU); progenitor panel: CD55 (Cell signaling, ref 31759, dil 1: 800), CD44 (Abcam, ref ab189524, dil 1: 3000), ALDH1 (BD Biosciences, ref 611194, dil 1: 1000), CD34 (Agilent, ref M716529, dil 1: 200), MDM2 (Diagomics, ref MAD-000682QD, RTU); TGF panel: TGF-β1 (Abcam, ref ab215715, dil 1: 100), CD3 (Agilent, ref ir503, RTU), CD14 (Abcam, ref ab182032, dil 1: 200), MDM2 (Diagomics, ref MAD-000682QD, RTU).
Tissue sections were coverslipped with Prolong™ Diamond Antifade Mountant (Thermo Fisher, ref P36970) and stored at 4 °C. Subsequently, slides were scanned using the Vectra® 3 automated quantitative pathology imaging system (Vectra 3.0.5; Akoya Biosciences). Multispectral images were unmixed and analyzed using the inForm Advanced Image Analysis Software (inForm 2.6.0; Akoya Biosciences) and the HALO software for immune subsets quantification.
Cell culture from PDX tumors, adipogenic and osteogenic differentiation assays
Single-cell suspensions were established from freshly collected PDX tumors and cultured in αMEM medium (Sigma, ref M6074) supplemented with 10% FBS and 1% Penicillin/streptomycin. For differentiation assays, cells were seeded either at 2 × 105 cells/ well (96-well-plates) (Bodipy or Alizarin staining) or 106 cells/well (6-well-plates) for 24 h. Cells were then differentiated with 66 nM insulin (Sigma, ref I9278), 3 µM rosiglitazone (Millipore, ref 250,557366 µM IBMX (Sigma, I5879), 0.1 µg/ml transferrin (Sigma, ref T8158), 1 nM T3 (Sigma, ref T6397), 100 nM hydrocortisone (Sigma, ref H0888) (complete adipogenic cocktail, ADC) for 14 days, or with StemMACS OsteoDiff medium (Milteny, ref 091,130,678) for osteogenic differentiation (ODC) for 21 days.
TGF-β (R&D, ref 240-B-002/CF) was added at 10 ng/ml and changed every 3 days. TGF-βR inhibitors SB431542 (R&D, ref 1614) or galunisertib (Selleckchem, ref S2230) were added at 10 µM, 24 h before adding TGF-ß and ADC.
Cells were either directly fixed for Bodipy or Alizarin red staining, or pelleted for RNA or protein extraction.
Bodipy staining
Cells were fixed for 24 h at 4 °C in PBS containing 4% PFA, and then permeabilized for 30 min at RT with PBS containing 0.2% Triton. The supernatant was removed and replaced by PBS containing 0.4 µg/ml DAPI (Thermo Scientific, ref 62248) and 10 µg/ml Bodipy (Invitrogen, ref D3922). Images were taken on a Leica microscope with a DFC 3000 G camera at 10 × magnification.
Alizarin red staining
Cells were fixed for 24 h at 4 °C in PBS containing 4% PFA, and then stained with 50 µl/well of Alizarin red (Merck, ref 2003999) for 30 min at RT. The supernatant was removed, and cells were then washed with distilled water. Images were taken using an Evos XL Core microscope (Magnification x10). Data were quantified using ImageJ.
Reverse transcription and qPCR
RNA was extracted from cultured cells using Qiazol (Qiagen, ref 79306). RNA obtained from human or murine white adipocytic tissue was used as a control. cDNA was generated using a reverse transcriptase kit (High Capacity cDNA Reverse Transcription Kit, Applied Biosystems, ref 4368814) from 1 μg of total RNA. Assays-on-Demand for assessing the expression level of human FABP4 (Hs1086177_m1), PPARG (Hs01115513_m1), LPL (Hs00173425_m1), as well as the control RPLP0 (Hs99999902_m1) genes were obtained from Applied Biosystems.
The human and murine specific GPX3 primers (Hs-GPX3 F: ACGGGCCAGTACATTGAACTGA; Hs-GPX3 R: GGAAAGCCCAGAATGACCAGAC; Mm-Gpx3 F: GGCTTGGTCATTCTGGGCTTC; Mm-Gpx3 R: CCACCTGGTCGAACATACTTGAGA) and the total (both human and murin) TBP primers (Total-TBP F: TGCACAGGAGCCAAGAGTGAA; Total-TBP R: CACATCACAGCTCCCCACCA) were chosen using the Oligo 6.0 program (National Biosciences, Plymouth, MN). The mouse and the human GPX3 gene primer pairs were selected to be unique when compared to the sequence of their respective orthologous gene. By contrast, the primer pair, referred to as the “Total-TBP” primer pair, was selected to amplify both the mouse and the human TBP genes.
To avoid amplification of contaminating genomic DNA, one of the two primers was always placed at the junction between two exons. For each human-specific primer pair validation, we performed no template control (NTC), no-human-reverse-transcriptase control (human RT negative), mouse-reverse-transcriptase control (mouse RT positive from a pool of normal and tumoral mouse RNAs extracted from various tissue types) assays, which produced negligible signals (Ct > 40), suggesting that primer–dimer formation, genomic DNA contamination and cross-species contamination effects were negligible. The same controls were realized for each mouse-specific primer pair.
Results expressed as N-fold differences in GPX3 target gene expression relative to TBP gene expression and termed “NGPX3” were determined as NGPX3 = 2ΔCtsample, where the ΔCt value of the sample was determined by subtracting the Ct value of the GPX3 gene from the Ct value of the TBP gene (used as endogenous control). NGPX3 values of the samples were also subsequently normalized to obtain a “basal mRNA level” (smallest quantifiable amount of mRNA (Ct = 35)) equal to 1.
RT-qPCR was carried out in a CFX96 touch-deep well real-time PCR detection system (Biorad) using TaqMan Master Mix (Applied Biosystems).
Western blotting
SIL10AS proteins were extracted with RIPA and an anti-protease cocktail (Roche, ref 11–836). Western blots were hybridized with anti-Smad2 (CST, ref CS3122, dil 1: 1,000), anti-Phospho-Smad2 (CST, ref CST3108, dil 1: 1,000), anti-FABP4 (Abcam, ref ab92501, dil 1 :1000), or anti-GAPDH (Proteintech, ref HRP-60004, dil 1: 10,000) at 4 °C overnight and revealed with anti-rabbit IgG horseradish peroxidase coupled (1:20,000, Biorad, ref 17505046). Proteins were detected using chemiluminescence (Biorad, ref 170–5060). Signals were acquired with a CCD camera (Chemidoc, Biorad). P-Smad2/Total Smad2 normalization was performed.
TGF-β dosage
Tissues were processed within 1 h after tumor resection. 2 cm2 of tumoral pieces were cut into small pieces and then dissociated in 2 ml of PBS (Gibco) using the gentleMACS™ tissue dissociators (Miltenyi). Dissociated cells were then filtered with a 70 mm cell strainer, and the supernatant was harvested for quantitative measurement of active TGF-β1, done by sandwich ELISA according to the manufacturer recommendations (Assay Genie Sandwich ELISA kit, AssayGenie, ref HUFI00248). Results were obtained in pg/cm2.
Statistical analyses
Pairwise comparisons between cell group proportions were backed up using Chi2 statistical tests on cell counts. The pairwise_prop_test function from the rstatix R package was used for single-cell data analysis. The propeller package in R was used to perform a moderated t test on logit transformed cell type proportions to test for statistically significant associations with the clinical covariates; FDR correction was applied for multiple testings.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
The scRNA-seq and bulk RNA-seq raw data generated in this study have been deposited at the Gene Expression Omnibus (GEO) under the accession code GSE221494. The shallow WGS and targeted DNAseq raw data generated in this study have been deposited with links to BioProject accession number PRJNA113972 in the NCBI BioProject database [https://www.ncbi.nlm.nih.gov/bioproject/1139721]. Source data are provided in this paper.
References
Lee, A. T. J., Thway, K., Huang, P. H. & Jones, R. L. Clinical and molecular spectrum of liposarcoma. J. Clin. Oncol. 36, 151–159 (2018).
Board, W.C.o.T.E. WHO Classification of Tumours: Soft Tissue and Bone Tumours, (International Agency for Research on Cancer, 2020).
de Pinieux, G. et al. Nationwide incidence of sarcomas and connective tissue tumors of intermediate malignancy over four years using an expert pathology review network. PLoS ONE 16, e0246958 (2021).
Coindre, J. M., Pedeutour, F. & Aurias, A. Well-differentiated and dedifferentiated liposarcomas. Virchows Arch. 456, 167–179 (2010).
Gronchi, A. et al. Soft tissue and visceral sarcomas: ESMO-EURACAN-GENTURIS Clinical Practice Guidelines for diagnosis, treatment and follow-up(☆). Ann. Oncol. 32, 1348–1365 (2021).
Blay, J. Y. et al. Surgery in reference centers improves survival of sarcoma patients: a nationwide study. Ann. Oncol. 30, 1143–1153 (2019).
Callegaro, D. et al. Development and external validation of two nomograms to predict overall survival and occurrence of distant metastases in adults after surgical resection of localised soft-tissue sarcomas of the extremities: a retrospective analysis. Lancet Oncol. 17, 671–680 (2016).
Gronchi, A. et al. Variability in patterns of recurrence after resection of primary retroperitoneal sarcoma (RPS): A report on 1007 patients from the multi-institutional collaborative RPS working group. Ann. Surg. 263, 1002–1009 (2016).
Jones, R. L., Fisher, C., Al-Muderis, O. & Judson, I. R. Differential sensitivity of liposarcoma subtypes to chemotherapy. Eur. J. Cancer 41, 2853–2860 (2005).
Petitprez, F. et al. B cells are associated with survival and immunotherapy response in sarcoma. Nature 577, 556–560 (2020).
Italiano, A. et al. Pembrolizumab in soft-tissue sarcomas with tertiary lymphoid structures: a phase 2 PEMBROSARC trial cohort. Nat. Med. 28, 1199–1206 (2022).
Thway, K. Well-differentiated liposarcoma and dedifferentiated liposarcoma: An updated review. Semin. Diagn. Pathol. 36, 112–121 (2019).
Dalal, K. M., Kattan, M. W., Antonescu, C. R., Brennan, M. F. & Singer, S. Subtype specific prognostic nomogram for patients with primary liposarcoma of the retroperitoneum, extremity, or trunk. Ann. Surg. 244, 381–391 (2006).
Micci, F., Teixeira, M. R., Bjerkehagen, B. & Heim, S. Characterization of supernumerary rings and giant marker chromosomes in well-differentiated lipomatous tumors by a combination of G-banding, CGH, M-FISH, and chromosome- and locus-specific FISH. Cytogenet. Genome Res. 97, 13–19 (2002).
Louis-Brennetot, C. et al. The CDKN2A/CDKN2B/CDK4/CCND1 pathway is pivotal in well-differentiated and dedifferentiated liposarcoma oncogenesis: an analysis of 104 tumors. Genes Chromosomes Cancer 50, 896–907 (2011).
Garsed, D. W. et al. The architecture and evolution of cancer neochromosomes. Cancer Cell 26, 653–667 (2014).
Momand, J., Zambetti, G. P., Olson, D. C., George, D. & Levine, A. J. The mdm-2 oncogene product forms a complex with the p53 protein and inhibits p53-mediated transactivation. Cell 69, 1237–1245 (1992).
Cisse, M. Y. et al. Targeting MDM2-dependent serine metabolism as a therapeutic strategy for liposarcoma. Sci. Transl. Med. 12, https://doi.org/10.1126/scitranslmed.aay2163 (2020).
Barretina, J. et al. Subtype-specific genomic alterations define new targets for soft-tissue sarcoma therapy. Nat. Genet. 42, 715–721 (2010).
Cancer Genome Atlas Research Network Electronic address, e.d.s.c. & cancer genome atlas research, N. comprehensive and integrated genomic characterization of adult soft tissue sarcomas. Cell 171, 950–965.e28 (2017).
Chibon, F. et al. ASK1 (MAP3K5) as a potential therapeutic target in malignant fibrous histiocytomas with 12q14-q15 and 6q23 amplifications. Genes Chromosomes Cancer 40, 32–37 (2004).
Mariani, O. et al. JUN oncogene amplification and overexpression block adipocytic differentiation in highly aggressive sarcomas. Cancer Cell 11, 361–374 (2007).
Amin-Mansour, A. et al. Genomic evolutionary patterns of Leiomyosarcoma and Liposarcoma. Clin. Cancer Res. 25, 5135–5142 (2019).
Mereu, E. et al. Benchmarking single-cell RNA-sequencing protocols for cell atlas projects. Nat. Biotechnol. 38, 747–755 (2020).
Zhou, Y. et al. Single-cell RNA landscape of intratumoral heterogeneity and immunosuppressive microenvironment in advanced osteosarcoma. Nat. Commun. 11, 6322 (2020).
Wei, Y. et al. Single-cell analysis and functional characterization uncover the stem cell hierarchies and developmental origins of rhabdomyosarcoma. Nat. Cancer 3, 961–975 (2022).
Patel, A. G. et al. The myogenesis program drives clonal selection and drug resistance in rhabdomyosarcoma. Dev. Cell 57, 1226–1240.e8 (2022).
Baslan, T. & Hicks, J. Unravelling biology and shifting paradigms in cancer with single-cell sequencing. Nat. Rev. Cancer 17, 557–569 (2017).
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 16, 1289–1296 (2019).
Schroeder, B. A. et al. CD4+ T cell and M2 macrophage infiltration predict dedifferentiated liposarcoma patient outcomes. J. Immunother. Cancer 9, https://doi.org/10.1136/jitc-2021-002812 (2021).
Dancsok, A. R. et al. Tumor-associated macrophages and macrophage-related immune checkpoint expression in sarcomas. Oncoimmunology 9, 1747340 (2020).
Alaluf, E. et al. Heme oxygenase-1 orchestrates the immunosuppressive program of tumor-associated macrophages. JCI Insight 5, https://doi.org/10.1172/jci.insight.133929 (2020).
Katzenelenbogen, Y. et al. Coupled scRNA-Seq and intracellular protein activity reveal an immunosuppressive role of TREM2 in cancer. Cell 182, 872–885 (2020).
Merrick, D. et al. Identification of a mesenchymal progenitor cell hierarchy in adipose tissue. Science 364, https://doi.org/10.1126/science.aav2501 (2019).
Emont, M. P. et al. A single-cell atlas of human and mouse white adipose tissue. Nature 603, 926–933 (2022).
Schwalie, P. C. et al. A stromal cell population that inhibits adipogenesis in mammalian fat depots. Nature 559, 103–108 (2018).
Kannan, S., Lock, I., Ozenberger, B. B. & Jones, K. B. Genetic drivers and cells of origin in sarcomagenesis. J. Pathol. 254, 474–493 (2021).
Cassinelli, G., Pasquali, S. & Lanzi, C. Beyond targeting amplified MDM2 and CDK4 in well differentiated and dedifferentiated liposarcomas: From promise and clinical applications towards identification of progression drivers. Front. Oncol. 12, 965261 (2022).
Hirata, M. et al. Integrated exome and RNA sequencing of dedifferentiated liposarcoma. Nat. Commun. 10, 5683 (2019).
Kanojia, D. et al. Genomic landscape of liposarcoma. Oncotarget 6, 42429–42444 (2015).
Hofvander, J. et al. Different patterns of clonal evolution among different sarcoma subtypes followed for up to 25 years. Nat. Commun. 9, 3662 (2018).
Liu, W. et al. Integrated genomic and transcriptomic analysis revealed mutation patterns of de-differentiated liposarcoma and leiomyosarcoma. BMC Cancer 20, 1035 (2020).
Rodeheffer, M. S., Birsoy, K. & Friedman, J. M. Identification of white adipocyte progenitor cells in vivo. Cell 135, 240–249 (2008).
Sanchez-Gurmaches, J. & Guertin, D. A. Adipocyte lineages: tracing back the origins of fat. Biochim. Biophys. Acta 1842, 340–351 (2014).
Ferrero, R., Rainer, P. & Deplancke, B. Toward a consensus view of mammalian adipocyte stem and progenitor cell heterogeneity. Trends Cell Biol. 30, 937–950 (2020).
Burl, R. B. et al. Deconstructing adipogenesis induced by beta3-adrenergic receptor activation with single-cell expression profiling. Cell Metab. 28, 300–309 (2018).
Raajendiran, A. et al. Identification of metabolically distinct adipocyte progenitor cells in human adipose tissues. Cell Rep. 27, 1528–1540.e7 (2019).
Beird, H. C. et al. Genomic profiling of dedifferentiated liposarcoma compared to matched well-differentiated liposarcoma reveals higher genomic complexity and a common origin. Cold Spring Harb. Mol. Case Stud. 4, https://doi.org/10.1101/mcs.a002386 (2018).
Binh, M. B. et al. Dedifferentiated liposarcomas with divergent myosarcomatous differentiation developed in the internal trunk: a study of 27 cases and comparison to conventional dedifferentiated liposarcomas and leiomyosarcomas. Am. J. Surg. Pathol. 31, 1557–1566 (2007).
Dantey, K., Schoedel, K., Yergiyev, O., Bartlett, D. & Rao, U. N. M. Correlation of histological grade of dedifferentiation with clinical outcome in 55 patients with dedifferentiated liposarcomas. Hum. Pathol. 66, 86–92 (2017).
Evans, H. L., Khurana, K. K., Kemp, B. L. & Ayala, A. G. Heterologous elements in the dedifferentiated component of dedifferentiated liposarcoma. Am. J. Surg. Pathol. 18, 1150–1157 (1994).
Matushansky, I. et al. A developmental model of sarcomagenesis defines a differentiation-based classification for liposarcomas. Am. J. Pathol. 172, 1069–1080 (2008).
Burns, J. et al. The proteomic landscape of soft tissue sarcomas. Nat. Commun. 14, 3834 (2023).
Grafe, I. et al. TGF-beta Family Signaling in Mesenchymal Differentiation. Cold Spring Harb. Perspect. Biol. 10, https://doi.org/10.1101/cshperspect.a022202 (2018).
Mazzu, Y. Z., Hu, Y., Shen, Y., Tuschl, T. & Singer, S. miR-193b regulates tumorigenesis in liposarcoma cells via PDGFR, TGFbeta, and Wnt signaling. Sci. Rep. 9, 3197 (2019).
Acknowledgements
This work was supported by funding from ANR-10-EQPX-03, Institut Curie Génomique d’Excellence (ICGex) (S.W.), Emergence Cancéropôle Ile-de-France 2020-1-EMERG-41-ICR-1 (S.W.), Sarcoma Foundation of America SFA13-19 (J.W. and S.W.), GEFLUC les entreprises contre le Cancer 2021 (S.W.), Fondation ARC pour la Recherche Contre le Cancer ARCPJA2021060004026 (S.W.) and INTERSARC 2022-008 (S.W.). S.W. was supported by a grant from the Institut National de la Santé et de la Recherche Médicale (INSERM) and the Foundation Bettencourt-Schueller, and by a grant from the Foundation Nuovo-Soldati. J.V. was supported by a grant from la Ligue Contre le Cancer and by Institut Curie. J.W. acknowledges support from SIRIC INCa-DGOS-INSERM_12554. The ICGex NGS platform of the Institut Curie was supported by the grants ANR-10-EQPX-03 (Equipex) and ANR-10-INBS-09-08 (France Génomique Consortium) from the Agence Nationale de la Recherche (“Investissements d’Avenir” program), by the ITMO-Cancer Aviesan (Plan Cancer III) and by the SiRIC-Curie program (SiRIC Grant INCa-DGOS-4654). S.W. acknowledges support from the non-profit organization 1,2,3 Sarcom’ence. We thank the patients, their family members, and the clinicians involved in their care. We also acknowledge support from Institut Curie for sample collection, banking, and processing: the Biological Resource Center and its members, the ICGEX platform and its members, the LIP platform and its members, the Unité de Génétique Somatique and its members, and the Department of Pathology and its members.
Author information
Authors and Affiliations
Contributions
N.G., C.Q., G.P., and S.W. conceived the study, performed the experiments, analyzed the results, and wrote the manuscript. L.L. performed pathological, immunohistochemical, and immunofluorescence experiments. S.E.Z. performed pathological review and characterization of tumor samples and analyzed pathological and immunohistochemical experiments. S.Bo. and D.T. provided access to patient samples. J.V. and F.Q. participated in bioinformatic analyses. I.B., K.A., B.M., and L.C. participated in experimental procedures. N.E.T. acted as project manager. A.D., H.D., D.D., L.K.L., and L.G. performed PDX establishment and management. O.M. provided access to archived tumor samples. S.Ba. provided support for NGS experiments. J.J.W., G.P., and O.D. provided scientific guidance. All authors read and approved the final version of the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Communications thanks Junyue Cao, Ping Chi, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Source data
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Gruel, N., Quignot, C., Lesage, L. et al. Cellular origin and clonal evolution of human dedifferentiated liposarcoma. Nat Commun 15, 7941 (2024). https://doi.org/10.1038/s41467-024-52067-1
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-024-52067-1
- Springer Nature Limited
This article is cited by
-
A metabolic crosstalk between liposarcoma and muscle sustains tumor growth
Nature Communications (2024)