Evolution of Dermatofibrosarcoma Protuberans to DFSP-Derived Fibrosarcoma: An Event Marked by Epithelial–Mesenchymal Transition–like Process and 22q Loss

Dermatofibrosarcoma protuberans is a rare cutaneous sarcoma known for its indolent course, its tendency to recur mostly at the site of the primary tumor and its high cure-rate when it can be completely resected (1–3). The superficial location and the very low metastatic potential of classic dermatofibrosarcoma protuberans (DFSP) may justify a less aggressive surgical approach as compared with what applied to soft tissue sarcomas (STS). However, a fibrosarcomatous (FS) transformation may occur in 5% to 15% of DFSP, and fibrosarcomatous dermatofibrosarcoma protuberans (FS-DFSP) is associated to an increased risk of metastases, in the range of 10% to 15% (4, 5).

In most cases, the presence of fibrosarcomatous features can be detected in the primary tumor, but in some patients, it is identified only on the relapsed tumor (4, 5). Data on the proportion of patients who experience a fibrosarcomatous evolution from a pure DFSP are scanty.

DFSP and FS-DFSP are characterized by the reciprocal chromosome translocation t(17;22)(q22;q13.1) or, more often, supernumerary ring chromosome or markers derived from t(17;22), in which the collagen type Iα1 gene (COL1A1) on chromosome 17 is fused to the platelet-derived growth factor β-chain gene (PDGFB) on chromosome 22. The resulting upregulation of the PDGFB protein activates the PDGFB receptor (PDGFRB), inducing tumor growth through an autocrine–paracrine loop (4, 6, 7).

COL1A1–PDGFB rearrangement is of major help in confirming the diagnosis in those cases in which the classic component is completely lost (4, 8). COL1A1–PDGFB retainment in FS-DFSP suggests that other still undetermined genetic abnormalities account for progression of DFSP to FS-DFSP.

Therefore, we investigated the natural history of DFSP evolution to FS-DFSP in patients presenting to the Fondazione IRCCS Istituto Nazionale dei Tumori (Milan, Italy; INT) with a primary lesion, from 1993 to 2012. In addition, we molecularly profiled a set of DFSP and FS-DFSP by a combined approach, including whole transcriptome sequencing, copy number analysis, FISH, qRT-PCR, and immunohistochemical investigations, to gain insights into the molecular bases of fibrosarcomatous progression and to identify possible prognostic factors and druggable targets.

All consecutive patients affected by primary DFSP, located at any site, and treated at INT from January 1993 to December 2012 were considered, including only patients admitted for primary DFSP, either for initial treatment or reexcision after a previous inadequate surgery performed elsewhere. Patients with recurrent disease, defined as tumor regrowth at the original tumor site at least 6 months after treatment at another institution, were excluded. Data were extracted from a prospectively maintained database, including all adult patients with STS treated at INT.

In all cases, diagnosis was confirmed by an expert pathologist (S. Pilotti). All surgical excisions were macroscopically complete and classified according to the closest surgical margin, which was microscopically categorized as positive (tumor within 1 mm from the inked surface, R1) or negative (absence of tumor within 1 mm from the inked surface, R0). Before surgery, all patients underwent staging with chest X-ray or CT scan.

After surgery, all patients were followed up every 6 months for the first 5 years and then every year for, at least, 5 more years. Generally, follow-up included a clinical exam, local ultrasound or MR, along with a chest X-ray or chest CT every other visit. Last follow-up was defined as a patient's last recorded visit at INT at the time of data collection. Local recurrence (LR) and distant metastasis (DM) were confirmed with a biopsy in all cases.

Overall survival (OS) and crude incidence of LR and DM were evaluated. Events were considered the first evidence of local or distant relapse for crude cumulative incidence (CCI) of LR or CCI of DM and death related to any cause for OS. Event times were computed from the date of surgery to the date of the event occurrence.

To gain insights into mechanisms involved in fibrosarcomatous progression, tumor samples from 5 patients suffering from DFSP and 5 patients with FS-DFSP were collected for molecular profiling. Peripheral blood sample collection from each patient included in the translational study was also planned upfront in case the assessment of somatic/germline status of putative mutation(s) had to be determined. These cases were selected on the basis of the availability of frozen samples and peripheral blood and did not belong to the surgical series, having been operated on after 2012.

This retrospective study was approved by the ethics committee of INT.

Fibrosarcomatous change was defined by the appearance in a classical storiform DFSP of at least 5% of high cellular area made up of spindle cells arranged in an herringbone pattern with a mitotic index >7/10 high power fields (9); necrosis, pleomorphic features, and myofibroblastic differentiation were also recorded and considered in making FS-DFSP diagnosis, as well as decrease/disappearance of CD34 immunoreactivity (10). Notably, in the current series, with the exception of case D6 in Table 1, which presented both components, fibrosarcomatous change accounted for more than 90%. IHC was performed using antibodies directed against the following proteins: Ki67, INI1, TWIST1, SNAI2/SLUG, ZEB1, EZH2, and p16. Detailed antibody information and staining conditions are provided in Supplementary Information S1.

FISH analyses were performed on 2-μm thick formalin-fixed paraffin-embedded (FFPE) tissue sections by using the indicated BAC probes (Children Hospital Research Institute, Oakland, CA): PDGFB, RP11-630N12 at the telomeric end and RP11-506F7 at centromeric end; COL1A1, RP11-131M15 at the telomeric end and RP11-93L18 at centromeric end.

BACs were labeled in SpectrumGreen (SG) or SpectrumOrange (SO; Abbott Molecular) by nick translation (Abbott Molecular). Probe labeling and FISH slide preparation were carried out according to standard protocols. The presence of the rearrangement was revealed by using either SG-labeled COL1A1 BAC probes together with SO-labeled PDGFB probes (COL1A1–PDGFB fusion) or differentially labeled (centromeric in SO, telomeric in SG) PDGFB probes (PDGFB break apart).

CDKN2A/p16 gene status was assessed by FISH on 2-μm thick FFPE tissue sections using the Vysis LSI CDKN2A/CEP 9 (Abbott Molecular).

Detailed methodologic procedures are provided in Supplementary Information S1. Briefly, whole transcriptome sequencing (RNA-seq) was performed on RNA isolated from fresh-frozen samples of 5 DFSP and 5 FS-DFSP using the TruSeq RNA Sample Prep v2 protocol (Illumina). An average of 94.5 million reads per sample was obtained.

The short reads were mapped on the human reference genome by TopHat/Bowtie pipeline. Large chromosomal rearrangements were detected with DeFuse, ChimeraScan, and FusionMap tools. For gene expression profile, after the calculation of normalization factors to scale the raw library size (calcNormFactors method), the functions lmFit followed by eBayes were adopted to perform the DE computation. Unsupervised principal component analysis (PCA) was performed with TM4-MEV (http://www.tm4.org) and visualized with the R package lattice.

Gene enrichment and prioritization were analyzed by using ToppGene, GSEA, WebGestalt, and PGE suites.

Genomic DNA was extracted with QIAamp DNA Mini Kit (Qiagen), labeled, and hybridized to Cytoscan HD Arrays (Affymetrix) following the manufacturer's instructions. Quality was checked by SNP QC and MAPD calculation. Copy number analysis was performed with Chromosome Analysis Suite (ChAS) Software, by applying hidden Markov model algorithm to detect amplified and deleted segments, with respect to a normal reference model file. To control for hyperfragmentation, adjacent segments separated by <50 probes were combined into one single segment, and only segments >50 probes were considered.

Expression analysis of EZH2, HOTAIR, H19, let-7a, and let-7b was performed by qRT-PCR on a CFX96 Real-Time Apparatus (Bio-Rad). All experiments were performed in triplicate and repeated at least twice. Detailed methodologic procedures are provided in Supplementary Information S1.

A total of 263 patients were identified. A total of 243 (92.4%) patients were affected by DFSP and 20 (7.6%) by FS-DFSP. Main patient characteristics are listed in Table 2. COL1A1–PDGFB fusion gene was confirmed by FISH in 18 of 20 FS-DFSP cases, whereas in 2 of 20 cases, FISH analysis was not performed due to lack of material.

The two groups of patients shared similar clinical characteristics. Microscopic surgical margins were positive in 23 (8.8%), 21 of 243 (8.6%) DFSPs, and 2 of 20 (10%) FS-DFSPs, respectively. Among patients with positive surgical margins, 13 cases (56.5%) had the tumor located in the head and neck, 5 in the trunk (2 of them at sternoclavicular area, partially involving the neck), 5 in the extremities (1 involved the foot and 2 the groin/urogenital area).

Median follow-up from the time of definitive surgery was 85 (range 12–194) months. At the time of the current analysis, 261 of 263 patients are alive and 2 of 263 are dead from other diseases. None of the patients is dead of disease and 2 patients are alive with metastatic disease.

Overall, 5/10/15–year OS were 99%/99%/93%, respectively.

LR occurred in 4 patients (1.5%), being in all the first event, following an R1 resection of the primary tumor in all 4 cases. Local relapse occurred in 2 of 21 (9.5%) DFSPs and in 2 of 2 (100%) FS-DFSPs operated on with positive microscopic margins. None of these patients subsequently developed DM. The other 18 patients operated on with positive margins remained disease free at a median follow-up of 96 months after surgery (range 19–147).

Two patients affected by FS-DFSP developed DM (0.76% of the whole series; 10% of FS-DFSP), whereas none of the patients affected by DFSP ever developed metastatic disease. The 5-, 10-, 15-year incidence of DM was 0%/0%/0% and 5%/10%/10% in DFSP and FS-DFSP, respectively. In all cases, DM was the first event. Metastases were located to the lung (1) and the pancreas (1). DM occurred after 26 and 125 months. The presence of the fusion gene (COL1A1–PDGFB) in the metastatic tissue was confirmed in all cases.

Tumors and peripheral blood from 10 patients suffering from DFSP (5) and FS-DFSP (5), operated on at INT after 2012, were collected. The characteristics of the tissue samples molecularly evaluated are summarized in Table 1 and in Supplementary Table S1. In particular, they were all derived from different patients (no pair-matched samples are included), from the primary tumor in 6 cases, local relapse in 2, and distant metastasis in 2. All cases but one were histologically homogeneous throughout the tumor sample; one case (D6, Table 1) exhibited instead both classic and fibrosarcomatous components (Fig. 1). At frozen section control, performed in all the cases, sampling of this case felt in the DFSP area.

FISH confirmed the presence of COL1A1–PDGFB in all cases. High levels of COL1A1–PDGFB fusion copy number/amplification was observed in 3 cases: 2 FS-DFSPs (D2, D4) and one DFSP (D6). No strong evidence of intratumoral heterogeneity for the fusion was observed, with the exception of the cases showing COL1A1–PDGFB amplification.

The gene expression profile was computed from whole transcriptome data and compared between the 5 DFSP and 5 FS-DFSP tumors. Unsupervised PCA of gene expression data showed that the DFSP group had a transcriptome profile highly homogenous. A considerable degree of heterogeneity was instead observed among FS-DFSP, with one case (D3/L168) clustering with DFSP (Supplementary Fig. S1). Gene expression analysis identified 439 differentially expressed transcripts, with 174 upregulated and 265 downregulated genes in FS-DFSP compared with DFSP (log2 fold change greater than +0.6; Supplementary Table S2).

Pathway analysis (Supplementary Table S2) highlighted an enrichment in GO-biological process terms related to extracellular matrix (ECM) organization, extracellular structure organization, cell migration, collagen fibril organization, locomotion, and collagen metabolic process, beside morphogenesis, neuro- and vasculogenesis, and cell proliferation. These analyses also pinpointed to a role for ECM and ECM–receptor interaction. Highly overrepresented Reactome terms included ECM organization and collagen formation, and ECM–receptor interaction was the top list term in KEGG pathways. Pathway analysis failed to underscore a significant involvement of an adaptive immune response in fibrosarcomatous evolution.

Twenty-nine genes differentially expressed in FS-DFSP versus DFSP belonged to the “epithelial–mesenchymal transition (EMT) hallmark gene signature,” according to the GSEA Hallmark gene sets (P = 6.97E−24). Moreover, our dataset overlapped with the DNA methylation map of pluripotent and differentiated cells (66 genes; ref. 11) and with GSEA C2 curated datasets related to stemness (Supplementary Table S2). On these grounds, we investigated by IHC the expression of three major EMT regulators, that is, TWIST1, SNAI2/SLUG, and ZEB1. Besides acting as EMT regulators, these transcription factors are also involved in stem cell maintenance, mesoderm development, and neural crest specification. Intriguingly, FS-DFSP displayed an increased number of decorated nuclei for any of the three EMT master regulators tested (Supplementary Fig. S5). These genes did not emerge as differentially expressed by transcriptome sequencing analysis, but their protein product is known to undergo posttranscriptional regulation (12, 13).

Positional gene enrichment analysis unveiled that a significant fraction of genes modulated during the transition from DFSP to FS-DFSP belonged to the 22q13 and 12q chromosome regions (Supplementary Table S2). In particular, of the 24 genes significantly modulated mapping on 22q, 16 were downregulated in FS-DFSP. These downregulated genes were all telomeric to PDGFB, whereas those centromeric to PDGFB tended to be upregulated.

To get insights on these findings, 5 DFSPs and 5 FS-DFSPs were analyzed by SNP array to identify copy number variations (CNV). CNV data were then integrated with the fusion transcript detection performed on RNA-seq data to build the complete picture of genomic structural variations (Fig. 2; Supplementary Table S3). Several alterations were estimated as mosaic and thus were representative of the clonal architecture of the tumor.

DFSPs were characterized by several macroscopic and few focal amplifications or deletions (mean: 7.7 regions for sample). On the contrary, FS-DFSP carried numerous alterations with a relevant increase of focal amplifications and deletions (mean: 27.2 regions for sample). Interestingly, the only fibrosarcomatous patient carrying a lower number of genomic alterations (D3) was the one showing a gene expression profile clustering with DFSP. Among the 182 altered genomic regions, the majority (69%) were gains of one or more copies. DFSP and FS-DFSP shared gains of 3 or more copies of the chromosomal regions flanking the COL1A1–PDGFB breakpoint. Namely, gain of the entire 17q21.33-q25.3 of chromosome 17 region (telomeric to COL1A1) and gain of the 22q12.3-q13.1 region (centromeric to PDGFB) were detected in all cases. FS-DFSP tended to show a higher frequency of copy number losses compared with DFSP. Gain of the whole chromosome 12 was observed in 2 of 5 FS-DSFPs, a finding that is in line with RNA-seq cytoband gene enrichment analysis. Macroscopic amplification of chromosome 5 was detected in 3 FS-DFSPs and 2 DFSPs, these latter carrying also gain of chromosome 1 and 4. One DFSP carried an extra copy of chromosome 8. Gains of chromosome 4, 5, 8, and 12 have been previously reported in DSFP (14).

Interestingly, in line with in silico predictions based on RNA-seq data, SNP array analysis highlighted the deletion of the 22q13-ter chromosome region, just telomeric to the PDGFB breakpoint, in 4 FS-DFSP. Loss of 22q13-ter was observed also in the classic component of the tumor (case D6) that displayed both components and was the only locally relapsed DFSP included in the study (Fig. 3A).

Apart from few discrepancies (likely due to intratumoral heterogeneity, nontumoral cell contamination, complex, and cryptic rearrangements), FISH analyses supported SNP array data and corroborated the involvement of chromosome 22 loss in fibrosarcomatous evolution. Precisely, beside the translocated alleles (one in case D5, 2–3 in cases D1, D3, D7, D8, D9, D11, and more than three in cases D2, D4, D6) and two intact COL1A1 genes, all but one DFSP carried two nontranslocated PDGFB alleles (chromosome 22), whereas all FS-DFSP retained just one copy (Fig. 3B and C and Supplementary Fig. S3). The patterns observed fit well with the features of the DFSP t(17;22) rearrangement as described in classic cytogenetics (see Cancer Genome Anatomy Project, http://cgap.nci.nih.gov/), which, with the exception of the juvenile forms (not included in this study), is typically unbalanced in origin. Indeed, the most frequent abnormality reported is the presence of one or more copies of a der(22) t(17;22) or related rings/marker chromosomes, together with two copies of normal chromosome 17. Moreover, in the majority of DFSP karyotypes, two copies of a normal chromosome 22 are also present, while the remaining karyotypes (about 1/3 in the Cancer Genome Anatomy Project files) show a single intact copy of chromosome 22. Intriguingly, in our series, the presence of one single copy of nontranslocated PDGFB/nontranslocated chromosome 22 segregated with the fibrosarcomatous form. Taken together, our results indicate that the detection of chromosome 22q13-ter loss by SNP array likely reflects the retention of a single intact chromosome 22 in FS-DFSP.

A scrutiny of the potential tumor suppressor genes mapping within the 22q13-ter deleted region in FS-DFSP pointed to the presence of two members of the let-7 family of miRNA, let-7a-3 and let-7b. qRT-PCR confirmed the downregulation of let-7a and let-7b in 22q deleted FS-DFSP (Fig. 3D). Intriguingly, these miRNAs have been shown to target, among the others, EZH2 (see below).

Both SNP array and FISH analyses pinpointed a possible role of 9p21 loss in metastatic evolution of FS-DFSP. In fact, SNP array revealed high instability of the 9p chromosome region, with multiple deletions of one copy and homozygous loss of the 9p21.3 cytoband, in the 2 FS-DFSP metastatic cases (D2, D9). This chromosome region harbors the p16/INK4A/CDKN2A locus. The deletion of CDKN2A was confirmed by FISH and correlated with loss of p16 expression, as detected by IHC (data not shown).

Finally, integration of SNP array and RNA-seq data indicated that, beside the pathognomonic COL1A1–PDGFB chimera, FS-DFSP expressed several additional fusion genes (Supplementary Table S3), indicating a high degree of genomic instability.

The overlap of the transcriptional profile of FS-DFSP with the DNA methylation map of pluripotent and differentiated cells (Supplementary Table S2; ref. 11) suggested a role for epigenetic remodeling in fibrosarcomatous evolution. Immunohistochemical analyses ruled out a role for INI1 loss in FS-DFSP pathogenesis (data not shown). Instead, the integration of molecular analyses suggested a role for the H3K27 methyltransferases EZH2, a component of the polycomb repressive complex 2. In fact, H19 and HOTAIR, two noncoding RNAs involved in the regulation of the activity of EZH2, were among the genes most significantly upregulated in FS-DFSP. The elevated expression of these long noncoding RNAs in FS-DFSP was confirmed by qRT-PCR (Supplementary Fig. S4). Although EZH2 was not included in the set of genes detected as differentially expressed by RNA-seq, immunohistochemical analyses revealed an increased number of EZH2 immunolabeled nuclei in FS-DFSP with respect to DFSP cases, which was paralleled by a higher Ki67 score (Fig. 4 and Supplementary Fig. S5). qRT-PCR confirmed higher expression levels of EZH2 among FS-DFSP (Supplementary Fig. S4). Intriguingly, EZH2 is among the targets of two miRNAs, let-7a and let-7b, that map in the 22q region that was found deleted in FS-DFSP.

In this retrospective series, fibrosarcomatous transformation was confirmed to be a rare event that affected only 20 (7.6%) of 263 patients surgically treated for a primary DFSP. All FS-DFSP carried the COL1A1–PDGFB fusion gene. In DFSP, negative margin resection was always curative. No DFSP patient suffered from distant relapses. Local relapses were observed only in patients treated with R1 resection. Metastases occurred only in 2 patients with FS-DFSP (0.76% of the whole series; 10% of FS-DFSP), and it was the first event. The molecular profiling of a subset of 10 cases (5 DFSP and 5 FS-DFSP) indicated that the fibrosarcomatous progression is sustained by a transcriptional reprogramming, which is paralleled by loss of genomic material from a nontranslocated chromosome 22q. In addition, loss of the short arm of chromosome 9, including the p16/CDKN2A/INK4A locus, was detected in the 2 metastatic FS-DFSP samples. EZH2 emerged as a promising druggable target.

We found fibrosarcomatous changes in a minority of cases (7.6%), with available series being consistent with a 5% to 20% rate of fibrosarcomatous transformation (1–4). Of course, institutional referral biases as well as sampling issues especially in oldest cases may be relevant. No differences were detected in sex, age, and anatomic location between DFSP and FS-DFSP. As we had already described (2), in this more recent and larger series, focused only on primary cases, we confirm that negative margin resection was always curative for DFSP and prevented LR (not DM) in FS-DFSP as well. The only factor associated to a metastatic risk was the presence of fibrosarcomatous features. However, this risk was low, and the majority of patients with a fibrosarcomatous transformation were also cured by surgery alone. By contrast, none of DFSP patients developed metastasis. Interestingly, none of our FS-DFSP patients with DM had an LR, suggesting that DM may be inherently related to tumor biology. By contrast, quality of surgical margin was not related to the histology but to the presence of a critical anatomic location. In particular, the LR risk was nil also for FS-DFSP patients, when margins were negative. This supports the notion that a wide surgical resection still remains the gold standard in both DFSP and FS-DFSP.

Cases of fibrosarcomatous transformation occurring over time in DFSP after multiple local relapses are described in other reports (8). We cannot challenge this possibility, as the proportion of patients who relapsed was very low. On the other side, a wide resection or a reexcision in case of positive margins should always be considered in FS-DFSP, given the risk of developing metastatic disease (2, 4, 8), while positive surgical margins may still be accepted in DFSP, especially when cosmesis is an issue. It is therefore evident the importance of an adequate tumor sampling and of correct pathologic diagnosis in decision making.

We ran a set of molecular investigations aimed at shedding light on the mechanisms of fibrosarcomatous evolution and, ideally, at identifying markers of prognostic and predictive value. To this end, we initially compared the transcriptome profile of 5 DFSPs and of 5 FS-DFSPs. All samples expressed the COL1A1–PDGFB fusion transcript, with no significant variation in its expression levels between DFSP and FS-DFSP, as previously reported (4, 8). No other recurrent fusion transcripts that may hallmark the FS-DFSP variant were detected.

Pathway analysis ruled out a major role for the immune response in fibrosarcomatous evolution of DFSP and suggested instead an involvement of tumor microenvironment and ECM remodeling. In particular, the switch from DFSP to FS-DFSP was paralleled by an enrichment in genes belonging to the EMT gene signature, as defined by Gröger and colleagues (15). The EMT program has been linked to aggressive tumor behavior, and modulation of mesenchymal traits toward a more undifferentiated state has been previously reported to play a pivotal role in sarcoma development and progression (16–21). EMT is governed by a set of transcription factors, including TWIST1, SNAI2/SLUG, and ZEB1. Here, we demonstrate that these proteins are markedly increased in FS-DFSP compared with DFSP. Noteworthy, the EMT program involves epigenetic remodeling with DNA and histone modifications (21).

Our study suggests that DFSP evolution to FS-DFSP might be correlated to loss of chromosome material involving the 22q region telomeric to PDGFB. In fact, a significant fraction of genes detected by RNA-seq as modulated in FS-DFSP compared with DFSP turned out to belong to the long arm of chromosome 22, which bears the PDGFB gene. In particular, FS-DFSP exhibited downregulation of genes telomeric to PDGFB (22q13-ter). This was is in keeping with SNP array data, which highlighted in FS-DFSP recurrent loss of material from the 22q-ter chromosome region. FISH analyses indicated that the pattern observed was compatible with the loss of one copy of nontranslocated chromosome 22. To the best of our knowledge, 22q loss, although occasionally reported (22–28), has never been associated to fibrosarcomatous evolution.

Previously, fibrosarcomatous transformation of DFSP had been correlated to the amplification of COL1A1–PDGFB fusion, an association that we and others failed to confirm (4, 22). Also in the series here analyzed, both DFSP and FS-DFSP shared extra copies of the chromosome material flanking the COL1A1–PDGFB breakpoint, ruling out a major role for COL1A1–PDGFB amplification in fibrosarcomatous evolution.

Chromosome 22q loss, in form of deletion or monosomy, is a recurrent event in different types of cancer, comprising GIST and rhabdoid tumors (29–32), and several minimal deleted regions, including 22q13-ter (33), have been identified harboring putative tumor suppressor genes. Interestingly, a scrutiny of the genes targeted by the 22q13-ter loss in FS-DFSP pointed to 2 members of the let-7 family of miRNAs, let-7a and let-7b. Let-7 genes encode tumor suppressor miRNAs that control the expression of a number of proteins involved in proliferation and apoptosis, EMT, adhesion, and migration (34, 35). Among others, let-7 miRNAs have been implicated in the regulation of components of the epigenetic machinery, namely EZH2 (36). EZH2 is the catalytic subunit of the polycomb repressive complex 2 (PRC2) that typically controls histone methylation–mediated gene repression via local chromatin reorganization, but it may also act as a gene activator through a PRC2-independent route. EZH2 plays a critical role in cancer development and progression by epigenetically altering the gene expression program (37, 38). Molecular analyses confirmed the downregulation of let-7a and let-7b in FS-DFSP, and IHC showed an increase of EZH2-immunolabeled nuclei in FS-DFSP. Moreover, 2 noncoding RNAs involved in the regulation of EZH2, H19, and HOTAIR (39–41) turned out to be among the RNAs most significantly upregulated in FS-DFSP. This finding suggests that overexpression of H19 and HOTAIR, together with let-7 downregulation, all concur to induce EZH2 and, in turn, chromatin remodeling and mesenchymal transdifferentiation in FS-DFSP (19, 39–41).

A role for epigenetic reprogramming in fibrosarcomatous transformation was also supported by the finding that 66 genes enriched in FS-DFSP versus DFSP are reported to undergo epigenetic regulation (11), and recent evidence is consistent with the involvement of chromatin remodeling in translocation-associated sarcomas (42). Overall, our analyses point at EZH2 as a promising target in DFSP. Remarkably, selective inhibitors of EZH2 are now available in the clinic. In particular, the EZH2 inhibitor tazemetostat is entering a phase III clinical study in INI1 deleted sarcoma by virtue of a functionally antagonistic relationship between PcG proteins and SWI/SNF complex (43). Our study suggests that this class of drug could be interesting in advanced FS-DFSP, even though EZH2 upregulation is mediated here by mechanism other than the deletion of INI1/SMARCB1.

Structural variation analysis revealed a loss of the short arm of chromosome 9 (9p21), that harbors the CDKN2A locus, in the two metastatic samples included in the study. Interestingly, none of the other patients of the current series has developed metastasis to date. The CDKN2A locus encodes p16/INK4A, an inhibitor of cyclin D–CDK4/6 complexes, and p14ARF that negatively controls HDM2-mediated ubiquitination and degradation of p53. Loss of the CDKN2A promotes uncontrolled cell proliferation and tumor progression (44). In our series, CDKN2A homozygous deletion correlated with loss of p16 expression, as indicated by FISH and IHC. Our finding, which is in line with a recent report by Eilers and colleagues (45), indicates that p16 loss might identify a subset of DFSP/FS-DFSP with poor prognosis. Noteworthy, an epigenetic link between p16 loss and EZH2 expression has been recently suggested (46). The evaluation of p16 gene copy number in the primary tumor of FS-DFSP patients included in the surgical series is ongoing to investigate to what extent this alteration is present at disease onset and correlates with the development of distant metastasis.

Overall, our study confirms that fibrosarcomatous evolution is a rare event in DFSP, marked by a low metastatic risk. Interestingly, none of DFSP patients suffered from distant relapse, and all cases were cured by negative margin surgery. This should be used to inform treatment decision for both DFSP variants. The molecular profiling of DFSP and FS-DFSP implicated EMT and epigenetic remodeling and unveiled a possible role for chromosome 22q13-ter loss in fibrosarcomatous evolution and CDKN2A/p16 loss in dismal prognosis, findings that are worthy of further investigation in larger series. The finding that EZH2 is upregulated in FS-DFSP discloses novel therapeutic opportunities for this subset of tumors.

S. Stacchiotti reports receiving a commercial research grant from Novartis. A. Gronchi has received speakers bureau honoraria from and is a consultant/advisory board for Novartis. M.A. Pantaleo reports receiving a commercial research grant from Novartis. P.G. Casali has received speakers bureau honoraria from and is a consultant/advisory board for Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Stacchiotti, A. Astolfi, A. Gronchi, M.A. Pantaleo, A.P. Dei Tos, C. Castelli, G.P. Dagrada, S. Pilotti, R. Maestro

Development of methodology: S. Stacchiotti, M. Urbini, A.P. Dei Tos, G.P. Dagrada

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Stacchiotti, A. Astolfi, A. Gronchi, M.A. Pantaleo, M. Brenca, M. Urbini, C. Colombo, S. Brich, P.G. Casali, G.P. Dagrada, R. Maestro, A. Fontana

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Stacchiotti, A. Astolfi, A. Gronchi, T. Negri, M. Urbini, V. Indio, S. Radaelli, A.P. Dei Tos, C. Castelli, G.P. Dagrada, S. Pilotti, R. Maestro, A. Fontana, M. Tazzari

Writing, review, and/or revision of the manuscript: S. Stacchiotti, A. Astolfi, A. Gronchi, M.A. Pantaleo, T. Negri, M. Brenca, S. Radaelli, S. Brich, A.P. Dei Tos, P.G. Casali, C. Castelli, G.P. Dagrada, S. Pilotti, R. Maestro, A. Fontana, M. Tazzari

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Stacchiotti, A. Fontana

Study supervision: S. Stacchiotti, A. Gronchi

This work was supported by the Italian Association for Cancer Research (AIRC) and by the Italian Ministry of Health (Ricerca Finalizzata). M. Brenca is the recipient of an FIRC fellowship.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.