The landscape of genetic aberrations in myxofibrosarcoma

Myxofibrosarcoma (MFS) is a rare subtype of sarcoma, whose genetic basis is poorly understood. We analyzed 69 MFS cases using whole‐genome (WGS), whole‐exome (WES) and/or targeted‐sequencing (TS). Newly sequenced genomic data were combined with additional deposited 116 MFS samples. WGS identified a high number of structural variations (SVs) per tumor most frequently affecting the TP53 and RB1 loci, 40% of tumors showed a BRCAness‐associated mutation signature, and evidence of chromothripsis was found in all cases. Most frequently mutated/copy number altered genes affected known disease drivers such as TP53 (56.2%), CDKN2A/B (29.7%), RB1 (27.0%), ATRX (19.5%) and HDLBP (18.9%). Several previously unappreciated genetic aberrations including MUC17, FLG and ZNF780A were identified in more than 20% of patients. Longitudinal analysis of paired diagnosis and relapse time points revealed a 1.2‐fold mutation number increase accompanied with substantial changes in clonal composition over time. Our study highlights the genetic complexity underlying sarcomagenesis of MFS.

What's new?
The genetic basis of myxofibrosarcoma, a rare subtype of sarcoma, remains poorly understood.
This large-scale integrated genetic study of 185 myxofibrosarcoma cases reveals ubiquitous genetic complexity, including the common occurrence of chromothripsis accompanied with local hypermutation. The results also highlight mutually exclusive alterations in CDKN2A/B and HDLBP on the one hand, and co-occurrence of mutations in TP53 and other genes implicated in DNA double-strand break repair on the other hand. Taken together, the findings offer a strong rational for investigating PARP inhibition and/or restoration of normal p53 function as potential treatment avenues for myxofibrosarcoma.

| INTRODUCTION
Soft tissue sarcomas (STS) are a diverse group of tumors with remarkable histologic diversity leading to more than 50 recognized subtypes. 1 Identification of subtype-specific translocations, including SS18-SSX in synovial sarcoma, FUS-DDIT3 in myxoid/round cell liposarcoma and BCOR-CCNB3 in Ewing-like and undifferentiated sarcomas, has revolutionized the diagnostics of sarcoma and has provided new insight into oncogenesis. 2,3 In addition, discovery of activating mutations in KIT or PDGFR in gastrointestinal stromal tumors 4,5 led to routine application of tyrosine kinase inhibitors for these sarcomas, highlighting the great value of genetics for both diagnostics and targeted treatment approaches. Another group of sarcomas are characterized not only by a recurring, tumor-specific genetic alteration, but also by complex karyotypes that are characteristic of severe genetic and chromosomal instability. Most common among this latter group of STS is myxofibrosarcoma (MFS), which typically occurs in late adult life, peaking in the seventh decade and is mainly encountered in the lower extremities. 1 While originally classified as a myxoid-type malignant fibrous histiocytoma, MFS was reclassified as a distinct entity in the WHO classification of 2002 because of its characteristic biological behavior and clinical features, including an infiltrative growth pattern and a high propensity to local recurrence. Besides frequent complex karyotypes, MFS shares many genetic commonalities with leiomyosarcoma (LMS) and undifferentiated pleomorphic sarcoma (UPS), including recurrent alterations affecting known tumor suppressor genes such as CDKN2A/B, TP53 and NF1. [6][7][8] However, the entire molecular pathogenesis of MFS remains incompletely understood due to limited cohort sizes investigated so far.
Here, we conducted a large-scale integrated genetic study of 185 MFSs, including 69 cases from current study and 116 cases from external data sets 7,9 and identified recurrent driver genes, remarkable complex structural variations and high intratumor heterogeneity in MFS.

| Patients and samples
The study cohort comprised 185 cases with MFS: 69 cases from the current study, 17 from TCGA data set and 99 from previous report. 7,9 The number of the samples and analysis were summarized in Figure S1.
In the current study, fresh-frozen tumor and normal tissues/ peripheral blood were provided by Charité Universitätsmedizin Berlin ( For targeted deep sequencing (TS), tumor (fresh frozen and FFPE tissues) and germline (tumor-free tissue or peripheral blood) samples were collected. Libraries were prepared from 100 to 500 ng of genomic DNA and generated using a SureSelect custom kit (Agilent Technologies), followed by massively parallel sequencing of enriched exon fragments on a HiSeq 2500 with 125-base pair paired-end mode, as previously described. 13 Table S2. Mutations were called using the in-house pipeline Genomon2 (version 2.5.2.) and EBCall, as previously described. 13 Validation of mutations identified by WES were performed by amplicon-based deep sequencing and targeted deep sequencing, as previously described. 16 The overall validation rate was 92.1% for fresh samples (35/38 mutations) and 91.4% for FFPE samples (127/139 mutations). Validation of mutations identified by WGS were performed by amplicon-based deep sequencing, TS and WES, as previously described. 16 The overall validation rate was 95.4% (267/280 mutations).
Detection of structural variations was performed by Genomon2.
Briefly, Genomon2 uses information from chimeric reads (containing breakpoints) and discordant read pairs. For each candidate structural variation, it realigns reads to the assembled contig sequence con- Telomere lengths (TLs) were calculated using TelSeq software (https://github.com/zd1/telseq) from WES data, which was validated for the use on the WES data in the original article. The relative TL ratio was defined as tumor telomere length/normal telomere length (tTL/nTL), that is, the tTL divided by the nTL, corresponding to the pair-matched TL ratio information.

| Computational analysis
Statistical analyses were performed using R (http://www.R-project. org). Multiple significance testing was adjusted according to the method described by Hochberg and Benjamini. 18 The methods used for the statistical analyses are described in detail in Section 3. According to an NMF-based decomposition, "clock-like" signatures, namely SBS1, 5 and 40, were predominant, explaining 57.9% of all SNVs. In two cases (ID UPN015 and UPN017; Figure 2), a substantial fraction of SNVs were assigned to SBS3, which is implicated in defective homologous recombination-based DNA damage repair, typically caused by defective BRCA1 or BRCA2. 19 While none of these two cases carried pathogenic somatic/germline variants in  (Table S3). Driver genes significantly mutated or positively selected in MFS were investigated using MutSigCV and dNdSCV with a P-value <.01 (Tables S4 and S5). Besides genes previously known as driver genes in MFS 7-9 such as TP53 (25/102, 24.5%),  (Table S6). Combining 17 MFS cases from TCGA database 9 and 99 from previous report, 7 we analyzed a total of 185 MFS cases. In accordance with the results from WES, FLG and ZNF780A were shown to be recurrently mutated (22.8% and 20.5%, respectively) ( Figure 3B). The incidence and distribution of driver mutations and CNAs are presented in Figure 4. Although infrequent, actionable hotspot mutations were identified in KRAS, BRAF and PIK3CA ( Figure S3). As expected, TP53 alterations were almost mutually exclusive with MDM2 amplifications (P-value = .020, Fisher's exact test).
We also identified recurrent mutations in regulators of the telomeres in 51 cases (27.6%), which included ATRX, SP100, DAXX and RBL2 ( Figure S4). ATRX cooperates functionally with DAXX, whose mutations are associated with altered telomeres. 45,46 While genetic DAXX alterations have not yet been reported in MFSs, 7 we identified two cases with mutations and seven cases with copy number alterations. Deletion of RBL2 showed a significant positive correlation (P-value: .016, by unequal variances t-test) with relative telomere length, which was evaluated by the ratio of tumor telomere length to normal telomere length and calculated by TelSeq ( Figure S5). RBL2, which is also known as Rb-related p130 protein, is reported to suppress telomerase-independent telomere lengthening 47     were marginally associated with poor prognosis (P-value = .046).
No other genetic alterations or clinical features (surgical margin, histological grade) were significantly associated with patients' prognosis ( Figure S7). In addition, we did not identify any significant association between genetic alterations and clinical features

| DISCUSSION
Here we present a comprehensive genetic study, which to our knowledge included the largest number of MFS cases ever studied. We unravel genetic aberrations commonly observed in MFS and report a remarkable intra-/intertumor heterogeneity in MFS tumor tissue. In addition to known genetic alterations, we uncovered several recurrent driver genes, which have not previously been identified as driver genes in MFS. These not only included genes that have been found in other cancer and sarcoma entities such as HDLBP, 48 but also other genes whose role in carcinogenesis is poorly understood (eg, ZNF780A). HDLBP encodes vigilin, an RNA-binding protein that has been implicated in the induction of heterochromatin, and was reported as candidate target for chromosomal 2q37. With respect to prognostic impact of genetic alterations in MFS, we were able to confirm an inferior overall survival for patients harboring RB1 alterations. 7 However, in contrast to previous reports, we did not observe an association between survival and alterations in TP53 nor CDKN2A/B. It should be noted that substantial heterogeneity with respect to presurgical application of chemotherapy with or without hyperthermia and radiotherapy may account for these differences. Even larger cohorts, preferentially enrolled on prospective trials, are warranted to further establish prognostic significance of genetic lesions for individual risk-guidance in MFS.
Taken together, our data provide a comprehensive genetic atlas of MFS sarcomagenesis and suggest at least three avenues for precision medicine guided treatment approaches to further improve patient outcome. access to the datasets is restricted. Note that the EGA provides secure access to restricted data for authorized researchers and clinicians. Further information is available from the corresponding author upon request.

ETHICS STATEMENT
All samples were collected with informed consent following approval of the institutional review boards of the respective institutions.