FET-rearranged Myoepithelial Tumors are Clinically Heterogeneous and Epigenetically Distinct from PLAG1-rearranged Adnexal and Salivary Gland Myoepithelial Tumors

Abstract

Purpose:

Fusion-positive myoepithelial tumors (MET) are clinicopathologically heterogeneous and variably termed mixed tumors and myoepithelial carcinomas. Since FET-rearranged MET lack ductal/epithelial differentiation, we test whether FET-rearranged MET are epigenetically distinct from adnexal PLAG1-rearranged MET, which we hypothesize to be analogues of salivary gland MET.

Experimental Design:

DNA methylation profiling from a multi-institutional cohort of 52 fusion-positive skin, soft tissue and bone MET cases was performed and compared to diverse tumor types, including salivary MET. The MET subgroups harbored EWSR1::KLF15, EWSR1/FUS::KLF17, EWSR1::PBX1, EWSR1::PBX3, EWSR1/FUS::POU5F1, SS18::POU5F1, EWSR1::ZNF444 and PLAG1 rearrangements. Pooled clinicopathological and outcome analysis with new and published cases (total 185) was performed.

Results:

The MET subgroups showed significant heterogeneity in age, site, and histology. Specifically, EWSR1::KLF15 MET affected predominantly young children (<5 years old); EWSR1::PBX1/PBX3 MET were enriched in skin/bone; EWSR1/FUS::POU5F1, SS18::POU5F1 and EWSR1::KLF15 MET tended to display malignant histology. Conversely, PLAG1-rearranged tumors were predominantly benign, arising in older adults and located in the skin. DNA methylation profiling revealed that FET-rearranged MET were epigenetically related to SS18::POU5F1 MET and FET::NFATC2 sarcomas, but entirely distinct from PLAG1-rearranged adnexal and salivary MET. Histologic features were correlated with the degree of genome-wide copy number variation. Median disease-specific survival was shortest in SS18::POU5F1 (31 months), EWSR1::PBX3 (38 months), and EWSR1::KLF15 (45 months) MET. On multivariate analysis, age < 25 years old was a significant predictor of worse progression-free survival.

Conclusions:

FET-rearranged MET are epigenetically unrelated to cutaneous and salivary gland MET, and their malignant counterparts are best classified as sarcomas rather than carcinomas.

1. Introduction

The general concept of myoepithelial tumors (MET) of skin, soft tissue and bone, i.e., mixed tumor, myoepithelioma and myoepithelial carcinoma, was introduced in 1997 by Drs. Kilpatrick and Fletcher who for the first time defined their diagnostic morphological and immunohistochemical criteria (1). The former currently include trabecular, reticular, nested, and/or solid growth of variably spindled or epithelioid cells embedded in (chondro-)myxoid or hyalinized stroma, with mixed tumors often additionally exhibiting ductal differentiation. Immunohistochemically, the tumors should be positive for so called myoepithelial markers, which include broad spectrum cytokeratins (CK) and/or EMA coupled with the expression of S100, SOX10, or GFAP (2). According to Hornick and Fletcher, the distinction between benign MET or mixed tumors and their malignant counterparts has been primarily based on the degree of cytologic atypia. Cases with no or mild cytological atypia were classified as benign and were shown to mostly behave in an indolent fashion. In contrast, moderate to severe atypia, usually accompanied by increased mitotic activity was associated with metastatic potential in about one third of cases (3).

In 2010, the first systematic molecular study of MET revealed that a large subset of deep-seated soft tissue or bone cases harbor rearrangements of EWSR1 gene with various fusion partners, such as POU5F1, PBX1 or ZNF444 (4). Further studies of MET have expanded the spectrum of EWSR1 fusion partners to include KLF15, KLF17 or PBX3 and showed that specific fusions are often associated with distinct morphological features (5–8). It has also been observed that occasional cases harbor alterations of FUS instead of EWSR1 (5), and that rare MET may be driven by other fusions such as SS18::POU5F1 (9–12).

In contrast, superficially located cases tend to lack EWSR1 alterations and are usually benign tumors exhibiting ductal differentiation (i.e. representing adnexal mixed tumors). These were eventually shown to harbor PLAG1 gene rearrangements (13–15). The topic was further complicated by the finding of SMARCB1 loss by immunohistochemistry in a subset of fusion-negative MET (16), the fact that some tumors diagnosed as “MET” do not harbor any detectable molecular alterations (17, 18), and that the term “myoepithelial” is also applied to salivary gland tumors in the head and neck, including pleomorphic adenomas, myoepitheliomas, myoepithelial carcinomas, etc. (19).

Overall, thanks to an improved understanding of their molecular background, we believe it has become evident that while they share some degree of overlapping morphological and immunohistochemical features, the group of MET as currently defined includes a heterogeneous mixture of neoplasms. In addition, all systematic studies focusing on the outcome of MET were from the pre-molecular era. It is therefore currently uncertain whether the various molecular subgroups differ in their biological behavior. In this multi-institutional study, we aimed to collect fusion-positive MET spanning the most common gene rearrangements, perform a pooled clinicopathological analysis by combining new cases with all molecularly confirmed cases reported previously, with a particular emphasis on clinical outcome. Additionally, using DNA methylation profiling, we compare the epigenetic profiles of each MET subgroup to assess their mutual relatedness and relationship with salivary gland MET.

1. Materials and methods

1.1. Case selection

The archives of 10 institutions were searched for genetically confirmed MET harboring either PLAG1 or EWSR1/FUS fusions. Regarding the latter, we included only cases with previously published fusion partners: POU5F1, KLF15, KLF17, PBX1, PBX3 and ZNF444. Additionally, a case with SS18::POU5F1 fusion was included as well. Molecular testing was done prospectively in a minority of cases suspected to represent MET. In 2 cases that were morphologically compatible with the PLAG1 subgroup, diffuse staining with PLAG1 immunohistochemistry (IHC) and congruent methylation clustering was used as a confirmatory method. Twenty cases were included in earlier publications (4, 5, 8, 15, 20–22). Clinical data, immunohistochemistry and molecular results were retrieved from pathology reports, when available. Follow-up data was obtained from medical reports or from referring physicians. Hematoxylin and eosin-stained slides were reviewed and the cytomorphology (epithelioid, spindle, etc.), histologic growth pattern (solid sheets, fascicular growth etc.) and the stromal features (e.g. myxoid, myxo-chondroid, sclerotic) were recorded for each case.

For the cutaneous MET which commonly exhibit epithelial structures in the form of ducts, the terminology of mixed tumors (or chondroid syringomas) vs. MET has been variably applied in literature. These tumors span a broad spectrum from epithelial-predominant apocrine mixed tumors (15, 23) to myoepithelial-predominant or so-called hyaline cell-rich cases (15, 24). To be included in this study, we used the cut-off of less than 5% of ductal differentiation as used in recent studies (15, 25).

For all cases, the criteria of malignancy included moderate to severe nuclear atypia (nuclear enlargement and hyperchromasia) and easily discernible prominent nucleoli (3). Mitotic activity per 10 high-power fields (HPF) and the presence of necrosis were recorded as well. In most cases, at least some immunohistochemical results of myoepithelial markers, such as CK, EMA, S100, and GFAP were available for review. Although most cases fulfilled the diagnostic criteria of a positive “myoepithelial immunoprofile”, the few cases that did not were not excluded, as our main inclusion criterion was the confirmation of a MET-compatible molecular alteration (8). These cases were also reviewed histologically to confirm presence of characteristic morphologic features. The molecular results were obtained by FISH in 13 cases as previously reported (8). In the remaining FET(EWSR1/FUS)- or PLAG1-rearranged MET cases, the fusion genes were detected by various clinical targeted RNA sequencing platforms such as Archer FusionPlex (26), Illumina TruSight RNA Pan-Cancer (27) or MSK-Fusion (28) as part of the diagnostic workup from the contributing academic institutions.

1.2. Review of the literature and pooled statistical analysis

We searched the English literature listed on PubMed for MET of skin, soft tissue, bone or visceral sites confirmed to harbor PLAG1 fusions (or, as a surrogate marker, diffuse nuclear expression of PLAG1 by immunohistochemistry in the proper context), and EWSR1/FUS rearrangements with one of the following partners: POU5F1, KLF15, KLF17, ZNF444 or PBX1/3. We also included for investigation rare tumors with SS18::POU5F1 fusions. Clinical data, including age, gender, location, size, treatment, disease progression, and survival data, were extracted and recorded. Reports with follow-up data of less than 6 months were excluded from the survival studies. Pathologic findings such as benign vs. malignant histology, mitotic count and the presence or absence of necrosis were collected for each case. Cases from the current study with available follow-up data were added to the pooled analysis. Whenever possible, follow-up data in cases reported previously and re-included in the current study was reviewed and updated. Progression of disease was defined as the first event of tumor recurrence/metastasis after surgical resection and/or therapy with no evidence of residual disease. Survival analyses were performed using R packages “survminer” version 0.5.0 (RRID:SCR_021094) and “survival” version 3.6.4 (RRID:SCR_021137), by comparison of hazard ratios using the log-rank test and plotted as Kaplan-Meier curves for disease-specific survival (DSS) and progression-free survival (PFS), with follow-up time defined as the number of months from baseline to last contact and event status captured by the survival indicator. Right-censoring was applied where appropriate. The clinicopathologic covariates that were evaluated for univariate and multivariate Cox regression survival analysis included age group (≤ 25 vs > 25 years old), sex (female vs male), benign vs malignant morphology, mitosis: ≤ 10 vs > 10 / high power fields, necrosis: absent vs present, size: ≤ 5 vs > 5 cm in greatest dimension) and molecular subgroups. For molecular subgroups, pairwise log-rank tests were performed to assess differences between groups. To control for multiple comparisons, p-values were adjusted using the Benjamini–Hochberg procedure. The cutoffs for the continuous variables age, mitosis and size categories are based on the median statistics of these categories. Insufficient information regarding nodal status and treatment modalities exist to consider these as variables. Since each covariate contains non-overlapping missing data, missing data imputed using Multiple Imputation by Chained Equations (MICE) from the mice R package version 3.18.0 (RRID:SCR_026363) using 5 imputations and the default predictive mean matching (pmm) method. When multiple imputed datasets were analyzed, parameter estimates and variances were combined using Rubin’s rules, as implemented in the “MIcombine” function from R package mitools version 2.4. Covariates of interest were included in the model, and log hazard ratios along with their variances were extracted from the fitted models. Hazard ratios were reported by exponentiating the regression coefficients. Ninety-five percent confidence intervals were constructed as $\exp (\widehat{\beta }\pm 1.96\times \sqrt{Var\left(\widehat{\beta }\right)})$, where $\widehat{\beta }$ (estimated beta coefficient) is the log hazard ratio associated with a given covariate. P-values were derived from the Wald statistic, calculated as twice the probability of observing a standard normal value greater than or equal to the absolute ratio of each coefficient to its standard error, i.e., $2\times \mathrm{\Phi }\left(-\left|\frac{\widehat{\beta }}{\sqrt{Var\left(\widehat{\beta }\right)}}\right|\right)$, where $\mathrm{\Phi }$ denotes the cumulative distribution function of the standard normal distribution. For disease-specific survival (DSS), Cox regression with Firth’s bias-reduced penalized maximum likelihood from the coxphf R package version 1.13.4 was used, which was applied to reduce bias in parameter estimates arising from small sample sizes or sparse events. This was adopted because there were no events for the benign morphology category, and Cox regression analysis would fail with infinite coefficient (hazard ratio). For PFS, Cox proportional hazard regression analysis using the “coxph” function from the survival R package version 3.6.4. was used. Right censoring is used for event status at the end of observation (at last follow-up).

1.3. DNA methylation profiling

For DNA methylation profiling, genomic DNA was extracted from formalin-fixed paraffin-embedded tissue sections for each of the samples. Next, 250 ng of genomic DNA was subjected to bisulfite conversion and processed on the Illumina Infinium Methylation EPICv2 platform with over 900,000 methylation sites according to the manufacturer’s instructions. Methylation profiling was successfully performed in the following MET: 1 EWSR1::KLF15, 4 EWSR1/FUS::KLF17, 5 EWSR1::PBX1, 6 EWSR1::PBX3, 16 EWSR1::POU5F1, 1 SS18::POU5F1, 2 EWSR1::ZNF444, and 11 PLAG1-rearranged cases. Additional IDAT files were obtained from cases 1–7, 10, 16–18 and 30 from the Malik et al study (EPICv1 platform) to enrich our dataset with fusions positive MET (4 EWSR1::KLF15, 2 EWSR1::PBX3, 1 EWSR1 POU5F1), along with 2 SMARCB1 deficient MET, and 1 EWSR1::NFATC2 sarcoma (17). We also supplemented this analysis with our internal IDAT files (EPICv2 platform) for 8 ossifying fibromyxoid tumors (OFMT) with EP400:PHF1 and 5 EWSR1/FUS::NFATC2 sarcomas.

Additionally, we obtained raw IDAT files for 230 samples from the Heidelberg sarcoma methylation classifier reference cohort (Gene Expression Omnibus study accession number GSE140686), which included a mix of samples profiled by the 450k and EPICv1 platforms, including the methylation classes angiomatoid fibrous histiocytoma (AFH), CIC-rearranged sarcoma, desmoplastic small round cell tumor (DSRCT), extraskeletal myxoid chondrosarcoma (EMC), epithelioid sarcoma, Ewing sarcoma (EwS), inflammatory myofibroblastic tumor (IMT), low-grade fibromyxoid sarcoma (LGFMS), melanoma, malignant peripheral nerve sheath tumor (MPNST), malignant rhabdoid tumor (MRT), schwannoma, sclerosing epithelioid fibrosarcoma (SEF), solitary fibrous tumor (SFT), and synovial sarcoma (SS) (29). Head and neck salivary gland tumors, including 19 pleomorphic adenomas, 15 myoepitheliomas, and 10 myoepithelial carcinomas were obtained from GSE243075 (19), profiled with the EPICv1 platform.

IDAT processing and data analysis on all samples was performed using the R software (version 4.4.1) (RRID:SCR_001905) and the “minfi” package (version 1.52.1) (RRID:SCR_012830). Raw IDAT files were normalized by preprocess SWAN function and filtered in for detection P value < 0.05. Only CpG probes common to Infinium 450k, EPIC and EPICv2 were kept. CpG probes containing single nucleotide polymorphisms (SNPs) and those on sex chromosomes are filtered out (234,047 probes remaining in final beta value matrix). After converting the raw methylated and unmethylated values into ratios, beta values are extracted by the getBeta function, defined as methylated signal/(methylated + unmethylated signals). For dimensionality reduction, per-probe variance was calculated across all case samples, and the top 25,000 most variable CpGs by variance were analyzed by principal component analysis (PCA) (prcomp function). The top principal components (PCs) that explained at least 80% of cumulative variance (top 100 PCs for entire cohort and top 60 PCs for focused analysis) were selected for downstream analysis. These components were subjected to Uniform Manifold Approximation and Projection (UMAP) for non-linear dimensionality reduction, using the umap2 implementation (uwot R package 0.2.2) with the following parameters: number of neighbors = 5, Euclidean distance metric, and a minimum distance parameter of 0.1. Suppl. Fig. S1A shows the scree plot; Suppl. Fig. S1B–C shows the PCA plots before and after UMAP dimensionality reduction. Cluster assignment was performed using the density-based machine learning clustering HDBSCAN (“hdbscan”) algorithm from the dbscan package version 1.2.2 with a minimum cluster size of 6. Observations identified as noise (cluster assignment = 0) were subsequently reassigned to the nearest cluster center. Cluster centers were defined as the mean UMAP coordinates of non-noise clusters. For each noise point, squared Euclidean distances to all cluster centers were computed, and the point was reassigned to the cluster with the nearest center. The resulting UMAP embeddings were combined with sample-level metadata, which included tumor type classifications. Each sample was annotated with both its tumor type and HDBSCAN-derived cluster assignment. Additionally, for unsupervised hierarchical clustering, per-probe variance was calculated across all case samples, and the top 5,000 probes ranked by variance were retained. The resulting beta-value matrix was standardized to zero mean and unit variance per feature. Hierarchical clustering of both rows and columns was performed using Euclidean distances and Ward’s minimum variance linkage (ward.D2), and visualized using the pheatmap R package version 1.0.12 (RRID:SCR_016418).

For differential methylation analysis, M values were extracted from FET- or SS18-rearranged MET, cutaneous PLAG1-rearranged MET, and salivary gland (head and neck) MET (pleomorphic adenomas, myoepitheliomas, myoepithelial carcinomas). M value was calculated by the log₂ ratio of methylated / unmethylated signal intensities with an offset of 100 added to both signals: log₂[(methylated signal + 100)/(unmethylated signal + 100)]. A positive M-value implies hypermethylation and a negative M-value implies hypomethylation. After removing SNPs and sex-chromosome probes from the M matrix using “rmSNPandCH” function from the DMRcate package (version 3.2.1), 610,092 probes remain in the M matrix. Differential methylation analysis was performed using the “cpg.annotation”, “dmrcate” (lambda = 1000, C=2) and “extractRanges” functions to identify differentially methylated regions (DMRs) between the cutaneous PLAG1-rearranged and salivary gland MET vs FET- or SS18-rearranged MET was performed using cpg.annotate followed by dmrcate functions from the R package DMRcate version 3.2.1. The corresponding M values of the CpG probes mapping to the genes in the top DMRs are represented by pHeatmap. Selected DMR was graphically represented by DMR.plot function. M values were filtered for significant DMRs as follows: 1) min_smoothed_fdr (minimum false discovery rate/FDR of the smoothed estimate), FDR-corrected Stouffer (Stouffer summary transform of the individual CpG p-values), FDR-corrected HMFDR (Harmonic mean of the individual CpG values) and FDR-corrected Fisher combined probability transform of the individual CpG p-values FDR < 0.01, 2), meandiff (mean differential/coefficient across the DMR) > ±0.2. Pathway enrichment analysis of significant DMRs was performed using the “goregion” function from the missMethyl package version 1.40.3 from the Kyoto Encyclopedia of Genes and Genomes (KEGG) PATHWAY database. After filtering for FDR < 0.05, the top 10 KEGG pathways ordered by number of DMRs is visualized with ggplot2 package (RRID:SCR_014601). Detailed annotations of the cpg sites contained in the DMRs (genomic regions, distance to TSS) are derived from the “cpg.annotate” function from the DMRcate package

For copy number variation (CNV) analysis, the Mset files after IDAT files preprocessing were processed by “conumee2” R package version 2.1.2. For normal controls, a set of 20 normal blood samples from GSE286313 processed by the same EPICv2 platform were downloaded. To ensure comparability, only the 34 FET-rearranged MET samples processed by EPICv2 are included. Both sets of idat files were processed by “preprocessIllumina” function of the minfi package. The conumee2 annotation object was created using “CNV.create.anno” function form conumee2 package setting array_type to “EPICv2”. The experimental and control data are loaded with “CNV.load” function, followed by CNV calling with “CNV.fit”, “CNV.bin”, and “CNV.segment” functions. The CNV.segment output includes chromosomes, start, end, num.mark and segmental mean information. For visualization of summary level CNV, “CNV.summaryplot” function from “conumee2” was used. For visualization of case-level CNV segments, overlapping segments were derived using the “CNTools” package version 1.52.0 (RRID:SCR_000281) with “CNSeg” followed by “getRS” function. The average copy number signals were then calculated at 1 MB genomic windows and visualized using the R package “ComplexHeatmap” version 2.8.0 (RRID:SCR_017270). CNV was considered present if the absolute segmentation mean is greater or equal to 0.15. Fraction of CNV across the genome was calculated by dividing the sum of all segments with CNV by the sum of all segments across the genome.

1.4. Compliance with Ethical Standards

The study was conducted following the rules set by the Faculty Hospital in Pilsen Ethics Committee and Institutional Review Boards from the participating institutions. Written informed consent was waived according to the respective IRB protocols and Ethics Committees. The study was conducted in accordance with ethical guidelines from the Council for International Organizations of Medical Sciences (CIOMS).

1.5. Data availability statement

The raw and processed DNA methylation profiling data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE300102. Additional data, such as the raw data behind the figures, is available upon request to the corresponding author.

2. Results

Fifty-five new cases of myoepithelial neoplasms were collected in this study, of which 53 had PLAG1/EWSR1/FUS/SS18 rearrangements confirmed by molecular testing, while 2 cases were confirmed by PLAG1 IHC as mentioned. A thorough review of the literature yielded an additional 130 cases of MET with confirmed PLAG1/EWSR1/FUS/SS18 fusions. Across the entire cohort, DNA methylation profiling data was available in 52 cases (45 new cases, 7 previously published in Malik et al).¹⁷ In the following sections, the clinicopathological features of all fusion subgroups will be addressed separately, summarizing together cases observed in this study as well as cases reported previously. A pooled DSS and PFS analysis combining all cases is included as well. Detailed clinicopathological features of all 185 cases included in our analysis are listed in Suppl Tables S1–S7.

2.1. Clinicopathological features of fusion subgroups

2.1.1. EWSR1/FUS::POU5F1

A total of 33 cases with EWSR1/FUS::POU5F1 fusions (16 new, 17 previously reported) involved 19 females and 13 males (1 unknown), with a median age of 26 years (range 3–73) (4, 8, 17, 18, 30–32). Tumors were most often on extremities (n=19; 5 acral), followed by trunk (6), bone (5), and kidney (3). Median size was 3.8 cm (range 0.6–17 cm) (Suppl. Table S1).

Most tumors (28/33) were malignant, showing a spectrum from hypocellular spindled/epithelioid (Fig. 1A) to high-grade undifferentiated round cells (Fig.1B). Most commonly though, they formed sheets/nests of large epithelioid cells with clear cytoplasm (Fig. 1C). Mitotic rates ranged 0–20/10 HPF (median 2); necrosis was seen in 5/19 cases. IHC showed cytokeratin (CK) (19/24), EMA (14/17), and S100 (24/25) positivity. GFAP (2/9), p63 (0/3), SOX10 (0/6) were infrequent/negative. PAX8 was positive in 2/2 renal cases. Four tumors lacked all myoepithelial markers.

Figure 1.

Histology of myoepithelial tumors. EWSR1/FUS::POU5F1 tumors showed a spectrum from hypocellular spindled/epithelioid (A) to high-grade undifferentiated round cells (B). Most cases, however, formed sheets/nests of large epithelioid cells with clear cytoplasm (C). Cases with SS18::POU5F1 were composed of solid sheets/nests of undifferentiated round to epithelioid cells (D). Tumors with EWSR1::KLF15 typically showed undifferentiated epithelioid to spindled cells in myxoid/fibromyxoid stroma (E). In contrast, EWSR1/FUS::KLF17 tumors showed spindled to epithelioid cells in trabecular/corded patterns within myxohyaline stroma, resembling parachordoma (F). EWSR1::ZNF444 neoplasms were composed of poorly differentiated round/epithelioid cells in fibrous or fibromyxoid stroma (G). Benign tumors with EWSR1::PBX1/3 fusion showed syncytial myoepithelioma morphology — solid, sheet-like growth of uniform ovoid to spindle cells, perivascular lymphocytes, low mitotic activity and no necrosis (H). Malignant cases had more variable, often fascicular morphology, with atypia (I), higher mitotic rates, and occasional necrosis. PLAG1-rearranged cutaneous myoepitheliomas were composed of epithelioid/plasmacytoid cells in myxoid stroma (J) with occasional ductal structures (K) and/or hyaline cartilage (L).

Molecularly, 31/33 had EWSR1/FUS::POU5F1 and 2 had FUS::POU5F1 fusions. Follow-up (n=11; median 24 months) showed no local recurrences but 6 metastases. Six were disease-free, 2 had disease, 2 died of disease, and 1 of unrelated causes. Three others with metastasis were lost to follow-up. Median DSS and PFS were not reached.

2.1.2. SS18::POU5F1

Six cases with SS18::POU5F1 fusion (1 new, 5 previously reported) included 4 females and 2 males (median age 23.5 years, range 11–40) (9–12, 21). Tumors involved the proximal lower extremities (2), viscera (kidney, pancreas), trunk, and head/neck. Median size was 5 cm (range 2.5–5 cm) (Suppl. Table S2).

All were malignant, composed of solid sheets/nests of undifferentiated round to epithelioid cells (Fig. 1D). One case showed ganglioneuromatous differentiation. Mitotic rates ranged from >5 to 20/10 HPF; necrosis was present in all 5 evaluable cases. Neoplastic cells expressed EMA (6/6), S100 (6/6), and CK (5/6), but were negative for SOX10 (0/3) and p63 (0/3).

Follow-up (n=4; 6–31 months, median 9) showed 2 recurrences and 3 metastases (2 at presentation). Two patients were disease-free and 2 died. Two additional patients with metastases were lost to follow-up. Median DSS was 31 months; median PFS was 0 month.

2.1.3. EWSR1::KLF15

Thirteen cases with EWSR1::KLF15 fusions (1 new, 12 previously reported) included 7 females and 6 males (median age 3 years, range 7 weeks–37 years) (7, 8, 17, 33–36). Tumors arose in extremities (4), trunk (3), head/neck (2), and viscera (4: kidney [2], heart, bladder). One case was acral. Tumor sizes ranged from 2.5–10.5 cm (median 5 cm) (Suppl. Table S3).

Histologically, most showed undifferentiated epithelioid to spindled cells in myxoid/fibromyxoid stroma (Fig. 1E). All but one were malignant. Mitotic activity ranged from 2 to > 5/10 HPF; necrosis was present in 7/9 cases. All tested cases were positive for S100 and SOX10; CK, EMA, and GFAP were positive in 10/10, 7/10, and 4/5, respectively. One case was negative for p40/p63. No FUS::KLF15 fusions were found.

Outcomes were available for 9 patients (15–60 months, median 34.5). Four recurred, six metastasized. Five were disease-free, four died of disease. Median DSS was 45 months; median PFS was 11.5 months.

2.1.4. EWSR1/FUS::KLF17

Sixteen cases with EWSR1/FUS::KLF17 fusions (6 new, 10 previously reported) (refs) included 8 males and 8 females (median age 29, range 5–58) (5, 8, 18, 37). Most tumors were in the lower extremities (9 cases, 4 acral), with others on the trunk (2), shoulder (1), and visceral sites (lung/pleura: 3; jejunum: 1). Tumor sizes ranged from 2–9.5 cm (median 4.3 cm) (Suppl. Table S4).

Histologically, tumors showed spindled to epithelioid cells in trabecular/corded patterns within myxohyaline stroma, resembling “parachordoma” (Fig. 1F). Six of 16 were malignant. Mitotic rates ranged from 0–18/10 HPF (median 0); necrosis was seen in 2/12 cases. All tested cases were S100 (15/15) and CK (15/15) positive; EMA and SOX10 were positive in 8/10 and 4/5, respectively. p63 (0/1) and brachyury were negative (0/2). One case showed HMB45-positive pigmented cells.

Molecular testing showed FUS::KLF17 in 11/16 cases and EWSR1::KLF17 in 5/16, with no clear differences between groups. Of 7 patients with follow-up (8–228 months), 2 had recurrence and metastasis, 5 were disease-free, 1 was alive with disease, and 1 died. One additional case had metastasis at presentation but was lost to follow-up. Median DSS was 108 months; median PFS was 67 months.

2.1.5. EWSR1::ZNF444

Four cases with EWSR1::ZNF444 fusions (2 new, 2 previously reported) were all female patients (median age 42, range 2–64) (8, 38). Tumor sites included the head (2), lung/pleura (1), and thigh (1). One tumor measured 12.5 cm; sizes of others were unknown. All were malignant, composed of poorly differentiated round/epithelioid cells in fibrous or fibromyxoid stroma (Fig. 1G). Mitotic rates were 5 and 30/10 HPF in two cases; necrosis was seen in 1 of 3 evaluable cases. IHC showed S100, EMA, and CK positivity in 2/4 cases; both tested cases were GFAP-positive. One case was negative for myoepithelial markers. All harbored EWSR1::ZNF444 fusions (Suppl. Table S5).

Three patients had follow-up (72–132 months); two had recurrence and metastasis—one died of disease, the other was alive with widespread disease. The third had no evidence of disease. Median DSS was 114 months; median PFS was 39 months.

2.1.6. EWSR1::PBX1/PBX3

Sixty six cases with EWSR1::PBX1/PBX3 fusions (13 new, 53 previously reported) involved 33 females and 32 males (1 unknown), median age 35 (range 2–75) (4, 6, 8, 17, 22, 39–43). Tumors were located in skin/soft tissue (40), bone (21), and viscera (5). The skin/soft tissue lesions occurred on the upper extremity (10), lower extremity (24), and trunk (4); location was unknown in 1. The visceral tumors were all thoracic. The skeletal cases involved long bones (16), flat bones (5), and small foot bones (2). Median tumor size was 0.7 cm (range 0.3–19 cm); bone tumors tended to be larger, with no age difference between superficial and bone cases (Suppl. Table S6).

Most skin/soft tissue tumors (37/40) were benign, whereas only 18/26 bone/visceral cases were benign. Benign tumors showed syncytial myoepithelioma morphology—solid, sheet-like growth of uniform ovoid to spindle cells, perivascular lymphocytes, low mitotic activity (0–2), and no necrosis (Fig. 1H). Malignant cases had more variable, often fascicular morphology, with atypia (Fig. 1I), higher mitotic rates, and occasional necrosis.

Immunohistochemically, nearly all cases were S100+ (66/66) and EMA+ (59/60); CK was positive in 9/22, p63 in 1/2, GFAP in 0/3, and all tested cases were SOX10-negative (0/19). Molecularly, 55 tumors had EWSR1::PBX3 fusions; 11 had PBX1 fusions, with no major differences observed.

Limited follow-up showed 2/3 cutaneous/visceral cases recurred, but none metastasized or were fatal. Among 11 bone cases (follow-up 7–180 months, median 22), 5 recurred, 1 metastasized, 7 were disease-free, 2 were alive with disease, 1 died of unrelated causes, and 1 died of disease. For the EWSR1::PBX1 subgroup: median DSS was not reached; median PFS was 48 months. For the EWSR1::PBX3 subgroup: median DSS was 38 months; median PFS was 31 months.

2.1.7. PLAG1-altered subgroup

Forty-seven PLAG1-rearranged cutaneous/adnexal tumors (16 new, 31 previously reported) were included (13–15, 25, 39). Patients (31 females, 16 males) had a median age of 53 years (range 18–84). Most tumors arose from extremities (n=35; 30 lower, 27 acral), followed by head and neck (9), trunk (2), and 1 unknown. Median tumor size was 2 cm (range 0.4–7 cm) (Suppl. Table S7).

Histologically, 45 tumors were benign and 2 malignant, typically presenting as well-circumscribed dermal/subcutaneous nodules of epithelioid/plasmacytoid cells in myxoid stroma (Fig. 1J) with occasional ductal structures (<5%) (Fig. 1K) and/or hyaline cartilage (Fig. 1L). Atypia was absent or mild, and mitoses ranged from 0–8/10 HPF (median 0). Small necrosis foci (1–5%) were found in 2/22 cases. IHC showed consistent expression of S100, SOX10, CK, and PLAG1. GFAP (7/12), p63 (10/17), and EMA (7/14) were variably positive. PLAG1 alterations were confirmed by RNA-seq (n=11), FISH (n=25), or IHC (n=9), with fusion partners including TRPS1 (n=6), LIFR (n=2), and one case each of CHCHD7, NCALD, YWHAZ.

Follow-up (≥ 6 months; n=13, median 96 months) showed 2 recurrences after incomplete excision, including 1 case with lung metastasis 30 years post-diagnosis. All other patients were disease-free. Median DSS and median PFS were not reached.

2.2. Comparison of clinicopathologic features of FET-rearranged myoepithelial tumors

Next, a pooled clinicopathologic analysis of age, sex, primary site, and histology from new and published cases (n = 185) was performed. Across soft tissue EWSR1/FUS-rearranged MET, while EWSR1::PBX1, EWSR1::PBX3, EWSR1::KLF17 and EWSR1/FUS::POU5F1 MET affected broad age ranges (2–73 years), the first three groups were found in older adults (> 40 years of age) and younger adults (≤ 40 years of age) at roughly equal proportions (median 36.5, 35 and 28.5 years old, respectively), with a minority of cases occurring in older children (5 to 18 years of age), whereas patients with EWSR1/FUS::POU5F1 and SS18::POU5F1 MET skewed significantly younger (median 23.5 years old for both groups), with a majority seen in younger adults or older children. On the other hand, EWSR1/FUS::KLF15 MET arose predominantly in young children (median 3 years old) (AVOVA p < 0.0001). Patients with cutaneous PLAG1-rearranged MET, in contrast, were usually older adults (median 53 years old) (Figure 2A, Suppl. Table S8). In terms of sex ratios, while POU5F1- and ZNF444-rearranged soft tissue MET and cutaneous PLAG1-rearranged MET skewed towards female (1 to 1.5, all female, 1 to 2, respectively), the other groups affected male and female at approximately equal ratios (chi-square P < 0.0001) (Figure 2A, Suppl. Table S8).

Figure 2.

Molecular subgroups of myoepithelial tumors and their clinicopathologic associations. (A) Stacked bar plots illustrating the distribution of soft tissue myoepithelial tumors across molecular subgroups (FET::KLF15, FET::KLF17, FET::PBX1, FET::PBX3, FET::POU5F1, SS18::POU5F1, FET::ZNF44) and cutaneous PLAG1-rearranged tumors, and their associated clinical and pathologic features, including fusion partner, sex, age distribution, anatomic site, depth, and size (greatest dimension) distribution. (B) Stacked bar plots illustrating the distribution of clinical and pathologic features across soft tissue myoepithelial tumors molecular subgroups, including tumor morphology (benign vs. malignant), mitotic activity (per 10 high power fields/hpf), tumor cell necrosis, local recurrence, distant metastasis, and survival status. (C) Comparison of immunohistochemical marker expression (EMA, cytokeratin/CK, S100, SOX10, p63, GFAP) across molecular subgroups. White/blank denote missing data.

While almost all EWSR1/FUS-rearranged MET affected both soft tissue sites and visceral organs, most osseous MET were EWSR1::PBX1/3 MET (chi-square P < 0.0001). Cutaneous PLAG1-rearranged MET, on the other hand, arose from superficial sites (skin and subcutis), and tended to be smaller in terms of greatest tumor dimension (median 1.2 cm vs 4–5 cm for most FET-rearranged MET, ANOVA P = 0.0075) (Figure 2A, Suppl. Table S8). In terms of histology, applying the histologic classification criteria from Hornick et al (3) EWSR1/FUS::POU5F1, SS18::POU5F1, EWSR1::ZNF444 and EWSR1::KLF15 were significantly more likely to display malignant morphology, as well as mitotic activity > 10 per 10 high power fields and presence of tumor cell necrosis, while cutaneous PLAG1-rearranged MET, EWSR1::PBX1 and EWSR1::PBX3 tended to show benign morphology, as well as mitotic activity < 10 per 10 high power fields and absence of tumor cell necrosis, (chi-square P < 0.0001) (Figure 2B, Suppl. Table S8).

While almost all MET subgroups experienced local recurrence (15–67%), EWSR1/FUS::POU5F1, SS18::POU5F1, EWSR1::ZNF444 and EWSR1::KLF17, EWSR1::KLF15 were significantly more likely to undergo distant metastasis compared to EWSR1::PBX1/3 and PLAG1-rearranged MET (Figure 2B).

Immunophenotypically, FET-rearranged soft tissue MET usually coexpressed EMA and S100, variably expressed CK, and were negative for SOX10 (with the exception of EWSR1::KLF15 and EWSR1/FUS::KLF17, which were usually CK and SOX10 positive). Similarly, SS18::POU5F1 MET also co-expressed EMA, S100 and CK, but were consistently negative for SOX10. In contrast, cutaneous PLAG1-rearranged MET were consistently positive for CK, S100 and SOX10 (Figure 2C, Suppl. Table S8).

2.3. DNA methylation and copy number variation (CNV) analysis

To better understand the relationship and classification of various fusion-driven MET (n = 52) in relation to each other and to potentially related entities, we analyzed the methylomes of fusion-driven MET in comparison to a spectrum of other potentially related entities (see below). Infinium DNA methylation array data was available in EWSR1::KLF15 (n = 5), EWSR1/FUS::KLF17 (n = 4), EWSR1::PBX1 (n = 5), EWSR1::PBX3 (n = 8), EWSR1/FUS::POU5F1 (n = 16), SS18::POU5F1 (n = 1), EWSR1::ZNF444 (n = 2), and PLAG1-rearranged MET (n = 11). Of these 52 samples, 7 were previously published from Malik et al (2 EWSR1::PBX3, 4 EWSR1::KLF15, 1 EWSR1::POU5F1). This group of tumors was compared to SMARCB1-deficient MET (n = 3) [and by extension epithelioid sarcoma (n = 18) and malignant rhabdoid tumors (n = 25)], EWSR1/FUS::NFATC2 sarcomas (n = 6), salivary gland myoepitheliomas (n = 15), salivary gland myoepithelial carcinomas (n = 10) and salivary gland pleomorphic adenomas (n = 19), as well as various (mostly) mesenchymal entities (n = 230) (AFH, CIC-rearranged sarcoma, DSRCT, EMC, Ewing sarcoma, IMT, LGFMS, melanoma, OFMT, schwannoma, SEF, SFT, and SS).

To investigate the epigenetic relationships among the various MET classes with other entities, UMAP dimensionality reduction analysis on the first 100 principal components followed by density-based clustering showed that despite the shared “myoepithelial” designation, FET(EWSR1/FUS)-rearranged MET were epigenetically unrelated to cutaneous PLAG1-rearranged MET or salivary gland (head and neck) pleomorphic adenomas and myoepitheliomas/myoepithelial carcinomas (Figure 3A). On the contrary, FET(EWSR1/FUS)-rearranged MET shared a closer epigenetic relationship with EWSR1/FUS::NFATC2 sarcomas, which despite its variable/null immunophenotype, display substantial histologic overlap with FET-rearranged MET (44), as well as SMARCB1-deficient MET and epithelioid sarcomas. This was corroborated by unsupervised hierarchical clustering of the top 5000 most variable CpG sites across FET(EWSR1/FUS)-rearranged MET compared to EWSR1/FUS::NFATC2 sarcomas, OFMT, epithelioid sarcomas, cutaneous PLAG1-rearranged MET, and salivary gland myoepitheliomas/myoepithelial carcinomas (Suppl. Fig. S2).

Figure 3.

DNA methylation-based clustering. (A) Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction of the first 100 principal components followed by HDBSCAN density-based clustering of the DNA methylation profiles of FET-rearranged soft tissue myoepithelial tumors, cutaneous PLAG1-rearranged and salivary gland myoepithelial tumors and a spectrum of soft tissue and sarcoma types. The sample points were colored by histologic diagnosis. Ellipses and shades indicate clusters computed by HDBSCAN. Dashed line encircles entities in subsequent focused analysis in panel B. (B) Focused DNA methylation analysis of myoepithelial tumor molecular subgroups. UMAP dimensionality reduction of the first 60 principal components followed by HDBSCAN density-based clustering of the DNA methylation profiles of soft tissue myoepithelial tumors with defined fusion partners (FET::KLF15, FET::KLF17, FET::PBX1, FET::PBX3, FET::POU5F1, SS18::POU5F1, FET::ZNF44) and comparison with selected soft tissue tumor types including angiomatoid fibrous histiocytoma, epithelioid sarcoma, inflammatory myofibroblastic tumor, low-grade fibromyxoid sarcoma, malignant rhabdoid tumor, NFATC2-rearranged sarcoma, ossifying fibromyxoid tumor, and sclerosing epithelioid fibrosarcoma. The sample points were colored by histologic diagnosis. Ellipses and shades indicate clusters computed by HDBSCAN.

Next, a focused comparison of the methylomes of various FET-rearranged MET molecular subgroups with each other, as well as SEF/LGFMS, AFH, IMT, MRT, epithelioid sarcomas, SMARCB1-deficient MET, EWSR1/FUS::NFATC2 sarcomas and OFMT was performed by UMAP dimensionality reduction of the first 60 principal components followed by density-based clustering. This analysis showed that although this group of tumors were epigenetically more closely related to one another than with other mesenchymal entities, the methylomes of FET-rearranged MET largely segregated based on their respective underlying molecular subgroups (Figure 3B). This finding was substantiated by unsupervised hierarchical clustering of the top 1000 most variable CpG sites across the FET-rearranged MET molecular subgroups and EWSR1/FUS::NFATC2 sarcomas (Suppl. Fig. S3). Of note, the single case of SS18::POU5F1 clustered with the other FET-rearranged MET involving a POU5F1 partner, in keeping with a shared pathogenesis, and separate from synovial sarcoma (which harbors SS18::SSX fusions).

To interrogate the differentially methylated loci between salivary gland/cutaneous PLAG1-rearranged MET versus FET-rearranged MET, differential methylation analysis was performed comparing these two classes of MET. There were 2,443 significant differentially methylated regions (DMR) with a mean difference in methylation levels of at least 20%. Among the significant DMRs, 1,610 were hypermethylated, and 833 were hypomethylated in FET-rearranged MET relative to salivary gland/cutaneous PLAG1-rearranged MET, respectively (Figure 4A, Suppl. Fig. S4, Suppl Table S9). One of the top hypermethylated DMRs in FET-rearranged MET contained KRT15, which encoded for a low molecular weight cytokeratin. One of the top hypermethylated DMRs in salivary gland/cutaneous PLAG1-rearranged MET contained PRRX1, which encodes for a homeobox transcription co-regulator (Figure 4B). KEGG pathway enrichment analysis of the coding genes annotated to the CpGs covered by the significant DMRs showed that signaling pathways relevant for keratinocyte differentiation, cytoskeleton in muscle cells and focal adhesion were enriched (Figure 4C). Compared to DMRs that were hypomethylated in FET-rearranged MET or uninvolved CpGs, hypermethylated DMRs tended to be closer to transcription start site (TSS) (36% vs 28% vs 19%, respectively, ≤ 10 kb from TSS) and enriched in CpG islands (15% vs 6% vs 5%, respectively) and promoter regions (39% vs 24% vs 20%, respectively) (chi-square P < 0.0001 for all comparisons) (Figure 4D).

Figure 4.

Differential DNA methylation in salivary gland/PLAG1-rearranged versus FET-rearranged myoepithelial tumors. (A) Volcano plot of significant differentially methylated regions (DMRs) (FDR-corrected p values < 0.001, absolute mean methylation difference at least 20%) salivary gland/PLAG1-rearranged myoepithelial tumors from soft tissue FET-rearranged myoepithelial tumors. Coding genes annotated to CpG sites covered by the significant DMRs are labeled. (B) Genomic view of DMRs on chromosomes 17 and 1 near the KRT15 (hypermethylated) and PRRX1 (hypomethylated) loci, respectively. Plot displays beta value group means and confidence intervals. (C) Top 10 significant pathways from KEGG pathway enrichment analysis of the significant DMRs. FDR: false discovery rate. DMG: differentially methylated genes. (D) Stacked summary plots comparing the genomic features of CpG sites among hypermethylated and hypomethylated DMRs compared to uninvolved CpGs: distance to nearest transcription start site (TSS), CpG annotation (island, shore, shelf, open sea), and gene annotation (promoter, coding, intergenic).

Next, leveraging probe coverage sum intensities across the genome on the Infinium methylation array, copy number variation (CNV) analysis was performed against a set of reference normal blood samples. Interestingly, compared to those with benign morphology, MET with malignant morphology harbored increased chromosomal arm-level CNV, including gains on chr1q, 4, 17q and 19, and losses on chr1p, 3, 6, 11, 16–17, 19, and 20 visualized by case-level and summary plots (Figure 5A–B), as well as significantly higher percentage of genome involved by copy number variation (CNV) on average (t-test p = 0.03) (Figure 5C). Of note, SS18::POU5F1 MET also typically harbored malignant histologic features (10, 11). However, since only one case of SS18::POU5F1-rearranged MET had available methylomic data, we are unable to draw definitive conclusions regarding the correlation between histology and CNV in this tumor type.

Figure 5.

Copy number variation (CNV) analysis of FET-rearranged myoepithelial-like tumors. (A) Case-level genome-wide heatmap depicting chromosomal arm level copy number (CN) variation (log₂ ratio), annotated by histology (benign vs malignant) molecular subgroup. (B,C) Copy number summary plots depicting chromosomal arm level copy number (CN) variation aggregated by histology (B, Benign; C, Malignant). (D) Percent genome altered by copy number variation (CNV) in tumors across molecular subgroups.

2.4. Survival analysis

Finally, a pooled outcome analysis of fusion-positive MET from new and published cases with available survival data (n = 69) was performed: EWSR1::KLF15 (n = 8), EWSR1/FUS::KLF17 (n = 7), EWSR1::PBX1 (n = 4), EWSR1::PBX3 (n = 14), EWSR1/FUS::POU5F1 (n = 15), SS18::POU5F1 (n = 6), EWSR1::ZNF444 (n = 3), and PLAG1-rearranged MET (n = 12). The clinicopathologic covariates that were evaluated for univariate and multivariate Cox regression survival analysis included age group (<= 25 vs > 25 years old), sex (female vs male), benign vs malignant morphology, mitosis: <= 10 vs > 10 / high power fields, necrosis: absent vs present, size: <= 5 vs > 5 cm in greatest dimension, and molecular group. Malignant histologic features were defined as moderate to severe cytologic atypia (coarse chromatin, nuclear hyperchromasia, prominent / large nucleoli, nuclear pleomorphism).

Across molecular groups, median DSS was shortest in SS18::POU5F1 (31 months), EWSR1::PBX3 (38 months), and EWSR1::KLF15 (45 months) MET. Patients with EWSR1/FUS::KLF17 and EWSR1::ZNF444 MET had intermediate median DSS (108 and 114 months, respectively). In contrast, patients with EWSR1/FUS::POU5F1 and EWSR1::PBX1 MET and those with cutaneous PLAG1-rerranged MET had superior DSS (Figure 6A, Suppl. Table S8, S10). On pairwise comparison, the following MET molecular subgroups showed significant difference in DSS: PLAG1 vs SS18::POU5F1 (Benjamini–Hochberg-corrected p = 0.033).

Figure 6.

Histologic and molecular parameters associated with outcome in myoepithelial tumors. (A-D) Kaplan–Meier curves illustrating disease-specific survival (DSS) stratified by molecular subgroup (A), tumor morphology (B), presence of necrosis (C), and mitotic activity (D) (log-rank p-values and univariate Cox regression hazard ratios, HR). Statistically significant pairwise comparison of DSS between molecular subgroups included: PLAG1 vs SS18::POU5F1 (p = 0.033) (Benjamini–Hochberg-corrected p-value) (E-H) Kaplan–Meier curves illustrating progression-free survival (PFS) stratified by molecular subgroup (E), tumor morphology (F), presence of necrosis (G), and mitotic activity (H). (log-rank p-values and univariate Cox regression hazard ratios, HR). Statistically significant pairwise comparison of PFS between molecular subgroups included: PLAG1 vs SS18::POU5F1 (p = 0.0019), PBX3 vs SS18::POU5F1 (p = 0.0013), KLF17 vs SS18::POU5F1 (p = 0.0016) (Benjamini–Hochberg-corrected p-values).

For disease-specific survival (DSS), on univariate analysis, molecular alterations (log-rank p = 0.0012), age ≤ 25 years old (log-rank p = 0.043, univariate HR 3.38), malignant histology (log-rank p = 0.0013, univariate HR 25.34), presence of necrosis (log-rank p = 0.02, univariate HR 5.60), and elevated mitotic activity (≥ 10/10 high power fields) (log-rank p = 0.00027, univariate HR 10.28) were all significantly associated with worse outcome. Tumor size ≥ 5 cm in greatest dimension was borderline associated with worse DSS (log-rank p = 0.062, univariate HR 3.47) (Figure 6A–D, Suppl. Fig. S5A,B). However, on multivariate analysis, none of the covariates remained significant for worse DSS; mitotic activity > 10 per 10 hpf was borderline significant for worse DSS (Cox regression p = 0.0896, HR = 6.93) (forest plot shown in Suppl. Fig. S5C, Suppl. Table S10), which were likely attributed to the significant multicollinearity of these covariates (kappa condition number were 37 for DSS).

Across molecular groups, median PFS was shortest in SS18::POU5F1 (< 1 month) and EWSR1::KLF15 (11.5 months) MET. Patients with EWSR1::PBX3, EWSR1::PBX1, EWSR1::ZNF444 and EWSR1/FUS::KLF17 and MET had intermediate median PFS (31, 48, 39 and 67 months, respectively). In contrast, patients with cutaneous PLAG1-rerranged MET had superior PFS. Median PFS for EWSR1/FUS::POU5F1 was not evaluable as < 50% experienced event (Figure 6E, Suppl. Table S8, S10).

For progression-free survival (PFS), on univariate analysis, molecular alterations (log-rank p < 0.0001), age ≤ 25 years old (log-rank p = 0.00015, univariate HR 3.81), malignant histology (log-rank p < 0.0001, univariate HR 6.27), presence of necrosis (log-rank p = 0.00034, univariate HR 4.62), elevated mitotic activity (≥ 10/10 high power fields) (log-rank p = 0.00087, univariate HR 4.62), and tumor size ≥ 5 cm (log-rank p = 0.00014, univariate HR 3.02) were all significantly associated with worse outcome (Figure 6E–H, Suppl. Fig. S6A,B). On multivariate analysis, only age ≤ 25 years old remained significant for worse PFS (Cox regression p = 0.032, multivariate HR 3.13) (forest plot shown in Suppl. Fig. S6C, Suppl. Table S10). On pairwise comparison, the following MET molecular subgroups showed significant difference in PFS: PLAG1 vs SS18::POU5F1 (p = 0.0019), PBX3 vs SS18::POU5F1 (p = 0.0013), KLF17 vs SS18::POU5F1 (p = 0.0016) (Benjamini–Hochberg-corrected p-values).

3. Discussion

This study presents the largest cohort of fusion-positive MET to date. In a pooled analysis with the 130 cases reported previously, we attempted to demonstrate the striking heterogeneity of the various molecular subgroups that are currently included as a single entity in the WHO classification (2). Although previous studies already partially addressed some of the morphological or epigenetic differences (8, 17, 23, 45), thanks to the larger size of our cohort and more detailed clinical and outcome data, combined with extensive literature review, we were able to provide a more comprehensive account of the different clinicopathological features, including outcome, and more thoroughly investigate the epigenetic relationships among these neoplasms.

Despite partially overlapping morphology and immunophenotype, the various molecular subgroups of MET show significantly different clinicopathological features. The EWSR1/FUS::POU5F1 fusion-positive subgroup predominantly occurs in young adults or children and has a strong predilection for the extremities, with less common occurrence in the kidneys, bones or trunk. Morphologically, there seems to be a spectrum ranging from benign-appearing tumors to undifferentiated round cell sarcomas. However, most cases are somewhere in between the two extremes, exhibiting instead nests of malignant epithelioid cells with abundant clear cytoplasm and moderate atypia. Clinically, they represent aggressive neoplasms with metastases occurring in about half of all cases. Using unsupervised hierarchical clustering of RNA sequencing data, the SS18::POU5F1-rearranged MET were previously shown to cluster together with EWSR1/FUS::POU5F1 cases (9). In our study, the SS18::POU5F1-rearranged case displayed similar methylome as EWSR1/FUS::POU5F1 MET, corroborating prior reports classifying SS18::POU5F1-rearranged tumors as being related to EWSR1/FUS::POU5F1 MET (9, 11). Both fusion types affect similar age groups, but the SS18::POU5F1 subset so far spared extremities and occurred only in the groin, back, viscera and parotid gland. Histologically, they are composed of sheets or nests of high-grade undifferentiated epithelioid to rounds cells. Despite very limited follow-up, the SS18::POU5F1 subset showed worse PFS and DSS compared to EWSR1/FUS::POU5F1 cases, with 5/6 cases metastasized and 2/4 patients died of disease. As the number of cases is quite limited, additional studies are needed to evaluate if SS18::POU5F1-MET represents a molecular variant of FET::POU5F1 MET.

The EWSR1::KLF15 subgroup is typically found on the extremities or in the viscera of infants or young children under the age of 5 years. They characteristically show sheets or cords of undifferentiated epithelioid to spindled cells embedded in myxoid stroma. Two-thirds of cases metastasized and almost half of patients died of disease. This subgroup demonstrated significantly worse PFS and DSS compared to cases with EWSR1/FUS::KLF17, where 3 out of 8 cases metastasized, and 1 patient died of disease. Further, EWSR1/FUS::KLF17-rearranged MET predominantly occur in young adults with a predilection for the lower extremity and visceral sites (typically pleura/lungs). The microscopic appearance of most cases (10/16) was benign, and the morphology, as previously noted (8), is reminiscent of the original description of so-called parachordoma (46). On the other hand, the EWSR1::ZNF444 subgroup has very limited data. All 4 cases had an undifferentiated round/epithelioid cell morphology and 2/3 followed an aggressive clinical course.

Most of the EWSR1::PBX1/3 MET arise in the skin (or much less commonly soft tissues) of the extremities or in the long bones, with a minority found in thoracic visceral sites. Regardless of anatomic sites, they predominantly occur in young to middle aged adults and are rare in children. Importantly, we suspect that the cutaneous cases, which are usually diagnosed as so-called syncytial myoepitheliomas, may be underreported. Anecdotally, they represent the most common MET we see in routine and consultation practices. Morphologically, EWSR1::PBX1/3 MET span a broad spectrum. Most cases across all locations appear histologically benign as syncytial myoepitheliomas (22, 47). A smaller subset of cases, especially in the bones, may present as malignant cellular round to spindle cell proliferations with increased mitotic activity (25). However, even such cases sometimes show areas with the bland syncytial morphology, indicating tumor progression. The cutaneous cases included in our analysis (i.e. with molecular confirmation) have very limited follow-up information. Nevertheless, morphological studies on syncytial myoepitheliomas without fusion confirmation indicate that cases with benign syncytial morphology are also clinically benign and rarely recur even if incompletely excised (22, 40, 47). In contrast, 11 bone cases in our study with available follow-up indicated increased tendency for recurrence and, in a few cases with malignant morphology, capacity for metastasis and even tumor-related death.

Immunohistochemically, although the extent varied from focal to diffuse, most soft tissue FET- or SS18-rearranged MET express myoepithelial markers such as S100 protein and EMA. CK expression is highly variable. However, there were stark differences in SOX10 expression; some fusion subgroups (EWSR1::KLF15 or EWSR1/FUS::KLF17) showed consistent expression, whereas others were always negative (EWSR1/FUS::POU5F1, SS18::POU5F1, EWSR1::PBX1/PBX3), further highlighting the heterogeneity of these tumors. Moreover, while GFAP, p40 and p63 are considered and clinically used as MET markers, p40 and p63 were (except for 1 case) almost always negative in soft tissue FET- or SS18-rearranged MET, while GFAP is highly variable.

Nonetheless, the most distinctly different MET subgroup is cutaneous PLAG1-altered MET. Compared to soft tissue FET-rearranged MET, PLAG1-altered MET occur in significantly older individuals and typically do not affect children: many above the age of 40 years—none of the 47 cases in our analysis occurred below the age of 18 years. All were located in the skin or subcutis and the vast majority arose from acral extremities or the face. Apart from 2 cases, all were histologically benign. The majority showed minor foci with epithelial differentiation in the form of small ducts and some cases also contained well-developed hyaline cartilage, both features that were absent in all other studied subgroups. Immunohistochemically, PLAG1-altered MET consistently co-expressed CK, S100 protein and SOX10 diffusely and strongly, in contrast to any other MET subgroups. The available outcome information confirmed that most PLAG1 MET are benign, as only 1/11 metastasized (median follow-up length 96 months).

From the epigenetic perspective, our unsupervised dimensionality reduction and hierarchical clustering analysis of the methylomes of MET and a large group of reference tumor types confirmed the heterogeneity of METs. A detailed comparison of various FET-rearranged MET tumors revealed that, while these tumors are epigenetically more similar to each other than to other mesenchymal entities, the methylomes of soft tissue FET-rearranged MET tend to segregate according to their specific molecular subgroups. Our analysis also included SMARCB1-deficient MET and, due to a significant morphological overlap, EWSR1/FUS::NFATC2-rearranged sarcomas, the latter of which are listed among undifferentiated round cell sarcomas in the current WHO classification (48). Interestingly, SMARCB1-deficient MET clustered with epithelioid sarcomas and separately from all other MET subgroups. Since epithelioid sarcomas represent another major class of SMARCB1-deficient tumors and alterations in the SWI/SNF complex are known to significantly alter the epigenetic landscape of neoplastic cells (49), these two tumor types may be closely related to each other. In contrast, the NFATC2-rearranged sarcomas shared overlapping methylomes with FET-rearranged MET. Although this also warrants further explorations, given its MET-like morphology, variable EMA and CK expression (44) and presence of EWSR1/FUS-rearrangements, this finding may suggest that NFATC2 positive cases might be better classified among FET-rearranged MET. Of note, many of the above discussed findings are in line with results of previous smaller scale analyses of methylation or RNA sequencing data (17, 23, 45).

It has long been claimed that the myoepithelial neoplasms of skin and soft tissues represent analogues of the salivary gland myoepithelial neoplasms (13, 14). It is therefore notable in this regard that the methylomes of PLAG1-altered MET of the skin overlapped with those of PLAG1-altered myoepithelial neoplasms of the salivary glands, i.e., pleomorphic adenomas, as well as salivary gland myoepitheliomas/myoepithelial carcinomas, and that both were completely distinct from the methylomes of the entire group of FET-rearranged soft tissue and bone MET subgroups, corroborating the findings from two recent methylation studies (17, 45). In our study, KRT15, was hypermethylated in soft tissue FET-rearranged MET, but hypomethylated in salivary/cutaneous PLAG1-altered MET. Since cytokeratins are consistently expressed in salivary gland/cutaneous PLAG1-rearranged MET, but variably expressed in FET-rearranged MET, and based on the conventional dogma that hypomethylation and increased RNA expression are correlated and vice versa, we infer that hypermethylation of KRT15, which encodes for a low molecular weight Type 1 cytokeratin, may play a role in the differential keratin expression between these two classes of MET, one of the key clinicopathologic differences we wish to highlight in this paper. On the other hand, PRRX1 was in one of the hypermethylated DMRs in salivary gland/cutaneous PLAG1-rearranged MET. PRRX1 is a homeobox transcription co-regulator that plays key roles in mesodermal development and epithelial-mesenchymal transition (50, 51). Its hypomethylation in FET-rearranged MET may lead to increased expression and give rise to a mesenchymal phenotype in this group of tumors. Further, pathway enrichment analysis of the DMRs showed that pathways relevant for cytoskeleton in muscle cells and focal adhesion are enriched---biologic functions of myoepithelial cells, which act as structural components and are contractile cells that help expel secretions from exocrine salivary and sweat gland. Because of this, we infer that these findings support the hypothesis that despite both being referred to as “myoepithelial” tumors, salivary gland/cutaneous PLAG1-rearranged MET are bona fide myoepithelial tumors while FET-rearranged MET are mesenchymal tumors lack a true myoepithelial phenotype. A future extension of our study would entail RNA-sequencing of METs to integrate the potential transcriptomic correlates of the differential methylomic profiles.

The epigenetic link coupled with their highly overlapping morphological, immunohistochemical and molecular features already noticed in some previous reports further supports the notion that cutaneous PLAG1-altered MET are true analogs of salivary gland MET (13–15). This claim is further reinforced by the anatomical origins of the PLAG1-altered cutaneous MET arising in proximity to where cutaneous apocrine-type sweat glands are found. These glands share histologic similarities with salivary glands (14, 23), including the presence of true myoepithelial cell layer that co-express CK, SOX10 and S100. This is supported by the fact that most salivary gland pleomorphic adenomas are also PLAG1-rearranged (52). Such association with non-neoplastic myoepithelial cells and their immature precursors within sweat glands is lacking in myoepithelial neoplasms of deep soft tissue and bone, which typically do not harbor PLAG1 alterations and are instead often associated with EWSR1/FUS-rearrangements.

Further, the question of how the various FET fusions give rise to clinicopathologic and epigenetically diverse entities is an important and fascinating topic. FET-fusions are presumed to function as aberrant transcription factors via their C-terminal fusion partners (53). Notably, KLF15, KLF17, and ZNF444 are C2H2 zinc finger proteins, which are transcriptional regulatory proteins that bind to DNA, RNA and proteins (54), whereas POU5F1 and PBX1/3 belong to the homeodomain family. POU5F1 (also known as OCT4) is a transcription factor that maintains stem cell pluripotency, while PBX1/PBX3 are transcription factors involved in embryonic development (55). These different classes of transcriptional factors/regulators bind to different DNA binding motifs. For example, ZNF444 binds to the sequence CCCCCTCCCC, while POU5F1 targets the octamer motif ATTTGCAT. Conversely, PLAG1—also a zinc finger protein with a distinct DNA-binding motif—has been implicated in activating IGF2 in PLAG1-rearranged salivary gland tumors (56). However, biological and functional characterization of how chimeric fusion oncoproteins drive oncogenesis in these entities falls beyond the scope of this study, and requires functional experiments such as in vitro expression of fusion oncoproteins in relevant cell culture models. Such functional experiments will serve as important future studies to better understand the differences in transcriptional targets or regulatory mechanisms in FET(EWSR1/FUS)- vs PLAG1-rearranged MET.

We believe that the above discussed stark differences between PLAG1-altered MET and soft tissue FET-rearranged MET raise important nosological questions. Besides partially overlapping morphology and immunophenotype, available evidence suggests that both groups are otherwise unrelated and display fundamentally different clinical behavior, which argues against their continued classification under the same “MET” umbrella term. Therefore, we advocate for the introduction of a clear nosological distinction. Due to the overlapping features with salivary gland MET and the close anatomical relationship with normal myoepithelial cells in the skin, we believe the PLAG1-altered subgroup is best classified as cutaneous mixed tumors/myoepitheliomas. In contrast, FET-rearranged MET typically do not arise from anatomic locations with non-neoplastic myoepithelial cell of origin nor display evidence of ductal/epithelial differentiation. Moreover, according to the current WHO classification, the terminology for malignant cases is “myoepithelial carcinoma”, which is a misnomer that often cause significant confusions for oncologists, as these neoplasms are not epithelial malignancies, but behave as and share many features with other fusion-associated sarcomas, such as ossifying fibromyxoid tumors, extraskeletal myxoid chondrosarcomas or, as shown, NFATC2-rearranged sarcomas. Therefore, malignant FET-rearranged MET are likely best classified as sarcomas rather than carcinomas. Given how myoepithelial differentiation is poorly defined and the clear differences in clinicopathologic features and outcome across molecular subgroups, individual FET-rearranged MET is best defined objectively based on the specific underlying fusion.

In summary, this study illustrates that the group of fusion-driven MET, as currently defined, includes a heterogeneous mixture of neoplasms with variable outcomes and different clinicopathological, molecular and epigenetic features. In particular, cutaneous/adnexal PLAG1-altered MET represents a distinct category with epithelial cell of origin that shares similar methylomes with salivary gland MET. In contrast, FET-rearranged MET are distinct from adnexal and salivary gland MET and should be objectively defined by their underlying molecular fusions.

Translational relevance.

In this study, we focus on soft tissue and bone myoepithelial tumors (MET), a poorly understood group of rare benign and malignant neoplasms that are traditionally diagnosed by morphological and immunohistochemical criteria. Recently, multiple MET subgroups were defined by EWSR1/FUS gene rearrangements with various fusion partners. It is currently uncertain whether the different molecular subgroups represent a single entity and whether they are related to salivary gland and cutaneous adnexal MET. Our study, the largest to date, uncovers significant clinicopathological heterogeneity, including major differences in biological behavior and tumor methylomes across the various molecular subtypes of MET. We provided evidence that while the PLAG1-rearranged adnexal subset represents a true myoepithelial analogue of salivary gland MET, the EWSR1/FUS-rearranged METs is most likely of mesenchymal origin, and their malignant forms are best classified as sarcomas rather carcinomas. These insights enable more accurate diagnosis and better clinical stratification of patients.