Background
Sarcomas of the skin and the adjacent soft tissue comprise a heterogeneous tumor group [
1]. The classification of these tumors follows the lineage differentiation of tumor cells, which is predominantly assessed by their expression of lineage specific markers. However, in many cases an unambiguous subtype assignment by histologic and immunohistochemical means is not possible, and molecular analyses for establishing a final diagnosis is required [
2]. Unfortunately, certain entities also lack unequivocal molecular traits, even if more sophisticated molecular approaches such as next generation sequencing are applied. Atypical fibroxanthomas (AFX) and pleomorphic dermal sarcomas (PDS) belong to the aforementioned group of ill-defined tumors and currently remain a diagnosis of exclusion [
3].
AFX and PDS exhibit overlapping histologic features making a reliable distinction in many cases problematic [
4]. The most important criterion in favor of the diagnosis PDS is an invading growth pattern into subdermal structures, which can be difficult to assess if small biopsies are provided for histopathological diagnostics [
5]. Other diagnostic histologic features include necrosis, lymphovascular and perineural invasion. However, general features of anaplasia such as nuclear pleomorphism and atypical mitoses are common to both AFX and PDS [
6,
7]. The distinction of AFX and PDS as different entities remains clinically important. AFX has an overall favorable biological behavior compared to the much higher potential for recurrence and metastasis in PDS [
3,
6,
7]. Novel diagnostic approaches allowing a clear distinction of AFX and PDS would be of great value considering the steadily increasing incidence of skin cancers [
8] and promising results of targeted therapies for certain dermal tumor subtypes [
9,
10].
DNA-methylation profiling has evolved as a powerful method for determining cell differentiation. Array-based epigenotyping technologies nowadays enable large-scale high-throughput studies of DNA methylation patterns. The study of DNA-methylation in different cancers has already revealed molecular subgroups within known histologically defined tumor types [
11‐
18] and led additionally to the discovery of new tumor types based on unique molecular features [
19,
20]. Recently it has been shown to have great diagnostic capabilities determining the lineage of small blue round cell tumors not otherwise specified [
21], cancers of unknown primary [
22] and nervous system tumors [
23].
AFX and PDS are generally believed to be of mesenchymal lineage, although a few studies have suggested an epithelial origin [
24,
25]. Detailed DNA-methylation patterns in AFX and PDS have not been reported yet. We therefore performed genome-wide methylation profiling and copy number analysis of AFX, PDS and potential histologic mimics, with a focus on cutaneous squamous carcinomas (cSCC) and basal cell carcinomas (BCC) of the head and neck, alongside of melanomas and 11 soft tissue tumor entities.
Materials and methods
Sample selection
In total, 228 tumor specimens from different patients, all prototypical examples of their corresponding subtype, were included (Additional file
1: Table S1). AFX, PDS, cSCC and BCC were collected from the Dermatopathology Bodensee in Friedrichshafen (Germany) and the Department of Dermatology of the University Hospital in Essen (Germany). Melanomas and soft tissue tumors were collected from the Institute of Pathology of the University Hospital in Heidelberg (Germany), in Kiel (Germany), in Jena (Germany), in Nijmegen and in Rotterdam (both the Netherlands), from the Institute of Pathology in Bamberg (Germany) and from the Department of Pathology of the Laboratoire National de Santé (Luxembourg). Diagnoses were based on standard histopathological criteria in conjunction with immunohistochemical and molecular analyses according to the current WHO classification [
1]. The methylation data of melanomas and some soft tissue tumors were published previously [
12,
15,
21].
DNA was extracted from formalin-fixed and paraffin-embedded (FFPE) tumor tissue, thereby only using representative tumor tissue with highest available tumor content was chosen for genomic DNA isolation. The Maxwell® 16FFPE Plus LEV DNA Kit was applied on the automated Maxwell device (Promega, Madison, WI, USA) according to the manufacturer’s instructions. Tumor DNAs had a total amount of > 100 ng and were suitable for the array-based DNA-methylation analysis.
Genome-wide DNA-methylation data generation and pre-processing
The tumors were subjected to Illumina Infinium 450 k BeadChip or the successor EPIC/850 k BeadChip (Illumina, San Diego, USA) analysis at the Genomics and Proteomics Core Facility of the German Cancer Research Center (DKFZ) Heidelberg. DNA-methylation data were normalized by performing background correction and dye bias correction (shifting of negative control probe mean intensity to zero and scaling of normalization control probe mean intensity to 20,000, respectively). Probes targeting sex chromosomes, probes containing multiple single nucleotide polymorphisms and those that could not be uniquely mapped were removed. Probes were excluded if the predecessor Illumina Infinium 450 k BeadChip did not cover them, thereby making data generated by both 450 k and EPIC comparable for subsequent analyses. In total, 438,370 probes were kept for analysis.
Unsupervised clustering, t-SNE analysis, cumulative copy number plotting and identification of differentially methylated regions
For unsupervised hierarchical clustering, we selected 10,000 probes that showed the highest median absolute deviation (MAD) across the beta values. Samples were hierarchically clustered using Euclidean distance and Ward’s linkage method. Methylation probes were reordered by hierarchical clustering using Euclidean distance and complete linkage. The unscaled methylation levels were shown in a heat map from unmethylated state (blue color) to methylated state (red color). For unsupervised 2D representation of pairwise sample correlations dimensionality reduction by t-distributed stochastic neighbor embedding (t-SNE) was performed using the 10,000 most variable probes, a perplexity of 20 and 2500 iterations. Copy-number assessment for segmental/entire chromosomal changes was done manually based on array data by a proprietary algorithm based on the R-package conumee after additional baseline correction (
https://github.com/dstichel/conumee).
Discussion
Our study demonstrates the predictive power of genome-wide methylation profiling in sarcomas of the skin (AFX/PDS) and their histologic mimics. Notably, all examined tumor subtypes exhibit specific epigenetic fingerprints with one exception. As expected, unsupervised clustering did not sort AFX and PDS into separate methylation groups. This finding is in line with the hypothesis that AFX and PDS are part of a common tumor spectrum with AFX potentially being a precursor lesion of PDS [
3].
The concept of AFX and PDS comprising a single entity is supported by genetic studies [
26,
27]. AFX and PDS carry similar, but yet unspecific patterns of
TP53 and
TERT promoter mutations associated with UV-exposure such as observed in melanoma, cSCC and BCC [
3,
26,
28,
29]. Recently, a next-generation sequencing based study on a considerable number of AFX and PDS expanded the overlapping mutational pattern to
NOTCH1/
2 and
FAT1 [
27]. However, only a single whole-exome study of AFX has been presented so far [
30]. Thus, further whole-exome/genome studies with larger sample numbers of both AFX and PDS will be required to fully understand the genetic underpinnings of these tumors.
Copy-number aberrations were found in a comparable frequency and overlapping distribution in AFX and PDS. This is in concordance with previous studies showing recurrent copy number alterations mostly involving chromosome 8 and 9 [
27,
31]. In addition, we found non-recurrent amplifications in 5/30 cases, which almost equally affected AFX and PDS. In contrast to our findings, a previous study detected amplifications only in PDS [
31]. Hence, they suggested such markers for a tumor progression towards PDS. However, the study cohort was mainly composed of PDS (n = 24) with only three AFX cases for comparison.
Beside amplifications, we also noticed recurrent homozygous
CDKN2A deletions in PDS (40%) and less frequently in AFX (12%).
CDKN2A deletions have been recognized as an adverse prognostic marker in a number of tumors, i.e. in melanoma [
32,
33]. Furthermore, a link between the susceptibility to checkpoint inhibitors and deletions of
CDKN2A was discovered in some cell lines derived from SCC of the head and neck region (HNSCC) [
34]. It remains to be determined whether this finding may be adapted to AFX/PDS and cSCC. If validated in further studies,
CDKN2A status might prove as a valuable biomarker in AFX and PDS that might open new therapeutic avenues in a substantial portion of patients suffering from this disease.
Our study does not provide a final decision on the ongoing debate regarding the histogenesis of AFX and PDS. Many experts assume that AFX and PDS derive from a mesenchymal origin [
1,
4], whereas others suggest that AFX may derive from an epithelial origin [
30,
35]. This theory was initially introduced by older studies describing clinicopathological similarities between AFX and cSCC with a sarcomatoid dedifferentiation [
36,
37]. AFX similar to cSCC and BCC frequently shows an association with actinic skin damage and a close proximity between the epithelium and the neoplastic spindle cell population, however without an epithelial dysplasia or carcinoma in situ component, which are both features and arguments for the diagnosis of a cutaneous spindle cell carcinoma with loss of keratin expression [
1,
38,
39]. Although we noticed a separation of BCC from cSCC and AFX/PDS by epigenetic profiling and also a remarkable delineation between AFX/PDS and cSCC, DNA-methylation profiles of individual AFX, PDS and cSCC were overlapping. Thus, the DNA-methylation analysis primarily recapitulated the morphology of BCC, cSCC and AFX/PDS, which is usually quite distinct.
Correctly distinguishing AFX/PDS from other tumors is critical to allocate affected patients to the correct type of treatment and follow-up protocols. The current diagnosis of AFX/PDS based primarily on lack of expression of certain lineage markers. However, there is a constant risk that tumors of other lineages may have lost expression of diagnostically relevant markers due to dedifferentiation and then may be misclassified as AFX/PDS. For certain entities, such as the illustrated example where methylation and gene mutation signatures argue for a melanoma, misclassification could have significant consequences for the patient [
40].
Therefore, it would seem prudent to perform molecular testing of cutaneous neoplasms when making a definitive diagnosis is not possible based on histomorphological and immunohistochemical assessment alone.
Conclusion
Our study demonstrates a proof of concept that DNA-methylation may be a valuable aid in routine diagnostics of skin tumors posing a diagnostic challenge with conventional analytic methods. Our data support the concept that AFX and PDS are histologically and molecularly closely related and probably belong to a common tumor spectrum. We observed a CDKN2A deletion in AFX (12%) and PDS (40%), which may represent a potential biomarker, if validated in future studies. Copy number analysis and DNA methylation profiling can aid in distinguishing AFX/PDS from other histologic mimics, even though these analyses alone cannot reliably distinguish AFX from PDS. The assessment of histopathological features such as subcutaneous involvement, necrosis, and lymphovascular or perineural invasion still remain critical in differentiating PDS from AFX.
Authors’ contributions
CK and DSt contributed equally to this manuscript. CK, TM and AvD conceived the project. CK and AvD wrote the manuscript with input from all co-authors. CK coordinated data generation. MB supervised the DNA-methylation array analysis. CK, DSt and DSch analyzed DNA-methylation array data. CK, KGG, DER, CV, WNMD, IP, MM, ACB, RB, SMP, UF, GM, TM and AvD provided tumor samples and metadata. MB supervised the methylation data generation. All authors analyzed the data and contributed to the final manuscript. All authors read and approved the final manuscript.