Undifferentiated and Dedifferentiated Metastatic Melanomas Masquerading as Soft Tissue Sarcomas: Mutational Signature Analysis and Immunotherapy Response

Abstract

The distinction between undifferentiated melanoma (UM) or dedifferentiated melanoma (DM) from undifferentiated or unclassifiable sarcoma can be difficult and requires careful correlation of clinical, pathologic, and genomic findings. In this study, we examined the utility of mutational signatures to identify patients with UM/DM with particular attention as to whether this distinction matters for treatment since the survival of patients with metastatic melanoma has dramatically improved with immunologic therapy, while durable responses are less frequent in sarcomas. We identified 19 cases of UM/DM that were initially reported as unclassified or undifferentiated malignant neoplasm or sarcoma and submitted for targeted next-generation sequencing analysis (NGS). These cases were confirmed as UM/DM by virtue of harboring melanoma driver mutations, UV signature, and high tumor mutation burden (TMB). One case of DM showed melanoma in situ. Meanwhile, eighteen cases represented metastatic UM/DM. Eleven patients had a prior history of melanoma. Thirteen of nineteen (68%) of the tumors were immunohistochemically completely negative for four melanocytic markers (S100, SOX10, HMB45, and MELAN-A). All cases harbored a dominant UV signature. Frequent driver mutations involved BRAF (26%), NRAS (32%), and NF1 (42%). In contrast, the control cohort of undifferentiated pleomorphic sarcomas (UPS) of deep soft tissue exhibited a dominant aging signature in 46.6% (7/15) without evidence of UV signature. The median tumor mutation burden for DM/UM vs. undifferentiated pleomorphic sarcoma (UPS) was 31.5 vs. 7.0 muts/Mb (P < 0.001). A favorable response to immune checkpoint inhibitor (ICI) therapy was observed in 66.6% (12/18) of patients with UM/DM. Eight patients exhibited complete response and were alive with no evidence of disease at the last follow-up (median: 45.5 mo). Our findings support the usefulness of UV signature in discriminating DM/UM vs. UPS. Furthermore, we present evidence suggesting that patients with DM/UM and UV signature can benefit from ICI.

Introduction

Melanoma is known for its histological plasticity. It has been termed the “great mimicker” for its ability to resemble other tumor types¹, in particular sarcomas.² This can create diagnostic challenges and can lead to misdiagnoses.^3-4 While immunohistochemistry can often firmly establish melanocytic differentiation in the work-up of morphologically challenging malignant tumors, in some undifferentiated melanomas or biopsies of dedifferentiated tumors, that may not always be possible. Undifferentiated melanomas (UM) exhibit a complete absence of melanocyte differentiation antigens (S100, SOX10, HMB45, and MELAN-A),⁵. Similarly, a biopsy sample of a dedifferentiated melanoma (DM), in which only a part of the entire tumor retains expression of melanocyte differentiation antigens, may also appear undifferentiated.^4,6 The correct diagnosis of DM/UM often requires molecular studies. With the increasing availability of sequence data, there have been a few publications in recent years documenting such cases. In a recent series of 35 patients, a predominance of undifferentiated pleomorphic sarcoma (UPS)-like histopathology was noted, which eventually were diagnosed as metastatic UM/DM.⁷ UPS is a diagnosis of exclusion for usually deeply located soft tissue neoplasms of an unknown cell of origin typically presenting in the trunk and extremities.⁸ While the etiology of UPS is largely unknown, approximately 5.1% are associated with radiation exposure.⁹ In contrast, cutaneous melanoma arises from melanocytes following the accumulation of genomic alterations often induced by ultraviolet radiation.^10-11 Historically, treatment approaches for undifferentiated sarcomas have included radiation, surgery, and standard chemotherapy.^12,13 Recent clinical trials have shown some promising results in certain sarcoma subtypes treated with immune checkpoint inhibitors,^14-15,16 whereas many melanomas respond favorably to immunotherapy.^17,18 However, whether UM/DM responds to immunotherapy similarly to conventional melanomas is not known.

In cancer genomics, mutational signatures are defined as a pattern of somatic mutation imprinted by events capable of inducing mutational processes, as detailed in the Catalogue Of Somatic Mutations In Cancer (COSMIC), a compendium of mutational signatures.^19-20 For instance, mutational analysis of lung adenocarcinomas obtained from known smokers revealed a recurrent C>A transversion, designated as smoking signature (COSMIC signature 4).²¹ Specific to our study, ultraviolet (UV) radiation exposure induces a characteristic set of C>T (or G>A) and CC→TT signature mutations at dipyrimidines sites.^22,23,24 This signature serves as an adjunct diagnostic marker for cutaneous melanomas and other tumors arising from the skin.^25-26 These mutational signatures are readily identified by next-generation sequencing (NGS) technologies and clinically used to inform tumor etiology and predict treatment response.^24,20 UPS is not enriched with UV-induced genomic signature (COSMIC signature 7); on the contrary, it is known to harbor an aging signature (COSMIC signature 1), characterized by C>T passenger mutations resulting from spontaneous deamination of 5-methylcytosine at CpG dinucleotides.²⁷ Importantly, the UV signature is a C>T mutation in a specific trinucleotide context, making it distinct from the aging signature.^28-29 As previously shown, bioinformatic signature analysis requires a high volume of somatic mutations per sample.²⁴ Therefore, it is practical to propose the implementation of large gene panels as a necessary prerequisite for signature determination when sequencing clinical samples.^30,31 Prior studies using small gene panels were limited in their ability to use mutational signatures to rule out UPS when interrogating DM/UM cases bearing UPS-like features.^5-7

Recent experimental studies have implicated melanoma dedifferentiation as a mechanism of immunotherapy resistance^32-33, and immunotherapy outcome data for UM and DM patients is scarce.³⁴ On the other hand, earlier studies have reported relatively lower or modest immunotherapy responses in multiple sarcoma types, including UPS patients.^15,35,36 Given the histologic similarity between UPS and UM⁷ and unavailability of specific immunohistochemical markers for UPS, and the scarcity of literature on UM response to immunotherapy, it is unknown if UM suffers similar immunotherapy outcome as the UPS.

One aim of our study is to investigate the utility of mutational signatures to discriminate DM/UM versus UPS of soft tissue following routine NGS of clinical samples using a large-scale sequencing platform. Equally important, we aim to evaluate outcomes and treatment responses in the DM/UM group.

In this study, we report our experience with metastatic UM/DM encountered in clinical practice, which were originally diagnosed as undifferentiated or sarcomatoid malignant neoplasms with differential diagnoses, which included high-grade sarcomas (See tables 1, 2). We document the importance of mutation analysis for the diagnosis and its implications for subsequent therapy decisions.

Table 1.

Clinical and radiologic characteristics of Dedifferentiated and Undifferentiated melanomas

Case No.	Age/Sex	Tumor Location (Superficial vs deep-seated)	Tumor Size (cm), Largest dimension	No. of lesions	History of melanoma	Time interval (months)	Details of known melanoma history (site, Breslow thickness and sentinel lymph node)
1	82/Male	Temple (superficial to deep)	3.2	1	YES	17	Cheek CPM, 0.3 mm, Nx
2	57/Male	Axilla (deep)	8.7	2	YES	0 (synchronous)	Concurrent melanoma in gluteal region (with conventional immunophenotype)
3	25/Male	Groin (deep)	11.9	1	NO	N/A	N/A
4	72/Male	Axilla (deep)	12.5	1	NO	N/A	N/A
5	81/Male	Thigh (deep)	4.7	1	YES	24	Leg CPM, 2 mm, N0
6	64/Female	Liver	6.5	1	YES	60	Cheek CPM, 0.8mm, Nx
7	71/Male	Small bowel	5.4	1	NO	N/A	N/A
8	68/Male	Arm (deep)	9.2	3	YES	0 (synchronous)	History of Back CPM (RU)
9	59/Male	Supradiaphragmatic mass	6	1	NO	N/A	N/A
10	55/Male	Back (superficial)	12	2	NO	N/A	N/A
11	82/Male	Lung	5.2	1	YES	24	Scalp CPM, 1.2 mm, N0
12	57/Male	Lung	3.3	2	YES	42	Neck CPM, 0.7 mm, N1
13	63/Male	Chest wall (superficial)	18.6	6	NO	N/A	N/A
14	69/Female	Abdomen wall (deep)	5.6	1	NO	N/A	N/A
15	54/Male	Axilla (deep)	14	2	YES	156	Shoulder CPM, 2.3 mm, N1
16	53/Female	Abdominal wall (deep)	3	4	YES	~192	Resection of pulmonary metastatic melanoma 16 years prior (conventional immunophenotype)
17	21/Female	Calf (deep)	13	2	YES	23	History of CPM, unknown site (RU)
18	82/Male	Axilla (deep)	5.2	1	YES	52	Back CPM, 0.4 mm, Nx
19	63/Male	Axilla (deep)	13.2	3	NO	N/A	N/A

CPM = Cutaneous primary melanoma; Time interval= Time interval between the prior known melanoma and dedifferentiated/undifferentiated melanoma metastasis; N/A = Not Applicable; Report Unavailable = RU.

Table 2.

Histological and immunohistochemical characteristics of Dedifferentiated and Undifferentiated melanomas.

Case No.	Specimen type	Cytomorphology	Immunophenotype of De-differentiated, Undifferentiated, and Trans-differentiated (divergent) melanoma					Initial Impression based on standard Pathologic evaluation
			S100	SOX10	HMB45	Melan A	Other markers of melanocytic and or divergent/trans-differentiation
1	Resection	Mixed epithelioid, spindle and pleomorphic cells	Pos (focal)	Pos (focal)	Pos (focal)	Pos (focal)	Diffuse CD10 (++), Focal [Tyrosinase (+), MiTF (+), CD31 (+) and CD68 (+)], Neg for (CD34, desmin, PRAME, AE1/AE3, CK5/6, and p40)	Malignant spindle and epithelioid cell neoplasm with sarcomatoid features and melanocytic differentiation, favor dedifferentiated malignant melanoma
2	Biopsy	Mixed epithelioid, spindle and pleomorphic cells	Neg	Neg	Neg	Neg	Vimentin (+), Neg for (CD45, CD30, AE1/AE3, Cam 5.2, p40, desmin, SMA, MDM2, and CDK4).	High grade sarcoma with spindle, pleomorphic and histiocytoid (epithelioid) features and large areas of necrosis; based on subsequent discovery gluteal melanoma, UM could not be excluded
3	Biopsy	Mixed epithelioid and clear cells	Pos (diffuse)	Pos (diffuse)	Neg	Neg	Retained BAF47(INI1). Neg for , CD3, CD30, CD20, ERG, CD34, TTF1 and OCT-4)	Malignant S100 protein-positive neoplasm with necrosis
4	Biopsy	Pleomorphic and epithelioid cells	Neg	Neg	Neg	Neg	Focal vimentin (+), CD30 (+), retained INI1 and SMARCA4, Neg for (CD45, CD3, CD20, CD2, CD5, TIA1, Granzyme B, MUM1, CD79a, CD15, Lysozyme, CD11c, CD163, ALK-D5F3, AE1/AE3, CK18, EMA, CD31, CD34, ERG, desmin, SALL4, WT1, CKIT, and BRAF and OCT-2)	High grade undifferentiated malignant neoplasm
5	Biopsy	Epithelioid cells	Neg	Neg	Neg	Neg	Vimentin (+), retained INI1, Neg for (Tyrosinase, AE1/AE3, CAM5.2, SMA, Desmin ERG, CD31, CD34, MDM2, NKX3.1, TTF-1, PAX8, CD5/6 and BRAF V600E)	High-grade malignant epithelioid neoplasm
6	Biopsy	Spindle cells	Pos (diffuse)	Pos (diffuse)	Neg	Neg	Loss of H3K27me3, Neg for (MiTF, BRAF, NTRK1)	Metastatic high grade malignant spindle cell neoplasm
7	Resection	Spindle cells	Neg	Neg	Neg	Neg	Focal pos for (caldesmon and CD34), Neg for AE1/AE3, CD31, ERG, CD117, DOG1 and ALK)	Undifferentiated spindle sarcoma
8*	Biopsy	Mixed spindle and pleomorphic cells	Neg	Neg	Neg	Neg	Neg for (SMA, desmin, CDK4, MDM2, and AE1/AE3)	Undifferentiated Pleomorphic Sarcoma, high-grade, with associated necrosis
9	Biopsy	Mixed spindle and pleomorphic cells	Neg	Neg	Neg	Neg	Diffusely (+) for (OSCAR, CD56 and TLE1). Focal (+) for (CD99, WT1). Neg for (Claudin 4, cytokeratin CAM 5.2, chromogranin, synaptophysin, INSM1, TTF-1, calretinin, NUT, CD3, CD20, CD45, NKX3.1, and STAT6). Retained INI1 and SMARCA4)	High grade spindle and epithelioid malignant neoplasm
10	Resection	Mixed epithelioid, spindle and clear cells	Pos (focal)	Pos (focal)	Neg	Neg	Loss of H3K27me3, Focal [MiTF (+), Tyrosinase (+)], and PRAME (+)	Poorly differentiated S100+ malignant spindle and epithelioid cell neoplasm
11	Resection	Mixed round and pleomorphic cells	Neg	Neg	Neg	Neg	Focal (+) (CD99, BCL2, and DOG1). Neg for (CD34, desmin, synaptophysin, tyrosinase, INSM1, MDM2, myoD1 and myogenin)	Metastatic sarcomatoid neoplasm
12	Biopsy	Pleomorphic and epithelioid cells	Neg	Neg	Neg	Neg	Focal (++) (CAM5.2, and AE1/AE3). Neg for (TTF1, p40, PAX8, and NKX3.1)	Poorly differentiated malignant neoplasm with extensive necrosis
13	Resection	Epithelioid cells	Neg	Neg	Neg	Neg	Vimentin (+), retained BAF47(INI1), Focal (Cam5.2 (+), EMA (+)), Neg for (CD3, CD20, CD117, CD31, CD34, ERG, CD79a, CD163, CD68, CD30, CD45, MUM-1, CD138, Napsin-A, TTF-1, NKX3.1, SMA, desmin, arginase, myogenin, CK5/6, CK7, AE1/AE3, CK20, PAX-8, calretinin, CDX2, inhibin)	Malignant epithelioid neoplasm
14	Biopsy	Spindle cell tumor in a myxoid background	Neg	Neg	Neg	Neg	Diffuse (++) for (Claudin 4, Oscar, Myogenin, MyoD1 and Desmin), focal PRAME (+), Neg for (TTF-1, P40, cam5.2, SMA, CD31, ERG, CEA, WT1, calretinin, MDM2, and CDK4. Retained BAP1, INI-1, and SMARCA4)	Metastatic poorly differentiated malignant neoplasm with epithelial and rhabdomyosarcomatous differentiation
15	Resection	Spindle and round cells	Neg	Neg	Neg	Neg	Focal (+) [CD34 and SMA], Neg for (EMA, AE1/AE3, desmin, ERG, STAT6, MUC4, CD45, CD21, CD35, CD23 and PRAME)	Malignant spindle cell neoplasm, NOS
16	Biopsy	Mixed epithelioid, spindle and pleomorphic	Pos (Focal)	Neg	Neg	Neg	Diffuse PRAME (+) and BRAFV600E	Metastatic melanoma with dedifferentiation
17	Resection	Round to epithelioid	Neg	Neg	Neg	Neg	Tyrosinase (−), Diffuse (++) for (PRAME, Desmin, myogenin and BRAFV600E), Focal MyoD1 (+)	High-grade sarcomatoid malignant neoplasm with necrosis
18	Resection	Epithelioid	Pos (focal)	Pos (focal)	Neg	Neg	Focal (+) (PRAME, AE1/AE3). Neg for (MiTF, BRAFV600E and RASQ61R)	Metastatic melanoma with dedifferentiation
19	Resection	Mixed epithelioid, spindle and pleomorphic	Neg	Neg	Neg	Neg	Focal PRAME (+), Neg for (AE1/AE3, EMA, CD31, Desmin, SMM, and CD30. CD45, CD3 and CD20)	High grade sarcomatoid neoplasm with spindle and epithelioid phenotype

Neg = Negative, Pos = Positive, + = Weakly positive; ++ = Strongly positive; N/A = not available; NOS = Not Otherwise Specified;

^*= Prior history of melanoma was unknown at the time of diagnosis

Materials and Methods

Case selection and clinical data collection

This study was approved by Institutional Review Board (IRB NO. 16 −1682). Pathology records were searched for immunohistologically unclassifiable malignant cases that were submitted for NGS-based testing as undifferentiated or dedifferentiated malignant neoplasms suspicious for melanoma versus sarcoma and subsequently classified as melanoma after testing with next-generation sequencing (MSK-IMPACT) from January 2014 to January 2022. The following parameters were recorded: sex, age, tumor location, and history of other tumors. Notably, only one sample was tested per patient, and all samples (except case No.1) were metastases. The cases were a mix of in-house and consultation cases. Nine of 19 cases were received as resection specimens.

Immunohistologic data review

All available pathologic specimens which included hematoxylin-and-eosin (H&E) slides, immunohistochemical (IHC) stains, and reported IHC results were re-reviewed by two pathologists (IK and MH) to evaluate tumor morphology and confirm the uncertainty of tumor lineage after reviewing the expression of melanocytic markers with at least four stains (S100, SOX10, Melan-A, and HMB-45). Immunohistochemistry (IHC) was performed as previously reported.³⁷

Comprehensive genomic profiling (CGP)

All samples included have undergone CGP with DNA-based next-generation sequencing (NGS) using MSK-IMPACT. MSK-IMPACT is a hybridization capture-based assay that assesses somatic mutations, copy number, and structural variants in 410-505 cancer-related genes against the patient’s matched blood sample (the size of gene panel analyzed was dependent upon the time when the tumor was sequenced and the version of MSK-IMPACT available).^24,38 These molecular alterations were further curated using OncoKB to identify clinically relevant cancer gene alterations.³⁹ The profiles of sequenced tumors were studied to identify melanoma-related recurrent oncogenic drivers, such as alterations affecting the MAPK signaling pathway, namely activating “hot spot” mutations of BRAS/NRAS or Neurofibromatosis Type 1 (NF1) loss of function (LoF) mutations as reported by large scale sequencing studies.^{40-41,42,43-44} In addition, targeted RNA-sequencing (MSK-Fusion targeted RNAseq)⁴⁵ utilizing Archer^™ FusionPlex^™ technology⁴⁶ was performed for possible discovery of rearrangements on selected cases which showed immunohistologic findings suspicious for fusion-driven sarcomas.

Mutational signature analysis

Similar to our previous study,²⁶ MSK-IMPACT results were used to identify a mutational signature pattern for each sample with greater than 20 mutational calls. Briefly, individual signature types were classified using the six substitution subtypes: C>A, C>G, C>T, T>A, T>C, and T>G.²⁸ The estimated contribution of each signature to the mutation spectrum was determined using publicly available computational methods.⁴⁷ The mutation spectrum of each sample was refitted to the 30 described mutational signatures.²⁰ Every signature was assigned a weight that corresponds to the proportion of mutations explained by each given signature. A mutational signature with greater than 40% contributing mutations was labeled as dominant.³⁸ Using the same method, MSK-IMPACT results of 853 samples of Sloan Kettering cutaneous melanomas (SKCM) and 167 samples of primary undifferentiated pleomorphic sarcoma (accessioned January 2014 to July 2021) were analyzed for mutational signatures to establish a controlled comparative analysis.

Post-genotyping classification criteria

Samples were classified as melanoma when found to harbor oncogenic driver mutations affecting either MAPK signaling pathway, such as activating “hot spot” mutations of BRAF/NRAS or NF1 loss of function (LoF) mutations in combination with UV signature (>40%) and high tumor mutation burden (TMB), that is >90^th percentile of MSK-IMPACT TMB (Across All Tumor Types) (Table. S4). Samples were further classified based on IHC staining as either UM or DM. As previously reported,^4-5 tumors with complete loss of melanocytic markers (S100, SOX10, HMB45, and MELAN-A) were labeled as UM. Meanwhile, DM was reserved for melanomas showing a biphasic profile, represented by immunonegative area (dedifferentiated component) transitioning to regions of conventional melanoma immunoreactive for any of the four (4) aforementioned melanocytic immunostains (S100, SOX10, HMB45, and MELAN-A).

Clinical outcomes and follow-up

Clinical outcome was assessed through January 30, 2022, or until death from disease or unrelated causes. The primary study outcomes analyzed were immunotherapy response and disease progression/recurrence. The response was assessed by the treating oncologist and documented as complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD) according to Response Evaluation Criteria in Solid Tumors (RECIST) guidelines.⁴⁸ Unfavorable response was defined as a progressive disease after receiving “optimal therapy” (at least 3 cycles of immune checkpoint inhibitors (ICI) therapy). Patients’ status (alive, dead, or lost to follow-up) at the date of the last follow-up was recorded.

Statistical analysis

Statistical analysis was performed using R version 4.1.3 (https://www.R-project.org/). Fisher’s exact test was used for the comparison of clinicopathogenomic parameters.

Results

Demographic and Clinical Features

NGS-based testing (MSK-IMPACT) was utilized to identify 19 cases with pathogenomic features compatible with either DM or UM. Clinicopathologic characteristics of all nineteen cases of malignant tumors initially reported as unclassified or undifferentiated malignant neoplasm or sarcoma are summarized in Table 1. There were 15 males and 4 females. The age at diagnosis ranged from 21 to 82 years (median = 63 years). Seventeen of nineteen (17/19) patients were aged >50 years. Fifty-eight percent (11/19) of patients had a history of cutaneous melanoma which was not always available at the time of diagnosis in the majority of cases. Notably, 84% (16/19) of cases presented as deep-seated masses, of which five cases involved visceral organs and the rest deep soft tissue. The axilla was the most common tumor site (5 cases), presenting as mass lesions with an average size of 10.7 cm. To our knowledge, all axillary masses were found in deep soft tissue, not in a lymph node basin (Table 1). Radiologically, tumor size at diagnosis ranged from 3 to 18.6 cm (median = 6.5), with fifteen tumors (79%) measuring > 5 cm. 47% (9/19) of patients had multiple lesions.

Immunohistologic characteristics

Undifferentiated melanoma (UM) subset

Overall, 68% (13/19) of cases were negative for all four melanocytic markers (S100, SOX10, HMB45, and MELAN-A) and without specific expression of sarcoma lineage markers, hence fitting for UM profile as per aforementioned post-genotyping classification criteria. Seven of the 13 cases were composed of mixed pleomorphic, spindled, and epithelioid cells with bizarre nuclei, features morphologically compatible with UPS (Fig. 1a-b). The remaining six cases showed a predominant epithelioid or spindled morphology, with rhabdomyoblastic differentiation in 2 cases. Three types of transdifferentiation (heterologous/divergent differentiation) were noted (i) epithelial (AE1/AE3, EMA, Claudin-4, and OSCAR expression) (Fig. 2b-c), (ii) rhabdomyoblastic (Myogenin, MyoD1 and desmin expression) (Fig. 1d-e) and (Fig. 2d-e), (iii) myofibroblastic/fibroblastic differentiation with expression of SMA and CD34 (case 15).

Fig. 1. Morphologic and immunohistochemical features of undifferentiated melanoma.

Case 19 (A) Spindle cell-rich areas overlapping with (B) zones of pleomorphic and epithelioid cells with bizarre mono- and multinucleated cells; Case 17 (C) Spindle cells in a myxoid background with rhabdomyosarcomatous differentiation with strong expression of (D) desmin and (E) myogenin; Case 15 (F) sheets of hyperchromatic oval to round cell alternating with (G) myxoid matrix rich spindle cell zones.

Fig. 2. Morphologic and immunohistochemical features of dedifferentiated melanoma/undifferentiated melanoma.

Case 14 (A) Epithelioid and round cell morphology with rhabdomyosarcomatous and epithelial differentiation demonstrated by expression of (B) OSCAR, (C) claudin-4, (D) myogenin, (E) desmin. Case 10 (F) Low power view showing dermal component with no evidence of an epidermal connection, (G) intersecting fascicles of spindled cells with brisk mitosis, alternating with (H) zones of clear cells.

Dedifferentiated melanoma (DM) subset

Using the aforementioned post-genotyping classification criteria, six cases were classified as DM. Melanoma in situ was identified in only one case (case 1). In the latter case, the dedifferentiated component showed pleomorphic dermal sarcoma (PDS)-like features with diffuse CD10 expression. One case (case 16) showed only focal expression of S100 with diffuse expression of PRAME. The remaining four cases had S100 and SOX10 expression without HMB45 and MELAN-A expression. The latter subset, Case 6, had intersecting fascicles of spindled cells and loss of H3K27me3, thus raising concern for malignant peripheral nerve sheath tumor (MPNST). Likewise, case 10 was a superficial cutaneous mass (without evidence of overlying melanoma in situ) showing a biphenotypic pattern consisting of clear cell zones (Fig. 2h) simulating clear cell sarcoma (CCS) and additional areas predominated by long fascicles of hyperchromatic, uniform spindled cells with brisk mitosis and loss of H3K27me3 (Fig. 2f-g). Microscopic findings are summarized in Table 2.

Next-generation sequencing: Mutational signature analysis

In our DM/UM series, all 19 cases passed the criteria for mutational signature analysis (> 20 mutations). In our control cohort, 85.4% (729/853) of cutaneous melanomas and 9% (15/167) of primary UPS harbored more than 20 mutations per sample. Cutaneous melanomas had a dominant UV signature in 93.9% (685/729) of the cases (Fig. 3c; Supplementary Fig. 1), while none of the primary undifferentiated pleomorphic sarcomas showing >20 mutations showed a dominant UV signature (0/15) (Fig. 3b, c). A dominant aging signature (COSMIC1) was most prevalent in primary UPS, accounting for 46.6% (7/15) of samples. Meanwhile, the remaining UPS cases did not show a dominant signature (Fig. 3b, c). Analysis of UV signature as a continuous variable across was performed on three (3) groups (DM/UM versus control [UPS and cutaneous melanoma], Fig. 3d). Mutations attributed to UV signature accounted for the majority of all mutations in DM/UM, which was similar to the distribution of UV-related mutations in cutaneous melanomas but sharply distinct from primary UPS (p < 0.001).

Fig. 3. UV signature in dedifferentiated (DM) and Undifferentiated melanoma (UM) with cutaneous melanoma (CM) and undifferentiated pleomorphic sarcoma (UPS) as control groups.

(A) Contribution of various signatures to each case of DM and UM. Each bar represents a single case, and Y-axis represent proportion attributed by each mutational signature per case. (B) Spectrum of mutational signature in UPS. (C) Accumulative fraction of evaluable cases characterized by dominant signatures for CM vs. (DM+UM) vs. UPS. (D) Dot plots showing percentage of aging (signature 1) and UV- related mutations (signature 7) in each sample as a continuous variable. Dominant signature threshold denoted by a dotted line at 40%.

Distribution of genomic alterations in undifferentiated and dedifferentiated melanomas

Next-generation sequencing of the nineteen cases (19) revealed a median of 34 nonsynonymous somatic mutations per case (range, 16 – 102). Tumors had a mean TMB of 35.1 total mutations per megabase (range = 13.2–90.4). Of note, the lowest TMB score recorded in this series (13.2) is equivalent to the 61st percentile of melanoma TMB in the MSK-IMPACT cohort (n=2447). Overall, 63% (12/19) of cases bore a TMB score > 21 (equivalent to > 76^th percentile of melanoma TMB in MSK-IMPACT cohort, and >90^th percentile of MSK-IMPACT TMB (Across All Tumor Types). The oncogenic driver mutations affecting signaling of the MAPK⁴⁹ (mitogen-activated protein kinase) pathway were identified in all cases: Codon 600 of BRAF (V600E/K) (n = 5), codon 61 of NRAS (Q61R/K/L) (n = 6) and NF1 loss of function (LoF) mutations (n = 8). TERT promoter mutation (g.1295228 C>T [C228T]) was identified in all cases. In addition, known melanoma frequent co-occurring mutations were detected; PTPRT⁵⁰ (n = 12), PREX2⁵¹ (n = 9), NOTCH4⁵² (n = 8), GRIN2A⁵³ (n = 8), ERBB4⁵⁴ (n = 8), RAC1⁴⁹ (n = 5) and ROS1⁵⁵ (n = 5). In addition to the aforementioned prevalent UV signature, all cases had a genomic profile compatible with cutaneous melanoma. Driver alterations of extra-cutaneous melanoma^56-42 involving genes such as GNA11, GNAQ, SF3B1, and KIT were not identified. No rearrangements or fusions were identified on five cases (Supplementary Table S4 [†]) that underwent RNA sequencing for the possible discovery of driver fusions.⁴⁵ Similarly, case 10 with clear cell sarcoma as one of the differential diagnoses, was negative for rearrangement of EWSR1 break-apart FISH.⁵⁷ Based on the aforementioned post-genotyping classification criteria, 13 cases were classified as UM, while 6 were labeled as DM. Key genomic alteration profiles are summarized in Figure 4 and Supplementary Table S4. A detailed list of genomic alteration is provided in Supplementary Table S1 (Intronic mutations attributing to a mutational signature are not listed). Notably, positive IHC staining was seen in two of 6 cases that underwent immunohistochemistry with anti-BRAF V600E (VE1). Meanwhile, only one case was subjected to RASQ61R IHC staining and showed negative immunostaining (Table 2). Overall, BRAFV600E and RASQ61R IHC results were compatible with subsequent NGS-based mutation testing.

Fig. 4.

Distribution of frequent mutations and copy number alterations in the dedifferentiated melanoma/undifferentiated melanoma group with parallel comparison to undifferentiated pleomorphic sarcoma group (using fisher exact test).

Distribution of mutations in undifferentiated pleomorphic sarcomas

Clinicopathologic characteristics of UPS control cohort (15 cases, each with > 20 mutations) are summarized in Supplementary Table S2, and a detailed list of genomic alteration is provided in Supplementary Table S3 (Intronic mutations attributing to mutational signature are not listed). Tumors had a median size of 8.2 cm (range, 4.2-15). The median TMB was 7 total mutations per megabase (range, 0.9-31.5). Of note, a sample with the highest TMB (31.5) was derived from a patient with known Lynch Syndrome. Overall, 73.3% (11/15) of cases had a TMB score < 12. Notably, the median TMB of soft tissue sarcomas in the MSK-IMPACT cohort is recorded as 1.1 (data not included). In the 15 UPS cases with >20 mut, oncogenic mutations were frequently seen in TP53 (n=11 cases), NF1 (n=5), RB1 (n=5), and ATRX (n=3). Also, deep deletion of CDKN2A/2B was identified in 2 cases. No recurrent oncogenic BRAF or NRAS driver alterations were identified.

Outcome following immune checkpoint inhibitors (ICI) treatment

Clinical follow-up and treatment response data are summarized in Table 3. Follow-up data were available for all patients, interval ranged from 3 to 181 months (median: 33 mo). Treatment with anti-PD-1 inhibitor with or without a CTLA-4 inhibitor was given in all except one patient who had no evidence of disease following resection of localized disease. A combination of nivolumab and ipilimumab was utilized in 13 patients, and anti-PD-1 monotherapy with nivolumab or pembrolizumab in 5 patients. Twelve of eighteen (66.6%) patients had favorable responses (CR [n = 8], PR [n = 2], SD [n = 2]). Meanwhile, 6 patients died following the progression of disease (PD). In the latter subgroup, 3 patients died after receiving at least >3 cycles of ICI. The remaining 3 patients received one (1) cycle of ICI each, two of them following initial treatment with combination BRAF-inhibitor and MEK-inhibitor therapy (cases 13 and 17), and all 3 cases initially presented with multiple lesions (n = 2-6). More specifically, case 17 presented with 2 lesions but suffered CNS metastasis after ICI cessation following severe dermatomyositis. Case 8 presented with 3 lesions leading to CNS metastasis within 3 months. Case 13 presented with metastases in six organs prior to therapy and died within six months. Overall, all eight patients with complete response (CR) were alive with no evidence of disease at the last follow-up (median: 45.5 mo). No recurrences were observed. Notably, 6 cases with UPS-like morphology received ≥ 3 cycles of ICI. In the latter subset, 4 cases had a favorable response: 3 showing complete response and 1 with stable disease. Finally, using Kaplan Meier analysis, no survival difference was observed following initiation of immunotherapy when UM/DM group (n=18) was compared with the stage-matched cutaneous melanoma (SKCM) control group (n=172) treated with Ipilimumab + Nivolumab, Nivolumab, or Pembrolizumab (Fig. 5).

Table 3.

Treatment and Clinical outcomes of Dedifferentiated/Undifferentiated melanomas

Case Number	Immunotherapy +/− Target therapy	Cycles of Immunotherapy (n) and duration of target therapy [months]	Response/Recurrence	Outcome	Follow-up (months)
1	Nivolumab	(8)	PD, perineural spread along mandibular (V3) nerve, with bilateral adrenal metastases	DOD	35
2	Ipilimumab and Nivolumab	(17)	Complete response, no recurrence	AFD	63
3	Ipilimumab and Nivolumab	(14)	Complete response, no recurrence	AFD	58
4	Ipilimumab and Nivolumab	(3)	PD following IO and RT, hemorrhage following rupture of axillary arteries	DOD	4
5	Ipilimumab and Nivolumab	(15)	Complete response, no recurrence	AFD	81
6	Ipilimumab and Nivolumab	(4)	Durable partial response (near complete response), no progression	AWD	54
7	Nivolumab	(21)	Complete response, no recurrence	AFD	33
8	Ipilimumab and Nivolumab	(1)	PD (CNS metastasis, multiple lesions, largest 2.6 cm)	DOD	3
9	Ipilimumab and Nivolumab	(3)	Stable disease, treatment c/b Colitis, pancreatitis, hypophysitis and stroke	AWD	14
10	Nivolumab	(7)	Complete response, no recurrence	AFD	12
11	Pembrolizumab	(3)	Stable disease	AWD	18
12	Pembrolizumab, Nivolumab, then Ipilimumab	(11), (15), then (3)	PD (progression of CNS metastasis; Basal ganglia, 5.8 cm), IO course c/b adrenal insufficiency following bilateral adrenal metastases	DOD	77
13	Dabrafenib and trametinib, then Nivolumab and Ipilimumab	[2], then (1)	Partial response, PD (High burden disease before treatment initiation; metastasis in six organs)	DOD	6
14	Ipilimumab and Nivolumab	(9)	Complete response, no recurrence	AFD	20
15	Ipilimumab and Nivolumab	(23)	Durable Partial response, no progression	AWD	168
16	Pembrolizumab	(10)	Complete response	AFD	181
17	Encorafenib and Binimetinib, then Ipilimumab and Nivolumab	[4], then (1)	Initial good response (50% decrease in splenic metastasis (7.2 to 3.1 cm), IO course c/b worsening dermatomyositis precluding further treatment, PD (brain metastasis – 3 cm)	DOD	19
18	None	n/a	No evidence of disease following resection	AFD	61
19	Ipilimumab and Nivolumab	(5)	Complete response, no recurrence	AFD	11

AFD = Alive and free of disease; AWD = alive with disease; DOD = dead of disease; PD = Progressive disease; M = months; IO = immunotherapy; c/b = complicated by; RT = Radiation therapy; [n] = duration of target therapy in months, (n) = Cycles of immunotherapy.

Fig. 5.

Kaplan-Meier overall survival curve of dedifferentiated melanoma/undifferentiated melanoma group (n=18) compared with stage-matched cutaneous melanoma (SKCM) group (n=172) treated with immunotherapy (Ipilimumab + Nivolumab, Nivolumab, or Pembrolizumab).

Discussion

In this study, we retrospectively analyzed nineteen cases of immunohistologically unclassifiable malignant tumors that were subsequently re-classified as melanoma after undergoing NGS-based CGP as part of routine clinical care. A recent study demonstrated the application of mutational signature analysis to reclassify lung-only melanomas as metastases of cutaneous origin.²⁶ Here, the application of the same methodology identified UV-signature and driver mutations that allowed classification as malignant melanoma of a series of tumors with otherwise unclear lineage. We provide thorough documentation of their clinicopathologic characteristics, treatment, and outcomes. Our study validates the utility of mutational signatures to discriminate DM/UM from UPS and provides evidence of implications this distinction carries for therapy decisions and clinical outcomes following ICI and targeted therapy.

In this series, we identified DM/UM cases that showed immunohistologic features simulating unclassifiable/undifferentiated sarcomas, and others mimicking specific sarcoma types such as rhabdomyosarcoma, malignant peripheral nerve sheath tumors (MPNST) and clear cell sarcoma (CCS) (1 case), and pleomorphic dermal sarcoma (PDS). For example, cases 19, 14, and 10, all had no known history of melanoma and demonstrated UV signature-induced high TMB, which tied well with complete response to ICI therapy.⁵⁸ To elaborate, case 19 presented with multiple lesions to include a 13.2 cm axillary mass that showed UPS-like features while bearing NRAS Q61L mutation. Agaimy et al.,⁷ recently published the largest series of DM/UM patients, in which five out of nine cases with driver NRAS codon 61 mutation showed UPS-like histopathology. In agreement with the latter study⁷, the majority of UM cases in our series were located in the axilla and showed UPS-like features. Notably, sarcoma spreads predominantly by hematogenous route, whereas lymph node metastases are typically uncommon.^59-60

Moreover, in case 14, a presumed diagnosis of unusual rhabdomyosarcoma was entertained in a 69-year-old female based on showing rhabdomyosarcomatous differentiation with desmin/myogenin positivity and aberrant epithelial expression. ^61-62 In general, rhabdomyosarcoma (RMS) is more common in children^63-65, and in this case, RNA seq was negative for fusion or rearrangements. In addition, the high TMB and complete response to ICI tied well with the diagnosis of melanoma.⁶⁶

Case 10 was a dermal-based lesion with biphasic features consisting of predominantly clear cell zones with focal melanocytic expression and clear cell-poor areas enriched with intersecting fascicles of spindled cells with brisk mitosis and loss of H3K27me3 by IHC⁶⁷, that altogether raised concern for CCS⁶⁸ versus MPNST⁶⁹ versus neural crest derived⁷⁰ collision tumor. However, FISH was negative for EWSR1 rearrangement, and loss of melanocytic expression is unusual for CCS. Despite harboring NF1 LoF mutation, the lesion was not anatomically associated with a nerve and there were no stigmata of neurofibromatosis which would have supported the possibility of cutaneous MPNST.⁷¹ Instead, a high TMB favored MPNST-like melanoma⁷² and occasional loss of H3K27me3 has been described in melanoma.⁷³

Using MSK-IMPACT results, all 19 cases with sarcoma-like morphology were found to bear a melanoma genomic profile.²⁴ More specifically, all cases showed hotspot mutations of melanoma arising from intermittent (BRAF V600, NRAS Q61)⁷⁴ and severe sun exposure (NF1 LoF mutations)⁷⁵, typical of cutaneous origin.⁴⁹ Further, all cases showed TERT promoter mutations with associated melanoma co-occurring mutations affecting genes such as PTPRT⁵⁰, PREX2⁹¹, ERBB4⁵⁴, RAC1⁴⁹, and ROS1⁵⁵. It’s worth noting that BRAF and NRAS mutations are rare in sarcomas and have mostly been reported in presumed UPS cases with a previous history of melanoma.^76-77 In support, prior studies have leveraged the latter molecular profiles to decipher tumors of unknown primary that are suspicious for sarcoma versus melanoma.⁷⁸ In addition, our series consisted of 42% (8/19) of melanomas of unknown primary (MUP), which had a similar genomic profile as cases with a history of cutaneous melanomas. Comparably, these results are directly in line with a large-scale study showing similar MAPK driver alterations between cutaneous melanoma and MUP, thus suggesting a cutaneous origin for the latter.⁴¹

Further underscoring cutaneous origin of our DM/UM series, all cases harbored a dominant UV mutational signature with associated high TMB.⁵⁸ This observation was in agreement with our cutaneous melanoma control group, in which 93.9% of the samples harbored a dominant UV signature. Additionally, our series was enriched with older males and NF1 mutant melanomas that consequently accounted for five times higher median TMB compared to the selected highly mutant UPS control group. In the same manner, Shoushtari et al.,⁴¹ showed a predominance of high TMB in NF1 mutant cutaneous melanomas and MUP, in which the TMB positively correlated with advancing age. Conversely, large-scale sarcoma studies have described an enrichment of low TMB and aging signature in UPS samples, without evidence of UV signature.²⁷ Altogether, the genomic profile detected in our DM/UM series was in favor of melanomas of cutaneous origin.⁴⁰

Similar to a TCGA publication,²⁷ a UV signature was absent in the UPS control group, and instead, 46.6% (7/15) of cases harbored a dominant aging signature. In addition, a distinct genotype devoid of MAPK oncogenic (hotspot) alterations and enriched with TP53, RB1, and ATRX mutations was identified in the primary UPS control cohort.⁷⁹ As shown by others,^80-81 the preferential predominance of a UV signature in cutaneous melanoma, pleomorphic dermal sarcoma (PDS), and cutaneous angiosarcomas is linked to their superficial cutaneous origin. Likewise, the lack of a UV signature in UPS and other subfascial soft tissue tumors is due to their deep-seated site of origin.^27,82 Notably, Cheung et al.,⁸³ have used the term ‘’undifferentiated pleomorphic sarcomas” to describe primary scalp tumors bearing UPS-like immunohistopathology and UV mutational signature, which are features fitting for pleomorphic dermal sarcoma (PDS).⁸² To clarify, PDS commonly arises from the scalp, and its lineage of origin is unknown. Hence, its historically referred to as superficial malignant fibrous histiocytoma or UPS of the skin.⁸² To that end, PDS should not be confused with deep-seated UPS of the extremities and trunk, which we have shown to be rich in COSMIC1 but poor in UV signature.^27,82 Nevertheless, cutaneous pleomorphic spindle cell tumors⁸⁴, such as PDS, are important differential diagnoses when considering the diagnosis of UM. Unlike melanoma, the majority of PDS lack hotspot MAPK alterations and bear recurrent NOTCH1/2 and FAT1 inactivation mutations.⁸⁵ The latter mutations were not identified in our DM/UM series. Overall, our analysis confirms the specificity of UV signatures in DM/UM cases as compared to UPS of deep soft tissue.

In this series, immunohistochemistry for BRAF V600E (n=6) and NRAS Q61R (n=1) showed concordance with NGS results (2 for BRAFV600E and 1 for NRASQ61R). To employ BRAFV600E or NRASQ61R IHC for all tumors with morphological features of UPS is not very practical. Besides, as rare sarcomas do harbor these mutations^91-93, it has limited utility. Additionally, the RASQ61R IHC stain lacks specificity for NRAS Q61R mutations and has been shown to stain positive in KRAS Q61R mutant tumors.⁹⁴ Considering that KRAS Q61R mutations are common in carcinomas and rare in melanoma, RASQ61R IHC stain may not be ideal for accurately discovering UM/DM.^95-96 Two patients (cases 13 and 17) received BRAF/MEK inhibitors with short-lived benefit for 2 and 4 months, respectively. Case 13 had initial high burden of disease (metastases in six organs) and died only six months following diagnosis despite receiving an initial cycle of ICI. Similarly, patient 17 had an initial response to BRAF/MEK inhibitor but died of disease following CNS metastases and cessation of ICI treatment due to severe dermatomyositis. Down-regulation of master regulator of melanocytic differentiation (MITF) has been described as a mechanism of resistance to BRAF⁸⁶ and other MAPK⁸⁷ pathway inhibitors. Rare reports of DM/UM patients have shown continued response to BRAF and MEK inhibitors,^88-89 including those showing rhabdomyosarcomatous differentiation,^7,90 a known marker of poor prognosis.⁹⁰ Overall, the adverse outcome in our two patients is likely related to their burden of disease and therapeutic limitations as a result of treatment complications.

A review of published literature identified only a small number of patients with ICI-treated DM/UM.^88,90,34 In the latter reports, two patients with rhabdomyosarcomatous differentiation⁹⁰ exhibited only partial response, while one patient showed complete response of recurrent disease that initially masqueraded as a carcinoma.³⁴ It has been speculated by some authors that a dedifferentiated phenotype could represent an inflammation-induced mechanism of resistance in adoptive T-cell transfer therapy.^32-33 Of note, all patients in our series had evidence of de-/undifferentiation prior to therapy, except for one patient (case 16) who initially received ipilimumab for 5 years due to lung metastatic melanoma (conventional phenotype), leading to 10 years of remission before the discovery of DM at multiple sites. However, her recurrence showed a complete response to pembrolizumab, and therefore it remains uncertain if the dedifferentiated phenotype was treatment induced.

Of note, we observed a favorable response in 12 out of 18 patients (66.6%). No significant survival difference was observed following initiation of immunotherapy when UM/DM group (n=18) was compared with stage-matched cutaneous melanoma (SKCM) control group (Fig.5). In particular, eight patients exhibited complete response and were alive with no evidence of disease at the last follow-up (median: 45.5 mo). Of interest, four out of seven UM patients bearing UPS-like features showed a favorable response. Furthermore, a favorable response was associated with NRAS (5/6), NF1(4/8), and BRAF(3/5) driver mutations. Of note, a previous analysis of treatment outcomes in cutaneous melanomas harboring MAPK alterations showed a significantly longer time to treatment failure (TTF) in the high TMB - NF1 mutant subset as compared to NRAS Q61 and BRAF V600 mutant groups, which showed an intermediate TMB and inferior TTF.⁴¹ Surprisingly, in our series, 4 out of 8 patients who showed complete response harbored NRAS Q61 mutations. This effect is likely mediated by the unusually high mean fraction of UV – related mutations (86%) and high TMB harbored by this NRAS Q61-mutant subset. To emphasize, UV mutational signature enrichment is proportionally linked to the immunogenicity of tumor neoantigens and is predictive of better immunotherapy response and survival.^41,58 Our findings are comparable to previously observed clinical outcomes in metastatic melanoma treated with immune checkpoint inhibitors.^18,66

In summary, our study confirms the diagnostic value of mutational signatures, in particular the presence of a UV signature, to discriminate metastatic de- or undifferentiated melanoma from deep soft tissue undifferentiated sarcomas. The clinical follow-up of our patients who were then treated for metastatic melanoma indicates that immunotherapy is an effective treatment for patients with undifferentiated/de-differentiated melanomas, highlighting the importance of the diagnostic distinction from sarcoma and the likely relevance of a high TMB for predicting treatment response. These findings support the practice of using next-generation sequence analysis of any patient with undifferentiated malignancy, in which metastatic melanoma is considered in the differential diagnosis.

Data Availability Statement:

Data generated or analyzed during this study are included in this published article. Additional datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.