Distinct genomic features across cytolytic subgroups in skin melanoma

Background Skin melanoma is a highly immunogenic cancer. The intratumoral immune cytolytic activity (CYT) reflects the ability of cytotoxic T and NK cells to eliminate cancer cells, and is associated with improved patient survival. Despite the enthusiastic clinical results seen in advanced-stage metastatic melanoma patients treated with immune checkpoint inhibitors, a subgroup of them will later relapse and develop acquired resistance. We questioned whether CYT associates with different genomic profiles and thus, patient outcome, in skin melanoma. Methods We explored the TCGA-SKCM dataset and stratified patients to distinct subgroups of cytolytic activity. The tumor immune contexture, somatic mutations and recurrent copy number aberrations were calculated using quanTIseq, MutSigCV and GISTIC2. Chromothriptic events were explored using CTLPScanner and cancer neoepitopes were predicted with antigen garnish. Each tumor's immunophenoscore was calculated using Immunophenogram. Mutational signatures and kataegis were explored using SigProfiler and compared to the known single or doublet base substitution signatures from COSMIC. Results Metastatic skin melanomas had significantly higher CYT levels compared to primary tumors. We assessed enrichment for immune-related gene sets within CYT-high tumors, whereas, CYT-low tumors were enriched for non-immune related gene sets. In addition, distinct mutational and neoantigen loads, primarily composed of C > T transitions, along with specific types of copy number aberrations, characterized each cytolytic subgroup. We found a broader pattern of chromothripsis across CYT-low tumors, where chromosomal regions harboring chromothriptic events, contained a higher number of cancer genes. SBS7a/b, SBS5 and SBS1 were the most prevalent mutational signatures across both cytolytic subgroups, but SBS1 differed significantly between them. SBS7a/b was mutually exclusive with SBS5 and SBS1 in both CYT subgroups. CYT-high patients had markedly higher immunophenoscore, suggesting that they should display a clinical benefit upon treatment with immune checkpoint inhibition therapy, compared to CYT-low patients. Conclusions Overall, our data highlight the existence of distinct genomic features across cytolytic subgroups in skin melanoma, which might affect the patients' relapse rate or their acquisition of resistance to immune checkpoint inhibition therapies. Supplementary Information The online version contains supplementary material available at 10.1007/s00262-021-02918-3.


Background
Cutaneous melanoma is a very aggressive and highly immunogenic cancer with extensive genetic and transcriptional diversity [1][2][3][4][5][6][7][8].The tumor usually results after exposure to UV radiation, and has the highest rate of somatic mutations and neoantigens among all cancer types [9].The Cancer Genome Atlas (TGCA) Network recently classi ed skin melanoma into mutant BRAF, mutant RAS, mutant NF1, and Triple-WT genetic subtypes, based on the pattern of the top mutated genes [10].The tumor's microenvironment comprises a very heterogeneous cell population, including broblasts, lymphocytes, macrophages and other immune cells, as well as, adipocytes and cells that form the structural elements of skin blood vessels dashed in the extracellular matrix [11].A high content of tumor-in ltrating lymphocytes (TILs) in the tumor's microenvironment associates with a favorable prognosis and improved overall survival of the patients [12,13].An increased immune cytolytic activity (CYT), de ned by the attempt of the cytotoxic T cells and natural killer (NK) cells to eliminate cancer cells via the secretion of GZMA and PRF1, has also been associated with improved patient survival [14].
Due to the complexity of immune regulation in vivo, combination immunotherapies are expected to provide a better therapeutic bene t for the patients [19].Nevertheless, a subgroup of responders to immunotherapy, later relapse and develop acquired resistance; whereas, others do not respond at all (primary resistance) [33].Therefore, a better understanding of the underlying resistance mechanisms is emergently needed.
Here, we questioned whether CYT associates with different genomic pro les in skin melanoma.We investigated how the immune landscape in these tumors relates to somatic mutations, mutational signatures, copy number aberrations, chromothriptic events, the expression of immune checkpoints or other immune-related markers, and the presence of different types of immune cells within the tumor microenvironment of primary or metastatic tumors.We also predicted how skin melanoma patients of different cytolytic activity respond to immune checkpoint blockade therapy.Overall, we provide enough evidence of the existence of distinct genomic features across cytolytic subgroups of skin melanoma.

Cytolytic Activity Calculation And Downstream Rna-seq Analysis
We calculated each SKCM patient's levels of activity (CYT), as the geometric mean of GZMA and PRF1 [14,35,36].GZMA leads to caspase-independent apoptosis, while PRF1 forms pores in the target tumor cells facilitating the entry of granzymes into them.Gene expression values were presented in Transcripts Per Million (TPM).
We then divided patients into two cohorts, the upper 25th quartile of the cytolytic index (CYT-high) and the lower 25th quartile (CYT-low), each with an identical admixture of histology-stage combinations.All subsequent comparisons were made between CYT-high and CYT-low (metastatic or primary) skin melanomas.P-values were adjusted using the Benjamini-Hochberg (BH) method.
We used gene set variation analysis (GSVA) to re ne alterations in pathway activity [37] and clustered tumors hierarchically, with complete linkage as the distance metric.We compared gene expression between the two cytolytic subsets of melanomas, rst by estimating the mean-variance relationship of the log-counts using voom [38] and then analyzing them with limma [39].We considered differentially expressed genes, those having a BH-adjusted p-value < 0.1.Graphs were plotted with ggplot2.Immunohistochemistry (IHC) data of GZMA and PRF1 protein expression were extracted from the Human Protein Atlas [40][41][42], and further analyzed.The antibodies used in IHC were as follows: rabbit pAb anti-GZMA, HPA054134, 1:200 dilution, Sigma-Aldrich, Atlas Antibodies Cat#HPA054134, RRID:AB_2682395; Antigen retrieval was performed using HIER pH6; and mouse mAb anti-PRF1, CAB002436, 1:10 dilution, Leica Biosystems, Cat#NCL-PERFORIN, RRID:AB_563955; Antigen retrieval was performed using HIER pH6.

Detection Of Somatic Mutations And Copy Number Aberrations (scna)
We calculated the most signi cantly mutated genes (SMG) in each CYT subtype of primary and metastatic skin melanomas, using MutSigCV (v1.3.01)[46,47] with an FDR (q-value) = 0.1 as threshold of signi cance.We further employed GISTIC (v2.0.22) [34,48] to identify genomic regions with signi cant ampli cations or deletions in each sample.A G-score considering the amplitude of the aberration, as well as the frequency of its occurrence across each tumor sample was assigned.We calculated the somatic copy number alterations in each sample by taking the sum of segment mean changes ≥ 0.6 and ≤ -0.4 between somatic and normal samples.Regions with FDR q-values < 0.1 were considered signi cant.Both mutational and copy number analyses were performed using the Broad Institute's GenePattern platform (http://software.broadinstitute.org/cancer/software/genepattern).

Detection Of Chromothriptic Events
Chromothriptic events in CYT-high or low skin melanomas were investigated using CTLPScanner [49].In brief, after downloading segmentation data (level 3) of single nucleotide polymorphism (SNP) arrays from the TCGA-SKCM dataset, we implemented DNA copy number segmentation data, using the circular binary segmentation algorithm [50], and the NCBI38/hg38 genome assembly.For the detection of chomothripsis or chomothripsis-like regions, we used copy number status change ≥ 20 times, log10 of likelihood ratio ≥ 8, minimum segment size = 10 kb, and signal distance between adjacent segments = 0.3.Shattered chromosomomal regions were visualized based on the signal value for genomic gains (≥ 0.15) or losses (≤ -0.15) and further highlighted.Finally, chromothripsis-located genes were annotated using the COSMIC database [51].
Consensus centromeric regions were extracted from http://hgdownload.cse.ucsc.edu/goldenpath/hg38/database/centromeres.txt.gz.Telomeric regions were de ned as 20 kb from each side of the chromosomal ends.Intersection between chromothripsis events and centromeric or telomeric regions was de ned as any overlap between the event coordinates and the region coordinates.
Histograms of copy number segment switches were drawn to identify events of chromothripsis that arise in conjunction with telomere crisis, i.e., a period of genome instability during tumorigenesis when depletion of the telomere reserve generates unstable dicentric chromosomes [52][53][54].The patterns and distribution of copy number state switches indicate the emergence and temporal order of the major genomic rearrangement events.

Detection Of Cancer Neoepitopes And Immunophenoscores
We predicted the cancer neoepitopes using "antigen.garnish",with the following MHC molecules, as previously described [35,55,56] We classi ed the peptides that were predicted to bind MHC with high a nity (IC 50 < 50 nM) or with greatly improved a nity compared to their wild-type counterparts, either as classically (CDNs) or alternatively de ned neoepitopes (ADNs).

Mutational Signatures And Rainfall Plots
We extracted mutational signatures as previously described [58, 59] and compared them against the validated mutational signatures provided by COSMIC (http://cancer.sanger.ac.uk/cosmic/signatures).We further applied cosine similarity against COSMIC's validated signatures to identify the best matches within signatures.We reconstructed the mutational spectrum of CYT-high and CYT-low skin melanomas, using the residual sum of squares (RSS) to measure the e ciency of the reconstruction of the original mutational pro le.

Tumor Heterogeneity
Tumor heterogeneity in SKCM tumors was inferred by clustering variant allele frequencies (VAF).The extent of each skin melanoma's intra-tumor heterogeneity was quantitatively calculated by the width of the VAF distribution.We assigned a mutant-allele tumor heterogeneity (MATH) score to each SKCM sample, as previously described [35].

Patient Survival And Synergistic Target Analysis
The overall survival of the melanoma patients was performed using data extracted from the human skin cutaneous melanoma dataset (TCGA-SKCM).Kaplan-Meier curves using log rank (Mantel Cox) test estimated the patient survival.We examined whether GZMA and PRF1 act synergistically on patient survival, using SynTarget [62].

Immune cytolytic activity in skin cutaneous melanoma
We assessed the intratumoral immune cytolytic activity (CYT) in melanoma patients, measuring the TPM values of their GZMA and PRF1 [14].Table S1 lists the exact TCGA-SKCM dataset samples that were included in our analysis, along with their clinical information.We then strati ed patients by de ning skin melanomas in upper quartile of the cytolytic index, as CYT-high, and those in the lower quartile, as CYT-low.GZMA and PRF1 were tightly co-expressed across both primary and metastatic tumors (Spearman rank correlation, rho ~ 0.9) (Fig. 1a).
We and others have previously shown that CYT is markedly higher in skin melanoma relative to the normal skin [14,36].Here, we found that metastatic melanomas showed signi cantly higher cytolytic activity compared to primary tumors (p = 8.11e-06).Importantly, this was not observed between primary and metastatic tumors falling within the same extreme CYT percentiles, suggesting the existence of two main subgroups of tumors: CYT-high and CYT-low (Fig. 1b).
At the protein level, it was evident that neither toxin was highly expressed in skin tumors.In speci c, GZMA protein expression (nuclear or cytoplasmic/membranous) was medium only in one skin melanoma; whereas, it was low in six and absent in ve out of twelve skin tumors.The corresponding intensity scores were as follows: moderate, 2/12; weak, 7/12; and negative, 3/12.The absence of PRF1 protein was more evident, as it was not detected in any of the 11 skin melanomas (Fig. 1c and Table S2).
As expected, metastatic melanomas had signi cantly higher total number of mutations against primary melanomas (p = 0.00178) (Fig. 1d).However, when we assessed the CYT-high and CYT-low skin tumors separately, the metastatic subgroup did not exhibit any difference in the mutation load (p = 0.772) and this was not correlated with the cytolytic index.Among primary tumors on the other hand, CYT-high melanomas accumulated signi cantly more mutations (p = 0.0186), and the mutational burden was signi cantly correlated with the tumors' cytolytic activity, suggesting a primary association between cytolytic activity and mutation load in these tumors (Fig. 1e-f).
In tune with the mutagenic role of UV radiation in melanoma, the most prevalent mutational signatures that we detected had higher cosine-similarity with COSMIC's signatures 7 (CC > TT dinucleotide mutations at dipyrimidines due to UV exposure) and 11 (strong transcriptional strand-bias for C > T substitutions, resembling treatment with alkylating agents).This preference for signature 7 (RSS < 1e-03, cosine similarity > 0.99) did not differ between the two cytolytic subgroups, whatsoever (Figure S1a-b).They also resembled with signature 30 (C > T, unknown aetiology, observed in a small subset of breast cancers) and to a less degree (cosine similarity, 0.724) to mutational signature 6 (defective mismatch repair system) (Fig. 1g-h).

Kataegis is equally distributed across different cytolytic subgroups of skin melanoma
We hypothesized that mutation showers (or kataegis) are associated with cytolytic-high cutaneous melanomas.Kataegis were recently uncovered by whole-genome sequencing of B cell and non-hematopoietic tumors [58, 60, 63-65].We identi ed 74 kataegic sites across 26 tumors, associated with 1,567 mutations.Five (19.2%) of these tumors were CYT-high metastatic SKCMs and six (23.07%)CYT-low (3 metastatic and 3 primary tumors), suggesting that kataegis is equally distributed across skin tumors of different cytolytic activity.
Importantly, 1,395 of the mutations (89%) were C > T transitions, which is consistent with the notion that kataegis results from DNA replication over cytidine deamination of resected DNA [65,66].Considering the established role of the APOBEC cytidine deaminases in kataegis [67], these data support a partial APOBEC involvement in cutaneous melanomas, irrespective of their immune cytolytic index.
We next investigated whether CYT associates with transcriptional or genomic differences between different cytolytic subgroups in primary and metastatic skin melanomas.
To better evaluate the sources of gene expression across all SKCM tumors, we calculated tumor purity and ploidy, using ABSOLUTE [45].Overall, CYT-low tumors had higher purity compared to the CYT-high ones; whereas, the ploidy values of most tumors clustered around 2.4-2.7 (genome-wide duplication), without differences between the two cytolytic subsets (Figure S4).
Taken together, the above results provide further proof that the tumor microenvironment in CYT-high skin melanomas is more in amed and immunogenic compared to that in CYT-low tumors.
To assess the relationship between CYT markers and immune checkpoint molecules, we run correlation analysis of their gene expression levels.Both cytolytic genes were notably correlated with the expression of at least ve different inhibitory checkpoints in SKCM tumors (p < 0.0001), corroborating that combination therapies with immune checkpoint inhibition should effectively be used to overcome resistance and broaden the clinical bene t for these patients (Fig. 2e).
Cell type fraction analysis revealed that CYT-high (primary and metastatic) tumors are signi cantly enriched in B cells, M1 macrophages and CD8 + T cells; whereas CYT-low tumors contain signi cantly higher levels of monocytes, NK cells and CD4 + T cells (Fig. 2d).Further CIBERSORT analysis revealed that the majority of (primary and metastatic) CYT-high tumors were signi cantly enriched in γδT cells, follicular helper T cells (Tfh) and Tregs, compared to their CYT-low counterparts (Figure S7).These cells were previously shown to participate in cancer immunosurveillance [77], autoimmunity [78] and immunosuppression [79].

Cyt Correlates With Discrete Mutational Events In Melanoma
We next focused on the exome-seq data of the TCGA-SKCM and detected the signi cantly mutated genes in each cytolytic subgroup of skin melanoma.CYT-low primary tumors were signi cantly associated with mutations in BRAF, DMRT3, GSTA5, TP53, CRYBA4, STRA13, PRAMEF12, RNF32, SLC1A6, TEKT2 and NRAS among others; whereas, CYThigh primary tumors with a totally different group of genes, including LCK, GSTSF1L, BMF, CSTL1, DRGX, ENTPD3, CAMK4, GPR151, LTF and others (Fig. 3a).This discrepancy hints that the two cytolytic subsets are correlated with different somatic mutations; but we should also take into consideration the small sample number of CYT-high primary tumors (n = 10) compared to the CYT-low ones (n = 31).
In contrast, the SMGs in both cytolytic subgroups of metastatic melanomas included previously described [80] driver mutations in oncogenes and tumor suppressors (NRAS, BRAF, CDKN2A, TP53 and PTEN), as well as COL4A4.
The fraction of C > T transitions at dipyrimidines was the highest across all skin melanomas.The frequency of speci c substitutions did not differ between CYT-high and -low metastatic tumors.However, across the cohort of primary tumors, CYT-high melanomas had more C > T transitions against CYT-low tumors (p < 0.01); but this could probably be due to the difference in sample number (103 primary vs 368 metastatic skin melanomas) (Fig. 3b).
Likewise, there was no association between CYT and BRAF, NRAS, TP53 or NF1 mutations, indicating that immunerelated alterations within the tumor microenvironment, and therefore immunotherapies, are independent of the tumor's genotype (Figure S8).

Cyt Associates With Different Structural Changes In Melanoma
Skin melanoma is characterized by increased chromosomal instability (CIN) with extensive gains and losses, which associate with poor patient prognosis [81-83].However, whether these chromosomal aberrations correlate with the cytolytic index, is unknown.Therefore, we assessed the somatic copy number aberrations (SCNA) within each cytolytic subtype in primary and metastatic skin melanomas.
Taken together, the above ndings show that speci c types of somatic mutations and copy number aberrations are characteristic for each cytolytic subgroup in primary and metastatic melanoma.
To eliminate the possibility of assessing lower con dence of detecting somatic mutations and SCNAs due to tumor cellularity, we performed ABSOLUTE analysis [45] and found no variance in the calculated cellularity estimates between CYT-high and -low SKCM samples.MATH scores ranged from 14.70-47.17 in primary and 14.59-62.22 in metastatic SKCMs.Neither the total mutation load, nor the total copy number events, correlated with tumor heterogeneity in melanoma.In addition, the MATH scores were similar between the two cytolytic subgroups (in primary SKCM, CYT-high vs. CYT-low, 31.04 ± 11.52 vs 29.86 ± 12.97, p > 0.05; in metastatic SKCM, 34.43 ± 12.02 vs 29.94 ± 13.31, p > 0.05), signifying that the dissimilarities in copy number and mutational load, are not due to a variable intratumor heterogeneity (Fig. 4c).Therefore, we propose that distinctive mutational and structural changes can discriminate skin melanomas of different cytolytic activity.

Overexpression of GZMA and PRF1 synergistically affects patient overall survival
To determine whether CYT-high is a good prognostic indicator, we explored the overall survival of skin melanoma patients having high or low expression levels of GZMA and PRF1, using SynTarget [62].
High expression of each cytolytic gene, individually, was positively correlated with the overall survival in melanoma patients (p = 0.02).The subgroup of non-metastatic melanoma patients having both genes over-expressed showed a signi cant positive effect in survival compared to the remaining patients ("other") (p = 0.0282); whereas, simultaneous low levels of both genes shifted signi cantly (p = 0.0048) towards a negative effect (Fig. 4d).This observation implies that the overexpression of both GZMA and PRF1 genes, can synergistically affect the overall survival of skin melanoma patients.

Chromothriptic Events In Cytolytic Subsets Of Skin Melanoma
Local chromosome shattering (chromothripsis) is a mechanism proposed to cause clustered chromosomal rearrangements, following chromosomal breaks at multiple locations.Chromothripsis has been detected in 2-3% of cancers [84] and involves impaired DNA repair [85].Such genomic rearrangements may drive the development of cancer through the deletion of tumor suppressor genes or an increase in copy number of oncogenes, among other mechanisms.The occurrence of chromothriptic events in skin melanoma, and their effect on genes associated with checkpoint inhibition and immune cytolytic activity is still unclear.
The existence of different chromothriptic patterns has been proposed to be due to different mechanisms, including dysfunction of telomeres and DNA damage in micronuclei [54,[86][87][88][89].
The telomere regions were not affected by chromothripsis in the vast majority of events (93,521 out of 93,535 events, 99.9%) across all skin melanomas.Focusing on the two cytolytic subgroups though, the telomere region affected one CYT-high and seven CYT-low skin tumors out of the 14 events, in total (Fig. 5b).
Chromothriptic events not directly affecting telomere regions can also result from telomere fusions, since the genomic regions included in chromatin bridges can be distant from the telomeres, depending on the structure of the dicentric chromosomes formed in telomere crisis [54,86].In more than one fourth of the tumors (58/211; 27.5%), the centromere was included in the segment that was affected by chromothripsis, including 16 CYT-high and 68 CYT-low tumors (Fig. 5b).
Overall, our ndings support the existence of a broader pattern of chromothripsis across cytolytic low skin melanomas.
Increased neoantigen load associates with high cytolytic levels in primary skin melanomas Cancer neoantigents arise from mutated peptides and favorably drive T-cell recognition of cancer cells [90,91].They are therefore, an attractive immune target because their selective expression on cancer cells can minimize immune tolerance [92].
It was of interest to con rm whether cancer neoantigents associate with high cytolytic levels in skin melanoma.Analyzing mutation data from the Cancer Immunome Database (TCIA) [57], it was evident that primary CYT-high skin tumors contained higher number of mutations and neoantigens compared to their CYT-low counterparts.This was not evident, however, in metastatic CYT-high tumors (Fig. 6a).
To investigate this further, we predicted the missense mutations that could potentially function as T-cell neoepitopes in skin melanomas, using antigen.garnish.Our results con rmed that CYT-high primary (and not metastatic) tumors are signi cantly enriched in (classically and alternatively de ned) neoantigens, but these, were not correlated with a high cytolytic activity (Fig. 6b).The tumors' MATH scores also did not correlate with their neoantigen load.Overall, these data support that a high cytolytic activity is driven by a high mutational/neoantigenic load only in primary skin melanomas.

Prediction of skin melanoma patients' response to immune checkpoint inhibition
Since just a subgroup of skin melanoma patients responds to immune checkpoint inhibition, the need to elucidate the mechanisms of resistance and predict markers, is high.TILs, the expression of PD-1 or PD-L1, the mutational load [93], or clonal neoantigens [91], have all been proposed as markers; however, none of them has been fully validated, yet [94].
Hypothesizing that CYT-high skin melanomas have higher immunophenoscore due to increased immunogenicity, resulting in better prognosis and response to therapy, we analyzed two data sets containing skin melanoma patients treated with anti-CTLA-4 [95] and anti-PD-1 inhibitory molecules [96], and used each patient's IPS to predict their response.
Assessed globally, CYT-high skin melanomas (both primary and metastatic) had signi cantly higher HLA levels compared to the intratumoral mean expression.In contrast, CYT-low melanomas were characterized by downregulation of several immune checkpoints, compared to the intratumoral mean expression.Importantly, we could observe similar patterns across patients treated with anti-CTLA-4 alone or combined with anti-PD-1/PD-L1/2, or anti-PD-1/PD-L1/2 monoclonal antibodies, alone (Fig. 7a-b).In addition, CYT-high SKCMs contained signi cantly higher numbers of cytotoxic cells (CD8 + T cells, γδT cells, NK cells), and lower numbers of MDSC and Treg cells.
As expected, CYT-high SKCM patients had a markedly higher IPS (and consequently, an expected clinical bene t) compared to CYT-low patients who did not receive checkpoint inhibition.Interestingly, the IPS scores were signi cantly higher across all CYT-high tumors (both primary and metastatic), upon treatment with CTLA-4 or PD-1/PD-L1/2 blockers, or a combination of both (p < 0.0001) (Fig. 7c).
Our data indicate that the IPS has a predictive value in CYT-high melanoma patients who received CTLA-4 and PD-1/PD-L1/2 inhibition therapy, and are in accordance with previous observations that patients with higher levels of tumor cytolytic activity, and expression of immune checkpoints, bene ted more from the corresponding immune checkpoint blockade [95].

Discussion
A better understanding of the immunobiology of cancer immunosurveillance and immunoediting in skin melanoma, will con dently lead to the development of more effective immunotherapeutic approaches [97].
Here, we extensively analyzed the gene expression and genomic landscape of skin melanoma in the context of intratumoral immune cytolytic activity.We strati ed primary and metastatic tumors according to a validated cytolytic gene expression signature, and discriminated a subgroup of them with noticeable T-cell reactivity.We show that CYThigh skin melanomas are signi cantly enriched for immune-related gene sets associated with activated CD8 + T cells, B cells, M1 macrophages, activated dendritic cells and NK cells, among others, corroborating the existence of an in amed tumor microenvironment in these patients.In contrast, CYT-low skin melanomas were enriched for nonimmune-related gene sets, primarily associated with monocytes and CD4 + T cells.
Abundant CD8 + T cell in ltrates are well known to exist in in amed metastatic melanoma and drive the upregulation of PD-L1, IDO and Tregs in the tumor microenvironment [98][99][100].Additionally, CD8 + T-cell in ltration associates with a better response of cancer patients to chemotherapy [101], neoadjuvant therapy [102] and anti-PD-1 immunotherapy [103].A key role for tumor-in ltrating B cells in modulating the anti-tumor immune response, was recently proposed in skin melanoma [104].B cells play an important prognostic role and can predict non-response to an immune checkpoint inhibitor in metastatic melanoma [104].The enrichment of CYT-high tumors in M1 macrophages is in line with their suppressive role in cancer progression, while that of M2 macrophages in CYT-low tumors, underscores their role in favoring tumor growth and dissemination [105].Activated dendritic cells are also important in the immune response against cancer cells and can be used as a strong independent prognostic factor [106]; whereas, activated NK cells can also e ciently kill malignant melanoma cells [107].
Taken together, these ndings suggest that the strati cation of skin melanoma patients according to their gene expression pro ling can distinguish them between those having a strong cytolytic T-cell response, and those with a less privileged immune microenvironment.A similar, CYT-based strati cation was previously performed for other types of cancer [14,35,36,108].
We also showed that the mutational burden is higher among metastatic tumors, and that the increased mutational and neoantigen load that elicits an immune response by TILs, associates with high cytolytic levels in primary skin melanomas.In a similar study, a high mutational burden was previously associated with response in melanoma patients treated with immune checkpoint inhibition, but not with overall survival [109].
Although many of these mutations have been previously characterized in skin melanoma (e.g., BRAF, NRAS, TP53, PPP6C, CDKN2A, RAC1, etc.) [10,80], this is the rst report that appreciates them in association with tumors of different cytolytic activity.
We also provide evidence that different genomic structural variations are implicated in the progression of distinct cytolytic subgroups in skin melanoma.We show that CYT-low skin melanomas have a higher number of ampli cations and deletions in their genome.These, include mainly recurrent NOTCH2 ampli cations, and non-silent mutations and/or deletions in PD-L1, PD-L2, CDKN2A/B, PAX5, ETS1, BCL7, RAD51, JAK2, APAF1, FOXO3, CTNNA1 and IGF1, among others.These genomic structural variations have a profound effect on immune activation in these tumors.
Following these observations, we detected marked differences in the chromothriptic pattern between CYT-high and CYT-low tumors.These regarded the prevalence of chromothriptic events, the number of chromothriptic chromosomes per tumor and the cancer genes that they harbor, or the involvement of telomeres and centromeres in the chromothriptic regions.Such differences suggest that distinct mechanisms could give rise to chromothripsis across the two cytolytic subgroups in skin melanoma.
In addition, we highlight several targetable immune-related genes, other than PD-1, PD-L1/2 and CTLA-4, which may coordinately contribute to immunosuppression.IDO1 and LAG3 overexpression, which was evident in CYT-high skin melanomas, provides an alternative immunosuppressive barrier that is needed to hamper the antitumor activity of CTL and NK cells, in these tumors.This is something that was initially proposed in skin melanoma cell lines [110], and later corroborated at the human level, by us and others [14,36].We suggest that combined immune inhibition of these markers along with the PD-1/PD-L1 and/or the CTLA-4 axis, could provide a better therapeutic outcome for these patients [19].
Together with the synergistic effect of GZMA and PRF1 in patient survival, we provide evidence that CYT-high skin melanoma patients who received CTLA-4 or PD1/PD-L1/PD-L2 checkpoint blockade therapy (or a combination of both), have a markedly higher immunophenoscore and consequently, an expected clinical bene t, compared to CYTlow patients who did not receive checkpoint inhibition.

Conclusions
In the we highlight new links between certain expressional or genomic changes and the activation of an antitumor immune response in skin melanoma.Our ndings also corroborate that combination therapies with PD-1, PD-L1/2 and/or CTLA-4 blockade, should be able to overcome resistance and broaden the clinical bene t for these patients.SCNAs compared to primary tumors, but also CYT-low tumors had more SCNAs compared to CYT-high ones.(c) Tumor enriched in cancer genes, as these are registered in COSMIC.In addition, chromothriptic events broadly did not directly affect telomere regions, apart from 7 CYT-high and 1 CYT-low tumor.More CYT-low skin tumors also contained chromothriptic events affecting centromere regions (68 CYT-low vs 16 CYT-high).(c) The majority of chromothriptic events affected more chromosomes in CYT-low tumors.In speci c, chromothriptic events largely affected Figure 6 Increased neoantigen load associates with high cytolytic levels in primary skin melanomas.(a) Both the number of total mutations and neoantigens, was signi cantly higher within CYT-high primary SKCMs, compared to CYT-low tumors.On the other hand, the total mutation load, as well as the neoantigenic load did not differ between the two cytolytic subgroups of metastatic skin melanomas.(b) CYT-high primary (and not metastatic) tumors contain signi cantly more classically (CDNs) and alternatively de ned (ADNs) neoantigens; but these, were not correlated with a high cytolytic activity.

chromosomes 1 ,
5 and 22 among CYT-high tumors, and 1, 3, 5 and 6among CYT-low SKCM tumors.(d) The general view of copy number aberration pro le in a representative SKCM tumor (TCGA_SKCM_955602).Green dots depict genomic gains; whereas, red dots, genomic losses in this tumor sample.(e) Indicative chromothripsis region in chromosome 9 of a skin melanoma (TCGA_SKCM_955602) with chromosomal gains and losses.COSMIC cancer genes are denoted.(f) Histogram of copy number segment switches in sample TCGA_SKCM_955602.The copy number changes oscillate among three or more states, indicating that chromothripsis may be the initiating event in this tumor.