1 Introduction
Immune checkpoint inhibitors (ICIs) have revolutionized melanoma treatment. According to the EADO/EORTC/EDF and ESMO guidelines, patients with inoperable metastatic melanoma should receive first-line ICI therapy regardless of the mutation status because of prolonged long-term disease control with ICIs over targeted therapies [
11,
12]. However, half of ICI-treated patients with melanoma do not benefit from treatment, underscoring the urgent need for reliable biomarkers.
Tumor mutational burden [
8,
10,
30] and tumor neoantigen burden [
25] are widely discussed biomarkers. However, analysis in a clinical routine is challenging, as protocols differ among papers, cut-off values are objects of discussion, and costs are rather high. Tumor tissue methylation is a well-established diagnostic biomarker for brain tumors and sarcomas [
2,
22,
24]. Methylation analysis, however, including the required laboratory set-up, can be easily established, and thereby represents a promising method for biomarker evaluation in everyday clinical use. Alterations in DNA methylation are a key feature of tumorigenesis that result in dysregulated transcriptional gene activity. However, epigenetic changes in immune cells have also been reported to contribute to response and resistance to ICI-based therapies [
6]. Indeed, tumor tissue methylation was previously shown to serve as a potential biomarker for ICI response in patients with sarcoma, head and neck cancer, lung cancer, and metastatic melanoma [
7,
29,
39,
40]. Here, we report that DNA methylation profiling of treatment-naïve tumor samples independent of the tumor origin from patients with metastatic melanoma could be an additional useful tool for predicting response to ICIs and emphasizes that tissue-based DNA methylation profiling should be incorporated in future biomarker research studies.
2 Methods
2.1 Patients’ and Clinical Characteristics
A total of 71 patients (27 [38.0%] female and 44 [62.0%] male) with a median age of 65 years (range: 32–84 years) were retrospectively identified from the patient records of the Department of Dermatology, Medical University of Vienna, from 2015 to 2022. The inclusion criteria were as follows: (i) age ≥18 years; (ii) stage IVM1a–IVM1d metastatic melanoma according to the American Joint Committee on Cancer, 8th Edition [
20]; (iii) sufficient formalin-fixed paraffin-embedded tumor tissue material prior to ICI initiation available; (iv) first-line therapy with ICIs including monotherapy with a programmed cell death protein 1 (PD-1) inhibitor (nivolumab, pembrolizumab), monotherapy with a cytotoxic T-lymphocyte associated protein 4 inhibitor (ipilimumab), or combination therapy of a PD-1 inhibitor (nivolumab) and cytotoxic T-lymphocyte associated protein 4 inhibitor (ipilimumab) and an ICI-based clinical trial study medication; (v) administration of a minimum of one cycle of ICI; (vi) radiologically measurable disease (lesion size ≥10×10 mm) prior to ICI therapy start; and (vii) availability of at least one radiological restaging under ICI therapy or the occurrence of death before the first performed radiological restaging. Patient and clinical characteristics, including age, sex, tumor mutation status (
BRAF,
NRAS,
cKIT, or wild type), Eastern Cooperative Oncology Group (ECOG) performance status, baseline S100 and lactate dehydrogenase, and survival data (overall survival [OS], progression-free survival [PFS]) were obtained from patient records. Radiological assessment of response was defined according to the immune-based Response Evaluation Criteria in Solid Tumors (iRECIST) [
38] and was performed by an independent radiologist blinded to patient outcomes. Objective response was determined by immune complete response or immune partial response, and disease control by immune complete response, immune partial response, or immune stable disease as best achieved response. Disease progression was defined as iPD and further subclassified as immune unconfirmed progressive disease and immune confirmed progressive disease. The date of immune unconfirmed progressive disease was classified as the timepoint of progression if no subsequent restaging was followed by a clear clinical progression or the occurrence of death. The date of the first radiologically confirmed progressive disease was defined as the immune confirmed progressive disease [
38]. Patients with immune complete response and immune partial response were classified as responders, and patients with immune stable disease and iPD (immune unconfirmed progressive disease or immune confirmed progressive disease) were classified as non-responders. Patients who died due to early progression and received only one cycle of ICIs before the first restaging were classified as non-responders.
2.2 Genome-Wide DNA Methylation Analysis
Tumor tissue samples from patients with melanoma before to the first cycle of ICI therapy were collected retrospectively. A total of 81 tumor tissue samples from 71 patients were included in the final methylation analysis. These were categorized into four groups: 28 lymph node metastases (34.6%), 20 organ metastases (24.7%), 24 cutaneous metastases (29.6%), and nine primary tumors (11.1%) (Table
1). Of note, in a subset of patients, 8/71 (11.3%) had more than one treatment-naïve tumor sample from different organ sites available. In detail, one patient had a lymph node and an organ metastatic lesion (12.5%), two patients had a lymph node metastatic lesion and a primary tumor tissue (25%), three patients had one lymph node and one cutaneous metastatic lesion (37.5%), one patient had a cutaneous metastatic lesion and a primary tumor tissue (12.5%), and one patient had a lymph node metastatic lesion, two cutaneous metastatic lesions, and a primary tumor tissue (12.5%) [Table
S1 of the Electronic Supplementary Material (ESM)]. The 5×10-μm slides were obtained from formalin-fixed paraffin-embedded tumor tissue after macrodissection with ≥70% tumor cells, as evaluated by an independent certified dermatopathologist blinded to the patients’ clinical outcomes. Genomic DNA was extracted and treated with sodium bisulfite following standard procedures, and genome-wide DNA methylation analyses were performed using Infinium MethylationEPIC BeadChip microarrays (Illumina, San Diego, CA, USA) as previously described [
39]. Raw Microarray Data (.idat files) were loaded into R (version R 4.0.4, R Foundation for Statistical Computing, Vienna, Austria) using the RnBeads package (.idat files). Single nucleotide polymorphism-associated, non-specific/cross-hybridizing, and sex chromosome-specific probes were excluded from further analysis. The SWAN algorithm was applied for data normalization [
26]. Calculation of differential methylation between groups was conducted using RnBeads implementation of the limma package as well as computation of a combined rank score, which depends on the difference in mean methylation levels of two groups, the mean methylation quotient, and statistical significance. For subsequent analyses, the top 500 differentially methylated CpG sites (DMPs) were selected. To evaluate the robustness of the results, we have used a validation cohort [
29]. Analyses of gene ontology (GO) were conducted using the geometh function from the missMethyl package [
33] and heatmaps were generated using ClustVis [
27].
Table 1
Patients’ clinical and tumor characteristics
Sex (n; %) | |
Female | 27 (38.0%) |
Male | 44 (62.0%) |
Age (years) | |
Median | 65 (range: 32–84 years) |
Metastatic melanoma, stage IV according to AJCC 8th edition (n; %) | |
Stage IVM1a–IVM1d | 71 (100%) |
Mutation status (n; %) | |
BRAF | 25 (35.2%) |
NRAS | 15 (21.1%) |
cKIT | 3 (4.2%) |
wt | 28 (39.5%) |
Baseline ECOG (n; %) | |
0 | 64 (90.1%) |
1 | 7 (9.9%) |
Baseline LDH (U/L) | |
Normal | 47 (66.2%) |
>ULN | 20 (28.2%) |
NA | 4 (5.6%) |
Baseline S100 (μg/L) | |
Normal | 36 (50.7%) |
> ULN | 32 (45.1%) |
NA | 3 (4.2%) |
Origin of treatment naïve tumor tissue for methylation profiling (n; %) | |
Lymph node metastases | 28 (34.6%) |
Organ metastases | 20 (24.7%) |
Cutaneous metastases | 24 (29.6%) |
Primary tumors | 9 (11.1%) |
First-line IO (n; %) | |
Nivolumab monotherapy | 24 (33.8%) |
Pembrolizumab monotherapy | 12 (16.9%) |
Ipilimumab monotherapy | 1 (1.4%) |
Nivolumab and ipilimumab | 27 (38.0%) |
ICI-based study medication | 7 (9.8%) |
Number of cycles applied (cycles) | |
Median | 12 (range: 1–168 cycles) |
Treatment duration (months) | |
Median | 11 (range: 0–80 months) |
Concomitant therapy (n; %) | |
No | 44 (62.0%) |
Yes | 27 (38.0%) |
Surgery | 9 (12.7%) |
Radiation therapy | 10 (14.1%) |
Gamma knife surgery | 8 (11.3%) |
PFS and OS (months) (n; %) | |
Median PFS | 8.5 (range 0–104.1 months) |
Median OS | 30.6 (range: 0–104.1 months) |
iBOR according to iRECIST (n; %) | |
iCR | 14 (19.7%) |
iPR | 15 (21.1%) |
iSD | 17 (23.9%) |
iPD (iUPD or iCPD) | 25 (35.2%) |
Response (n; %) | |
Responder | 29 (40.8%) |
Non-responder | 42 (59.2%) |
Systemic FU-therapy (n; %) | |
Nivolumab | 13 (18.3%) |
Pembrolizumab | 2 (2.8%) |
Ipilimumab | 9 (12.7%) |
Nivolumab and ipilimumab | 12 (16.9%) |
Nivolumab and relatlimab | 2 (2.8%) |
Pembrolizumab and lenvatinib | 1 (1.4%) |
Dabrafenib and trametinib | 7 (9.9%) |
Zelboraf and cobimetinib | 1 (1.4%) |
Encorafenib and Binimetinib | 7 (9.9%) |
Trametinib | 1 (1.4%) |
Dacarbazine | 1 (1.4%) |
Temozolomide | 1 (1.4%) |
Imatinib | 1 (1.4%) |
Duration of FU (months) | |
Median | 17.3 (range: 0–56.6 months) |
2.3 Statistics
Follow-up was calculated from the start date of ICI therapy until last contact or death, whatever occurred first. Descriptive statistics were used to describe patient characteristics, such as age, sex, and mutation status. For nominally scaled variables, absolute numbers and percentages were presented, and for metric variables, the mean, standard deviation, median, minimum, and maximum were applied. The Kaplan–Meier method was used to estimate PFS and OS from the first ICI application until the occurrence of disease progression or death. Patients without disease progression were censored on the date of their last visit. Kaplan–Meier curves were compared using the log-rank test and Cox’ proportional hazards models. To test the sensitivity-specificity trade-off, receiver operating characteristic classifier curve analyses were performed. The Kruskal–Wallis test and Mann–Whitney U test were used for comparisons between the groups. Statistical significance was determined by a p-value <0.05 (two-sided), multiple testing was corrected using the Bonferroni method. Statistical analyses were performed with Statistical Package for the Social Sciences (SPSS®) 23.0 software (SPSS Inc., Chicago, IL, USA) and standard R functions (version R 4.0.4; R Foundation for Statistical Computing, Vienna, Austria). The data cut-off date was 30 August, 2022.
4 Discussion
This study demonstrates that DNA methylation signatures of treatment-naïve tumor samples, regardless of the tumor tissue origin, can predict responses to ICIs in patients with stage IV metastatic melanoma, and is warranted for further analysis in prospective cohorts. It has been shown that epigenetic changes can contribute to response and failure to ICIs and DNA methylation profiling has been proposed to be a useful tool to predict responses to ICIs in different solid tumor entities [
39,
40,
42]. There is still a lack of data regarding DNA methylation profiling and responses to ICIs in the field of melanoma. Newell et al. and Filipinski et al. reported that DNA methylation might be useful for predicting the response to ICIs in metastatic melanoma [
7,
29]. In contrast, these studies did not evaluate whether the tumor tissue origin affected the DNA methylation signature and prediction to ICIs, nor did they use exclusively treatment-naïve tumor tissues for analysis. We have utilized the study of Newell et al. as a validation cohort, because of the biggest similarity to our cohort, although differences certainly limited the usage as a validation dataset. The Newell et al. cohort differed in the radiological response assessment, using RECIST and not iRECIST criteria, and responders were classified as complete response, partial response, and stable disease (>6 months) and non-responders as stable disease (<6 months) and progressive disease. Furthermore, no detailed information on the origin of tumor tissue used for the methylation analysis was available. The whole study cohort’s distribution of tumor tissues (
n = 71) mainly consisted of subcutaneous tissue (56%), lymph nodes (30%), brain metastases (9%), and primary tumors (4%). Sixty-six percent were treatment-naïve tumor tissue samples and 34% had an intervening systemic therapy before the start of ICI treatment. Nevertheless, we observed similar trends with hypomethylation being present mainly in responders. However, the heterogenous distribution of responders and non-responders observed in these two cohorts was not unexpected because of the above-mentioned variety in the cohorts. Therefore, tumor tissue characteristics alone might be insufficient to reliably predictive ICI responses and incorporating a multimodal concept including systemic inflammatory responses in combination with tissue-based parameters might be needed for a reproducible response prediction to ICIs in metastatic melanoma.
Our study using exclusively treatment-naïve tissue samples adds to the current knowledge that there is no tissue-dependent influence on DNA methylation profiling or prediction of responses to ICIs. It has been reported that location-dependent, global DNA methylation patterns are shared between mucosal and cutaneous melanoma from primary sites, whereas uveal melanoma exhibits a different methylation signature [
19]. Therefore, and because of the limited response rates of <10% to ICIs compared with cutaneous melanoma, we excluded uveal and mucosal melanomas from our analysis [
1,
4,
36,
49]. Consequently, this further reduced potential batch effects when evaluating the DNA methylation signatures of different tissue sites, including primary tumors and affected body sites, by metastatic diseases of organs (including brain metastases), lymph nodes, and skin. To our knowledge, no such comparison regarding the DNA methylation signature and response to ICIs have been performed, and our results indicate that tissue samples from different sites may likewise predict the response to ICIs to a similar degree.
In addition, our pathway analysis provides further insights into the potentially differentially regulated pathways between responders and non-responders. Enrichments were detected in olfactory and G-protein coupled receptor signaling pathways, which were recently associated with migration and proliferation processes in melanoma [
14,
34], as well as enrichments in the humoral response, which were shown to be supportive for ICI responses in metastatic melanoma [
17].
Furthermore, our data imply that DNA methylation profiling has a higher sensitivity and specificity than established clinical markers such as LDH [
47] and ECOG performance status [
44]. An LDH ≥1 upper limit of normal is a well-established negative prognostic marker for reduced PFS and OS but has a low predictive value for a beneficial treatment outcome [
47]. The ECOG performance status was initially established and used as a predictor of response and toxicity in patients receiving chemotherapy [
31]. Regarding ICIs, a baseline ECOG performance status of zero has been associated with disease control [
44], whereas other studies did not find a correlation between an ECOG score of zero and ICI-based survival benefit [
48]. This confirms our findings that non-responders compared to responders showed a trend towards a higher ECOG status ≥1 (
p = 0.0887), albeit not significant. However, we did not detect differences between non-responders and responders in baseline LDH (
p = 0.6646) and S100 levels (
p = 0.7500). We further observed that hypomethylation was associated with a beneficial response to ICIs. It was shown that global hypomethylation was associated with the expression of programmed death-ligand 1 (PD-L1) on melanoma cells in vitro [
6,
15]. This is important because immunosuppressive receptors such as PD-L1 and PD-1 can inhibit the function of effector T cells [
18]. However, the role of PD-L1 expression and response to ICIs in melanoma is discussed controversially and could not predict a positive clinical outcome [
3,
28], Robert et al. 2015; [
46], which led to the approval of PD-1 inhibitors for melanoma in metastatic [
23,
45] and adjuvant [
43] settings by the US Food and Drug Administration irrespective of the PD-L1 expression status of the tumor tissue.
Several limitations must be considered in the current analysis, such as the heterogeneity of the patient cohort in terms of patient characteristics and applied treatments. In comparison with other studies, we could demonstrate that tumor tissues of different origins can be used for DNA methylation profiling, which will potentially contribute to the clinical routine as a useful and easily applicable biomarker in the near future. Tissue-based DNA methylation profiling together with for example, systemic inflammatory signatures, should be addressed by future studies.
Declarations
Conflicts of Interest
Julia Maria Ressler received honoraria for lectures from Bristol-Myers Squibb, Roche, Amgen, and Novartis and travel support from Sanofi, Roche, and Bristol-Myers Squibb. Angelika Martina Starzer received honoraria for lectures from AstraZeneca and travel and congress support from PharmaMar, MSD, and Lilly. Christoph Höller reports speaker honoraria from Amgen, BMS, MSD, Novartis, and Roche, participation in advisory boards for Amgen, Astra Zeneca, BMS, Incyte, MSD, Novartis, Pierre Fabre, and Roche, and research funding by Amgen to the institution. Matthias Preusser has received honoraria for lectures, consultation, and advisory board participation from the following for-profit companies: Bayer, Bristol-Myers Squibb, Novartis, Gerson Lehrman Group (GLG), CMC Contrast, GlaxoSmithKline, Mundipharma, Roche, BMJ Journals, MedMedia, Astra Zeneca, AbbVie, Lilly, Medahead, Daiichi Sankyo, Sanofi, Merck Sharp & Dome, Tocagen, Adastra, Gan & Lee Pharmaceuticals, and Servier. Anna Sophie Berghoff has research support from Daiichi Sankyo and Roche, honoraria for lectures, consultation, and advisory board participation from Roche Bristol-Meyers Squibb, Merck, Daiichi Sankyo, AstraZeneca, and CeCaVa, and travel support from Roche, Amgen, and AbbVie. Philipp Tschandl received honoraria from Silverchair, unrestricted grants for education projects from Lilly, and honoraria for lectures from AbbVie, Lilly, FotoFinder, and Novartis. Erwin Tomasich, Teresa Hatziioannou, Helmut Ringl, Gerwin Heller, Rita Silmbrod, Lynn Gottmann, and Nina Zila have no conflicts of interest that are directly relevant to the content of this article.