Treatment for metastatic melanoma includes targeted and/or immunotherapy. Although many patients respond, only a subset has complete response. As late-stage patients often have multiple tumors in difficult access sites, non-invasive techniques are necessary for the development of predictive/prognostic biomarkers. PET/CT scans from 52 patients with stage III/IV melanoma were assessed and CT image parameters were evaluated as prognostic biomarkers. Analysis indicated patients with high standard deviation or high mean of positive pixels (MPP) had worse progression-free survival (P = 0.00047 and P = 0.0014, respectively) and worse overall survival (P = 0.0223 and P = 0.0465, respectively). Whole-exome sequencing showed high MPP was associated with BRAF mutation status (P = 0.0389). RNA-sequencing indicated patients with immune “cold” signatures had worse survival, which was associated with CT biomarker, MPP4 (P = 0.0284). Multiplex immunofluorescence confirmed a correlation between CD8 expression and image biomarkers (P = 0.0028).

Implications:

CT parameters have the potential to be cost-effective biomarkers of survival in melanoma, and reflect the tumor immune-microenvironment.

The worldwide incidence of cutaneous melanoma has steadily increased and represents a significant health problem (1). For patients with resected stage III/IV disease, standard treatment includes systemic therapies. However, only a proportion of patients experience long-term survival (2). Not all patients have equal risk of relapse and biomarkers are needed to more precisely deliver care.

An effective immune response is necessary for long-term survival in melanoma (3). Tumors with favorable immune responses have CD8+ cell infiltrates (4) and distinct gene expression signatures consistent with an immune “hot” microenvironment (5). Patients with immune hot tumors have longer overall survival (OS) with immunotherapies (6).

For patients with metastatic melanoma, PET/CT imaging is routinely performed as standard-of-care (7). This provides key clinical information and enables radiomics analysis, a non-invasive tumor assessment using statistical parameters derived from medical imaging.

Emerging data suggest that radiomics might provide prognostic biomarkers across several cancer types. In colorectal cancer, multi-parametric PET/CT analysis identified KRAS mutant tumors with hypoxic or proliferative phenotypes (8). Similarly in non–small cell lung cancer, quantitative CT analysis was significantly associated with KRAS mutations (8). A pan-cancer study developed a machine-learning process based on CT images to assess CD8 expression and PD-L1 immunotherapy response to predict tumor immune phenotype and survival (9). In melanoma, a study of 42 patients, undergoing anti-VEGF therapy, showed CT image characteristics were highly accurate in predicting OS (10).

The primary aim of this study was to identify PET/CT image characteristics as prognostic biomarkers of survival in patients with stage III/IV melanoma (n = 52). To understand the underlying molecular mechanisms, we combined whole-exome sequencing data (WES), RNA-sequencing (RNAseq) analysis, and multiplex immunofluorescence.

Study group

We performed a single center study comprising 52 patients with stage III/IV melanoma. Tumor tissue and blood were collected, at surgery. Pre-treatment PET/CT scans were available. Patients were recruited through the Cancer Evolution Biobank (HREC/10/PAH/153, UQ/2011001286) between July 2014 and July 2018. All participants provided written informed consent. Project approval was granted by the Metro South Human Research Ethics Committee (HREC/16/QPAH/671) and by the University of Queensland (UQ/2017000149). The study was conducted in accordance with the Declaration of Helsinki.

Patients were treated at the Princess Alexandra Hospital melanoma unit (Queensland, Australia). Treatment included BRAF inhibitor and/or immunotherapy, detailed in Supplementary Table S1. Clinical follow-up data (HREC/18/QMS/48596) were used to determine OS and progression-free survival (PFS; Supplementary Table S2). OS was the time from surgery until death from disease. PFS was the time from surgery until the first recurrence of disease confirmed through biopsy or conclusive radiological evidence. BRAF p.V600 mutation status was obtained through clinical pathology records. No patients were lost to follow-up.

Tissue was stored in RNA later. DNA/RNA were extracted using the Qiagen AllPrep DNA/RNA mini kit according to the manufacturer's protocol (Qiagen). Tumors were also formalin-fixed paraffin-embedded (FFPE). Hematoxylin and eosin slides were reviewed by an anatomical pathologist (G. Strutton) for tumor content.

Positron emission tomography and computed tomography

PET/CT scans were assessed retrospectively to extract imaging parameters for each tumor. CT texture analysis was performed using a commercially available filtration-histogram method (Feedback PLC; ref. 11). A region of interest (ROI) was drawn around the tumor. It was then measured using several filters (fine to course) described previously as spatial scaling factors (SSF) ranging between 2 and 6 mm (Visual Overview, step 1). Lesions comprising at least 20 pixels (CT pixel size 1.0 mm) were analyzed to ensure sufficient spatial variation in image intensity for extraction of meaningful texture features. For each tumor, a quantitative dataset was compiled, consisting of: mean (M), mean of positive pixels (MPP), standard deviation (SD), entropy (E), kurtosis (K), and skewness (S). All parameters were measured using each SSF.

PET parameters: Metabolic tumor volume (MTV), maximum standard uptake volume (SUVmax), total lesion glycolysis (TLG), and number of pixels were also included in the analysis.

Statistical analysis

Survival analysis was performed using a tree-structured model to stratify patients into low or high subgroups based on a single PET or CT parameter. Recursive binary partitioning determined the association between OS/PFS and covariates (R Foundation for Statistical Computing) and identified the optimal cutoff to stratify patients into survival groups (Supplementary Table S3). Log-rank Kaplan–Meier analysis assessed the predictive/prognostic value of each covariate using the cutoff values (GraphPad Prism 8.3.1). The Benjamini–Hochberg (BH) procedure corrected for multiple testing. The significance level for all analysis was 0.05.

Whole-exome sequencing

WES on matching tumor/buffy coat samples was available from a previous study for 33 patients (12). Data are available through the European Genome–phenome Archive, study ID EGAS00001004619 (dataset ID: EGAD00001006375; Supplementary Table S2). For 32 patients, the matched tumor deposit was also assessed by PET/CT.

WES was performed on the Illumina Hiseq4000 to a depth of ×500 in the tumor and ×100 in the buffy coat (n = 29). Tumor/buffy coat pairs from an additional 4 patients were sequenced on the Illumina NextSeq. Tumor mutation burden (TMB) was reported as the number of mutations per megabase (Mut/Mb) in the coding region. Detailed methods for mutation detection and calling are described in Aoude and colleagues (12). QIMR Berghofer Human Research Ethics Committee granted approval for genomic analysis (QIMR Berghofer HREC P3577).

SNP array

For samples sequenced on the HiSeq4000, SNP arrays (2.5M Illumina) and the qpure bioinformatics tool were used to assess the tumor content (13). For samples sequenced on the NextSeq500, cellularity was assessed using the mean allele fraction. All samples contained >20% tumor content.

RNA sequencing

RNAseq was performed on tumor RNA with a RIN score >5 and tumor cellularity >20% (n = 21; Supplementary Table S2). Libraries were generated using the TruSeq Stranded mRNA kit and sequenced with 100bp paired end reads. Reads were aligned to GRCh37 using STAR (version 2.5.2a; ref. 14) and Cutadapt (version 1.11). Quality control metrics were computed using RNA-SeQC (version 1.1.8; ref. 15) and gene expression estimated using RSEM (version 1.2.30; ref. 16). Gene read counts were normalized to transcripts-per-million (TPM). RNAseq data are available through the European Genome-phenome Archive study ID EGAS00001004619 (dataset ID: EGAD00001006375).

Published immune-modulator genes were assessed to identify patients with immune hot/cold signatures (17). Unsupervised clustering was performed using the log2 TPM expression values using Euclidian distance and complete-linkage clustering. The heatmap shows the row-wise centered and standardized z-scores.

Multiplex immunofluorescence and histomorphometry

Multiplex immunofluorescence analysis was performed on 31 FFPE tumors (Supplementary Table S2) to assess expression of immune markers: CD4, CD8, PD-L1, and PD-L2. Tumors selected had >50% tumor content.

After deparaffinization, hydration, and antigen retrieval (Dako pH9, 100°C, 20 minutes), sections were treated with peroxidase and blocked. Slides were stained using the Ventana BenchMark Special Stains platform (Roche Diagnostics).

Nuclei were DAPI stained. The antibodies used were: CD4 (#M7310, Dako, 1:800, mouse secondary-Opal 650); CD8 (#M7103, Dako, 1:4000, mouse secondary-Opal 690); PD-L1 (#51296S; Cell Signaling Technology; 1:400, rabbit secondary-Opal 570); PD-L2 (#MAB1224–100, R&D Systems, 1:400, mouse secondary-Opal 520).

Slides were scanned (×20) on a Zeiss AxioScan Z1 and images analyzed and size reduced using Zen3.1 (Zeiss). The Visiopharm Image Analytical System (Version 2017.2.4.3387) quantified the staining. DAPI nuclei were automatically segmented defining the areas. All markers were quantified using the imbedded Visiopharm system workflow. The marker staining levels were reported as a ratio of the marker intensity to the assessed area.

Clinicopathologic analysis

We examined a cohort of 52 patients with melanoma (Supplementary Table S1) to determine whether PET/CT image characteristics were associated with survival. In stage III patients (n = 47), we assessed the largest positive lymph node. In resected stage IV patients (n = 5), we analyzed the largest metastatic deposit. A subset of patients, 19/52 (37%) harbored a BRAF p.V600E mutation. Two patients had an alternative BRAF variant (p.L601E or p.T599I) and were considered wild-type for this study. The median PFS was 9.1 months (range, 1–46 months). The median OS was 27.6 months (range, 3–50 months). Median follow-up for survivors was 32.3 months (range, 12–50 months).

PET/CT variables and survival outcomes

We assessed individual PET (MTV, SUVmax, TLG, pixels) and CT parameters (M, MPP, SD, E, K, S). Each parameter was measured using a range of filters described previously as SSF ranging from 2 to 6 mm (Visual Overview). As an example SD2 denotes SD measured at SSF2 (2 mm).

We performed univariable survival analyses using each PET/CT parameter as a continuous variable, and included clinicopathologic features (BRAF status, ulceration of primary lesion, stage and TMB). SD was significantly associated with PFS and OS (Fig. 1A, cox regression) regardless of the SSF used (SD2–6). MPP was also associated with PFS using all filters, but not OS. Both the SD and MPP parameters were found to have greater prognostic significance than clinicopathologic features in this cohort and passed correction for multiple testing (BH<0.05). From the filter set, MPP at SSF 4 (MPP4) and SD at SSF 3 (SD3) were most significantly associated with survival. Downstream analysis focused on these markers.

Figure 1.

Survival in patients with metastatic melanoma stratified using CT texture analysis. A, Cox regression analysis integrating survival with continuous variables from PET/CT imaging. PET parameters were: Metabolic tumor volume (MTV), maximum standard uptake volume (SUVmax), total lesion glycolysis (TLG), and number of pixels. CT parameters were: Mean (M), mean of positive pixels (MPP), standard deviation (SD), entropy (E), kurtosis (K), and skewness (S). Each CT parameter was measured using spatial scaling factor (SSF 2–6 mm) where SSF denotes the size of the feature highlighted in the scan. Additional biomarkers were analyzed: BRAF status, ulceration of primary tumor, stage, and tumor mutation burden (TMB). B, Kaplan–Meier (log-rank) survival analysis showing MPP4 low/high groups and progression-free survival (PFS), **, P = 0.0014; MPP4 and overall survival (OS), *, P = 0.0465; SD3 and PFS, ***, P = 0.00047; SD3 and OS, *, P = 0.0223. C, Hazard ratio combining clinicopathologic information and radiomics markers. Clinicopathologic variables were compared with CT parameters using multivariate cox-regression analysis. Assessment was performed for both PFS and OS.

Figure 1.

Survival in patients with metastatic melanoma stratified using CT texture analysis. A, Cox regression analysis integrating survival with continuous variables from PET/CT imaging. PET parameters were: Metabolic tumor volume (MTV), maximum standard uptake volume (SUVmax), total lesion glycolysis (TLG), and number of pixels. CT parameters were: Mean (M), mean of positive pixels (MPP), standard deviation (SD), entropy (E), kurtosis (K), and skewness (S). Each CT parameter was measured using spatial scaling factor (SSF 2–6 mm) where SSF denotes the size of the feature highlighted in the scan. Additional biomarkers were analyzed: BRAF status, ulceration of primary tumor, stage, and tumor mutation burden (TMB). B, Kaplan–Meier (log-rank) survival analysis showing MPP4 low/high groups and progression-free survival (PFS), **, P = 0.0014; MPP4 and overall survival (OS), *, P = 0.0465; SD3 and PFS, ***, P = 0.00047; SD3 and OS, *, P = 0.0223. C, Hazard ratio combining clinicopathologic information and radiomics markers. Clinicopathologic variables were compared with CT parameters using multivariate cox-regression analysis. Assessment was performed for both PFS and OS.

Close modal

None of the PET parameters were associated with survival (Fig. 1A).

The patient cohort was divided into low/high groups using the optimal cutoffs for survival (Supplementary Table S3). High MPP4 (>34.3) was significantly associated with poor PFS [hazard ratio, 3.050; 95% confidence interval (CI), 1.572–5.918, P = 0.0014, log-rank (Mantel-Cox) test; Fig. 1B]. MPP was significantly associated with PFS using all filters (MPP2-MPP6), indicating that this was a robust measurement (Supplementary Table S3). A high MPP4 score was also significantly associated with poor OS [hazard ratio, 3.319; 95% CI, 1.201–9.173; P = 0.0465, log-rank (Mantel-Cox) test; Fig. 1B].

A high SD3 (>55) was significantly associated with poor PFS [hazard ratio, 2.876; 95% CI, 1.363–6.070; P = 0.00047, log-rank (Mantel-Cox) test; Fig. 1B]. Importantly, SD remained statistically significant when measured across a range of filters (Supplementary Table S3). In addition, high SD3 was significantly associated with poor OS (P = 0.0223; hazard ratio, 3.086; 95% CI; 1.049–9.075; log rank; Fig. 1B). The survival analysis for MPP4 and SD3 remained significant when only stage III patients were included (Supplementary Fig. S1), indicating this result was not driven by stage.

Multivariate cox-regression analysis assessed clinicopathologic variables and compared them with MPP4 low/high and SD3 low/high as prognostic factors (Fig. 1C). SD3 and MPP4 were significantly associated with PFS, P = 0.007 and P = 0.012, respectively. In addition, MPP4 was associated with OS, P = 0.035. Though a clear trend was seen with stage, the clinicopathologic variables were not statistically significant in this small cohort.

BRAF status is associated with PET/CT markers

WES analysis determined whether tumor genomic features were associated with PET/CT parameters (n = 33 patients). BRAF was the most frequently mutated gene (48%) followed by NF1 (39%) and H/K/NRAS (33%; Fig. 2A) in alignment with other studies (18).

Figure 2.

Genomic analysis. Whole-exome Sequence analysis was performed. A, Alterations (missense, nonsense and splice site mutations) in the most frequently mutated driver genes are represented. Frequencies of each mutation are indicated. Mutations in NRAS, KRAS, and HRAS are combined under N/K/HRAS. Samples are arranged according to their BRAF p.V600E mutation status that is associated with B, MPP4 (P = 0.0389, Fisher's exact) and SD3 (P = 0.0437, Fisher's exact). C, RNAseq Immune hot/cold signatures. Unsupervised clustering of immune modulator gene panel showing MPP4 high is associated with “cold” immune signature and MPP4 low is associated with a “hot” immune signature. Red indicates immune hot. Blue indicates immune cold. D, Unpaired t test showing MPP4 values in immune hot/cold groups, *, P = 0.0284 and SD3 values in immune hot/cold groups, not significant.

Figure 2.

Genomic analysis. Whole-exome Sequence analysis was performed. A, Alterations (missense, nonsense and splice site mutations) in the most frequently mutated driver genes are represented. Frequencies of each mutation are indicated. Mutations in NRAS, KRAS, and HRAS are combined under N/K/HRAS. Samples are arranged according to their BRAF p.V600E mutation status that is associated with B, MPP4 (P = 0.0389, Fisher's exact) and SD3 (P = 0.0437, Fisher's exact). C, RNAseq Immune hot/cold signatures. Unsupervised clustering of immune modulator gene panel showing MPP4 high is associated with “cold” immune signature and MPP4 low is associated with a “hot” immune signature. Red indicates immune hot. Blue indicates immune cold. D, Unpaired t test showing MPP4 values in immune hot/cold groups, *, P = 0.0284 and SD3 values in immune hot/cold groups, not significant.

Close modal

Both MPP4 high (P = 0.0389, Fishers exact) and SD3 high (P = 0.0437, Fishers exact; Fig. 2B) were associated with the presence of a BRAF p.V600 mutation. TMB was not related to either MPP4 or SD3 (Supplementary Fig. S2).

Immune hot/cold signatures are associated with MPP4

RNAseq analysis assessed whether the expression of immune modulator genes was related to MPP or SD (n = 21 patients). Unsupervised clustering resulted in an immune cold and an immune hot cluster (Fig. 2C; ref. 17). The immune hot signature correlated with low MPP4, whereas the immune cold signature related to high MPP4 (P = 0.0284, unpaired t test; Fig. 2D). SD3 showed no association (Fig. 2D).

CD8+ cells correlate with MPP4 and patient survival

To define the association between the immune hot phenotype and MPP4, we performed multiplex immunofluorescence on four immune markers, CD4, CD8, PD-L1, and PDL-2. CD4 is found on the surface of T-helper cells and macrophages. CD8 is expressed by cytotoxic cells (T cells, NK cells, cortical thymocytes, dendritic cells). On the basis of the RNAseq data, CD4 and CD8 were more highly expressed in the immune hot group (CD4 fold-change = 2.9 and CD8 fold-change = 8.9).

PD-L1 (encoded by CD274), expressed by activated T cells, is a target of current immunotherapies. PD-L2 (encoded by PDCD1LG2) is linked to PD-L1, though its involvement is yet to be fully ascertained. RNAseq data showed that CD274 and PDCD1LG2 were more highly expressed in the immune hot group (CD274 fold-change = 4.8 and PDCD1LG2 fold-change = 8.0).

FFPE sections (n = 31) were stained with the immune markers, and intensity was quantified (Supplementary Fig. S3). The number of tumor-infiltrated CD8+ cells was 2.3 times higher in the MPP4 low patients (P = 0.0166, Fig. 3A). PD-L1, PD-L2, and CD4 were not differentially expressed. None of the immune markers were associated with SD3 (Fig. 3B).

Figure 3.

Validation of immune biomarkers by multiplex immunofluorescence. Tissue sections were stained for CD8 (Cy7), PD-L1 (Cy3), CD4 (Cy5), and PD-L2 (FITC). Positive cells were quantified using the intensity of the tumor area selected. A, Two-way ANOVA test comparing intensity/area ratios in MPP4 low/high patients. MPP4 low patients had significantly more CD8+ cells, *, P = 0.0166. PD-L1, PD-L2, and CD4 levels were not significantly different between MPP4 low/high patients. B, Two-way ANOVA test showing no association between immune biomarkers and SD3. C, The χ2 analysis showing high CD8 expression was associated with low MPP4 and low SD3. Conversely, low CD8 expression was associated with high MPP4 and high SD3, **, P = 0.0028, χ2 = 11.742. The median level of CD8 expression, 5%, was used as the cutoff. D, Multiplex immunofluorescence images representing MPP4 low (MelR062) and MPP4 high (MelR166) phenotypes. Visual Overview. Experimental overview and clinical significance. From left to right, Step 1 shows CT texture analysis highlighting the metastatic deposit in the lymph node in blue. Histogram analysis of the region of interest (ROI) has been undertaken using a fine filter at special scaling factor 2 (SSF 2, blue) and course filter at special scaling factor 6 (SSF 6, pink). Step 2 indicates the genomic and transcriptomic analysis, including whole-exome sequencing (WES), RNA sequencing (RNAseq), and immune signatures. Step 3 shows validation of immune markers using multiplex immunofluorescence staining of patient tumor tissue. Step 4 correlates immune markers with patient survival outcomes.

Figure 3.

Validation of immune biomarkers by multiplex immunofluorescence. Tissue sections were stained for CD8 (Cy7), PD-L1 (Cy3), CD4 (Cy5), and PD-L2 (FITC). Positive cells were quantified using the intensity of the tumor area selected. A, Two-way ANOVA test comparing intensity/area ratios in MPP4 low/high patients. MPP4 low patients had significantly more CD8+ cells, *, P = 0.0166. PD-L1, PD-L2, and CD4 levels were not significantly different between MPP4 low/high patients. B, Two-way ANOVA test showing no association between immune biomarkers and SD3. C, The χ2 analysis showing high CD8 expression was associated with low MPP4 and low SD3. Conversely, low CD8 expression was associated with high MPP4 and high SD3, **, P = 0.0028, χ2 = 11.742. The median level of CD8 expression, 5%, was used as the cutoff. D, Multiplex immunofluorescence images representing MPP4 low (MelR062) and MPP4 high (MelR166) phenotypes. Visual Overview. Experimental overview and clinical significance. From left to right, Step 1 shows CT texture analysis highlighting the metastatic deposit in the lymph node in blue. Histogram analysis of the region of interest (ROI) has been undertaken using a fine filter at special scaling factor 2 (SSF 2, blue) and course filter at special scaling factor 6 (SSF 6, pink). Step 2 indicates the genomic and transcriptomic analysis, including whole-exome sequencing (WES), RNA sequencing (RNAseq), and immune signatures. Step 3 shows validation of immune markers using multiplex immunofluorescence staining of patient tumor tissue. Step 4 correlates immune markers with patient survival outcomes.

Close modal

We assessed whether CD8 protein expression was correlated with MPP4 and SD3 by comparing patients with high CD8 expression (>5%) to those with low expression. The cutoff was the median level of CD8 expression across the sample group. Low CD8 expression was associated with a high MPP4/SD3 measurement (P = 0.0028, χ2 = 11.742, Fig. 3C). These findings confirmed the RNAseq results. Fig. 3D shows staining for MPP4 low (MelR062) and MPP4 high (MelR166) tumors.

Patients with melanoma undergo routine PET/CT scans as part of their clinical management. Using a non-invasive tumor assessment, widely available in radiology practices, we found a prognostic signature derived from PET/CT imaging that represented a “hot” tumor immune microenvironment.

We showed that CT parameters, MPP4 and SD3, were associated with patient survival. These biomarkers appeared to have better prognostic value than primary tumor ulceration status, which forms part of the staging system. This should be investigated in a larger cohort and expanded to include different disease stages as sample size is a limitation of this study.

RNAseq analysis indicated that patients with immune hot tumors had low MPP4 values (P = 0.0284). Furthermore, MPP4 low tumors had more CD8+ cells (P = 0.0166). High CD8+ infiltration correlated with better PFS (P = 0.0005). These results are concordant with other pan-cancer analysis using radiomics to quantify CD8+ cells in the tumor (9, 10, 19). Our findings build on previous studies showing that CT parameters correlate with the immune signatures and patient survival.

Identifying a correlation between the radiomics parameters and tumor biology (BRAF status, immune signatures, and CD8 expression) demonstrates the prognostic performance of this technique. However, this needs to be confirmed in validation cohorts.

An important outcome of this study would be to determine whether these biomarkers can be applied to unresectable stage IV patients where tumor tissue is unavailable. Moreover, stage IV patients may have multiple tumor sites with inter- and intratumor heterogeneity further complicating treatment outcomes (20, 21). This technique could allow patients with unresected disease to be treated with a precision medicine approach as radiomics parameters are derived from routine images. Furthermore, if risk of recurrence could be determined at stage III, a subset of patients with low risk of relapse, could avoid the potential toxicities of systemic therapy, which can be irreversible (22, 23).

Imaging is an established clinical tool. The quantitative imaging techniques in this study are simple and readily applied to existing protocols. This highlights the utility of radiomics in the clinical decision-making process. Prognostic or predictive biomarkers would be of significance for melanoma research and clinical practice. Inclusion of novel imaging biomarkers in clinical trials can enable more accurate stratification of patients and identification of sub-groups enriched for response.

The potential of this study is to leverage routine, readily accessible, non-invasive imaging technology to derive insight into the underlying immune landscape of melanoma. Providing a personalized approach to reduce treatment failure, morbidity, and costs associated with the treatment of metastatic melanoma.

J.V. Pearson reports other support from genomiQa Pty Ltd. outside the submitted work. V. Atkinson reports personal fees from BMS, MSD, Merck, Novartis, Nektar, Q Biotics, Pierre fabre, Roche, and other support from Oncosec outside the submitted work. N. Waddell reports grants from National Health and Medical Research Council (NHMRC) and a QIMR Berghofer clinical collaboration award during the conduct of the study, as well as other support from genomiQa pty Ltd. outside the submitted work. K. Miles reports family members are shareholders in Feedback PLC. No disclosures were reported by the other authors.

L.G. Aoude: Conceptualization, data curation, formal analysis, funding acquisition, methodology, writing–original draft, writing–review and editing. B.Z.Y. Wong: Data curation, writing–review and editing, PET/CT image analysis. V.F. Bonazzi: Formal analysis, methodology, writing–original draft, writing–review and editing. S. Brosda: Formal analysis, writing–original draft, writing–review and editing. S.B. Walters: Formal analysis, methodology, writing–review and editing. L.T. Koufariotis: Formal analysis, methodology, writing–review and editing, bioinformatics. M.M. Naeini: Data curation, writing–review and editing, bioinformatics. J.V. Pearson: Writing–review and editing, bioinformatics. H. Oey: Writing–review and editing, bioinformatics. K. Patel: Resources, writing–review and editing. J.J. Bradford: Formal analysis, writing–review and editing. C.J. Bloxham: Formal analysis. V. Atkinson: Writing–review and editing, patient recruitment. P. Law: Writing–review and editing, PET/CT image analysis. G. Strutton: Writing–review and editing, pathology review. G. Bayley: Writing–review and editing, patient recruitment. S. Yang: Writing–review and editing, patient recruitment. B.M. Smithers: Writing–review and editing, patient recruitment. N. Waddell: Supervision, writing–review and editing, bioinformatics. K. Miles: Conceptualization, formal analysis, writing–original draft, writing–review and editing, PET/CT image analysis. A.P. Barbour: Conceptualization, resources, funding acquisition, writing–original draft, writing–review and editing, patient recruitment.

L.G. Aoude is supported by a National Health and Medical Research Council of Australia (NHMRC) Early Career Fellowship (APP1109048). N. Waddell is supported by an NHMRC Senior Research Fellowship (APP1139071). Project funding was provided through the 2017 priority-driven Collaborative Cancer Research Scheme, Cure Cancer Australia with the support of Cancer Australia (1144639) and a QIMR Berghofer clinical collaborative grant. The Cancer Evolution Biobank is supported by the PA Research Foundation (RSS_2020_040). Additional funding was provided by The Gallipoli Medical Research Foundation. The authors wish to acknowledge the work of the Queensland Melanoma Project as well as the Melanoma Unit at the Princess Alexandra Hospital, Brisbane, Australia. Statistical support was provided by QFAB Bioinformatics. We are grateful to the patients who have participated in this project. We would like to acknowledge the support of the Estate of the late Alec Pearman and Di Jameson.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Bray
F
,
Ferlay
J
,
Soerjomataram
I
,
Siegel
RL
,
Torre
LA
,
Jemal
A
. 
Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries
.
CA Cancer J Clin
2018
;
68
:
394
424
.
2.
Ugurel
S
,
Rohmel
J
,
Ascierto
PA
,
Flaherty
KT
,
Grob
JJ
,
Hauschild
A
, et al
Survival of patients with advanced metastatic melanoma: the impact of novel therapies-update 2017
.
Eur J Cancer
2017
;
83
:
247
57
.
3.
Gide
TN
,
Wilmott
JS
,
Scolyer
RA
,
Long
GV
. 
Primary and acquired resistance to immune checkpoint inhibitors in metastatic melanoma
.
Clin Cancer Res
2018
;
24
:
1260
70
.
4.
Massi
D
,
Rulli
E
,
Cossa
M
,
Valeri
B
,
Rodolfo
M
,
Merelli
B
, et al
The density and spatial tissue distribution of CD8(+) and CD163(+) immune cells predict response and outcome in melanoma patients receiving MAPK inhibitors
.
J Immunother Cancer
2019
;
7
:
308
.
5.
Galon
J
,
Bruni
D
. 
Approaches to treat immune hot, altered, and cold tumours with combination immunotherapies
.
Nat Rev Drug Discov
2019
;
18
:
197
218
.
6.
Gajewski
TF
,
Corrales
L
,
Williams
J
,
Horton
B
,
Sivan
A
,
Spranger
S
. 
Cancer immunotherapy targets based on understanding the T-cell–inflamed versus non-t cell-inflamed tumor microenvironment
.
Adv Exp Med Biol
2017
;
1036
:
19
31
.
7.
Schroer-Gunther
MA
,
Wolff
RF
,
Westwood
ME
,
Scheibler
FJ
,
Schurmann
C
,
Baumert
BG
, et al
F-18-fluoro-2-deoxyglucose positron emission tomography (PET) and PET/computed tomography imaging in primary staging of patients with malignant melanoma: a systematic review
.
Syst Rev
2012
;
1
:
62
.
8.
Miles
KA
,
Ganeshan
B
,
Rodriguez-Justo
M
,
Goh
VJ
,
Ziauddin
Z
,
Engledow
A
, et al
Multifunctional imaging signature for V-KI-RAS2 Kirsten rat sarcoma viral oncogene homolog (KRAS) mutations in colorectal cancer
.
J Nucl Med
2014
;
55
:
386
91
.
9.
Sun
R
,
Limkin
EJ
,
Vakalopoulou
M
,
Dercle
L
,
Champiat
S
,
Han
SR
, et al
A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti–PD-1 or anti–PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study
.
Lancet Oncol
2018
;
19
:
1180
91
.
10.
Smith
AD
,
Gray
MR
,
del Campo
SM
,
Shlapak
D
,
Ganeshan
B
,
Zhang
X
, et al
Predicting overall survival in patients with metastatic melanoma on antiangiogenic therapy and RECIST stable disease on initial posttherapy images using CT texture analysis
.
AJR Am J Roentgenol
2015
;
205
:
W283
93
.
11.
Miles
KA
,
Ganeshan
B
,
Hayball
MP
. 
CT texture analysis using the filtration-histogram method: what do the measurements mean?
Cancer Imaging
2013
;
13
:
400
6
.
12.
Aoude
LG
,
Bonazzi
VF
,
Brosda
S
,
Patel
K
,
Koufariotis
LT
,
Oey
H
, et al
Pathogenic germline variants are associated with poor survival in stage III/IV melanoma patients
.
Scientific Reports
2020
;
10
:
17687
.
13.
Song
S
,
Nones
K
,
Miller
D
,
Harliwong
I
,
Kassahn
KS
,
Pinese
M
, et al
qpure:
A tool to estimate tumor cellularity from genome-wide single-nucleotide polymorphism profiles
.
PLoS ONE
2012
;
7
:
e45835
.
14.
Dobin
A
,
Davis
CA
,
Schlesinger
F
,
Drenkow
J
,
Zaleski
C
,
Jha
S
, et al
STAR: ultrafast universal RNA-seq aligner
.
Bioinformatics
2013
;
29
:
15
21
.
15.
DeLuca
DS
,
Levin
JZ
,
Sivachenko
A
,
Fennell
T
,
Nazaire
MD
,
Williams
C
, et al
RNA-SeQC: RNA-seq metrics for quality control and process optimization
.
Bioinformatics
2012
;
28
:
1530
2
.
16.
Li
B
,
Dewey
CN
. 
RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome
.
BMC Bioinformatics
2011
;
12
:
323
.
17.
Thorsson
V
,
Gibbs
DL
,
Brown
SD
,
Wolf
D
,
Bortone
DS
,
Ou Yang
TH
, et al
The immune landscape of cancer
.
Immunity.
2019
;
51
:
411
2
.
18.
Hayward
NK
,
Wilmott
JS
,
Waddell
N
,
Johansson
PA
,
Field
MA
,
Nones
K
, et al
Whole-genome landscapes of major melanoma subtypes
.
Nature
2017
;
545
:
175
80
.
19.
Katsoulakis
E
,
Yu
Y
,
Apte
AP
,
Leeman
JE
,
Katabi
N
,
Morris
L
, et al
Radiomic analysis identifies tumor subtypes associated with distinct molecular and microenvironmental factors in head and neck squamous cell carcinoma
.
Oral Oncol
2020
;
110
:
104877
.
20.
Tangella
LP
,
Clark
ME
,
Gray
ES
. 
Resistance mechanisms to targeted therapy in BRAF-mutant melanoma—a mini review
.
Biochim Biophys Acta Gen Subj
2020
;
1865
:
129736
.
21.
Attrill
GH
,
Ferguson
PM
,
Palendira
U
,
Long
GV
,
Wilmott
JS
,
Scolyer
RA
. 
The tumour immune landscape and its implications in cutaneous melanoma
.
Pigment Cell Melanoma Res
2021
;
34
:
529
49
.
22.
Kennedy
LB
,
Salama
AKS
. 
A review of immune-mediated adverse events in melanoma
.
Oncol Ther
2019
;
7
:
101
20
.
23.
Meirson
T
,
Asher
N
,
Bomze
D
,
Markel
G
. 
Safety of BRAF+MEK inhibitor combinations: severe adverse event evaluation
.
Cancers
2020
;
12
:
1650
.