CoChat

Tumour gene expression signature in primary melanoma predicts long-term outcomes

Open access

Published version of a preprint

The DOI matches its Crossref record; 66 of the 69 references it cites were found in a registry or an index.

Abstract

Adjuvant systemic therapies are now routinely used following resection of stage III melanoma, however accurate prognostic information is needed to better stratify patients. We use differential expression analyses of primary tumours from 204 RNA-sequenced melanomas within a large adjuvant trial, identifying a 121 metastasis-associated gene signature. This signature strongly associated with progression-free (HR = 1.63, p = 5.24 × 10 −5 ) and overall survival (HR = 1.61, p = 1.67 × 10 −4 ), was validated in 175 regional lymph nodes metastasis as well as two externally ascertained datasets. The machine learning classification models trained using the signature genes performed significantly better in predicting metastases than models trained with clinical covariates ( p AUROC = 7.03 × 10 −4 ), or published prognostic signatures ( p AUROC < 0.05). The signature score negatively correlated with measures of immune cell infiltration (ρ = −0.75, p < 2.2 × 10 −16 ), with a higher score representing reduced lymphocyte infiltration and a higher 5-year risk of death in stage II melanoma. Our expression signature identifies melanoma patients at higher risk of metastases and warrants further evaluation in adjuvant clinical trials.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41467-021-21207-2

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 The Cam_121 gene expression signature is strongly associated with survival in uni- and multivariate Cox regression analyses (AVAST-M primary melanoma cohort; n = 194). Kaplan–Meier survival plots comparing the survival probabilities (y-axes) as a function of time in years (x-axes) of groups with high and low “Cam_121” (quantile 0.33 split) for outcomes a overall survival (OS) and b progression-free survival (PFS). The p value of a two-sided logrank test comparing the survival distributions of both groups are indicated. c Forest plot indicating the hazard ratio (HR) estimates relating to the Cam_121 signature when predicting OS (green) and PFS (orange) by means of Cox proportional hazard models controlling for different (sets of) clinical variables (y-axis). The HR estimates are indicated by the dots at the centre of the error bars; the horizontal error bars correspond to the 95% confidence intervals of the HR. The two-sided Wald t test p values corresponding to the signature “Cam_121” parameter are indicated for each model and outcome. ECOG Eastern Cooperative Oncology Group Performance Status. TIL count Tumour-infiltrating lymphocyte count.
Fig. 1 The Cam_121 gene expression signature is strongly associated with survival in uni- and multivariate Cox regression analyses (AVAST-M primary melanoma cohort; n = 194). Kaplan–Meier survival plots comparing the survival probabilities (y-axes) as a function of time in years (x-axes) of groups with high and low “Cam_121” (quantile 0.33 split) for outcomes a overall survival (OS) and b progression-free survival (PFS). The p value of a two-sided logrank test comparing the survival distributions of both groups are indicated. c Forest plot indicating the hazard ratio (HR) estimates relating to the Cam_121 signature when predicting OS (green) and PFS (orange) by means of Cox proportional hazard models controlling for different (sets of) clinical variables (y-axis). The HR estimates are indicated by the dots at the centre of the error bars; the horizontal error bars correspond to the 95% confidence intervals of the HR. The two-sided Wald t test p values corresponding to the signature “Cam_121” parameter are indicated for each model and outcome. ECOG Eastern Cooperative Oncology Group Performance Status. TIL count Tumour-infiltrating lymphocyte count.CC BY · PubMed Central
Fig. 2 Validation of the Cam_121 in further datasets. We explored the Cam_121 signature in the entirely separate lymph node (LN) samples from the AVAST-M Cohort (a–c; n = 143), as well as in three independently acquired external datasets including the Leeds Melanoma Cohort (d, e; n = 677), TCGA-SKCM LN (f, g; n = 212) and TCGA-SKCM Skin (h, i; n = 156). AVAST-M LN Cohort: Kaplan–Meier (K–M) survival plots comparing the survival probabilities (y-axes) as a function of time in years (x-axes) of groups with high and low “Cam_121” (quantile 0.33 split) for outcomes a overall survival (OS) and b progression-free survival (PFS). The p value of a logrank test comparing the survival distributions of both groups is indicated on each curve. c Forest plot indicating the hazard ratio (HR) estimates (dots at the centre of the error bars) and corresponding 95% confidence intervals (horizontal error bars) related to the Cam_121 signature when predicting OS (green) and PFS (orange) by means of Cox proportional hazard models when controlling for different (sets of) clinical variables (y-axis). The two-sided Wald t test p values corresponding to the signature “Cam_121” parameter are indicated for each model and outcome. ECOG Eastern Cooperative Oncology Group Performance Status. Leeds Melanoma Cohort: d K–M curve comparing the melanoma-specific survival probabilities (y-axis) of groups with high and low “Cam_121” (quantile 0.33 split) through time in years (x-axis) and p value of a two-sided logrank test comparing the survival distributions. e Forest plot showing the HR estimates (dots at the centre of the error bars) and 95% confidence intervals (horizontal error bars) of the HR estimates corresponding to the continuous signature “Cam_121” parameter when predicting melanoma-specific survival by means of different Cox proportional hazard models (y-axis). Multivariate correction was undertaken for sex, age, Breslow thickness, ulceration, mitotic count, as well as BRAF and NRAS mutation status and (in the final row), correction was also undertaken for the tumour-infiltrating lymphocyte (TIL) score. TCGA-SKCM Cohort: f, h K–M curves comparing the overall survival probabilities of groups with high and low “Cam_121” (quantile 0.33 split) and p value of a two-sided logrank test comparing the survival distributions in TCGA-SKCM LN and TCGA-SKCM Skin datasets, respectively. TCGA-SKCM Cohort: g, i Forest plot showing the HR estimates (dots at the centre of the error bars) and 95% confidence intervals (horizontal error bars) corresponding to the of related to the continuous signature “Cam_121” parameter when controlling for different (sets of) clinical variables (y-axes) in the TCGA-SKCM LN and TCGA-SKCM Skin datasets, respectively. Multivariate correction was undertaken for sex, age, stage and Breslow thickness. The p values of two-sided Wald t tests corresponding to the signature “Cam_121” parameter are indicated for each model and outcome (no multiplicity correction used).
Fig. 2 Validation of the Cam_121 in further datasets. We explored the Cam_121 signature in the entirely separate lymph node (LN) samples from the AVAST-M Cohort (a–c; n = 143), as well as in three independently acquired external datasets including the Leeds Melanoma Cohort (d, e; n = 677), TCGA-SKCM LN (f, g; n = 212) and TCGA-SKCM Skin (h, i; n = 156). AVAST-M LN Cohort: Kaplan–Meier (K–M) survival plots comparing the survival probabilities (y-axes) as a function of time in years (x-axes) of groups with high and low “Cam_121” (quantile 0.33 split) for outcomes a overall survival (OS) and b progression-free survival (PFS). The p value of a logrank test comparing the survival distributions of both groups is indicated on each curve. c Forest plot indicating the hazard ratio (HR) estimates (dots at the centre of the error bars) and corresponding 95% confidence intervals (horizontal error bars) related to the Cam_121 signature when predicting OS (green) and PFS (orange) by means of Cox proportional hazard models when controlling for different (sets of) clinical variables (y-axis). The two-sided Wald t test p values corresponding to the signature “Cam_121” parameter are indicated for each model and outcome. ECOG Eastern Cooperative Oncology Group Performance Status. Leeds Melanoma Cohort: d K–M curve comparing the melanoma-specific survival probabilities (y-axis) of groups with high and low “Cam_121” (quantile 0.33 split) through time in years (x-axis) and p value of a two-sided logrank test comparing the survival distributions. e Forest plot showing the HR estimates (dots at the centre of the error bars) and 95% confidence intervals (horizontal error bars) of the HR estimates corresponding to the continuous signature “Cam_121” parameter when predicting melanoma-specific survival by means of different Cox proportional hazard models (y-axis). Multivariate correction was undertaken for sex, age, Breslow thickness, ulceration, mitotic count, as well as BRAF and NRAS mutation status and (in the final row), correction was also undertaken for the tumour-infiltrating lymphocyte (TIL) score. TCGA-SKCM Cohort: f, h K–M curves comparing the overall survival probabilities of groups with high and low “Cam_121” (quantile 0.33 split) and p value of a two-sided logrank test comparing the survival distributions in TCGA-SKCM LN and TCGA-SKCM Skin datasets, respectively. TCGA-SKCM Cohort: g, i Forest plot showing the HR estimates (dots at the centre of the error bars) and 95% confidence intervals (horizontal error bars) corresponding to the of related to the continuous signature “Cam_121” parameter when controlling for different (sets of) clinical variables (y-axes) in the TCGA-SKCM LN and TCGA-SKCM Skin datasets, respectively. Multivariate correction was undertaken for sex, age, stage and Breslow thickness. The p values of two-sided Wald t tests corresponding to the signature “Cam_121” parameter are indicated for each model and outcome (no multiplicity correction used).CC BY · PubMed Central
Fig. 3 The Cam_121 gene expression signature is predictive of metastases across multiple machine learning classification models and in an internal validation dataset (n = 143). a Plot showing the mean ± standard deviation of area under the ROC curve (AUROC) predicted by different classification models when trained using tenfold cross validation (CV; repeats = 1000) on the AVAST-M primary melanoma dataset (n = 194). The features used for training each classification model are indicated on the top grey panel. Within each panel, seven different machine learning classifiers were trained to predict metastases. Statistical comparison using one-sided two-sample Welch t tests are indicated (see also Supplementary Table S2a). b Area under the ROC curve plots, showing, for the best performing classification model selected in each panel of a, its performance on an entirely separate AVAST-M lymph node validation dataset (n = 143). The one-sided DeLong’s test p value is reported for each comparison (see also Supplementary Table S2b). c, d Venn diagrams comparing the number of correctly predicted relapse outcomes (yes/no) of 143 patients from the models described in b. c Venn diagram showing the number of correctly predicted relapse outcomes specific to or common between “Cam_121 + clinical covariates” (blue) vs “Clinical covariates” (pink). d Venn diagram showing the number of correctly predicted relapse outcomes specific to or common between “Cam_121”(blue) and “clinical covariates” alone (pink). Out of a total of 143 patients, 30 were wrongly predicted by both the models in c and 36 were wrongly predicted by both the models in d. e Plot showing the performance of different classification models in predicting metastases in terms of AUROC (mean ± standard deviation) when trained on the AVAST-M lymph node dataset (n = 143). Within each panel, 14 different machine learning classifiers were trained to predict metastases: seven using tenfold CV (repeats = 1000) and seven using bootstrap resampling method (repeats = 1000). The two horizontal lines indicated within each panel denote the median AUROC of these seven classifiers, respectively. Statistical comparison using one-sided two-sample Welch t tests are indicated (bootstrap in green and tenfold CV in orange), see also Supplementary Tables S2c-d. Gerami_27, LMC_150: Leeds Melanoma Cohort 150 gene signature. See “Methods” section “Machine learning analysis“ for details about the classification algorithms. Source data are provided as a Source data file.
Fig. 3 The Cam_121 gene expression signature is predictive of metastases across multiple machine learning classification models and in an internal validation dataset (n = 143). a Plot showing the mean ± standard deviation of area under the ROC curve (AUROC) predicted by different classification models when trained using tenfold cross validation (CV; repeats = 1000) on the AVAST-M primary melanoma dataset (n = 194). The features used for training each classification model are indicated on the top grey panel. Within each panel, seven different machine learning classifiers were trained to predict metastases. Statistical comparison using one-sided two-sample Welch t tests are indicated (see also Supplementary Table S2a). b Area under the ROC curve plots, showing, for the best performing classification model selected in each panel of a, its performance on an entirely separate AVAST-M lymph node validation dataset (n = 143). The one-sided DeLong’s test p value is reported for each comparison (see also Supplementary Table S2b). c, d Venn diagrams comparing the number of correctly predicted relapse outcomes (yes/no) of 143 patients from the models described in b. c Venn diagram showing the number of correctly predicted relapse outcomes specific to or common between “Cam_121 + clinical covariates” (blue) vs “Clinical covariates” (pink). d Venn diagram showing the number of correctly predicted relapse outcomes specific to or common between “Cam_121”(blue) and “clinical covariates” alone (pink). Out of a total of 143 patients, 30 were wrongly predicted by both the models in c and 36 were wrongly predicted by both the models in d. e Plot showing the performance of different classification models in predicting metastases in terms of AUROC (mean ± standard deviation) when trained on the AVAST-M lymph node dataset (n = 143). Within each panel, 14 different machine learning classifiers were trained to predict metastases: seven using tenfold CV (repeats = 1000) and seven using bootstrap resampling method (repeats = 1000). The two horizontal lines indicated within each panel denote the median AUROC of these seven classifiers, respectively. Statistical comparison using one-sided two-sample Welch t tests are indicated (bootstrap in green and tenfold CV in orange), see also Supplementary Tables S2c-d. Gerami_27, LMC_150: Leeds Melanoma Cohort 150 gene signature. See “Methods” section “Machine learning analysis“ for details about the classification algorithms. Source data are provided as a Source data file.CC BY · PubMed Central
Fig. 4 Weighted Cam_121 score negatively correlates with immune cell expression scores, indicating that a lower weighted signature expression score is associated with a richer immune microenvironment and better prognosis. Scatterplots and density plots showing the relationship between the standardised weighted Cam_121 score (x-axes) and a, b, the median activated B-cell gene expression (y-axes), c, d the median T-cell gene expression (y-axes) and e, f the total immune score (median gene expression of all cell types derived from Angelova et al.25, (y axes). Observations and lines of best fit are colour-coded according to their metastatic status, with red indicating relapse and blue indicating no relapse. The shaded region in the scatter plots of a, c and e corresponds to the 95% confidence interval of the line of best fit. The two-sided p values from the Pearson correlation coefficients are indicated for scatter plots. g Scatterplot of the scores of the observations on the two-first dimensions of a PCA analysis on the overall immune cell expression data. Observations are colour-coded according to their weighted Cam_121 expression score (“high”/”low” classification based on a quantile 0.33 split, indicated in red/yellow, respectively). Different symbols are used for observations with “high” (circles) and “low” (triangles) immune cell expression levels based on a median split. Source data are provided as a Source data file.
Fig. 4 Weighted Cam_121 score negatively correlates with immune cell expression scores, indicating that a lower weighted signature expression score is associated with a richer immune microenvironment and better prognosis. Scatterplots and density plots showing the relationship between the standardised weighted Cam_121 score (x-axes) and a, b, the median activated B-cell gene expression (y-axes), c, d the median T-cell gene expression (y-axes) and e, f the total immune score (median gene expression of all cell types derived from Angelova et al.25, (y axes). Observations and lines of best fit are colour-coded according to their metastatic status, with red indicating relapse and blue indicating no relapse. The shaded region in the scatter plots of a, c and e corresponds to the 95% confidence interval of the line of best fit. The two-sided p values from the Pearson correlation coefficients are indicated for scatter plots. g Scatterplot of the scores of the observations on the two-first dimensions of a PCA analysis on the overall immune cell expression data. Observations are colour-coded according to their weighted Cam_121 expression score (“high”/”low” classification based on a quantile 0.33 split, indicated in red/yellow, respectively). Different symbols are used for observations with “high” (circles) and “low” (triangles) immune cell expression levels based on a median split. Source data are provided as a Source data file.CC BY · PubMed Central
Fig. 5 Cam_121 negatively correlates with tumour-infiltrating lymphocyte counts. Violin plots showing the weighted (standardised) Cam_121 scores (y-axis) by levels of the TIL scores (x-axes) in the a AVAST-M primary melanoma dataset (n = 137) using Clark et al.26 TIL scores, b Leeds Melanoma Cohort dataset (n = 499) using Clark et al. TIL scores and c AVAST-M primary melanoma dataset (n = 139) using MIA27 TIL scores. The overall p values per plot are calculated using Fisher’s ANOVA, and pairwise comparison p values were defined using two-sided Student’s t test. The grey dotted line represents the 33% quantile cut off for the gene expression signature in each dataset. Source data are provided within the Source data file.
Fig. 5 Cam_121 negatively correlates with tumour-infiltrating lymphocyte counts. Violin plots showing the weighted (standardised) Cam_121 scores (y-axis) by levels of the TIL scores (x-axes) in the a AVAST-M primary melanoma dataset (n = 137) using Clark et al.26 TIL scores, b Leeds Melanoma Cohort dataset (n = 499) using Clark et al. TIL scores and c AVAST-M primary melanoma dataset (n = 139) using MIA27 TIL scores. The overall p values per plot are calculated using Fisher’s ANOVA, and pairwise comparison p values were defined using two-sided Student’s t test. The grey dotted line represents the 33% quantile cut off for the gene expression signature in each dataset. Source data are provided within the Source data file.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 66 references

Cited by

See all 70

Metadata

Authors
Garg, M., Couturier, D.-L., Nsengimana, J., Fonseca, N. A., Wongchenko, M., Yan, Y., Lauss, M., Jönsson, G. B., Newton-Bishop, J., Parkinson, C., Middleton, M. R., Bishop, D. T., McDonald, S., Stefanos, N., Tadross, J., Vergara, I. A., Lo, S., Newell, F., Wilmott, J. S., Thompson, J. F., Long, G. V., Scolyer, R. A., Corrie, P., Adams, D. J., Brazma, A., & Rabbie, R.
Affiliations
  • European Bioinformatics Institute (GB)
  • University of Cambridge (GB)
  • University of Leeds (GB)
  • Universidade do Porto (PT)
  • Genentech (US)
  • Lund University (SE)
  • Cambridge University Hospitals NHS Foundation Trust (GB)
  • University of Oxford (GB)
  • The University of Sydney (AU)
  • QIMR Berghofer Medical Research Institute (AU)
  • Wellcome Sanger Institute (GB)
Corresponding author
Rabbie, Roy
Venue
Nature Communications
Published
2021-02-18
Volume
12
Issue
1
Pages
1137-1137
ISSN
2041-1723 (ISSN-L)
OpenAlex
W3123792751
Cited by
70
References
69
Citations by year
2026: 7 · 2025: 9 · 2024: 15 · 2023: 11 · 2022: 20 · 2021: 7
FWCI
4.83
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Lund University Publications (Lund University) (submittedVersion)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • Apollo (cc-by)
  • Apollo
Funders
  • University of Sydney Medical Foundation
  • Department of Health | National Health and Medical Research Council
Topics
  • Health Sciences › Medicine › Oncology › Cutaneous Melanoma Detection and Management
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Cancer Research › Breast Cancer Treatment Studies
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene expression and cancer classification
Keywords
primary melanoma, gene expression signatures, metastasis prediction, prognostic factors, stage III melanoma, adjuvant therapy, lymph node metastasis, immune cell infiltration, overall survival, progression-free survival, stage II melanoma, machine learning classification
MeSH terms
Machine Learning, Progression-Free Survival, Humans, Melanoma, Neoplasm Staging, Prognosis, Time Factors, Reproducibility of Results, Gene Expression Regulation, Neoplastic, Multivariate Analysis, Proportional Hazards Models, Treatment Outcome, Gene Expression Profiling, Databases, Genetic

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Tumour gene expression signature in primary melanoma predicts long-term outcomes. https://app.cochat.ai/papers/10.1038/s41467-021-21207-2
The paper
Garg, M., Couturier, D.-L., Nsengimana, J., Fonseca, N. A., Wongchenko, M., Yan, Y., Lauss, M., Jönsson, G. B., Newton-Bishop, J., Parkinson, C., Middleton, M. R., Bishop, D. T., McDonald, S., Stefanos, N., Tadross, J., Vergara, I. A., Lo, S., Newell, F., Wilmott, J. S., . . . Rabbie, R. (2021). Tumour gene expression signature in primary melanoma predicts long-term outcomes. Nature Communications. https://doi.org/10.1038/s41467-021-21207-2

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.