Open AccessArticle

Leveraging Tumor Mutation Profiles to Forecast Immune Checkpoint Blockade Resistance in Melanoma, Lung, Head and Neck, Bladder and Renal Cancers

Guillaume Mestrallet

Division of Hematology and Oncology, Hess Center for Science & Medicine, Tisch Cancer Institute, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA

Onco 2024, 4(4), 439-457; https://doi.org/10.3390/onco4040031

Submission received: 11 October 2024 / Revised: 4 December 2024 / Accepted: 7 December 2024 / Published: 10 December 2024

Download

Browse Figures

Figure 1
Mutations associated with a difference in overall survival for patients with lung cancer following immunotherapy. N = 350 lung cancer patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05. "> Figure 2
Mutations associated with a difference in overall survival for patients with melanoma following immunotherapy. N = 320 melanoma patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05. "> Figure 3
Mutations associated with a difference in overall survival for patients with bladder cancer following immunotherapy. N = 215 bladder cancer patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05. "> Figure 4
Mutations associated with a difference in overall survival for patients with renal carcinoma following immunotherapy. N = 151 renal carcinoma patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05. "> Figure 5
Mutations associated with a difference in overall survival for patients with head and neck cancer following immunotherapy. N = 139 head and neck cancer patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05. "> Figure 6
Profiling of cancer patient mutations enables the prediction of their overall survival following ICB. (A) N = 350 lung cancer patients. (B) N = 320 melanoma patients. (C) N = 215 bladder cancer patients. (D) N = 151 renal carcinoma patients. (E) N = 139 head and neck cancer patients. Machine learning models were employed to predict patient overall survival in months after immune checkpoint blockade using previously identified mutational features. The dataset was partitioned into five different subsets and further split into training (80% of the patients) and testing (20% of the patients) subsets. Various algorithms, including Gradient Boosting, Random Forest, Decision Tree, Logistic Regression, Support Vector Classifier (SVC) and Multi-layer Perceptron (MLP), were trained on the mutational features using five-fold cross-validation. Additionally, a Mean Ensemble model combining the best-performing Random Forest and Gradient Boosting models was utilized to improve the accuracy of survival predictions. Hyperparameters of the models were optimized and their performance was evaluated using standard metrics on the test sets. "> Figure 7
Shapley values for each model predicting overall survival following ICB. (A) N = 350 lung cancer patients. (B) N = 320 melanoma patients. (C) N = 215 bladder cancer patients. (D) N = 151 renal carcinoma patients. (E) N = 139 head and neck cancer patients. ">

Review Reports Versions Notes

Simple Summary

Immune checkpoint blockade (ICB), radiotherapy, chemotherapy and surgery are currently used as therapeutic strategies against melanoma, lung, bladder, head and neck and renal cancers, but their efficacy is limited. This study examined 1175 patients who received ICB to understand treatment response and resistance. By analyzing tumor mutations, I found certain mutations associated with either improved or worsened survival outcomes. For example, in head and neck cancer, mutations in genes like PIK3CA and KRAS correlated with longer survival, while mutations in genes like TERT and TP53 were linked to shorter survival. I developed machine learning models that accurately predicted patient survival based on these mutations. These findings suggest that personalized immunotherapy, informed by individual tumor mutational profiles, could significantly enhance treatment outcomes for cancer patients.

Abstract

Immune checkpoint blockade (ICB), radiotherapy, chemotherapy and surgery are currently used as therapeutic strategies against melanoma, lung, bladder and renal cancers, but their efficacy is limited. Thus, I need to predict treatment response and resistance to address this challenge. In this study, I analyzed 350 lung cancer, 320 melanoma, 215 bladder cancer, 139 head and neck cancer and 151 renal carcinoma patients treated with ICB to identify tumor mutations associated with response and resistance to treatment. I identified several tumor mutations linked with a difference in survival outcomes following ICB. In lung cancer, missense mutations in ABL1, ASXL1, EPHA3, EPHA5, ERBB4, MET, MRE11A, MSH2, NOTCH1, PAK7, PAX5, PGR, ZFHX3, PIK3C3 and REL genes were indicative of favorable responses to ICB. Conversely, mutations in TGFBR2, ARID5B, CDKN2C, HIST1H3I, RICTOR, SMAD2, SMAD4 and TP53 genes were associated with shorter overall survival post-ICB treatment. In melanoma, mutations in FBXW7, CDK12, CREBBP, CTNNB1, NOTCH1 and RB1 genes predict resistance to ICB, whereas missense mutations in FAM46C and RHOA genes are associated with extended overall survival. In bladder cancer, mutations in HRAS genes predict resistance to ICB, whereas missense mutations in ERBB2, GNAS, ATM, CDKN2A and LATS1 genes, as well as nonsense mutations in NCOR1 and TP53 genes, are associated with extended overall survival. In head and neck cancer, mutations in genes like PIK3CA and KRAS correlated with longer survival, while mutations in genes like TERT and TP53 were linked to shorter survival. In renal carcinoma, mutations such as EPHA5, MGA, PIK3R1, PMS1, TSC1 and VHL were linked to prolonged overall survival, while others, including total splice mutations and mutations in B2M, BCOR, JUN, FH, IGF1R and MYCN genes were associated with shorter overall survival following ICB. Then, I developed predictive survival models by machine learning that correctly forecasted cancer patient survival following ICB within an error between 5 and 8 months based on their distinct tumor mutational attributes. In conclusion, this study advocates for personalized immunotherapy approaches in cancer patients.

Keywords:

lung cancer; bladder cancer; melanoma; renal carcinoma; head and neck cancer; biomarkers; machine learning; mutation profile; resistance to immunotherapy

1. Introduction

In 2020, approximately 2.2 million new cases of lung cancer and 1.8 million related deaths were reported [1]. Moreover, clear cell renal cell carcinoma (KIRC or ccRCC) is the most prevalent subtype of renal cell carcinoma (RCC) [2], and 175,000 deaths occur annually due to RCC [3]. Current treatments include therapies such as immune checkpoint blockade (ICB) or VEGFR tyrosine kinase inhibitors (TKIs) [4]. Clinical studies have utilized various immune checkpoints, such as anti-PD-1 (Nivolumab), anti-PD-L1 (Atezolizumab), and combinations like anti-CTLA4 with anti-PD-1 (Nivolumab and Ipilimumab) [5,6,7]. Immune checkpoint blockade has demonstrated significant efficacy in non-small cell lung cancer (NSCLC), now integrated into various treatment protocols such as first-line therapy for metastatic disease, consolidation therapy post-chemoradiation in unresectable locally advanced disease, and adjuvant therapy following surgical resection and chemotherapy in resectable disease [8]. An anti-IL-4/IL-13 antibody is also being tested in combination with an anti-PD-1 antibody in lung cancer (NCT05013450) [9]. The immune response, both under normal conditions and following ICB treatment, relies on antigen presentation to T cells via human leukocyte antigens (HLA) and associated cell-surface molecules (MICA, MICB and ULBP) [10]. T cell responses also involve genes that support T cell proliferation, cytotoxicity, and longevity (including BTN3A1, TNFRSF9, PRF1 and CD27 in renal cancer) [11,12,13,14]. Conversely, the expression of regulatory genes (such as ADORA2A, ARG1 and IL-1A) serves to dampen inflammation and maintain tissue equilibrium but can potentially promote cancer progression [15,16,17].

Importantly, a substantial proportion of patients, specifically 70% with advanced NSCLC and 80% with small cell lung cancer (SCLC), do not respond to current immune checkpoint blockade, with immune resistance being a persistent issue [8]. Similarly, a substantial number of KIRC patients exhibit immune resistance and do not fully respond to ICB [5,6,7]. This resistance may stem from the expression of various immune checkpoints and immunosuppressive pathways in KIRC patients, including SIGLEC15, PD-L1, TIM-3, PDCD1, CTLA4, LAG3, PDCD1LG2, TIGIT and HLA-G [2,18,19,20]. Furthermore, ICB therapies can lead to adverse effects such as adrenal insufficiency and autoimmune hepatitis [21].

Given these challenges, conducting a thorough meta-analysis across patient cohorts with cancer becomes crucial for understanding the complex mechanisms that determine responses and resistances to immune checkpoint blockade. Moreover, incorporating machine learning techniques into the analysis of these datasets emerges as a powerful approach to interpreting the underlying mechanisms. Drawing on successful applications in predicting patient outcomes across different cancers, machine learning tools have the potential to enhance the accuracy of predicting responses and resistances to immune checkpoint blockade [22,23,24,25,26,27]. This advancement holds the promise of improving diagnostic accuracy and customizing therapeutic approaches based on individual patient characteristics.

To advance the monitoring of cancer patients receiving immunotherapy, this proposal aims to identify critical genetic mutations linked to either resistance or positive response to treatment. These mutations will then be used as inputs for training several machine learning algorithms, enabling the design of personalized prediction models adapted to each patient’s unique genetic profile. Once developed, these machine learning models can predict how new patients are likely to respond to immunotherapy. Predicting a positive response could prompt the recommendation of immunotherapy for those specific patients, potentially offering significant benefits. Conversely, if resistance is predicted, considering alternative therapeutic strategies may be more beneficial. This approach seeks to refine treatment decisions, improving outcomes for cancer patients treated with immunotherapy.

2. Methods

2.1. Patient Datasets and Mutation Profiles

Patient datasets were selected using cBioPortal [28]. Patients from the MSK cohort received immune checkpoint blockade following melanoma (N = 320), lung (N = 350), bladder (N = 215), head and neck (N = 139) and renal (N = 151) cancer.

In lung cancer, 219 patients are deceased (62.6%) and 131 are alive (37.4%). A total of 170 patients are male (48.6%) and 180 are females (51.4%). A total of 329 patients received the PD1/PDL1 blockade (94%), and 21 received combination therapy (6%). A total of 272 patients have lung adenocarcinoma (77.4%), 45 have lung squamous cell carcinoma (12.9%), 13 have poorly differentiated non-small cell lung cancer (3.7%), 8 have large cell neuroendocrine carcinoma (2.3%), 8 have non-small cell lung cancer (2.3%), 2 have sarcomatoid carcinoma of the lung (0.6%), 2 have lung adenosquamous carcinoma (0.6%) and 1 has pleomorphic carcinoma of the lung (0.3%). For 171 patients (48.9%), the sample was primary, while for 179 patients (51.1%), it was metastasis. A total of 122 patients were older than 71 years old (34.9%), 119 were 61–70 years old (34%), 75 were 50–60 years old (21.4%) and 34 were 31–50 years old (9.7%).

In renal carcinoma, 58 patients are deceased (38.4%) and 93 are alive (61.6%). A total of 109 patients are male (72.2%) and 42 are females (27.8%). A total of 122 patients received the PD1/PDL1 blockade (80.8%), and 29 received combination therapy (19.2%). A total of 121 patients have renal clear cell carcinoma (80.1%), 9 have unclassified renal carcinoma (6%), 5 have chromophobe renal carcinoma (3.3%), 5 have papillary renal carcinoma (3.3%), 4 have renal cell carcinoma (2.6%), 3 have translocation-associated renal cell carcinoma (2%) and 2 have FH-deficient renal cell carcinoma (1.3%). For 80 patients (53%), the sample was primary, while for 71 patients (47%) it was metastasis. A total of 54 were 50–60 years old (35.8%), 25 were 31–50 years old (16.6%), 51 were 61–70 years old (33.8%), 3 were less than 30 years old (2%) and 18 were older than 71 years old (11.9%).

In melanoma, 125 patients are deceased (39.1%) and 195 are alive (60.9%). A total of 200 patients are male (62.5%) and 120 are females (37.5%). A total of 130 patients received the PD1/PDL1 blockade (40.6%), 75 received CTLA4 blockade (23.4%), and 115 received combination therapy (35.9%). A total of 187 patients have cutaneous melanoma (58.4%), 44 have melanoma of unknown primary (13.8%), 21 have acral melanoma (6.6%), 20 have uveal melanoma (6.3%), 17 have anorectal mucosal melanoma (5.3%), 11 have mucosal melanoma of the vulva (3.4%), 10 have head and neck mucosal melanoma (3.1%), 3 have desmoplastic melanoma (0.9%) and 3 have mucosal melanoma of the esophagus (0.9%). For 54 patients (16.9%), the sample was primary, while for 266 patients (83.1%) it was metastasis. A total of 73 were 50–60 years old (22.8%), 51 were 31–50 years old (15.9%), 85 were 61–70 years old (26.6%), 15 were less than 30 years old (4.7%) and 96 were older than 71 years old (30%).

In bladder cancer, 95 patients are deceased (44.2%) and 120 are alive (55.8%). A total of 164 patients are male (76.3%) and 51 are females (23.7%). A total of 192 patients received the PD1/PDL1 blockade (89.3%), and 23 received combination therapy (10.7%). A total of 147 patients have bladder urothelial carcinoma (68.4%), 47 have upper tract urothelial carcinoma (21.9%), 5 have urethral urothelial carcinoma (2.3%) and 4 have plasmacytoid/signet ring cell bladder carcinoma (1.9%). For 123 patients (57.2%), the sample was primary, while for 92 patients (42.8%) it was metastasis. A total of 43 were 50–60 years old (20%), 13 were 31–50 years old (6%), 75 were 61–70 years old (34.9%) and 84 were older than 71 years old (39.1%).

In head and neck cancer, 79 patients are deceased (56.8%) and 60 are alive (43.2%). A total of 109 patients are male (78.4%) and 30 are females (21.6%). A total of 131 patients received the PD1/PDL1 blockade (94.2%), and 8 received combination therapy (5.8%). A total of 37 patients have oropharynx squamous cell carcinoma (26.6%), 37 patients have head and neck squamous cell carcinoma (26.6%) and 25 patients have oral cavity squamous cell carcinoma (18%). For 42 patients (30.2%), the sample was primary, while for 97 patients (69.8%), it was metastasis. A total of 32 were 50–60 years old (23%), 27 were 31–50 years old (19.4%), 54 were 61–70 years old (38.8%), 4 were less than 30 years old (2.9%) and 22 were older than 71 years old (15.8%).

The patients included in the study were diverse, with some having undergone extensive prior treatments and others receiving various combination therapies. Additionally, the timing of MSK-IMPACT testing in relation to the start of ICI treatment varied [28]. 87% of the patients had stage IV or metastatic disease, and 117 patients had stage III melanoma, as reported in the original study. These tumors underwent targeted next-generation sequencing (NGS) (MSK-IMPACT panel). I identified shared mutations by analyzing the mutations expressed in at least two patients. All patient data are reported in Supplementary Table S1.

2.2. Statistics

I used Spearman and Pearson correlations as well as Cox Proportional Hazards models to determine the statistical significance of the observed differences. Kaplan–Meier curves and the log-rank test were also used for survival plots. The difference was considered to be significant when the p-value was below 0.05. * p < 0.05.

2.3. Machine Learning Algorithms to Predict the Response of Each Patient to Immune Checkpoint Blockade

Melanoma (N = 320), lung (N = 350), bladder (N = 215), head and neck (N = 139) and renal (N = 151) cancer patients from the MSK cohort received immune checkpoint blockade following cancer. The algorithms calculate the probability of patients being alive following immune checkpoint blockade according to mutational features. The dataset was divided into five subsets and further split into training (80% of the patients) and testing (20% of the patients) sets. Mutational features were used to train Gradient Boosting, Random Forest, Decision Tree, Support Vector Classifier (SVC), Logistic Regression and Multi-layer Perceptron (MLP) models on the training data, employing five-fold cross-validation to predict the response to ICB, specifically overall survival in months.

Gradient Boosting is an ensemble learning technique that incrementally constructs a strong predictive model by sequentially adding weak learners, such as decision trees, to the ensemble. The method aims to optimize a loss function by iteratively fitting new models to the residual errors of the previous ensemble. I utilized the Gradient Boosting algorithm from the scikit-learn library in Python. Random Forest is another ensemble learning approach that builds multiple decision trees based on random subsets of features and bootstrapped samples of the data. Predictions are aggregated through averaging (for regression) or voting (for classification) across all trees. Similarly, I implemented the Random Forest algorithm using the scikit-learn library. To capitalize on the strengths of both Gradient Boosting and Random Forest models, I employed a mean ensemble method. This approach combined predictions from both models by averaging them. The ensemble predictions served as the final output of this predictive model. Hyperparameter optimization for Random Forest and Gradient Boosting models was conducted independently using techniques like grid search or random search. This process aimed to fine-tune the model’s hyperparameters to increase performance based on suitable evaluation metrics. Additionally, these algorithms could be integrated into software to enhance user interfaces and manage data storage [29,30].

2.3.1. Data Preparation

I utilized a dataset sourced from a CSV file. The dataset was preprocessed by handling missing values through row removal and converting categorical variables into numerical format using one-hot encoding. The target variable for regression was ‘Overall_Survival_Months’, with other columns serving as features.

2.3.2. Data Splitting and Scaling

The dataset was divided into training (80%) and testing (20%) subsets using train_test_split. To standardize the features, I applied StandardScaler to both training and testing sets.

2.3.3. Model Training and Hyperparameter Tuning

I employed a variety of regression models, including Linear Regression, Lasso Regression, Support Vector Regression (SVR), k-Nearest Neighbors (KNN), Decision Tree Regression, Multi-layer Perceptron (MLP), Random Forest Regression and Gradient Boosting Regression. For the Random Forest and Gradient Boosting models, hyperparameters were optimized using RandomizedSearchCV and GridSearchCV.

2.3.4. Model Evaluation

Model performance was assessed using Mean Squared Error (MSE), Mean Absolute Error (MAE), Median Absolute Error and Max Error. Cross-validation with cross_val_score was performed to evaluate model robustness. Additionally, SHAP values were computed to interpret the predictions of the best-performing models.

2.3.5. Ensemble Model

A Voting Regressor, combining the top-performing Random Forest and Gradient Boosting models, was also trained and evaluated. This ensemble approach aimed to enhance the predictive performance.

2.3.6. Visualization

Scatter plots comparing actual versus predicted values were generated for each model and the ensemble model to visually assess prediction accuracy.

2.4. Alpha Fold Prediction of Protein Structure According to Mutation Status

Protein structures were predicted using Alpha Fold algorithms [31] and the amino acid sequences of the protein of interest. I showed the structure of the protein with the highest probability. The number of sequences per position is superior to 30 sequences per position. The predicted lDDT per position represents the model confidence (out of 100) at each position. I used the PyMol software Version 3.1 (Schrödinger, Inc., New York, NY, USA) to show the structures of the mutated and WT proteins aligned together.

3. Results

3.1. Missense and Nonstop Mutation Counts Predict Response to ICB in Lung Cancer

I proposed that the mutation characteristics of individuals with lung cancer could indicate their likelihood of responding to or resisting immunotherapy. To explore this, I analyzed the (Memorial Sloan Kettering Cancer Center) MSK Impact cohort dataset, which includes 350 lung cancer patients who underwent treatment with immune checkpoint inhibitors (ICB). I analyzed the total counts of various mutation types (5′ flank, nonsense, splice, nonstart, nonstop, missense, in-frame (IF) insertions and deletions and frameshift (FS) insertions and deletions) for each patient. By examining the Spearman and Pearson correlations between these mutation counts and overall survival in months, I found that patients with a higher number of missense and nonstop mutations tended to have longer overall survival (p < 0.05) (Supplementary Figure S1A). No other mutation types (such as splice site, 5′ flank, non-start, nonsense, IF indels, or FS indels) were linked to response or resistance within this cohort. Therefore, missense and nonstop mutations appear to be predictive of extended survival following ICB therapy.

3.2. Missense, Nonsense and 5′ Flank Mutation Counts Forecast Response Following ICB in Melanoma

Then, I investigated whether the mutation profile of melanoma patients could predict their response to immunotherapy using the MSK Impact cohort dataset, which included 320 patients treated with ICB. Through Spearman and Pearson correlations between mutation counts of each type and overall survival in months, I found a significant association indicating that patients with more missense, nonsense and 5′ Flank mutations experienced longer overall survival (p-value < 0.05) (Supplementary Figure S1B). Notably, no other mutation types showed associations with response or resistance in this cohort. Thus, missense, nonsense and 5′ Flank mutations appear to predict a favorable response to ICB therapy in terms of longer patient survival in melanoma.

3.3. Missense, Nonsense, Nonstart and Frameshift Deletion Mutation Counts Predict Response Following ICB in Bladder Cancer

I also investigated whether the mutation profile of bladder cancer patients could predict their response to immunotherapy using the MSK Impact cohort dataset, which included 215 patients treated with ICB. Through Spearman and Pearson correlations between mutation counts of each type and overall survival in months, I found a significant association indicating that patients with more missense, nonsense, nonstart and FS del mutations experienced longer overall survival (p-value < 0.05) (Supplementary Figure S1C). Notably, no other mutation types showed associations with response or resistance in this cohort. Thus, missense, nonsense, nonstart and FS del mutations appear to predict a favorable response to ICB therapy in terms of longer patient survival in bladder cancer.

3.4. Splice Mutation Count Forecasts Response Following ICB in Renal Carcinoma

I investigated whether the mutation profile of renal carcinoma patients could predict their response to immunotherapy using the MSK Impact cohort dataset, which included 151 patients treated with ICB. Through Spearman and Pearson correlations between mutation counts of each type and overall survival in months, I found a significant association indicating that patients with more splice mutations experienced longer overall survival (p-value < 0.05) (Supplementary Figure S1D). Notably, no other mutation types, including nonsense mutations, 5′ flank mutations, nonstart mutations, nonstop mutations, missense mutations, IF indels or FS indels, showed associations with response or resistance in this cohort. Thus, splice mutations appear to predict a favorable response to ICB therapy in terms of longer patient survival in renal carcinoma.

3.5. 5′Flank Mutation Count Forecasts Resistance Following ICB While Frameshift Deletion Count Predicts Response in Head and Neck Cancer

Finally, I investigated whether the mutation profile of head and neck cancer patients could predict their response to immunotherapy using the MSK Impact cohort dataset, which included 139 patients treated with ICB. Through Spearman and Pearson correlations between mutation counts of each type and overall survival in months, I found a significant association indicating that patients with more 5′Flank mutations experienced shorter overall survival (p-value < 0.05) (Supplementary Figure S1E). On the contrary, patients with more FS del mutations have better survival according to the log-rank test. Notably, no other mutation types, including nonsense mutations, splice mutations, nonstart mutations, nonstop mutations, missense mutations, IF indels, or FS ins, showed associations with response or resistance in this cohort. Thus, FS del mutations appear to predict a favorable response to ICB therapy, while 5′ Flank mutations were associated with a shorter overall survival post-ICB in head and neck cancer.

3.6. In Lung Cancer, Missense Mutations in ABL1, ASXL1, EPHA3, EPHA5, ERBB4, MET, MRE11A, MSH2, NOTCH1, PAK7, PAX5, PGR, ZFHX3, PIK3C3 and REL Genes Predict Response to ICB

I proposed that specific mutations might be linked to the correlation between longer overall survival in lung cancer and the presence of missense and nonstop mutations. To investigate this, I analyzed the mutation profiles of 350 lung cancer patients, examining their overall survival in months and mutation status. I employed the Cox Proportional Hazards model, along with Spearman and Pearson correlation analyses and log-rank tests, to evaluate the relationship between mutations and overall survival. These findings revealed that the REL missense mutation had a hazard ratio of 0.1358 (p < 0.05), indicating a significantly reduced risk of mortality compared to those without this mutation (Figure 1, Supplementary Figure S2). Additionally, I observed various hazard ratios for other mutations: PIK3C3 missense (0.1826), ARID5B frameshift deletion (32.6906), CDKN2C missense (50.8121), HIST1H3I translation start site (49.9026), RICTOR frameshift insertion (29.9149), SMAD2 missense (14.8336), SMAD4 in-frame deletion (16.8104) and TP53 frameshift deletion (3.6108), all of which were statistically significant (p < 0.05) (Figure 1, Supplementary Figure S2).

Notably, ARID5B, CDKN2C, HIST1H3I and RICTOR mutations were associated with shorter overall survival according to the log-rank test, whereas REL and PIK3C3 missense mutations were linked to longer survival. Furthermore, Spearman and Pearson correlations indicated that missense mutations in ABL1, ASXL1, EPHA3, EPHA5, ERBB4, MET, MRE11A, MSH2, NOTCH1, PAK7, PAX5, PGR, ZFHX3, PIK3C3, and REL genes, as well as nonsense mutations in ARID2 and MAX genes, were associated with longer overall survival (Supplementary Figure S2). Conversely, nonsense mutations in TGFBR2, ARID5B frameshift deletions, CDKN2C missense mutations, HIST1H3I translation start site mutations, RICTOR frameshift insertions, SMAD2 missense mutations, SMAD4 in-frame deletions, and TP53 frameshift deletions were linked to shorter overall survival. No significant differences were observed for other genes in this cohort. Overall, missense mutations in ABL1, ASXL1, EPHA3, EPHA5, ERBB4, MET, MRE11A, MSH2, NOTCH1, PAK7, PAX5, PGR, ZFHX3, PIK3C3 and REL genes are predictive of a favorable response to ICB, whereas mutations in TGFBR2, ARID5B, CDKN2C, HIST1H3I, RICTOR, SMAD2, SMAD4 and TP53 genes are associated with poorer overall survival.

3.7. In Melanoma, Mutations in FBXW7, NOTCH1, CREBBP and Other Genes Predict Response and Resistance Following ICB

I hypothesized that the correlation between longer overall survival in melanoma and increased missense, nonsense and 5′Flank mutations may be related to specific genetic alterations. I characterized the mutation profiles of 320 melanoma patients based on their overall survival in months and disease status. Using the Cox Proportional Hazards models, Spearman and Pearson correlations, and log-rank tests, I evaluated the association between mutations and overall survival. Interestingly, I found that individuals with FBXW7 missense mutations had a high hazard ratio of 5 (p < 0.05), indicating a significantly elevated risk of mortality compared to those without this mutation (Figure 2, Supplementary Figure S3). Additionally, missense mutations in CDK12, CREBBP, CTNNB1, NOTCH1 and RB1 genes were also associated with high hazard ratios of 2, 94, 3, 137 and 3, respectively (p < 0.05) (Supplementary Figure S3) and these FBXW7, CREBBP and NOTCH1 mutations were statistically linked to resistance to immunotherapy using the log-rank test.

Conversely, through Spearman and Pearson correlations, I showed that mutations such as FAM46C missense, PMS2 missense and other mutations were associated with longer overall survival, while missense mutations in RHOA were associated with shorter overall survival (Supplementary Figure S3). No significant differences were noted for other genes of the MSK Impact panel in this cohort. In addition, contrary to the other cancer types in this study, I found that histology was associated with differences in survival for melanoma (Supplementary Figure S4). In summary, mutations in FBXW7, CDK12, CREBBP, CTNNB1, NOTCH1 and RB1 genes predict resistance to immune checkpoint blockade (ICB), whereas missense mutations in FAM46C and RHOA genes are associated with extended overall survival in melanoma patients.

3.8. In Bladder Cancer, Mutations in ERBB2, GNAS, HRAS, ATM, CDKN2A, LATS1, NCOR1, SMARC4 and TP53 Genes Predict Response and Resistance Following ICB

I hypothesized that the correlation between longer overall survival in the bladder and increased missense, nonsense, FS del and nonstart mutations may be related to specific genetic alterations. I characterized the mutation profiles of 215 bladder cancer patients based on their overall survival in months and disease status. Using Cox Proportional Hazards models, Spearman and Pearson correlations and log-rank tests, I evaluated the association between mutations and overall survival. Interestingly, I found that individuals with HRAS splice mutations had a high hazard ratio of 41 (p < 0.05), indicating a significantly elevated risk of mortality compared to those without this mutation (Figure 3, Supplementary Figure S5). On the other hand, missense mutations in ERBB2 and GNAS genes were associated with hazard ratios of 0.45 and 0.07, respectively (p < 0.05) (Figure 3, Supplementary Figure S5), and these GNAS and HRAS mutations were also statistically respectively linked to response and resistance to immunotherapy using the log-rank test.

Conversely, through Spearman and Pearson correlations, I showed that mutations such as ATM missense, CDKN2A missense, LATS1 missense, NCOR1 nonsense and TP53 nonsense mutations were associated with longer overall survival (Supplementary Figure S5). No significant differences were noted for other genes of the MSK Impact panel in this cohort. In summary, mutations in HRAS genes predict resistance to immune checkpoint blockade (ICB), whereas missense mutations in ERBB2, GNAS, ATM, CDKN2A and LATS1 genes, as well as nonsense mutations in NCOR1 and TP53 genes, are associated with extended overall survival in bladder cancer patients.

3.9. In Renal Carcinoma, Mutations in B2M, BCOR, JUN, FH, IGF1R, MYCN and VHL Genes Forecast Resistance and Response Following ICB

I hypothesized that the correlation between longer overall survival in renal carcinoma and increased splice mutations may be related to specific genetic alterations. I characterized the mutation profiles of 151 renal carcinoma patients based on their overall survival in months and disease status. Using Cox Proportional Hazards models, Spearman and Pearson correlations and log-rank tests, I evaluated the association between mutations and overall survival. Interestingly, I found that individuals with B2M splice mutations had a high hazard ratio of 71 (p < 0.05), indicating a significantly elevated risk of mortality compared to those without this mutation (Figure 4, Supplementary Figure S6). Additionally, missense mutations in BCOR, FH, IGF1R, JUN and MYCN genes were also associated with high hazard ratios of 42, 136, 41, 86 and 26, respectively (p < 0.05) (Figure 4, Supplementary Figure S6) and these mutations were statistically linked to resistance to immunotherapy using the log-rank test.

Conversely, through Spearman and Pearson correlations, I showed that mutations such as EPHA5 missense, MGA frameshift deletion, PIK3R1 splice, PMS1 missense, TSC1 frameshift deletion and VHL mutations (predominantly VHL splice mutations) were associated with longer overall survival (Supplementary Figure S6). No significant differences were noted for other genes of the MSK Impact panel in this cohort. In summary, mutations in B2M, BCOR, JUN, FH, IGF1R and MYCN genes predict resistance to immune checkpoint blockade (ICB), whereas mutations in EPHA5, MGA, PIK3R1, PMS1, TSC1 and VHL genes are associated with extended overall survival in renal carcinoma patients.

3.10. In Head and Neck Cancer, Mutations in ANKRD11, ERCC5, FANCA, KMT2A, WT1, HIST1H1C, KRAS, PIK2CA, PTRT, TERT, TP53 and TSC1 Genes Predict Response and Resistance Following ICB

I hypothesized that the correlation between shorter overall survival in head and neck cancer and increased 5′ Flank mutations may be related to specific genetic alterations. I characterized the mutation profiles of 139 head and neck cancer patients based on their overall survival in months and disease status. Using Cox Proportional Hazards models, Spearman and Pearson correlations and log-rank tests, I evaluated the association between mutations and overall survival. Interestingly, I showed that 5′Flank mutations in TERT and missense mutations in TP53 were associated with a shorter overall survival using Spearman and Pearson correlations (Supplementary Figure S7). Interestingly, I found that individuals with FANCA nonsense mutations had a high hazard ratio of 441 (p < 0.05), indicating a significantly elevated risk of mortality compared to those without this mutation (Figure 5, Supplementary Figure S7). Additionally, missense mutations in ERCC5 and WT1, ANKRD11 IF del mutations and KMT2A FS del mutations were also associated with high hazard ratios of 405, 63, 20 and 197, respectively (p < 0.05) (Figure 5, Supplementary Figure S7). These ERCC5, FANCA, KMT2A and WT1 mutations were statistically linked to resistance to immunotherapy using the log-rank test. Conversely, through Spearman and Pearson correlations, I showed that mutations such as HIST1H1C missense, KRAS missense, PTRT FS ins, RAD51 missense, TSC1 missense and PIK3CA missense were associated with longer overall survival (Figure 5). No significant differences were noted for other genes of the MSK Impact panel in this cohort. In summary, mutations in TERT, TP53, ERCC5, WT1, ANKRD11 and KMT2A genes predict resistance to ICB, whereas mutations in HIST1H1C, KRAS, PTRT, RAD51, TSC1 and PIK3CA genes are associated with extended overall survival in head and neck cancer patients.

3.11. Profiling of Cancer Patient Mutations Enables the Prediction of Their Overall Survival Following ICB

I proposed that the mutational profiles associated with patient survival, as illustrated in Figure 1, Figure 2, Figure 3, Figure 4 and Figure 5, could be leveraged for predictive modeling using machine learning. Consequently, I developed several machine learning algorithms to forecast patient survival in months, as shown in Figure 6. To implement this, I partitioned the dataset into five subsets, dividing each into training (80% of the patients) and testing (20% of the patients) groups. I trained various models—including Random Forest, Gradient Boosting, Decision Tree, Support Vector Classifier (SVC), Logistic Regression and Multi-layer Perceptron (MLP)—on the mutational features using five-fold cross-validation to predict patient survival following immune checkpoint blockade (ICB). Additionally, I used a Mean Ensemble model that combined the most effective Random Forest and Gradient Boosting models to further improve prediction accuracy. The selected models, fine-tuned with optimal hyperparameters, were assessed with performance metrics on the test sets. Comparing actual versus predicted values, the Random Forest model, Gradient Boosting model and Mean Ensemble model exhibited the best performance (Figure 6).

For lung cancer, the models achieved a Mean Squared Error of 140, a Mean Absolute Error of 9 months, a Maximum Error of 47 months and a Median Absolute Error of 8 months (Figure 6A). Shapley value analysis identified a high Tumor Mutational Burden (TMB) and a high count of missense mutations as key survival predictors (Figure 7A). In melanoma, the models yielded a Mean Squared Error of 229, a Mean Absolute Error of 11 months, a Maximum Error of 59 months and a Median Absolute Error of 8.6 months (Figure 6B). Shapley value analysis highlighted the importance of BRAF, PITCH1, FOXP1, BRCA1 and ARID1A missense mutations, along with the total missense mutation count (Figure 7B). For bladder cancer, the results were a Mean Squared Error of 99, a Mean Absolute Error of 7 months, a Maximum Error of 32 months and a Median Absolute Error of 5.6 months (Figure 6C). Shapley values indicated that the total number of missense mutations and a high TMB were significant predictors (Figure 7C). In renal carcinoma, the models showed a Mean Squared Error of 183, a Mean Absolute Error of 10 months, a Maximum Error of 50 months and a Median Absolute Error of 8.6 months (Figure 6D). Shapley analysis pointed to VHL mutations and the total number of splice mutations as strong predictors (Figure 7D). For head and neck cancer, the models achieved a Mean Squared Error of 65, a Mean Absolute Error of 6 months, a Maximum Error of 25 months and a Median Absolute Error of 5 months (Figure 6E). Shapley values revealed that 5′Flank mutations, particularly in TERT and the total number of PIK3CA and TP53 missense mutations were influential (Figure 7E).

In summary, these machine learning models effectively predicted patient survival in months based on mutational signatures, with an average error of 5 to 8 months. This approach highlights the potential of using predictive mutational signatures and machine learning techniques to assess the likelihood of immunotherapy benefits for patients with melanoma, lung, bladder, head and neck and renal cancers.

The comparison of Shapley values between Gradient Boosting and Random Forest models provides insights into how each model interprets the importance of different mutational features in predicting patient response to immune checkpoint blockade (ICB). Shapley values are a method in machine learning used to explain the contribution of each feature to the prediction made by a model. Higher Shapley values indicate greater importance of a feature in influencing the model’s output. Comparing Shapley values between Gradient Boosting and Random Forest models offers a nuanced understanding of how different mutational profiles influence predictions of patient response to ICB, leveraging the strengths and interpretability of each machine learning approach.

3.12. Characterization of the Missense Mutations Driving the Prediction of the Response and Resistance by Machine Learning in Head and Neck Cancer Patients

I hypothesized that the PIK3CA E545K and TP53 R273C missense mutations strongly drive the prediction of these machine learning algorithms (Supplementary Figure S8B), which may be due to a structural change of the protein and its functionality. By performing Alpha Fold prediction of the 3D structure of PIK3CA E545K and TP53 R273C proteins, it seems that the mutations do not change a lot the structure of functional domains and helix domains (Supplementary Figure S8A). I also investigated if the prognosis value of these mutations was specifically associated with ICB therapy or any other condition. In TCGA head and neck cancer patient data, a PIK3CA E545K missense mutation was not associated with a difference in overall survival (Supplementary Figure S8B). On the contrary, TP53 R273C missense mutation was associated with a shorter overall survival in this cohort. Thus, as most of the TCGA patients were not treated with ICB, while all patients in the MSK cohort received ICB, it indicates that the prognosis value of a PIK3CA E545K missense mutation is specifically associated with response to ICB, while the poor survival outcome associated with TP53 R273C mutation is not specifically associated with ICB therapy.

4. Discussion

In the MSK cohort, only 37.4% of lung cancer patients, 60.9% of melanoma patients, 55.8% of bladder cancer patients, 43.2% of head and neck cancer patients and 61.6% of renal carcinoma patients survived following immune checkpoint blockade (ICB). Therefore, developing alternative strategies for patients resistant to ICB is crucial. Effective therapy selection hinges on identifying patients who may benefit from ICB versus those who might require alternative treatments. To achieve this, I identified specific mutations associated with either resistance or response to ICB in a cohort of 350 lung cancer, 320 melanoma, 215 bladder cancer, 139 head and neck cancer and 151 renal carcinoma patients.

In lung cancer, missense mutations in ABL1, ASXL1, EPHA3, EPHA5, ERBB4, MET, MRE11A, MSH2, NOTCH1, PAK7, PAX5, PGR, ZFHX3, PIK3C3 and REL genes were indicative of favorable responses to ICB. Conversely, mutations in TGFBR2, ARID5B, CDKN2C, HIST1H3I, RICTOR, SMAD2, SMAD4 and TP53 genes were associated with shorter overall survival post-ICB treatment. In melanoma, mutations in FBXW7, CDK12, CREBBP, CTNNB1, NOTCH1 and RB1 genes predict resistance to immune checkpoint blockade (ICB), whereas missense mutations in FAM46C and RHOA genes are associated with extended overall survival. In bladder cancer, mutations in HRAS genes predict resistance to immune checkpoint blockade (ICB), whereas missense mutations in ERBB2, GNAS, ATM, CDKN2A and LATS1 genes, as well as nonsense mutations in NCOR1 and TP53 genes, are associated with extended overall survival. In head and neck cancer, mutations in genes like PIK3CA and KRAS correlated with longer survival, while mutations in genes like TERT and TP53 were linked to shorter survival. Finally, in renal carcinoma, mutations in genes such as B2M, BCOR, JUN, FH, IGF1R and MYCN were predictive of resistance to ICB. Conversely, mutations in genes like EPHA5, MGA, PIK3R1, PMS1, TSC1 and VHL were linked to longer overall survival outcomes. These findings underscore the importance of personalized medicine in oncology, aiming to tailor treatment strategies based on individual genetic profiles to optimize therapeutic efficacy and patient outcomes.

MSH2 allows DNA mismatch/damage recognition and is frequently mutated in mismatch repair deficient tumors, which are more responsive to ICB than mismatch repair proficient tumors [32]. TGFBR2 encodes a receptor kinase critical for transmitting signals through the TGFß signaling pathway. Truncating mutations in TGBR2 frequently occur within the big adenine tract in the receptor type II (BAT-RII) region, leading to the loss of the cytoplasmic domain essential for catalytic activity [33]. TP53, a pivotal tumor suppressor involved in the DNA damage response pathway, is the most commonly mutated gene in cancer. Truncating mutations in TP53 are distributed across the gene and result in the production of various C-terminally truncated protein forms. These mutations are typically inactivating and are associated with unfavorable clinical outcomes [34,35,36,37,38]. Experimental investigations have demonstrated that these TP53 truncating mutations facilitate cancer cell proliferation, survival and metastasis. Ectopic expression of these mutations in melanoma cells has been shown to increase cell motility and promote tumor formation in vivo. These effects are partly attributed to the aberrant localization of truncated proteins to the mitochondria, where they modulate genes involved in cell survival pathways, including CypD [39].

B2M truncating mutations increase cancer cell evasion of immune surveillance [40,41,42]. BCOR, JUN and MYCN are epigenetic regulators and transcription factors altered in various solid and hematologic malignancies, including acute myeloid leukemia, but the underlying mechanisms are still under investigation [43,44,45]. Germline mutations in the FH gene, which encodes an enzyme responsible for converting fumarate to malate, are linked to a condition known as hereditary leiomyomatosis and renal cell cancer [46]. IGFR, an insulin growth factor receptor, is also altered by a mutation in various cancer types and may be targeted by tyrosine kinase inhibitors (TKIs), antisense oligonucleotides and monoclonal antibodies [47]. Thus, resistant patients with these mutations may benefit from combination therapy.

To enhance the predictive capability for immune checkpoint blockade (ICB) responses in cancer patients, I developed machine learning algorithms leveraging the mutational signature. I trained an Ensemble Model on mutational features identified as differentially expressed between patients exhibiting resistance or response to checkpoint blockade. These models achieved significant success, accurately predicting overall survival with an error margin between 5 and 8 months.

To further refine prediction accuracy, several strategies can be considered. First, I address class imbalance. Obtaining additional data, specifically from patients who exhibit resistance to ICB, can mitigate the impact of class imbalance in the dataset. This approach ensures that the model is equally trained on both response and resistance cases, improving its ability to generalize across different outcomes. Then, I integrate immune features. Combining tumor mutational data with matched immune features could provide a more comprehensive understanding of the immune landscape influencing ICB response. Previous research, including studies in glioblastoma, has highlighted associations between resistance to ICB and compromised immune responses such as altered monocyte, macrophage and T follicular helper cell activities, impaired antigen presentation, abnormal regulatory T cell responses, and increased expression of immunosuppressive genes like CD276, TGFB and IL2RA [22,27]. By integrating these approaches, I aim to enhance the precision and reliability of predicting ICB responses in cancer patients, ultimately optimizing therapeutic decision-making and improving patient outcomes. If MSK or other institutes increase the size of the cohort in the future, researchers may identify additional features associated with resistance.

5. Conclusions

In conclusion, this study underscores the critical importance of personalized approaches in the treatment of cancer, particularly in the context of ICB therapy. By thoroughly analyzing the mutational profiles of 1175 patients, I have identified specific genetic mutations associated with both favorable and poor survival outcomes. The development of predictive models using these mutational signatures has demonstrated significant potential in forecasting patient survival with notable accuracy. These findings advocate for the integration of individualized mutational data and advanced machine learning algorithms into clinical practice, promising to enhance the efficacy of immunotherapy and improve patient outcomes. Ultimately, this personalized approach offers a path toward more effective and targeted treatment strategies, paving the way for a new standard of care in cancer management.

Supplementary Materials

The following are available online at https://www.mdpi.com/article/10.3390/onco4040031/s1, Figure S1: A higher number of missense, nonsense, frameshift deletion, 5′ Flank and splice mutations predicts a longer survival following ICB. Figure S2: In lung cancer, mutations in ABL1, ASXL1, EPHA3, EPHA5, ERBB4, MET, MRE11A, MSH2, NOTCH1, PAK7, PAX5, PGR, ZFHX3, PIK3C3 and REL genes predict response to ICB. Figure S3: In melanoma, mutations in FBXW7, NOTCH1, CREBBP and other genes predict response and resistance following ICB. Figure S4: Impact of histology on response and resistance following ICB in melanoma patients. Figure S5: In bladder cancer, mutations in ERBB2, GNAS, HRAS, ATM, CDKN2A, LATS1, NCOR1, SMARC4 and TP53 genes predict response and resistance following ICB. Figure S6: In renal carcinoma, mutations in B2M, BCOR, JUN, FH, IGF1R, MYCN and VHL genes predict response and resistance following ICB. Figure S7: In head and neck cancer, mutations in ANKRD11, ERCC5, FANCA, KMT2A, WT1, HIST1H1C, KRAS, PIK2CA, PTRT, TERT, TP53 and TSC1 genes predict response and resistance following ICB. Figure S8: Characterization of the mutations driving the prediction of the response and resistance by machine learning in head and neck cancer patients.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

All patient and mutation data are available on cBioPortal [28] and in Supplementary Table S1 and Supplementary Data. https://www.cbioportal.org/study/summary?id = tmb_mskcc_2018 (accessed on 1 September 2024) Code is available on GitHub https://github.com/gmestrallet/Cancer_mutation (accessed on 1 September 2024).

Acknowledgments

I thank Paul Fremont for our helpful discussions.

Conflicts of Interest

The author declares no conflicts of interest.

Abbreviations

ABL1	Abelson Tyrosine Kinase 1
ARID2	AT-Rich Interaction Domain 2
ARID5B	AT-Rich Interaction Domain 5B
ASXL1	Additional Sex Combs Like 1
ATM	ATM Serine/Threonine Kinase
B2M	Beta-2-Microglobulin
BCOR	BCL6 Corepressor
CDK12	Cyclin Dependent Kinase 12
CDKN2A	Cyclin Dependent Kinase Inhibitor 2A
CDKN2C	Cyclin Dependent Kinase Inhibitor 2C
CREBBP	CREB Binding Protein
CTNNB1	Catenin Beta 1
EPHA3	EPH Receptor A3
EPHA5	EPH Receptor A5
ERBB2	Erb-B2 Receptor Tyrosine Kinase 2
ERBB4	Erb-B2 Receptor Tyrosine Kinase 4
FAM46C	Family With Sequence Similarity 46 Member C
FBXW7	F-Box and WD Repeat Domain Containing 7
FH	Fumarate Hydratase
GNAS	GNAS Complex Locus
HIST1H3I	Histone Cluster 1 H3 Family Member I
HIST1H1C	Histone Cluster 1 H1 Family Member C
HRAS	HRas Proto-Oncogene, GTPase
IGF1R	Insulin-Like Growth Factor 1 Receptor
JUN	Jun Proto-Oncogene, AP-1 Transcription Factor Subunit
KRAS	KRAS Proto-Oncogene, GTPase
LATS1	Large Tumor Suppressor Kinase 1
MAX	MYC Associated Factor X
MET	MET Proto-Oncogene, Receptor Tyrosine Kinase
MGA	MGA, MAX Dimerization Protein
MRE11A	MRE11 Homolog A
MSH2	MutS Homolog 2
MYCN	MYCN Proto-Oncogene, BHLH Transcription Factor
NCOR1	Nuclear Receptor Corepressor 1
NOTCH1	NOTCH Receptor 1
PAX5	Paired Box 5
PAK7	P21 (RAC1) Activated Kinase 7
PIK3CA	Phosphatidylinositol-4,5-Bisphosphate 3-Kinase Catalytic Subunit Alpha
PIK3C3	Phosphatidylinositol 3-Kinase Catalytic Subunit Type 3
PIK3R1	Phosphoinositide-3-Kinase Regulatory Subunit 1
PMS1	PMS1 Homolog 1
PMS2	PMS1 Homolog 2
PGR	Progesterone Receptor
RAD51	RAD51 Recombinase
REL	REL Proto-Oncogene, NF-KB Subunit
RICTOR	RPTOR Independent Companion Of MTOR Complex 2
RHOA	Ras Homolog Family Member A
RB1	Retinoblastoma 1
SMAD2	SMAD Family Member 2
SMAD4	SMAD Family Member 4
SMARC4	SWI/SNF Related, Matrix Associated, Actin Dependent Regulator of Chromatin, Subfamily A, Member 4
TERT	Telomerase Reverse Transcriptase
TP53	Tumor Protein P53
TGFBR2	Transforming Growth Factor Beta Receptor 2
TSC1	TSC Complex Subunit 1
VHL	Von Hippel-Lindau Tumor Suppressor
WT1	Wilms Tumor 1
ZFHX3	Zinc Finger Homeobox 3

References

Li, C.; Lei, S.; Ding, L.; Xu, Y.; Wu, X.; Wang, H.; Zhang, Z.; Gao, T.; Zhang, Y.; Li, L. Global Burden and Trends of Lung Cancer Incidence and Mortality. Chin. Med. J. 2023, 136, 1583–1590. [Google Scholar] [CrossRef] [PubMed]
Liao, G.; Wang, P.; Wang, Y. Identification of the Prognosis Value and Potential Mechanism of Immune Checkpoints in Renal Clear Cell Carcinoma Microenvironment. Front. Oncol. 2021, 11, 720125. [Google Scholar] [CrossRef]
Bray, F.; Ferlay, J.; Soerjomataram, I.; Siegel, R.L.; Torre, L.A.; 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. [Google Scholar] [CrossRef]
Lugand, L.; Mestrallet, G.; Laboureur, R.; Dumont, C.; Bouhidel, F.; Djouadou, M.; Masson-Lecomte, A.; Desgrandchamps, F.; Culine, S.; Carosella, E.D.; et al. Methods for Establishing a Renal Cell Carcinoma Tumor Spheroid Model With Immune Infiltration for Immunotherapeutic Studies. Front. Oncol. 2022, 12, 898732. [Google Scholar] [CrossRef] [PubMed]
McDermott, D.F.; Huseni, M.A.; Atskins, M.B.; Motzer, R.J.; Rini, B.I.; Escudier, B.; Fong, L.; Joseph, R.W.; Pal, S.K.; Reeves, J.A.; et al. Clinical Activity and Molecular Correlates of Response to Atezolizumab Alone or in Combination with Bevacizumab versus Sunitinib in Renal Cell Carcinoma. Nat. Med. 2018, 24, 749–757. [Google Scholar] [CrossRef] [PubMed]
Choueiri, T.K.; Powles, T.; Burotto, M.; Escudier, B.; Bourlon, M.T.; Zurawski, B.; Oyervides Juárez, V.M.; Hsieh, J.J.; Basso, U.; Shah, A.Y.; et al. Nivolumab plus Cabozantinib versus Sunitinib for Advanced Renal-Cell Carcinoma. N. Engl. J. Med. 2021, 384, 829–841. [Google Scholar] [CrossRef]
Miao, X.; Xu, R.; Fan, B.; Chen, J.; Li, X.; Mao, W.; Hua, S.; Li, B. PD-L1 Reverses Depigmentation in Pmel-1 Vitiligo Mice by Increasing the Abundance of Tregs in the Skin. Sci. Rep. 2018, 8, 1605. [Google Scholar] [CrossRef]
Mamdani, H.; Matosevic, S.; Khalid, A.B.; Durm, G.; Jalal, S.I. Immunotherapy in Lung Cancer: Current Landscape and Future Directions. Front. Immunol. 2022, 13, 823618. [Google Scholar] [CrossRef]
Mestrallet, G.; Sone, K.; Bhardwaj, N. Strategies to Overcome DC Dysregulation in the Tumor Microenvironment. Front. Immunol. 2022, 13, 980709. [Google Scholar] [CrossRef]
Mestrallet, G.; Rouas-Freiss, N.; LeMaoult, J.; Fortunel, N.O.; Martin, M.T. Skin Immunity and Tolerance: Focus on Epidermal Keratinocytes Expressing HLA-G. Front. Immunol. 2021, 12, 772516. [Google Scholar] [CrossRef]
Payne, K.K.; Mine, J.A.; Biswas, S.; Chaurio, R.A.; Perales-Puchalt, A.; Anadon, C.M.; Costich, T.L.; Harro, C.M.; Walrath, J.; Ming, Q.; et al. BTN3A1 Governs Antitumor Responses by Coordinating Aβ and Γδ T Cells. Science 2020, 369, 942–949. [Google Scholar] [CrossRef] [PubMed]
Fröhlich, A.; Loick, S.; Bawden, E.G.; Fietz, S.; Dietrich, J.; Diekmann, E.; Saavedra, G.; Fröhlich, H.; Niebel, D.; Sirokay, J.; et al. Comprehensive Analysis of Tumor Necrosis Factor Receptor TNFRSF9 (4-1BB) DNA Methylation with Regard to Molecular and Clinicopathological Features, Immune Infiltrates, and Response Prediction to Immunotherapy in Melanoma. eBioMedicine 2020, 52, 102647. [Google Scholar] [CrossRef]
Fan, C.; Hu, H.; Shen, Y.; Wang, Q.; Mao, Y.; Ye, B.; Xiang, M. PRF1 Is a Prognostic Marker and Correlated with Immune Infiltration in Head and Neck Squamous Cell Carcinoma. Transl. Oncol. 2021, 14, 101042. [Google Scholar] [CrossRef] [PubMed]
Hendriks, J.; Gravestein, L.A.; Tesselaar, K.; van Lier, R.A.W.; Schumacher, T.N.M.; Borst, J. CD27 Is Required for Generation and Long-Term Maintenance of T Cell Immunity. Nat. Immunol. 2000, 1, 433–440. [Google Scholar] [CrossRef]
Cekic, C.; Linden, J. Adenosine A2A Receptors Intrinsically Regulate CD8+ T Cells in the Tumor Microenvironment. Cancer Res. 2014, 74, 7239–7249. [Google Scholar] [CrossRef] [PubMed]
Bronte, V.; Serafini, P.; De Santo, C.; Marigo, I.; Tosello, V.; Mazzoni, A.; Segal, D.M.; Staib, C.; Lowel, M.; Sutter, G.; et al. IL-4-Induced Arginase 1 Suppresses Alloreactive T Cells in Tumor-Bearing Mice. J. Immunol. 2003, 170, 270–278. [Google Scholar] [CrossRef]
Mantovani, A.; Barajon, I.; Garlanda, C. IL-1 and IL-1 Regulatory Pathways in Cancer Progression and Therapy. Immunol. Rev. 2018, 281, 57–61. [Google Scholar] [CrossRef]
Wei, S.C.; Duffy, C.R.; Allison, J.P. Fundamental Mechanisms of Immune Checkpoint Blockade Therapy. Cancer Discov. 2018, 8, 1069–1086. [Google Scholar] [CrossRef]
Pardoll, D.M. The Blockade of Immune Checkpoints in Cancer Immunotherapy. Nat. Rev. Cancer 2012, 12, 252–264. [Google Scholar] [CrossRef]
Rouas-Freiss, N.; LeMaoult, J.; Verine, J.; Tronik-Le Roux, D.; Culine, S.; Hennequin, C.; Desgrandchamps, F.; Carosella, E.D. Intratumor Heterogeneity of Immune Checkpoints in Primary Renal Cell Cancer: Focus on HLA-G/ILT2/ILT4. Oncoimmunology 2017, 6, e1342023. [Google Scholar] [CrossRef]
Parikh, M.; Bajwa, P. Immune Checkpoint Inhibitors in the Treatment of Renal Cell Carcinoma. Semin. Nephrol. 2020, 40, 76–85. [Google Scholar] [CrossRef] [PubMed]
Mestrallet, G. Predicting Immunotherapy Outcomes in Glioblastoma Patients through Machine Learning. Cancers 2024, 16, 408. [Google Scholar] [CrossRef]
Sung, J.-Y.; Cheong, J.-H. Machine Learning Predictor of Immune Checkpoint Blockade Response in Gastric Cancer. Cancers 2022, 14, 3191. [Google Scholar] [CrossRef]
Tonneau, M.; Phan, K.; Manem, V.S.K.; Low-Kam, C.; Dutil, F.; Kazandjian, S.; Vanderweyen, D.; Panasci, J.; Malo, J.; Coulombe, F.; et al. Generalization Optimizing Machine Learning to Improve CT Scan Radiomics and Assess Immune Checkpoint Inhibitors’ Response in Non-Small Cell Lung Cancer: A Multicenter Cohort Study. Front. Oncol. 2023, 13, 1196414. [Google Scholar] [CrossRef]
Wiesweg, M.; Mairinger, F.; Reis, H.; Goetz, M.; Walter, R.F.H.; Hager, T.; Metzenmacher, M.; Eberhardt, W.E.E.; McCutcheon, A.; Köster, J.; et al. Machine Learning-Based Predictors for Immune Checkpoint Inhibitor Therapy of Non-Small-Cell Lung Cancer. Ann. Oncol. 2019, 30, 655–657. [Google Scholar] [CrossRef] [PubMed]
Mestrallet, G. Prediction of Glioma Resistance to Immune Checkpoint Inhibitors Based on Mutation Profile. Neuroglia 2024, 5, 145–154. [Google Scholar] [CrossRef]
Mestrallet, G. Predicting Resistance to Immunotherapy in Melanoma, Glioblastoma, Renal, Stomach and Bladder Cancers by Machine Learning on Immune Profiles. Onco 2024, 4, 192–206. [Google Scholar] [CrossRef]
Samstein, R.M.; Lee, C.-H.; Shoushtari, A.N.; Hellmann, M.D.; Shen, R.; Janjigian, Y.Y.; Barron, D.A.; Zehir, A.; Jordan, E.J.; Omuro, A.; et al. Tumor Mutational Load Predicts Survival after Immunotherapy across Multiple Cancer Types. Nat. Genet. 2019, 51, 202–206. [Google Scholar] [CrossRef]
Mestrallet, G. Software Development for Severe Burn Diagnosis and Autologous Skin Substitute Production. Comput. Methods Programs Biomed. Update 2022, 2, 100069. [Google Scholar] [CrossRef]
Mestrallet, G. Software Development to Estimate Herd Profitability According to Nutrition Parameters. agriRxiv 2023, 2023, 20230009899. [Google Scholar] [CrossRef]
Mirdita, M.; Schütze, K.; Moriwaki, Y.; Heo, L.; Ovchinnikov, S.; Steinegger, M. ColabFold: Making Protein Folding Accessible to All. Nat. Methods 2022, 19, 679–682. [Google Scholar] [CrossRef] [PubMed]
Mestrallet, G.; Brown, M.; Bozkus, C.C.; Bhardwaj, N. Immune Escape and Resistance to Immunotherapy in Mismatch Repair Deficient Tumors. Front. Immunol. 2023, 14, 1210164. [Google Scholar] [CrossRef] [PubMed]
Grady, W.M.; Myeroff, L.L.; Swinler, S.E.; Rajput, A.; Thiagalingam, S.; Lutterbaugh, J.D.; Neumann, A.; Brattain, M.G.; Chang, J.; Kim, S.J.; et al. Mutational Inactivation of Transforming Growth Factor Beta Receptor Type II in Microsatellite Stable Colon Cancers. Cancer Res. 1999, 59, 320–324. [Google Scholar] [PubMed]
Holstege, H.; Joosse, S.A.; van Oostrom, C.T.M.; Nederlof, P.M.; de Vries, A.; Jonkers, J. High Incidence of Protein-Truncating TP53 Mutations in BRCA1-Related Breast Cancer. Cancer Res. 2009, 69, 3625–3633. [Google Scholar] [CrossRef] [PubMed]
Lindenbergh-van der Plas, M.; Brakenhoff, R.H.; Kuik, D.J.; Buijze, M.; Bloemena, E.; Snijders, P.J.F.; Leemans, C.R.; Braakhuis, B.J.M. Prognostic Significance of Truncating TP53 Mutations in Head and Neck Squamous Cell Carcinoma. Clin. Cancer Res. Off. J. Am. Assoc. Cancer Res. 2011, 17, 3733–3741. [Google Scholar] [CrossRef]
Kawaguchi, T.; Kato, S.; Otsuka, K.; Watanabe, G.; Kumabe, T.; Tominaga, T.; Yoshimoto, T.; Ishioka, C. The Relationship among P53 Oligomer Formation, Structure and Transcriptional Activity Using a Comprehensive Missense Mutation Library. Oncogene 2005, 24, 6976–6981. [Google Scholar] [CrossRef]
DiGiammarino, E.L.; Lee, A.S.; Cadwell, C.; Zhang, W.; Bothner, B.; Ribeiro, R.C.; Zambetti, G.; Kriwacki, R.W. A Novel Mechanism of Tumorigenesis Involving pH-Dependent Destabilization of a Mutant P53 Tetramer. Nat. Struct. Biol. 2002, 9, 12–16. [Google Scholar] [CrossRef]
Bullock, A.N.; Fersht, A.R. Rescuing the Function of Mutant P53. Nat. Rev. Cancer 2001, 1, 68–76. [Google Scholar] [CrossRef]
Shirole, N.H.; Pal, D.; Kastenhuber, E.R.; Senturk, S.; Boroda, J.; Pisterzi, P.; Miller, M.; Munoz, G.; Anderluh, M.; Ladanyi, M.; et al. TP53 Exon-6 Truncating Mutations Produce Separation of Function Isoforms with pro-Tumorigenic Functions. eLife 2016, 5, e17929. [Google Scholar] [CrossRef]
D’Urso, C.M.; Wang, Z.G.; Cao, Y.; Tatake, R.; Zeff, R.A.; Ferrone, S. Lack of HLA Class I Antigen Expression by Cultured Melanoma Cells FO-1 Due to a Defect in B2m Gene Expression. J. Clin. Investig. 1991, 87, 284–292. [Google Scholar] [CrossRef]
Challa-Malladi, M.; Lieu, Y.K.; Califano, O.; Holmes, A.B.; Bhagat, G.; Murty, V.V.; Dominguez-Sola, D.; Pasqualucci, L.; Dalla-Favera, R. Combined Genetic Inactivation of Β2-Microglobulin and CD58 Reveals Frequent Escape from Immune Recognition in Diffuse Large B Cell Lymphoma. Cancer Cell 2011, 20, 728–740. [Google Scholar] [CrossRef] [PubMed]
Sucker, A.; Zhao, F.; Real, B.; Heeke, C.; Bielefeld, N.; Maβen, S.; Horn, S.; Moll, I.; Maltaner, R.; Horn, P.A.; et al. Genetic Evolution of T-Cell Resistance in the Course of Melanoma Progression. Clin. Cancer Res. Off. J. Am. Assoc. Cancer Res. 2014, 20, 6593–6604. [Google Scholar] [CrossRef] [PubMed]
Astolfi, A.; Fiore, M.; Melchionda, F.; Indio, V.; Bertuccio, S.N.; Pession, A. BCOR Involvement in Cancer. Epigenomics 2019, 11, 835–855. [Google Scholar] [CrossRef] [PubMed]
Ren, F.; Cai, X.; Yao, Y.; Fang, G. JunB: A Paradigm for Jun Family in Immune Response and Cancer. Front. Cell. Infect. Microbiol. 2023, 13, 1222265. [Google Scholar] [CrossRef] [PubMed]
Liu, R.; Shi, P.; Wang, Z.; Yuan, C.; Cui, H. Molecular Mechanisms of MYCN Dysregulation in Cancers. Front. Oncol. 2021, 10, 625332. [Google Scholar] [CrossRef]
Chen, Y.-B.; Brannon, A.R.; Toubaji, A.; Dudas, M.E.; Won, H.H.; Al-Ahmadie, H.A.; Fine, S.W.; Gopalan, A.; Frizzell, N.; Voss, M.H.; et al. Hereditary Leiomyomatosis and Renal Cell Carcinoma Syndrome-Associated Renal Cancer: Recognition of the Syndrome by Pathologic Features and the Utility of Detecting Aberrant Succination by Immunohistochemistry. Am. J. Surg. Pathol. 2014, 38, 627–637. [Google Scholar] [CrossRef]
Wang, P.; Mak, V.C.Y.; Cheung, L.W.T. Drugging IGF-1R in Cancer: New Insights and Emerging Opportunities. Genes Dis. 2022, 10, 199–211. [Google Scholar] [CrossRef]

Figure 1. Mutations associated with a difference in overall survival for patients with lung cancer following immunotherapy. N = 350 lung cancer patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05.

Figure 2. Mutations associated with a difference in overall survival for patients with melanoma following immunotherapy. N = 320 melanoma patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05.

Figure 3. Mutations associated with a difference in overall survival for patients with bladder cancer following immunotherapy. N = 215 bladder cancer patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05.

Figure 4. Mutations associated with a difference in overall survival for patients with renal carcinoma following immunotherapy. N = 151 renal carcinoma patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05.

Figure 5. Mutations associated with a difference in overall survival for patients with head and neck cancer following immunotherapy. N = 139 head and neck cancer patients. I investigated patient response or resistance to immune checkpoint blockade (ICB) based on mutation profiles. Only mutations included in the MSK IMPACT panel that showed statistically significant differences in survival were shown. I employed Cox regression and hazard ratios, with significance set at p-value < 0.05.

Figure 6. Profiling of cancer patient mutations enables the prediction of their overall survival following ICB. (A) N = 350 lung cancer patients. (B) N = 320 melanoma patients. (C) N = 215 bladder cancer patients. (D) N = 151 renal carcinoma patients. (E) N = 139 head and neck cancer patients. Machine learning models were employed to predict patient overall survival in months after immune checkpoint blockade using previously identified mutational features. The dataset was partitioned into five different subsets and further split into training (80% of the patients) and testing (20% of the patients) subsets. Various algorithms, including Gradient Boosting, Random Forest, Decision Tree, Logistic Regression, Support Vector Classifier (SVC) and Multi-layer Perceptron (MLP), were trained on the mutational features using five-fold cross-validation. Additionally, a Mean Ensemble model combining the best-performing Random Forest and Gradient Boosting models was utilized to improve the accuracy of survival predictions. Hyperparameters of the models were optimized and their performance was evaluated using standard metrics on the test sets.

Figure 7. Shapley values for each model predicting overall survival following ICB. (A) N = 350 lung cancer patients. (B) N = 320 melanoma patients. (C) N = 215 bladder cancer patients. (D) N = 151 renal carcinoma patients. (E) N = 139 head and neck cancer patients.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2024 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Mestrallet, G. Leveraging Tumor Mutation Profiles to Forecast Immune Checkpoint Blockade Resistance in Melanoma, Lung, Head and Neck, Bladder and Renal Cancers. Onco 2024, 4, 439-457. https://doi.org/10.3390/onco4040031

AMA Style

Mestrallet G. Leveraging Tumor Mutation Profiles to Forecast Immune Checkpoint Blockade Resistance in Melanoma, Lung, Head and Neck, Bladder and Renal Cancers. Onco. 2024; 4(4):439-457. https://doi.org/10.3390/onco4040031

Chicago/Turabian Style

Mestrallet, Guillaume. 2024. "Leveraging Tumor Mutation Profiles to Forecast Immune Checkpoint Blockade Resistance in Melanoma, Lung, Head and Neck, Bladder and Renal Cancers" Onco 4, no. 4: 439-457. https://doi.org/10.3390/onco4040031

APA Style

Mestrallet, G. (2024). Leveraging Tumor Mutation Profiles to Forecast Immune Checkpoint Blockade Resistance in Melanoma, Lung, Head and Neck, Bladder and Renal Cancers. Onco, 4(4), 439-457. https://doi.org/10.3390/onco4040031

Article Menu