Artificial intelligence.

Experts discuss the challenges and opportunities of using artificial intelligence (AI) to study the evolution of cancer cells and their microenvironment, improve diagnosis, predict treatment response, and ensure responsible implementation in the clinic.

LORIS robustly predicts patient outcomes with immune checkpoint blockade therapy using common clinical, pathologic and genomic features.

Despite the revolutionary impact of immune checkpoint blockade (ICB) in cancer treatment, accurately predicting patient responses remains challenging. Here, we analyzed a large dataset of 2,881 ICB-treated and 841 non-ICB-treated patients across 18 solid tumor types, encompassing a wide range of clinical, pathologic and genomic features. We developed a clinical score called LORIS (logistic regression-based immunotherapy-response score) using a six-feature logistic regression model. LORIS outperforms previous signatures in predicting ICB response and identifying responsive patients even with low tumor mutational burden or programmed cell death 1 ligand 1 expression. LORIS consistently predicts patient objective response and short-term and long-term survival across most cancer types. Moreover, LORIS showcases a near-monotonic relationship with ICB response probability and patient survival, enabling precise patient stratification. As an accurate, interpretable method using a few readily measurable features, LORIS may help improve clinical decision-making in precision medicine to maximize patient benefit. LORIS is available as an online tool at https://loris.ccr.cancer.gov/ .

The electrostatic landscape of MHC-peptide binding revealed using inception networks.

Predictive modeling of macromolecular recognition and protein-protein complementarity represents one of the cornerstones of biophysical sciences. However, such models are often hindered by the combinatorial complexity of interactions at the molecular interfaces. Exemplary of this problem is peptide presentation by the highly polymorphic major histocompatibility complex class I (MHC-I) molecule, a principal component of immune recognition. We developed human leukocyte antigen (HLA)-Inception, a deep biophysical convolutional neural network, which integrates molecular electrostatics to capture non-bonded interactions for predicting peptide binding motifs across 5,821 MHC-I alleles. These predictions of generated motifs correlate strongly with experimental peptide binding and presentation data. Beyond molecular interactions, the study demonstrates the application of predicted motifs in analyzing MHC-I allele associations with HIV disease progression and patient response to immune checkpoint inhibitors. A record of this paper's transparent peer review process is included in the supplemental information.

An immunogenetic basis for lung cancer risk.

Cancer risk is influenced by inherited mutations, DNA replication errors, and environmental factors. However, the influence of genetic variation in immunosurveillance on cancer risk is not well understood. Leveraging population-level data from the UK Biobank and FinnGen, we show that heterozygosity at the human leukocyte antigen (HLA)-II loci is associated with reduced lung cancer risk in smokers. Fine-mapping implicated amino acid heterozygosity in the HLA-II peptide binding groove in reduced lung cancer risk, and single-cell analyses showed that smoking drives enrichment of proinflammatory lung macrophages and HLA-II+ epithelial cells. In lung cancer, widespread loss of HLA-II heterozygosity (LOH) favored loss of alleles with larger neopeptide repertoires. Thus, our findings nominate genetic variation in immunosurveillance as a critical risk factor for lung cancer.

Clinical-genomic determinants of immune checkpoint blockade response in head and neck squamous cell carcinoma.

BACKGROUNDRecurrent and/or metastatic (R/M) head and neck squamous cell carcinoma (HNSCC) is generally an incurable disease, with patients experiencing median survival of under 10 months and significant morbidity. While immune checkpoint blockade (ICB) drugs are effective in approximately 20% of patients, the remaining experience limited clinical benefit and are exposed to potential adverse effects and financial costs. Clinically approved biomarkers, such as tumor mutational burden (TMB), have a modest predictive value in HNSCC.METHODSWe analyzed clinical and genomic features, generated using whole-exome sequencing, in 133 ICB-treated patients with R/M HNSCC, of whom 69 had virus-associated and 64 had non-virus-associated tumors.RESULTSHierarchical clustering of genomic data revealed 6 molecular subtypes characterized by a wide range of objective response rates and survival after ICB therapy. The prognostic importance of these 6 subtypes was validated in an external cohort. A random forest-based predictive model, using several clinical and genomic features, predicted progression-free survival (PFS), overall survival (OS), and response with greater accuracy than did a model based on TMB alone. Recursive partitioning analysis identified 3 features (systemic inflammatory response index, TMB, and smoking signature) that classified patients into risk groups with accurate discrimination of PFS and OS.CONCLUSIONThese findings shed light on the immunogenomic characteristics of HNSCC tumors that drive differential responses to ICB and identify a clinical-genomic classifier that outperformed the current clinically approved biomarker of TMB. This validated predictive tool may help with clinical risk stratification in patients with R/M HNSCC for whom ICB is being considered.FUNDINGFundación Alfonso Martín Escudero, NIH R01 DE027738, US Department of Defense CA210784, The Geoffrey Beene Cancer Research Center, The MSKCC Population Science Research Program, the Jayme Flowers Fund, the Sebastian Nativo Fund, and the NIH/NCI Cancer Center Support Grant P30 CA008748.

Immune selection determines tumor antigenicity and influences response to checkpoint inhibitors.

In cancer, evolutionary forces select for clones that evade the immune system. Here we analyzed >10,000 primary tumors and 356 immune-checkpoint-treated metastases using immune dN/dS, the ratio of nonsynonymous to synonymous mutations in the immunopeptidome, to measure immune selection in cohorts and individuals. We classified tumors as immune edited when antigenic mutations were removed by negative selection and immune escaped when antigenicity was covered up by aberrant immune modulation. Only in immune-edited tumors was immune predation linked to CD8 T cell infiltration. Immune-escaped metastases experienced the best response to immunotherapy, whereas immune-edited patients did not benefit, suggesting a preexisting resistance mechanism. Similarly, in a longitudinal cohort, nivolumab treatment removes neoantigens exclusively in the immunopeptidome of nonimmune-edited patients, the group with the best overall survival response. Our work uses dN/dS to differentiate between immune-edited and immune-escaped tumors, measuring potential antigenicity and ultimately helping predict response to treatment.

Validation of a Machine Learning Model to Predict Immunotherapy Response in Head and Neck Squamous Cell Carcinoma.

Head and neck squamous-cell carcinoma (HNSCC) is a disease with a generally poor prognosis; half of treated patients eventually develop recurrent and/or metastatic (R/M) disease. Patients with R/M HNSCC generally have incurable disease with a median survival of 10 to 15 months. Although immune-checkpoint blockade (ICB) has improved outcomes in patients with R/M HNSCC, identifying patients who are likely to benefit from ICB remains a challenge. Biomarkers in current clinical use include tumor mutational burden and immunohistochemistry for programmed death-ligand 1, both of which have only modest predictive power. Machine learning (ML) has the potential to aid in clinical decision-making as an approach to estimate a tumor's likelihood of response or a patient's likelihood of experiencing clinical benefit from therapies such as ICB. Previously, we described a random forest ML model that had value in predicting ICB response using 11 or 16 clinical, laboratory, and genomic features in a pan-cancer development cohort. However, its applicability to certain cancer types, such as HNSCC, has been unknown, due to a lack of cancer-type-specific validation. Here, we present the first validation of a random forest ML tool to predict the likelihood of ICB response in patients with R/M HNSCC. The tool had adequate predictive power for tumor response (area under the receiver operating characteristic curve = 0.65) and was able to stratify patients by overall (HR = 0.53 [95% CI 0.29-0.99], p = 0.045) and progression-free (HR = 0.49 [95% CI 0.27-0.87], p = 0.016) survival. The overall accuracy was 0.72. Our study validates an ML predictor in HNSCC, demonstrating promising performance in a novel cohort of patients. Further studies are needed to validate the generalizability of this algorithm in larger patient samples from additional multi-institutional contexts.

Unsupervised and supervised AI on molecular dynamics simulations reveals complex characteristics of HLA-A2-peptide immunogenicity.

Immunologic recognition of peptide antigens bound to class I major histocompatibility complex (MHC) molecules is essential to both novel immunotherapeutic development and human health at large. Current methods for predicting antigen peptide immunogenicity rely primarily on simple sequence representations, which allow for some understanding of immunogenic features but provide inadequate consideration of the full scale of molecular mechanisms tied to peptide recognition. We here characterize contributions that unsupervised and supervised artificial intelligence (AI) methods can make toward understanding and predicting MHC(HLA-A2)-peptide complex immunogenicity when applied to large ensembles of molecular dynamics simulations. We first show that an unsupervised AI method allows us to identify subtle features that drive immunogenicity differences between a cancer neoantigen and its wild-type peptide counterpart. Next, we demonstrate that a supervised AI method for class I MHC(HLA-A2)-peptide complex classification significantly outperforms a sequence model on small datasets corrected for trivial sequence correlations. Furthermore, we show that both unsupervised and supervised approaches reveal determinants of immunogenicity based on time-dependent molecular fluctuations and anchor position dynamics outside the MHC binding groove. We discuss implications of these structural and dynamic immunogenicity correlates for the induction of T cell responses and therapeutic T cell receptor design.

NKG2A and HLA-E define an alternative immune checkpoint axis in bladder cancer.

Programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1)-blockade immunotherapies have limited efficacy in the treatment of bladder cancer. Here, we show that NKG2A associates with improved survival and responsiveness to PD-L1 blockade immunotherapy in bladder tumors that have high abundance of CD8+ T cells. In bladder tumors, NKG2A is acquired on CD8+ T cells later than PD-1 as well as other well-established immune checkpoints. NKG2A+ PD-1+ CD8+ T cells diverge from classically defined exhausted T cells through their ability to react to human leukocyte antigen (HLA) class I-deficient tumors using T cell receptor (TCR)-independent innate-like mechanisms. HLA-ABC expression by bladder tumors is progressively diminished as disease progresses, framing the importance of targeting TCR-independent anti-tumor functions. Notably, NKG2A+ CD8+ T cells are inhibited when HLA-E is expressed by tumors and partly restored upon NKG2A blockade in an HLA-E-dependent manner. Overall, our study provides a framework for subsequent clinical trials combining NKG2A blockade with other T cell-targeted immunotherapies, where tumors express higher levels of HLA-E.

Functional landscapes of POLE and POLD1 mutations in checkpoint blockade-dependent antitumor immunity.

Defects in pathways governing genomic fidelity have been linked to improved response to immune checkpoint blockade therapy (ICB). Pathogenic POLE/POLD1 mutations can cause hypermutation, yet how diverse mutations in POLE/POLD1 influence antitumor immunity following ICB is unclear. Here, we comprehensively determined the effect of POLE/POLD1 mutations in ICB and elucidated the mechanistic impact of these mutations on tumor immunity. Murine syngeneic tumors harboring Pole/Pold1 functional mutations displayed enhanced antitumor immunity and were sensitive to ICB. Patients with POLE/POLD1 mutated tumors harboring telltale mutational signatures respond better to ICB than patients harboring wild-type or signature-negative tumors. A mutant POLE/D1 function-associated signature-based model outperformed several traditional approaches for identifying POLE/POLD1 mutated patients that benefit from ICB. Strikingly, the spectrum of mutational signatures correlates with the biochemical features of neoantigens. Alterations that cause POLE/POLD1 function-associated signatures generate T cell receptor (TCR)-contact residues with increased hydrophobicity, potentially facilitating T cell recognition. Altogether, the functional landscapes of POLE/POLD1 mutations shape immunotherapy efficacy.

A Targetable Myeloid Inflammatory State Governs Disease Recurrence in Clear-Cell Renal Cell Carcinoma.

It is poorly understood how the tumor immune microenvironment influences disease recurrence in localized clear-cell renal cell carcinoma (ccRCC). Here we performed whole-transcriptomic profiling of 236 tumors from patients assigned to the placebo-only arm of a randomized, adjuvant clinical trial for high-risk localized ccRCC. Unbiased pathway analysis identified myeloid-derived IL6 as a key mediator. Furthermore, a novel myeloid gene signature strongly correlated with disease recurrence and overall survival on uni- and multivariate analyses and is linked to TP53 inactivation across multiple data sets. Strikingly, effector T-cell gene signatures, infiltration patterns, and exhaustion markers were not associated with disease recurrence. Targeting immunosuppressive myeloid inflammation with an adenosine A2A receptor antagonist in a novel, immunocompetent, Tp53-inactivated mouse model significantly reduced metastatic development. Our findings suggest that myeloid inflammation promotes disease recurrence in ccRCC and is targetable as well as provide a potential biomarker-based framework for the design of future immuno-oncology trials in ccRCC. SIGNIFICANCE: Improved understanding of factors that influence metastatic development in localized ccRCC is greatly needed to aid accurate prediction of disease recurrence, clinical decision-making, and future adjuvant clinical trial design. Our analysis implicates intratumoral myeloid inflammation as a key driver of metastasis in patients and a novel immunocompetent mouse model. This article is highlighted in the In This Issue feature, p. 2221.

Pre-treatment serum albumin and mutational burden as biomarkers of response to immune checkpoint blockade.

The effects of cytokine and protein stabilizing carriers, such as serum albumin, on tumor response to immune checkpoint blockade (ICB) is not well understood. By examining 1714 patients across 16 cancer types, we found that high pretreatment serum albumin level predicts favorable tumor radiographic response following ICB treatment in a dose-dependent fashion. Serum albumin is a candidate biomarker that can be combined with tumor mutational burden (TMB) for additional predictive capacity, and the tumor response rate to ICB was ~49% in the albumin-high/TMB-high group.

Improved prediction of immune checkpoint blockade efficacy across multiple cancer types.

Only a fraction of patients with cancer respond to immune checkpoint blockade (ICB) treatment, but current decision-making procedures have limited accuracy. In this study, we developed a machine learning model to predict ICB response by integrating genomic, molecular, demographic and clinical data from a comprehensively curated cohort (MSK-IMPACT) with 1,479 patients treated with ICB across 16 different cancer types. In a retrospective analysis, the model achieved high sensitivity and specificity in predicting clinical response to immunotherapy and predicted both overall survival and progression-free survival in the test data across different cancer types. Our model significantly outperformed predictions based on tumor mutational burden, which was recently approved by the U.S. Food and Drug Administration for this purpose1. Additionally, the model provides quantitative assessments of the model features that are most salient for the predictions. We anticipate that this approach will substantially improve clinical decision-making in immunotherapy and inform future interventions.

Pharmacologic modulation of RNA splicing enhances anti-tumor immunity.

Although mutations in DNA are the best-studied source of neoantigens that determine response to immune checkpoint blockade, alterations in RNA splicing within cancer cells could similarly result in neoepitope production. However, the endogenous antigenicity and clinical potential of such splicing-derived epitopes have not been tested. Here, we demonstrate that pharmacologic modulation of splicing via specific drug classes generates bona fide neoantigens and elicits anti-tumor immunity, augmenting checkpoint immunotherapy. Splicing modulation inhibited tumor growth and enhanced checkpoint blockade in a manner dependent on host T cells and peptides presented on tumor MHC class I. Splicing modulation induced stereotyped splicing changes across tumor types, altering the MHC I-bound immunopeptidome to yield splicing-derived neoepitopes that trigger an anti-tumor T cell response in vivo. These data definitively identify splicing modulation as an untapped source of immunogenic peptides and provide a means to enhance response to checkpoint blockade that is readily translatable to the clinic.

High Response Rate and Durability Driven by HLA Genetic Diversity in Patients with Kidney Cancer Treated with Lenvatinib and Pembrolizumab.

Immune checkpoint blockade (ICB) therapy has substantially improved the outcomes of patients with many types of cancers, including renal cell carcinoma (RCC). Initially studied as monotherapy, immunotherapy-based combination regimens have improved the clinical benefit achieved by ICB monotherapy and have revolutionized RCC treatment. While biomarkers like PD-L1 and tumor mutational burden (TMB) are FDA approved as biomarkers for ICB monotherapy, there are no known biomarkers for combination immunotherapies. Here, we describe the clinical outcomes and genomic determinants of response from a phase Ib/II clinical trial on patients with advanced RCC evaluating the efficacy of lenvatinib, a multi-kinase inhibitor mainly targeting VEGFR and FGFR plus pembrolizumab, an anti-PD1 immunotherapy. Concurrent treatment with lenvatinib and pembrolizumab resulted in an objective response rate of 79% (19/24) and tumor shrinkage in 96% (23/24) of patients. While tumor mutational burden (TMB) did not predict for clinical benefit, germline HLA-I diversity strongly impacted treatment efficacy. Specifically, HLA-I evolutionary divergence (HED), which measures the breadth of a patient's immunopeptidome, was associated with both improved clinical benefit and durability of response. Our results identify lenvatinib plus pembrolizumab as a highly active treatment strategy in RCC and reveal HLA-I diversity as a critical determinant of efficacy for this combination. HED also predicted better survival in a separate cohort of patients with RCC following therapy with anti-PD-1-based combination therapy. IMPLICATIONS: These findings have substantial implications for RCC therapy and for understanding immunogenetic mechanisms of efficacy and warrants further investigation.

Single-cell sequencing links multiregional immune landscapes and tissue-resident T cells in ccRCC to tumor topology and therapy efficacy.

Clear cell renal cell carcinomas (ccRCCs) are highly immune infiltrated, but the effect of immune heterogeneity on clinical outcome in ccRCC has not been fully characterized. Here we perform paired single-cell RNA (scRNA) and T cell receptor (TCR) sequencing of 167,283 cells from multiple tumor regions, lymph node, normal kidney, and peripheral blood of two immune checkpoint blockade (ICB)-naïve and four ICB-treated patients to map the ccRCC immune landscape. We detect extensive heterogeneity within and between patients, with enrichment of CD8A+ tissue-resident T cells in a patient responsive to ICB and tumor-associated macrophages (TAMs) in a resistant patient. A TCR trajectory framework suggests distinct T cell differentiation pathways between patients responding and resistant to ICB. Finally, scRNA-derived signatures of tissue-resident T cells and TAMs are associated with response to ICB and targeted therapies across multiple independent cohorts. Our study establishes a multimodal interrogation of the cellular programs underlying therapeutic efficacy in ccRCC.

Putative Drivers of Aggressiveness in TCEB1-mutant Renal Cell Carcinoma: An Emerging Entity with Variable Clinical Course.

BACKGROUND: TCEB1-mutant renal cell carcinoma (RCC) is a rare variant of RCC with clear-cell features. Owing to its unique morphological and molecular features it has recently been proposed as a separate entity. Initial series suggested an indolent, early-stage phenotype. Here we expand our clinical cohort and describe a more detailed genomic analysis looking for potential drivers of aggressiveness. DESIGN, SETTING, AND PARTICIPANTS: We identified five new cases in our institutional sequencing cohort, four of whom were found to have high-stage disease (American Joint Committee on Cancer stage III/IV). Twelve previously reported cases were pooled for comparison purposes (Sato, The Cancer Genome Atlas, TRACERx Renal). OUTCOME MEASURES AND STATISTICAL ANALYSIS: We used our previously validated pipeline to analyze somatic mutations and copy number alterations (CNAs) in seven tumor samples with available data and estimated the number of cancer cells bearing each somatic mutation. The oncogenic potential of mutations was assessed using OncoKB and two other algorithms. Mann-Whitney U tests were used to evaluate differences in genomic markers between stage groups. RESULTS AND LIMITATIONS: All tumors showed biallelic inactivation of the TCEB1 gene according to a combination of somatic mutation and CNA analyses. Mutations were always found in residues involved in hydrophobic interactions with VHL. We found that high-stage tumors had additional oncogenic mutations (median 1, interquartile range [IQR] 1-1 vs 2, IQR 2-2; median difference 1, 95% confidence interval [CI] 1-1; p= 0.002) and showed whole-genome doubling events. They also seemed to have a higher burden of somatic CNAs (median fraction CNA genome 0.10, IQR 0.10-0.15 vs 0.63, IQR 0.58-0.68), however, this finding did not reach statistical significance (median difference 0.49, 95% CI 0.33-0.63; p=0.052). CONCLUSIONS: TCEB1-mutant RCC can show variable behavior ranging from very indolent to aggressive. Specific molecular events leading to high genomic instability seem to be associated with aggressiveness. This study expands the clinical spectrum of TCEB1-mutant RCC. PATIENT SUMMARY: We present four cases of aggressive TCEB1-mutant renal cell carcinoma, a rare type of kidney cancer. In-depth analysis of the genomes of these tumors revealed certain abnormalities that might explain this aggressive behavior.

Genetic and environmental determinants of human TCR repertoire diversity.

T cell discrimination of self and non-self is the foundation of the adaptive immune response, and is orchestrated by the interaction between T cell receptors (TCRs) and their cognate ligands presented by major histocompatibility (MHC) molecules. However, the impact of host immunogenetic variation on the diversity of the TCR repertoire remains unclear. Here, we analyzed a cohort of 666 individuals with TCR repertoire sequencing. We show that TCR repertoire diversity is positively associated with polymorphism at the human leukocyte antigen class I (HLA-I) loci, and diminishes with age and cytomegalovirus (CMV) infection. Moreover, our analysis revealed that HLA-I polymorphism and age independently shape the repertoire in healthy individuals. Our data elucidate key determinants of human TCR repertoire diversity, and suggest a mechanism underlying the evolutionary fitness advantage of HLA-I heterozygosity.

The Immune Microenvironment and Neoantigen Landscape of Aggressive Salivary Gland Carcinomas Differ by Subtype.

PURPOSE: Salivary gland carcinomas (SGC) are rare, aggressive cancers with high rates of recurrence and distant metastasis. These factors, and a lack of active systemic therapies, contribute to poor clinical outcome. Response rates with immune checkpoint blockade have been low, although clinical data remain sparse. To improve the efficacy of therapies, a more comprehensive understanding of relevant molecular alterations and immunologic processes is needed. EXPERIMENTAL DESIGN: To characterize the immune microenvironment and neoantigen landscape of SGCs, we performed RNA sequencing (RNA-seq) in 76 tumors representing the three most lethal histologies: adenoid cystic carcinoma (ACC), myoepithelial carcinoma (MECA), and salivary duct carcinoma (SDC). We analyzed transcriptomic profiles, tumor-infiltrating immune cell populations, and measures of T-cell activation/dysfunction. In 37 cases also undergoing exome sequencing, we analyzed somatic mutations and neoantigens. RESULTS: SDCs exhibited high levels of immune infiltration, with corresponding higher levels of T-cell dysfunction, and higher mutational load. In contrast, ACCs were characterized by an immune-excluded microenvironment, the presence of M2-polarized macrophages and myeloid-derived suppressor cells, and very low mutational load. MECAs were more heterogeneous, with both immune-low and immune-high phenotypes represented. Across all SGCs, levels of immune infiltration were associated with mutation- and fusion-derived neoantigens, and with aggressive clinical behavior. CONCLUSIONS: These findings provide new insights into the immune microenvironment and neoantigen landscape of SGCs, showing that mechanisms of immune escape appear to differ by histology. These data nominate potential immunologic vulnerabilities and may help guide the next steps of investigation in precision immunotherapy for these difficult-to-treat cancers.