Journal Club: Decoding Cancer's Blueprint - From Immunogenetic Risk to Single-Cell Regulatory Landscapes

· 10 min read · Paper Feed

Journal Club: Decoding Cancer's Blueprint - From Immunogenetic Risk to Single-Cell Regulatory Landscapes

Conceptual illustration: Decoding Cancer's Blueprint - From Immunogenetic Risk to Single-Cell Regulatory Landscapes

Introduction

Cancer initiation and progression are governed by a complex interplay of genetic mutations, environmental exposures, and the host's immune system. For decades, the central dogma of cancer biology has focused on driver mutations in protein-coding genes. A more nuanced picture is now emerging, one that integrates the role of immunosurveillance in cancer risk and the often noncoding regulatory rewiring that defines the malignant state. Understanding these layers matters for improving risk prediction, early detection, and therapeutic intervention.

This journal club synthesizes two 2024 Science papers that tackle these frontiers from complementary angles. The first, by Krishna et al., makes a population-scale genetic argument for adaptive immunity in cancer prevention, specifically linking HLA class II diversity to lung cancer risk in smokers. The second, by Sundaram et al., shifts to the epigenomic blueprint of established tumors, generating a single-cell chromatin accessibility atlas across eight cancer types and using deep learning to interpret the functional impact of noncoding mutations.

The common thread is a move beyond the exome. Both studies lean on large-scale genomic resources (biobanks and The Cancer Genome Atlas, TCGA) to interrogate non-coding regions (HLA loci, regulatory elements) and their functional consequences. Together, they offer a more integrated view of cancer etiology, spanning pre-neoplastic risk modulation to the regulatory logic of full-blown tumors.

Paper 1: HLA-II Heterozygosity as a Shield Against Lung Cancer

Key Findings

  1. HLA-II Heterozygosity Reduces Risk: Analysis of roughly 500,000 individuals from the UK Biobank and FinnGen found that heterozygosity at HLA class II loci (e.g., HLA-DRB1) is associated with a reduced risk of developing lung cancer. The protective effect held up over more than a decade of follow-up.
  2. Smoking-Dependent Protection: The benefit was specific to current and former smokers. Never-smokers showed no protective association, pointing to a mechanism tied to the presentation of smoking-derived neoantigens. The authors flag a striking lifetime figure: current smokers homozygous at HLA-DRB1 had a 13.9% risk of lung cancer in the UK Biobank cohort.
  3. Mechanistic Evidence from Tumors: Analysis of tumor genomes from TCGA, PCAWG, and the Hartwig Medical Foundation revealed widespread loss of heterozygosity (LOH) at HLA-II loci in lung adenocarcinomas and squamous cell carcinomas, at rates comparable to HLA-I LOH. The lost allele tended to be the one capable of presenting a larger repertoire of predicted neoantigens.
  4. Cellular Context from Single-Cell Data: Reanalysis of single-cell RNA-seq data from nonmalignant lungs showed that smoking drives enrichment of proinflammatory macrophages and induces HLA-II expression on lung epithelial cells, providing the cellular actors for potential CD4+ T cell surveillance.

Methodology & Study Design

  • Population Genetics: The team ran a genetic association study using genotype and longitudinal health data from two large biobanks (UK Biobank and FinnGen). They tested the association between HLA zygosity (homozygous vs. heterozygous) and lung cancer incidence, adjusting for polygenic risk scores and clinical factors.
  • Fine-Mapping: They went beyond gene-level zygosity to fine-map the effect to specific amino acid positions within the peptide-binding groove of HLA-II molecules, implicating functional peptide presentation diversity.
  • Multi-Cohort Genomic Analysis: Somatic HLA LOH was assessed in whole-exome and whole-genome sequencing data from three independent lung cancer cohorts, using tools like LOHHLA.
  • Integrative Single-Cell Analysis: Publicly available single-cell RNA-seq datasets of lung tissue were re-analyzed to characterize HLA-II expression and immune cell populations in smokers vs. non-smokers.

Significance

This is some of the strongest human population genetic evidence to date for the cancer immunosurveillance hypothesis. It repositions HLA beyond immunotherapy response and transplant matching, framing genetic variation in immunosurveillance as a fundamental component of cancer risk alongside smoking and inherited mutations. It also brings the HLA class II pathway and CD4+ T cells into focus for controlling early neoplasia, an area that has received less attention than HLA-I/CD8+ responses in oncology. The findings could eventually feed into risk-stratified lung cancer screening programs.

Paper 2: A Single-Cell Atlas of Cancer's Regulatory Genome

Key Findings

  1. A Pan-Cancer scATAC-seq Atlas: The study generated single-cell ATAC-seq data for 227,063 nuclei from 74 primary tumor samples across eight TCGA cancer types (e.g., breast, lung, colon), enabling deconvolution of malignant, stromal, and immune cell chromatin landscapes.
  2. Copy Number Alterations Shape Accessibility: Large-scale chromatin accessibility signals often mirror copy number alterations (CNAs). These megabase-scale accessibility patterns can be used to identify genetically distinct tumor subclones and reveal how subclone-specific CNAs affect transcription factor networks.
  3. Regulatory Identity of Cancer Subtypes: By comparing tumor cells to organ-matched healthy cell types, the authors defined each tumor's "nearest healthy" epigenetic state. A notable finding: the chromatin signature of basal-like breast cancer is most similar to secretory luminal epithelial cells rather than basal cells, which challenges the simple assumption that a tumor subtype's name tells you its cell of origin.
  4. Noncoding Mutations Have Functional Impact: The team trained interpretable deep neural networks using the BPNet architecture to learn a cancer-specific regulatory grammar. When they introduced somatic noncoding mutations from tumors into these models in silico, mutations predicted to strongly alter chromatin accessibility were enriched near known cancer driver genes such as TP53 and MYC. This provides systematic evidence that dispersed, non-recurrent noncoding mutations can be functional in cancer.

Methodology & Study Design

  • Single-Cell Epigenomics: The core experimental method was single-cell ATAC-seq (scATAC-seq) on archived TCGA tumor samples. Computational pipelines handled quality control, clustering, and cell type annotation based on accessibility at marker genes.
  • Integrative Genomics: Accessibility data was integrated with matched TCGA bulk DNA-seq to call CNAs and subclones, and with RNA-seq to link regulatory elements to gene expression.
  • Deep Learning for Regulation: BPNet models were trained on cancer-specific and healthy-tissue-specific accessibility data. The models learn the TF motif grammar predictive of accessibility and can score the effect of any DNA sequence variant.
  • In Silico Saturation Mutagenesis: The trained models predicted the effect of every possible single-nucleotide change in regulatory regions, as well as the specific impact of observed somatic noncoding mutations from cancer genomes.

Significance

This paper is a substantial resource and a methodological showcase. It gives the community a high-resolution map of the cis-regulatory landscape of primary human tumors at single-cell resolution. By coupling this atlas with interpretable AI, it provides a framework for moving from merely cataloging noncoding mutations to functionally interpreting them. It supports the idea that noncoding mutations contribute to oncogenesis by disrupting transcriptional programs, and offers a model for identifying which specific mutations are likely drivers.

Synthesis & Discussion

These two studies look at different stages of cancer (risk vs. established tumors) and different genomic features (germline variation vs. somatic epigenomic remodeling), but they are genuinely complementary. They are two sides of the same coin: how genomic information outside protein-coding exons dictates cancer fate.

Krishna et al. show that germline genetic diversity in immune genes (HLA-II) sets the stage for how effectively a host can respond to early, smoking-induced malignant transformations. The subsequent somatic loss of this protective heterozygosity in tumors reflects the selective pressure exerted by immune surveillance. Sundaram et al., in turn, provide the tools and maps to understand the somatic rewiring of gene regulation that defines the tumor cell itself, including how noncoding mutations might alter pathways critical for immune recognition or other hallmarks.

One connection worth highlighting is the shared use of single-cell genomics. Krishna et al. leaned on existing single-cell RNA-seq data to contextualize their genetic findings, showing where HLA-II is expressed in the smoking lung. Sundaram et al. generated a new single-cell epigenomic atlas that could, in future work, be directly integrated with immunogenetic data. For example, one could use their scATAC-seq data to ask whether HLA loci are differentially accessible in tumor subclones, or whether regulatory programs in cancer cells directly suppress antigen presentation pathways.

The papers do not contradict each other, but they do highlight different priorities. The immunogenetics study zeroes in on a specific, powerful mechanism of population-level risk. The chromatin atlas study is broader and more foundational, aimed at building generalizable models for regulatory interpretation. Together, they make a case for a more holistic view of cancer genomics that fully incorporates immunogenetics and regulatory genomics into the analytical framework.

Key Unanswered Questions:

  1. How do the noncoding regulatory programs discovered by Sundaram et al. directly influence the expression of HLA and other antigen presentation machinery studied by Krishna et al.?
  2. Beyond lung cancer, does HLA-II heterozygosity protect against other smoking-associated or inflammation-driven cancers?
  3. Can the deep learning models from Paper 2 be adapted to predict the immunogenicity of neoantigens presented by specific HLA allotypes from Paper 1?
  4. What are the specific CD4+ T cell clones and neoantigens that mediate the protective effect of HLA-II heterozygosity in pre-cancerous lesions?

For Your Lab Meeting

  1. Methodology & Generalizability: Krishna et al. relied on biobank data from predominantly European ancestries (UK Biobank, FinnGen). How might population-specific HLA haplotype structures affect the generalizability of their findings to global populations? What are the challenges in running similar immunogenetic studies in more diverse cohorts?
  2. Causality vs. Association: The HLA-II heterozygosity finding is an association. What experimental models (e.g., humanized mouse models carrying different HLA alleles and exposed to tobacco carcinogens) could establish a causal protective mechanism?
  3. Interpreting the AI: Sundaram et al. use interpretable neural networks (BPNet) to learn regulatory grammar. For a bioinformatician, what are the advantages and potential pitfalls of using such a model compared to a simpler method like motif enrichment analysis? How much should we trust the in silico predictions of mutation impact?
  4. Integration Potential: Imagine you have the resources to design a follow-up study that integrates the approaches of both papers. What would it look like (e.g., performing scATAC-seq on lung tumors from patients with known HLA genotypes to look for allele-specific accessibility and regulatory programs linked to immune evasion)?
  5. Clinical Translation: Paper 1 suggests incorporating immunogenetics into screening. What are the practical, ethical, and economic hurdles to adding HLA zygosity to a lung cancer risk model like the PLCOm2012? For Paper 2, how far are we from using models of noncoding mutation impact to distinguish driver from passenger mutations in a clinical sequencing report?

Key Terms

  • HLA Heterozygosity: The state of having two different alleles (variants) of a Human Leukocyte Antigen (HLA) gene, one inherited from each parent. This increases the diversity of peptides that can be presented to T cells.
  • Loss of Heterozygosity (LOH): A somatic genetic event in which one allele of a gene is lost (e.g., via deletion or mitotic recombination), resulting in a cell that is functionally homozygous for the remaining allele. In cancer, LOH often affects tumor suppressor genes or, as shown here, immune genes.
  • scATAC-seq (Single-Cell Assay for Transposase-Accessible Chromatin using sequencing): A technique that identifies regions of open, accessible chromatin in individual cells. It uses a transposase enzyme to insert sequencing adapters into DNA in nucleosome-free regions, providing a snapshot of the active regulatory genome in each cell.
  • cis-Regulatory Element: A non-coding DNA region (e.g., enhancer, promoter) that regulates the transcription of a nearby gene on the same chromosome. These are the regions mapped by ATAC-seq.
  • Interpretable Deep Learning (e.g., BPNet): A class of neural network models designed not only to make accurate predictions (e.g., of chromatin accessibility from DNA sequence) but also to provide human-understandable explanations for their predictions, such as which transcription factor motifs the model used to make its decision.

References

  1. Chirag Krishna, Anniina Tervi, M. Saffern, Eric Wilson, Seong-Keun Yoo, N. Mars, et al. (2024). An immunogenetic basis for lung cancer risk. Science. https://doi.org/10.1126/science.adi3808

  2. Laksshman Sundaram, Arvind Kumar, Matthew Zatzman, Adriana Salcedo, Neal G. Ravindra, Shadi Shams, et al. (2024). Single-cell chromatin accessibility reveals malignant regulatory programs in primary human cancers. Science. https://doi.org/10.1126/science.adk9217

Cancer Genomics Oncology Tumor