Germline-mediated immunoediting sculpts breast cancer subtypes and metastatic proclivity

Abstract

Cancer represents a broad spectrum of molecularly and morphologically diverse diseases. Individuals with the same clinical diagnosis can have tumors with drastically different molecular profiles and clinical response to treatment. It remains unclear when these differences arise during disease course and why some tumors are addicted to one oncogenic pathway over another. Somatic genomic aberrations occur within the context of an individual's germline genome, which can vary across millions of polymorphic sites. An open question is whether germline differences influence somatic tumor evolution. Interrogating 3,855 breast cancer lesions, spanning pre-invasive to metastatic disease, we demonstrate that germline variants in highly expressed and amplified genes influence somatic evolution by modulating immunoediting at early stages of tumor development. Specifically, we show that the burden of germline-derived epitopes in recurrently amplified genes selects against somatic gene amplification in breast cancer. For example, individuals with a high burden of germline-derived epitopes in ERBB2, encoding human epidermal growth factor receptor 2 (HER2), are significantly less likely to develop HER2-positive breast cancer compared to other subtypes. The same holds true for recurrent amplicons that define four subgroups of ER-positive breast cancers at high risk of distant relapse. High epitope burden in these recurrently amplified regions is associated with decreased likelihood of developing high risk ER-positive cancer. Tumors that overcome such immune-mediated negative selection are more aggressive and demonstrate an "immune cold" phenotype. These data show the germline genome plays a previously unappreciated role in dictating somatic evolution. Exploiting germline-mediated immunoediting may inform the development of biomarkers that refine risk stratification within breast cancer subtypes.

Figure 1 –. Germline-derived epitope burden in oncogenes selects against oncogene amplification

A) Schematic of germline-mediated immunoediting. Prior to transformation, differing germline genomes and HLA alleles result in differing numbers of epitopes derived from a gene of interest (e.g. Gene X). If during transformation, the tumor acquires additional copies of Gene X (i.e. somatic amplification), the number of epitopes increases further. As a result, individuals with a high burden of epitopes are more likely to be surveilled by the immune system triggering cell death. B) As proof of concept, GP2 is a well-characterized, naturally occurring (i.e., non-somatically mutated) immunogenic peptide derived from HER2. Schematic overview of analysis framework to investigate if the ability to present GP2, i.e., having MHC Class I alleles that bind GP2, is associated with HER2+ breast cancer. C) The ability to present GP2 is negatively associated with HER2+ breast cancer. Barplot shows the ratio of HER+ to HER2− in patients that have HLA alleles that can bind GP2 (GP2 presented) vs patients that do not (GP2 not presented). Odds ratio (OR) and p-value from logistic regression model correcting for first six genetic principal components. D) Schematic outlining methods to investigate germline-mediated immunoediting. Using four independent cohorts representing multiple stages of breast cancer, pre-invasive, primary invasive and metastatic invasive breast cancer, we investigated if the GEB in a gene of interest was associated with the likelihood of acquiring a somatic amplification of the gene using HER2 as a representative example. E) GEB in ERBB2 is negatively associated with HER2+ breast cancer. Barplot shows the ratio of HER2+ to HER2− patients with low, medium or high GEB. Odds ratio (OR) and p-value from logistic regression model correcting for the first six genetic principal components. F) Beyond ERRB2, we investigated amplicons that characterize four ER+/HER2− high risk of relapse subtypes (IC1, IC2, IC6 and IC9), where the percent of breast cancer cases they represent and the corresponding chromosome region and core genes is denoted for each subtype. G) GEB in recurrently amplified genes is negatively associated with gene amplification. Scatterplot shows odds ratio (x-axis) and 95% confidence intervals from logistic regression model correcting for the first six genetic principal components and somatic mutation burden. Covariates in the top panel indicate the direction of the effect, namely whether GEB is associated with increased or decreased likelihood of each subtype.

Figure 2 –. Germline-mediated immunoediting dictates breast cancer subtype early during tumorigenesis

A) Across five subtypes and three independent cohorts, high GEB in subtype-specific oncogenes is associated with a decreased likelihood of developing the cognate subtype of breast cancer. Forest plot shows the odds ratio and 95% confidence intervals from a meta-analysis across three cohorts: DCIS (N=341), TCGA (n=656) and ICGC (n=431). B) At the individual variant level, to avoid immunoediting, tumors should preferentially amplify the germline allele that produces a weaker epitope. For example, considering only heterozygous individuals, in scenario 1 the reference (ref) allele produces an epitope with higher binding affinity for MHC class I than the alternative (alt) allele. As a result, the tumor preferentially amplifies the alt allele as evidenced by the increased proportion of sequencing reads supporting the alt allele. By contrast, in scenario 2, the alt allele produces an epitope with higher binding affinity for MHC class I and the ref allele is preferentially amplified. C-D) Allele producing epitope with weaker MHC class I binding affinity is preferentially amplified. Boxplots of differential binding for epitopes derived from the alt allele vs ref allele (y-axis), i.e. a measure of alt allele binding affinity, for samples that preferentially amplified the alt or the ref allele. Effect size and p-value from Mann-Whitney rank sum test. Boxplots show analysis for rs1058808 derived from ERBB2 (C) and rs1292053 derived from TUBD1 (D).

Figure 3 –. Tumors that overcome a high burden of germline epitopes are more aggressive

A) Schematic of within-subtype comparison between GEB in primary tumors (TCGA and ICGC) vs metastatic tumors (Hartwig). B) Across five subtypes, metastatic tumors show an enrichment of epitopes compared to primary tumors. Forest plot shows odds ratio and 95% confidence intervals from meta-analysis of TCGA vs Hartwig and ICGC vs Hartwig. C) Schematic of within-subtype comparisons of GEB association with risk of relapse within five years in METABRIC. D-E) A high GEB is associated with increased risk of relapse in HER2+ (D) and ER+ (E) tumors. Hazard ratio and p-value from CoxPH model correcting for first two genetic principal components, percent genome altered and age. F) GEB in combination with the Integrative Clusters (IntClust) improves the accuracy of five-year relapse prediction in ER+ and HER2+ tumors. Forest plot shows c-index of predictive models considering the IntClust alone or in combination with GEB for 1,000 bootstrapped iterations. Fold change (FC) is calculated as the ratio of medians while the p-value is calculated as 1 – the proportion of iterations where the -index of the IntClust and GEB model was greater than the IntClust alone model.

Figure 4 –. A high germline epitope burden promotes an immunosuppressive phenotype

A) Schematic of within-subtype comparisons of the immune landscape between high GEB and low GEB tumors in TCGA. B-C) Unsupervised clustering of 23 immune features, selected to reflect broad immune cell populations, cytokine signaling and extracellular matrix composition, identified two dominant clusters within HER2+ (B) and ER+ (C) breast tumors driven by GEB. Heatmap shows the z-score of each immune feature (y-axis) for each tumor (x-axis). Covariates along the top indicate if the tumor has a high GEB or a low GEB along with clusters from two different clustering methods (Diana and Consensus). Immune features cluster into two broad categories, myeloid and lymphocyte predominant, as indicated by the covariate on the right. Statistics from Fisher’s exact test quantifying the enrichment of high GEB tumors in the myeloid predominant cluster for both clustering methods. D) Schematic of GEB association with progression to invasive breast cancer. E) DCIS lesions that do not progress to IBC are enriched for high GEB. Barplot shows the proportion of DCIS lesions that progress or not progress to IBC stratified by GEB. Statistics from a logistic regression model correcting for eight genetic principal components, HER2 and ER status. F) Myoepithelial integrity is negatively associated with GEB. Boxplot shows myoepithelial integrity (% of E-cadherin in myoepithelium), as defined by Risom et al. (43), in high vs low GEB lesions for a subset of lesions that had spatial proteomics data. Statistics are based on a Mann-Whitney Rank Sum test.

Figure 5 –. Germline-mediated immunoediting sculpts breast cancer subtypes and metastatic proclivity

Schematic of mechanistic molecular model of germline-mediated immunoediting and its implications for improving breast cancer risk stratification. Briefly, during tumorigenesis, lesions with a high GEB in a gene of interest are less likely to acquire somatic amplification of that gene. However, if the tumor gains additional copies of the gene, it is forced to develop an immune suppressive/evasive phenotype and is more aggressive. Conversely, low GEB has little impact. By the time the tumor has metastasized to distant sites, it develops immune suppression/evasion mechanisms and is refractory to immunoediting pressures. In the pre-cancerous setting, GEB may be indicative of risk of progression to an invasive cancer since lesions with high GEB would have to overcome stronger immune pressures. Once the lesion becomes invasive, within a breast cancer subtype, tumors may be further stratified into those with high and low risk of relapse based on GEB in subtype-specific genes.