This study was designed to identify genomic alterations, expression signatures, and interactions within tumor and the residents of the microenvironment, specifically immune cells, stromal elements, and the DCIS epithelium. The goal was to learn what enables large (> 5cm) clinically high-risk DCIS lesions to remain as DCIS as they grow and understand what characteristics are correlated with second events. Our cohort of 61 cases with long-term follow up (median 8 years) were all treated with standard of care (surgery +/- radiation). 14 of 61 (23%) had a subsequent disease recurrence event, 8 with invasive carcinoma (13%), and 6 with recurrent DCIS (10%). These are higher recurrence rates than those reported for DCIS, as expected given the cohort design including only high-risk DCIS6,22,23. We performed an array of tissue-based assays and microdissections accompanied by expert (ADB) annotation of the areas of DCIS, and also a whole tissue section (bulk) transcriptome analysis. Cross-platform correlations provide new insight into the underlying biologic interactions that appear to constrain or permit progression and may reflect patterns of immunosurveillance/recognition of these lesions. Even with this more homogeneous group of large lesions, we found surprising heterogeneity.
CNAs and mutations of breast cancer driver genes, most notably TP53, were found to be significantly associated with dense immune cell infiltrates. This is similar to previously reported observations in DCIS and invasive breast cancer14,24. Frequently observed CNAs in our cohort included gains on chromosomes 1q, 8q, 16p, and 17q and losses on chromosomes 8p, 16q, and 17p. Breast cancer is considered to be a copy number class tumor, and these CNAs in DCIS lesions are consistent with previously published reports19–21,25,26. While we did not find that the fraction of the DCIS genome altered correlated with immune infiltrates or recurrence events, CNAs at certain chromosomal regions were associated with various immune cell populations. Gains on chromosomes 6p, 6q, and 21q or losses on chromosomes 8p, 9p, and 17p were associated with higher TIL, T cell, CD3+CD8+ T cell, CD3+CD8− T cell, Treg, or B cell infiltrates. ERBB2 was the most prevalent gene with CNAs (53% of cases), which is consistent with a recent analysis of CNAs in DCIS26. However, our data agree with the findings of this study in that CNA at ERBB2 was not associated with recurrence events. We also found that copy number losses were common in various known breast cancer driver genes, including TP53 (> 40% of cases), BRCA1, BRCA2, ATRX, and ZMYM3. Loss of TP53 function leads to overall genomic instability and resistance to DNA damage-mediated apoptosis. ATRX functions to regulate the DNA damage response and ameliorates chromosomal instability27. ZMYM3 deficiency results in impaired homologous recombination repair and genome instability28. We observed higher immune infiltrates (TILs, T cells, CD8+ T cells, Treg, and B cells) in DCIS harboring TP53 mutations or losses of ATRX or ZMYM3, which is consistent with the concept that genomic instability may lead to the generation of neoantigens that can activate the immune system29, which could potentially prevent invasion and keep the lesion contained as DCIS.
In analyzing the presence, density, and spatial relationships of various immune cells within the tumor immune microenvironment, we identified significant associations with certain clinical characteristics and outcomes. Like their invasive counterparts, we found that HR-negative and/or HER2-positive DCIS was associated with high densities of TILs, T cells, CD3+CD8+ T cells, CD3+CD8− T cells, Treg cells, and B cells as well as proliferating (Ki67+) subgroups of T cells, Treg, and B cells. Several PD-1/PD-L1 immune cell populations were also associated with HR-negative or HER2+ DCIS. Gene expression signatures identifying immune cell populations or signaling pathways were associated with HR-negative or HER2+ DCIS as well. These associations are consistent with several published studies15,30–34; however, a recent study by Almekinders et al. that examined the relationship of DCIS immune infiltrates with recurrence events failed to find an association30. It should be noted that the cohort in the Almekinders study differed meaningfully from our own, in that the size of > 60% of the cases in their cohort was unknown compared to the requirement that our cases all be at least 5cm in size. Furthermore, more than half of our cases (57%) were HER2+ compared to < 30% of their cases30. Our cohort represents a higher risk population with overrepresentation of intrinsic subtypes known to possess a more immunologically active tumor microenvironment.
In examining 80 possible biomarkers derived from mIF, single-plex IHC, and gene expression assays, we were able to identify three that were associated with recurrence events. We found that a high density of CD3+CD8− T cells was significantly associated with a decrease in any recurrence event. High expression of a Treg gene signature and high expression of a proliferation gene signature were each associated with greater risk of overall recurrence; the proliferation gene signature was also associated with greater risk specifically for invasive recurrence. Similar associations of a Treg gene signature and higher proportion of Treg cells in the tumor immune infiltrate with increased recurrence has been observed in the setting of early stage (Ia and Ib) non-small cell lung cancer and non-metastatic renal cell carcinoma35,36. Furthermore, we found that close proximity of T cells to tumor cells was significantly associated with better outcomes, with respect to any recurrence or invasive recurrences. These findings suggest that it is not solely the presence of T cells in the tumor immune microenvironment but their spatial relationship to tumor cells that predicts clinical outcomes. This is similar to findings in invasive tumors where the same subtypes i.e., HR-negative and/or HER2-positive are the most likely to have immune infiltrates and are likely to respond to immunotherapy37. It may also provide us with clues to what can keep tumors contained, where despite their size they remain DCIS, at least for a time. The lesions with strong immune infiltrates may be more amenable to treatment with immune cell stimulation, as we have shown38.
The stromal compartment appears to plays a critical immunoregulatory role and is a critical part of the story. We used spatial proximity of αSMA+p63+ myoepithelial cells to calculate a myoepithelial continuity score. Consistent with recently published work17, we also found that myoepithelial continuity is significantly lower in DCIS compared to benign (normal or ADH) lesions and that higher myoepithelial continuity among DCIS lesions is significantly associated with a higher recurrence risk. In addition, low myoepithelial continuity was associated with a variety of immune related pathways. These findings raise an intriguing hypothesis that myoepithelial cells may limit host immune recognition of the DCIS cells harboring neoantigens. Indeed, HER2 amplification/overexpression itself may be one of these neoantigens—with high prevalence in our cohort, but also high prevalence in all DCIS compared to lower prevalence in invasive carcinomas39.
Aggregating all of this complex information, we identified patterns of immune microenvironment markers expression among DCIS that help to explain the associations, shown in the Supplementary Fig. 7. Of note, although we observed that HR-negative as well as HER2+ DCIS were associated with high densities of multiple immune cell populations (e.g. TILs, Tcells, Bcells), our unsupervised clustering analysis identified subsets of primarily HER2 + and HR-HER2- samples (Subtype C5) with relatively lower levels of these immune cell type markers. Unlike their Subtype C2 counterparts, which were also HER2 + or HR-HER2- and had high levels of multiple immune cell population, C5 samples exhibited higher levels of PD1pT and spatial measures of PDL1p immune cells to PD1p T cells. This analysis illustrates the heterogeneity in tumor immune microenvironment within high-risk DCIS both across and within HR/HER2 defined subtypes; and suggests classification of high risk DCIS by their microenvironment profile may offer insights beyond evaluation of single markers alone. For example, HER2 + DCIS with a strong immune infiltrate where PD1 T cells are in close proximity to PDL-1 + cells may indicate an environment where ongoing immune activity is preventing invasive recurrence. These insights are the basis for initiating immunotherapy trials for DCIS30,38–41.
In summary, this unique DEFENSE study of large DCIS that Nevertheless Stay Encapsulated demonstrates that DCIS can be stratified based upon features of the tumor immune microenvironment which, in turn, correlates with intrinsic epithelial subtypes including propensity for high neo-antigenicity, along with myoepithelial continuity and finally immune activation. It is the interplay among these features that determine the fate of these lesions. The observations we describe have the potential to guide clinical decision-making including the safety of active surveillance (watchful waiting) for lower risk lesions as well as improved therapy including immunomodulation that can be targeted to higher risk lesions which are most likely to be responsive and permit trafficking of immune cells past the myoepithelial layer to the lesions within the ducts.