Patterns of immune equilibrium and escape in indolent and progressing tumors

Cancers display a wide spectrum of clinical behavior: on one end of the spectrum are aggressive, rapidly-progressing tumors; on the other end, indolent and slowly- or non-progressing tumors. The former (e.g., anaplastic thyroid carcinoma, pancreatic adenocarcinoma) are frequently unresectable or metastatic at the time of clinical detection, and often ultimately prove fatal. In contrast, the latter (e.g., some small papillary thyroid or prostate tumors) can sometimes remain asymptomatic and not progress during a patient’s lifetime. Indolent tumors that, if never diagnosed, would not progress to cause symptoms or death, are referred to with the epidemiologic term “overdiagnosed.” Overdiagnosis has been documented in small tumors of the prostate, thyroid, breast and several other sites. Because overdiagnosed tumors pose no risk of harm, the treatment of these tumors is unnecessary, exposing patients only to risks and toxicities of interventions, without benefit (Figure S1A).¹ However, not all early-stage tumors of these sites have been overdiagnosed, and there are no reliable techniques to differentiate the subset that remain indolent from those that may progress; therefore, nearly all such cases are definitively treated with surgical resection and/or radiation therapy. Mechanisms underlying indolent tumor behavior are currently unknown. Understanding these mechanisms is a necessary first step to devising more precise cancer screening approaches and developing prognostic biomarkers to help guide clinical decision-making; for example, triaging such tumors to either immediate treatment or observational strategies.

Schreiber and colleagues have described a process in which the growth of occult cancer in a mouse model is suppressed by immunological surveillance, resulting in a state of immune equilibrium and tumor dormancy.² Tumors that eventually grow are observed to have adapted to immune pressure and evolved to a state of dampened immunogenicity – termed “immunoediting.” This is theorized to possibly occur via the loss of tumor antigens or their presentation by HLA, or establishment of an immunosuppressive state in the microenvironment.^2,3 However, these processes – immunologic equilibrium and escape – have not been directly observed during the natural history of human tumors. Cancer-immune equilibrium has been difficult to study in human tumors, very few of which are left untreated and observed, and even fewer of which have tissues available for analysis.

We hypothesized that some indolent or overdiagnosed tumors may exist in a state of cancer-immune equilibrium, with growth constrained by immune surveillance. In these tumors, eventual progression (if it occurs) would reflect the development of mechanisms to evade or adapt to immune predation. Here, we evaluated this hypothesis by profiling the immune microenvironmental and antigen presentation landscapes of tumors in overdiagnosis-prone cancer types, compared to non-overdiagnosed cancer types.

We first selected overdiagnosis-prone cancer types based on available literature estimating the prevalence of overdiagnosis via several techniques, such as autopsy data, screening trial data, incidence trends, and lead-time estimates (Supplements S1, S1C).^1,4 These data revealed 6 cancer types with consistent evidence that a subset of tumors exhibit indolent, non-progressing behavior: prostate adenocarcinoma (PRAD), papillary thyroid carcinoma (THCA), ER+/Her2− breast cancer (ER+BRCA), non-small cell lung cancer (NSCLC: adenocarcinoma [LUAD] or squamous cell carcinoma [LUSC]), and cutaneous melanoma (SKCM). Median estimates of the prevalence of overdiagnosis ranged from 25% (NSCLC) to 50% (THCA, Supplement S1C). In contrast to these overdiagnosis-prone cancer types, we compared 5 non-overdiagnosed cancer types in which there is no reported evidence of overdiagnosis, and nearly all early-stage tumors are expected to progress to symptomatic and potentially lethal disease if left untreated: pancreatic adenocarcinoma (PAAD), esophageal squamous cell cancer (ESCA), head and neck squamous cell carcinoma (HNSC), triple-negative breast cancer (TNBC), and gastric adenocarcinoma (STAD, Supplement S1C).

In tumors of each of these cancer types, we examined immunogenomic changes associated with progression, by comparing advanced-stage, larger tumors to early-stage, localized tumors. We analyzed features indicative of dampened anti-tumor immunity or tumor adaptation to immune pressure: a sparsely immune-infiltrated microenvironment, immunosuppressive cell populations, somatic loss of human leukocyte antigen via loss of heterozygosity (HLA LOH), downregulated expression of HLA class I/II genes, and T cell clonal expansion.⁵ We analyzed RNA-sequencing (RNA-seq) data from 4950 tumors in The Cancer Genome Atlas (TCGA): BRCA, n=1087 (ER+/PR+/Her2− [n=696] and TNBC [n=192]); ESCA, n=183; HNSC, n=523; LUAD, n=575; LUSC, n=490; PAAD, n=184; PRAD, n=495; SKCM, n=470; STAD, n=440; THCA, n=503).⁶ RNA-seq data were used to deconvolve infiltrating immune populations, quantify immunity signatures, and profile TCR complementarity-determining region 3 (CDR3) sequences. HLA LOH was assessed in whole exome sequencing data. Nine additional features were obtained from the analysis of Thorsson et al.⁷ (Complete methods and source datasets, Supplement). Small/localized tumors (T1/T2 or Gleason group 1) were compared to large/advanced (T3/T4) tumors within each cancer type (Supplement S2). Associations between these pre-defined immunologic features and tumor size/stage were tested in multivariable logistic regression with cancer histology as a covariate. The Benjamini–Hochberg (B-H) procedure was used for false discovery rate (FDR) correction.⁸

In the overdiagnosis-prone cancer types, we observed a consistent pattern: compared to small/localized tumors, larger/advanced tumors were characterized by significantly lower proportions of multiple measures of immunity, including effector immune cell infiltration (e.g., lymphocyte, T-cell signatures), immune activation (e.g., cytolytic activity, IFNγ pathway enrichment), and significantly higher proportions of immunosuppressive cell populations (e.g., M2 macrophages, myeloid-derived suppressor cells, cancer-associated fibroblasts). Associations are presented with FDR correction (Figure S1B, Table S1A). In addition, larger/advanced tumors had significantly more frequent loss of HLA, and lower entropy and richness of the TCR repertoire. These associations were robust across a wide range of tumor purity cutoffs (Table S1B).

In contrast, none of these large vs. small tumor differences were observed in the group of non-overdiagnosed cancer types (Figure S1B, Table S1A). There were also no significant differences in mutation count (silent or non-silent mutations) between large/advanced and small/localized tumors in either the overdiagnosed or non-overdiagnosed groups (Table S1A).^7,9

To validate these findings, we conducted a systematic review and identified 6 additional independent transcriptomic datasets for overdiagnosis-prone tumor types. We reviewed all publicly available transcriptomic datasets, to assemble cohorts with available raw data and sufficient numbers of tumors of each category (Supplements S3–S1C detail included and excluded [n=3] datasets). First considering each feature nominated in the discovery cohort of overdiagnosed cancers, we found that nearly all (22 of 24) were significantly associated with tumor size/extent across the class of overdiagnosis-prone cancer types (at FDR<.05) – larger/advanced tumors had decreased immune cell infiltration and activation, and more immunosuppressive microenvironments (with fibroblasts and cytolytic activity score the only features not validated; Figure S1C). To assess overall validation across all studies, we considered 43 statistically significant associations in the TCGA cohorts (Table S1A) and found 22 were validated in the same cancer type (the same feature in the same cancer type was significantly associated with tumor size; p=7 × 10⁻²², binomial distribution, Figure S1C).

For further validation, we identified 10 additional studies that profiled these features, but without raw data available, using a variety of methods such as immunohistochemistry (IHC), single-cell RNA-seq, whole exome sequencing, β–chain CDR3 sequencing, and microsatellite amplification (Supplements S4, S1C). These datasets additionally validated our observations of lower levels of immune effector cells, higher levels of suppressive immune populations, more HLA LOH, and increased TCR clonality, in large/advanced compared to small/localized tumors (Figure S1D). While all 5 features nominated in the discovery cohort that could be tested in this 2^nd validation set were validated, a limitation of this second group of validation datasets is that, because raw data were not available, we were unable to assess all features, and only analyzed published data elements (Supplement S4).

Taken together, these findings provide insight into the immunogenomic differences associated with tumor progression in different categories of cancer. In those cancers in which early-stage tumors can sometimes exhibit indolent or non-progressing behavior, we observe that larger tumors differ markedly from smaller tumors: larger tumors have an immune microenvironment that tends to be less infiltrated with effector immune cells, less inflamed, and more immunosuppressive; and a higher likelihood of dampened antigen presentation through allelic loss or downregulation of HLA – all mechanisms of escape from immune surveillance. Larger tumors also have a more clonal TCR repertoire, consistent with adaptation (via clonal expansion of T cell clonotypes) under immune pressure. These results are consistent with a process of dampened anti-tumor immunity and tumor cell adaptation to immune pressure, associated with tumor progression.

These findings suggest that small tumors that remain indolent and do not progress, are more likely to be constrained by immune surveillance, as had been suggested by preclinical models of cancers in equilibrium. In contrast, in non-overdiagnosed cancer types, none of these trends were observed, suggesting that these aggressive cancers do not escape from immune equilibrium as they progress from small to large; most likely, they escaped from immune equilibrium at a much earlier stage – possibly even prior to clinical presentation.

We note several important caveats. First, these comparisons are not longitudinal tumor samples. While such an analysis would characterize tumors with more precisely-documented behavior, it is not generally feasible to obtain such samples, because nearly all diagnosed cancers are surgically resected or otherwise definitively treated. Second, only a subset (perhaps 25–50%) of early-stage prostate, thyroid, ER+/Her2− breast, melanoma, and lung tumors have, in fact, been overdiagnosed and will remain indolent. Therefore, a comparison of advanced-stage to early-stage tumors is not a pure comparison of progressing vs. non-progressing tumors and would be expected to underestimate the differences between these 2 categories. Third, the strength of evidence for immune escape or adaptation varied across different tumor types and appeared weaker for LUSC than other cancer types. Further studies will be helpful in further illuminating immune escape in individual cancer types. Finally, it is unlikely that these immunologic findings explain the entirety of indolent tumor behavior, and it is quite possible that other mechanisms, such as genetic evolution (e.g., accumulation of new driver mutations), or other mechanisms may also play a role, although these factors have not yet been observed in the context of indolent/non-progressing cancers, to our knowledge. Future studies analyzing genetic, transcriptional, or metabolic alterations that are enriched in larger, immune-escaped tumors of overdiagnosed cancer types, may shed further light on this phenotype.

Nevertheless, the consistency of findings across multiple datasets – and the absence of these differences in non-overdiagnosed cancers – are suggestive that the mechanisms of immune equilibrium and immune escape explain, at least in part, the epidemiologic phenomenon of overdiagnosis. Indolent or overdiagnosed tumors – while evident at the population level – are not well characterized at the cellular or genetic level, and by remaining in immune equilibrium for longer than aggressive cancers, can provide a unique window into the process of immune escape.

With active surveillance now becoming an accepted alternative to immediate surgery in low-risk prostate and thyroid cancers, and being investigated for ductal carcinoma in situ of the breast, we anticipate that in the future, untreated tumors will become available for longitudinal analyses. Validation of these features of immune escape would be valuable in the development of prognostic biomarkers, to inform clinical decision-making and more personalized treatment options for patients with early-stage, potentially indolent cancers. These approaches may also help define characteristics of patients most likely to benefit from cancer screening, or investigate mechanisms of escape from immune predation at the earliest stages of evolution of preneoplasia to invasive cancer.

In conclusion, these findings provide human evidence for the heretofore only preclinical observation of cancer-immune equilibrium, which may explain, at least in part, the indolent behavior of some types of small tumors. We propose that the epidemiologic phenomenon of overdiagnosis is associated with a process of immune surveillance.