Immune Responses Vary in Preinvasive Colorectal Lesions by Tumor Location and Histology

Research has highlighted the importance of immune response in the prognosis of colorectal cancer: (1–4) patients whose tumors display a robust lymphocyte infiltrate (i.e., “hot” tumors; ref. 5) have a better prognosis than those with nonimmunogenic (i.e., “cold”) or immunosuppressive environments (6, 7). Indicators of tumor immunity may have a stronger relationship to colorectal cancer survival than clinical cancer stage (1). Several studies have identified marked variation in the immune composition of colorectal cancers of different molecular phenotypes, anatomic locations, and clinical stages (5, 8–12). The number of innate and adaptive immune also cells vary in the normal mucosa by anatomic location (13–15). However, less is known about the immune milieu in preinvasive colorectal lesions, especially in aggressive lesions (e.g., villous histology, high grade dysplasia) associated with increased risk of colorectal cancer (16).

Emerging evidence suggests there is a shared immunologic lineage between precancerous lesions and their invasive counterparts (17), yet how the immune responses differ in early phases of carcinogenesis is not well described. A few studies have examined aspects of the immune cell contexture within large bowel adenomas (18–22). The preliminary findings suggest that as adenomas progress, expression of inflammatory Th17-related cytokines increases, whereas that of cytotoxic Th1-associated cytokines decreases along with infiltration of CD8 T cells and NK cells (18, 20, 22). Emerging work has suggested that mast cells (23, 24) and NK cell ligands (25) may play key roles in host defense and immune surveillance which could be important in early carcinogenesis but there have been conflicting results regarding their role in inhibiting or promoting tumorigenesis (24–27). Further, there is a paucity of data characterizing the immune infiltrate in early lesions of different histologic types, colorectal locations, or degrees of dysplasia, and it may differ in early carcinogenesis compared with later (28–30).

Most investigations of immune infiltrate in colorectal precursor lesions have been small studies and only examined a few different immune markers (18, 21, 30). To better characterize the immune environment in early carcinogenesis, we studied an assortment of immune cell types (i.e., CD4, Th17, mast cells, NK-cell ligand), and cytokines (IL6, IFNγ, IL17A) in preinvasive lesions using whole slide automated immunofluorescence analysis. Our analysis contrasted immune counts for each analyte by clinicopathologic features while adjusting for batch, age, sex, race, potentially confounding variables known to be associated with immune responses (31–33).

Using the Medical University of South Carolina (MUSC) pathology laboratory information system CoPath (Cerner Corporation), we identified colorectal adenomas or serrated lesions (sessile serrated lesion SSL) or traditional serrated adenomas (TSA) excised from patients who underwent a sigmoidoscopy or colonoscopy with polypectomy between October 2012 and May 2016. Our study design was cross-sectional. We excluded patient samples if the lesion was <5 mm, as estimated by the study pathologist, or there was a known familial hereditary syndrome (FAP or Lynch syndrome). The MUSC Institutional Review Board II has approved the research study (IRB #PRO-00007139). Our research was approved under U.S. Common Rule 45 as expedited research. Approval from the IRB was obtained prior to start of our study. All analyses were performed on archival tissue specimens and no written informed consent was obtained.

For all cases, we abstracted personal characteristics (age at diagnosis, sex, race) from the electronic medical records and clinicopathologic data (anatomic location, grade, degree of dysplasia) variables from CoPath. To ensure uniformity of diagnoses, an independent pathologist (CB) reviewed all cases using a newly prepared hematoxylin and eosin (H&E) slide. The pathologist was blind to any patient or clinical information associated with the lesions. For each case, the study pathologist documented the dominant histologic pattern within the lesion [tubular adenoma, tubulovillous adenoma, villous adenoma, sessile serrated lesion (34), traditional serrated adenoma, other] and graded the villous component (0–100%). The pathologist also identified the grade of the lesion according to the most dysplastic area on the slide (i.e., none, low, focally high, high).

The University of North Carolina Translational Pathology Laboratory (TPL) performed the immunofluorescence (IF) multiplex staining. Prior to the start of the IF procedures, all antibodies were optimized using positive (e.g., tonsil) and negative control tissues as recommended by the vendor. The TPL also stained a TMA, which included many different types of cancers (e.g., pancreas, colon, breast, kidney). A small number of colorectal polyps (similar to the polyps in study cohort) and invasive colorectal cancers were also stained and analyzed to demonstrate feasibility. All stains were reviewed by the study immunologist (JW), the pathologist (DL), cancer epidemiologist (KW), and TPL Director (NNF) to ensure agreement and proper staining of cell types (e.g., lymphocytes for CD4).

Consecutive dual (CD117-MICA and CD4-RORC) or triple (IFN, IL6, IL17A) IF stains were performed in the Leica Bond-Rx fully automated staining platform (Leica Biosystems Inc.). Slides were dewaxed in Bond Dewax solution (#AR9222) and hydrated in Bond Wash solution (#AR9590). The application order of the pretreatment and staining steps, including epitope retrieval (ER), peroxidase and protein blocking, primary and secondary antibodies, and the Tyramide Signal Amplification (TSA), are shown in Supplementary Table S1. The ER for the 1st targets (MICA/B, IL-6) was maintained for 20 minutes in Bond ER solution 1 at pH 6.0 (#AR9661) and in ER solution 2 at pH 9.0 (#AR9640) for CD4; and for all other targets (2nd and 3rd) for 10 minutes in ER solution 1. The ER was followed with 10 minutes of endogenous peroxidase blocking using freshly made 3% H₂0₂ (Thermo Fisher Scientific, BP2633–500) in methanol (Thermo Fisher Scientific, A433P-4). Stains were completed in succession from first to third. Slides were counterstained with Hoechst 33258 (#H3569, Life Technologies) and mounted with ProLong Gold Antifade Mountant (#P36930, Life Technologies). Positive and negative controls (in which all primary antibodies were omitted) were included in each staining run. The single stain controls (when one primary antibody is omitted) were also done to check the cross-reactivity between the antibodies and detection.

We imaged slides in the Leica Aperio-FL (Leica Biosystems) in the Hoechst (blue), Cy2 (green/cyan), Cy3 (green), Cy5 (red) channels. The pathologist and analyst manually annotated regions containing tumor or nontumor glandular colon tissue on the entire images, only excluding regions containing tissue artifacts or noncellular areas. Annotated tissue regions were digitally analyzed using Tissue Studio Composer software version 2.7 (Tissue Studio Library version 4.4.2; Definiens Inc.). The software determined whether the average staining intensity was above the background threshold determined by negative regions on test slides for each immune protein marker. Then we used Tissue Studio software to identify all nucleated cells from the Hoechst stain and then calculated the area of the tissue on the image (mm²). For each slide, Tissue Studio software identified cells that expressed the targeted immune markers (see Supplementary Figs. S1–S3 for representative images of stained lesions). For example, to determine the number of mast cells per case, we counted any cell as positive if it stained positive for CD117. For Th17, however, the cell, by definition, is defined as CD4 positive and RORC positive, so we only counted cells as Th17 positive if the cell stained positive for both CD4 and RORC markers.

The primary endpoint was the count of positively stained immune markers within colorectal lesions. We modeled the log count of positive cells using a negative binomial generalized linear model (GLM), with the logarithm of the sample area included as a model offset for each analyte (35). In GLMs for count response variables, the offset is an adjustment term (equivalent to a covariate with unit slope) included in the model whenever the expected count is proportional to an index. Here, we expect the labeled cell counts to be proportional to the spatial area of nucleated cells. This feature must be accounted for to isolate the effects of model covariates appropriately. Clinicopathologic variables included categories of villous histology [(0–24%, 25–74%, 75%+), % villousness (0–100%), histologic type (SP, TA, TV, VA)], degree of dysplasia, and anatomic location were the independent variables of interest.

GLM models estimated the mean density counts (MDC) and their 95% confidence intervals (CI) for each clinicopathologic variable for each of the seven marker types; group comparisons of mean immune marker counts were performed using model-based contrasts. All models were adjusted for IF batch (indicator variables for six batches), age, sex, and race (base model) due to the possible differences in personal characteristics by adenoma type, colonic location, and immune infiltrate (32, 36–40). The multivariable models were constructed by adding each of the following clinicopathologic variables individually to the base model one at a time: lesion anatomic location (proximal colon, distal colon, rectum), categories of villous histology, percent of villous component (continuous), and degree of dysplasia (not high, high-grade). We include univariate models (batch adjusted only) in Supplementary Table S2 and the histologic type analysis in Supplementary Table S3. P values for the differences in models were based on Wald tests. All tests were two-sided, with a significance level (α) of P < 0.05. All statistical analyses were conducted in SAS 9.4 (SAS Institute).

Table 1 summarizes the characteristics of the 101 patients that provided the lesions for our study and the immune expression in these lesions, adjusted for batch and area. The mean age of patients with preinvasive lesions was 64 years (SD ±10). African Americans comprised 49% of our patients, and women represented 42% of the cases. The immune cell density were most abundant in those expressing NK ligands and IFNγ (Table 1). Multivariable models adjusted for age, sex, race, and batch are presented throughout but were broadly similar to batch only adjusted models (Supplementary Table S2).

Categories of percent of villous component exhibited significant differences in density counts across several innate markers (Table 2). Lesions with a high percentage of villous component relative to low (i.e., 76% vs. 0 to 25%) had significantly lower mast cell counts (P = 0.0005), IFNγ (P = 0.04), IL17A (P = 0.002), IL6 (P = 0.0003) but no difference in CD4 or Th17 T-cells (Table 2). When percent villous histology (0%–100%) was included as a continuous variable, we observed significant declines in IL17a (P = 0.02), IL6 (P = 0.004), mast cells (P = 0.0003), and NK ligands (P < 0.05; P values presented in Table 3). There was a trend of decreasing counts for IFNγ as percent villousness increased (P = 0.06). The analysis by histologic type were broadly similar to these findings (Supplementary Table S3).

The immune infiltrates tended to be greater in preinvasive lesions from the proximal colon than in the distal colon and rectum. For Th17 labeled cells, we observed higher mean counts in proximal compared with the distal colon (P < 0.0001) or rectal lesions (P = 0.004); we saw a similar pattern for CD4 positively stained cells. Cells labeled with IFNγ and IL6 also had significantly higher mean counts in the lesions of the proximal colon compared with the rectum. We also detected a significant decline in average IL6 cell counts from the distal colon to the rectum (Table 2). NK-cell ligand was significantly higher in the proximal compared with the distal colon (P for difference = 0.04) and five times higher than in the rectal lesions (P < 0.0001).

For six of the seven immune markers, we did not observe significant differences in immune counts in lesions with high-grade dysplasia and those without high-grade histology. We observed a modest decline in mast cells in patients with high-grade compared with non-high-grade lesions (0 = 0.04).

We observed that tumor-associated immune infiltrate varied by innate and adaptive marker type, villous histology, anatomic location, and degree of dysplasia. IFNγ and NK-cell ligands, two markers important in cytotoxic response, displayed the highest average density counts in preinvasive lesions; on average, serrated polyps showed higher density count expression than conventional adenomas. As the percentage of villousness in polyps increased, the density counts declined for several markers (mast cells, IL6, IL17A, and NK ligands), suggesting that aggressive histology correlates with immune suppression is consistent with a “cold” immune environment. Average density counts were significantly higher in the proximal colon than the rectal location for all markers except IL17a, consistent with that observed in invasive disease, yet the reasons are not fully understood. Surprisingly, no differences were observed between low- and high-grade lesions except for mast cells, which exhibited a modest decrease in expression. Overall, our results point to the need for additional studies interrogating a larger number of preinvasive lesions with a wide array of markers.

Colorectal cancers are often classified into different consensus molecular subtypes (CMS; ref. 41) or phenotypes (8, 42, 43), which also have different immune infiltrate profiles (8–12). In invasive lesions, a robust tumor immune response often indicates better outcomes (2, 8, 44). A few studies have investigated the consensus molecular subtypes in adenomas (17, 45). Most serrated polyps are believed to be the precursor lesions to MSI-high or the CMS-1 cancers, which tend to be highly immunogenic (11, 17, 29). In our small subset of cases, we observed higher immune infiltrate for NK-cell ligands and IL17a in serrated polyps compared with conventional adenomas, a pattern consistent with studies showing higher immune activation in serrated pathway lesions (8, 17, 41). On the other hand, most conventional adenomas are believed to be most closely associated with the "canonical" CMS-2 colorectal cancers (17, 45), which have a paucity of cytotoxic lymphocytes and appear more immunologically quiet (or “cold”; refs. 10, 11). Few previous studies have contrasted the immune responses among different conventional adenoma phenotypes (TA, TVA, VA). Our results suggest immune contexture of villous containing adenomas may be more immune-suppressed compared with tubular adenomas. Some have suggested that villous-containing lesions may develop into CMS3 colorectals (45) known for having a metabolically rich phenotype (41, 46). Immunologically, CMS-3 cancers are also cold (46), but unlike CMS-2, may have a high presence of naïve immune cell populations (41, 46), which suggests other factors (such as tumor mucins) may hamper the active immune microenvironment of these adenomas/cancers. Tumor-associated mucins appear to be enriched in villous histology lesions compared with tubular adenomas or normal mucosa (47–49). Mucins and glycans in tumors can bind to immune cells (e.g., NK, dendritic cells, macrophages) to suppress immune responses, leading to tumor escape and progression (50, 51). The lower immune infiltrate density detected in the villous-containing lesions could contribute to their aggressiveness by allowing these early neoplasms to subvert or escape immune surveillance activities and grow relatively unfettered (52–55). A more thorough understanding of the relationship between colorectal cancer phenotypes, histology, and the immune environment is warranted, especially with respect to markers of progression and recurrence. Many studies have identified notable differences in the immune responses by anatomic location in invasive cancers and normal mucosa, but we know less about immune responses in the preinvasive phase. The results of our study are consistent with previous reports of colorectal cancers, documenting the more robust immune responses in the proximal versus distal colon and rectum (56, 57). To our knowledge, only one previous study compared preinvasive lesions by colorectal location (22). The authors observed no differences in CD8⁺ or Treg cells in adenomas from the proximal and distal colon. Still, they found a significantly higher density of CD8⁺ cells in the tissue adjacent to adenomas in the proximal than in the distal colorectum. Others have shown that inflammation-inducing Th17 cells and mast cells (58) have higher expression in the normal mucosa in the proximal compared with distal colorectum (14, 59, 60). The reason for the higher immune marker abundance in the proximal colonic lesions may stem from the greater presence of microbial antigens within the mucosa. The mucosa is more porous in the proximal colon (compared with the distal colon and rectum) and is subject to more significant bacterial translocation (61). Breakages in the mucosal barrier promote contact between the bacterial content in the lumen and mucosal epithelium. Neoplasms common to the proximal colon (MSI-H and serrated lesions) appear to contain more pathogenic bacteria (62–64). The host reaction to the tumor-associated microbes can produce an immune response, which may drive Th1 or Th17 type inflammation, depending on the bug (65, 66). Immune rich or “hot” tumor are also more likely to be responsive to immune therapies which could have implications for immune prevention strategies (57).

Previous studies comparing the immune infiltrate densities along the adenoma-carcinoma sequence have illustrated a general decline in the Th1 cytotoxic responses [CD8⁺ (22), IFNγ (18)] with increased expression of Treg (22) and inflammatory signals (refs. 20, 30; e.g., macrophages, IL17A) and a decrease in CD8 positive cells (22). Neoplasms generate immunosuppressive signals as adenomas transition from advanced precursor lesions to invasive disease (67). However, in our data, we did not observe a significant change in immune infiltrate counts by degree of dysplasia except for mast cells, which declined. The role of mast cells in the tumor microenvironment during early carcinogenesis is not clear, with some indicating a protective role (68) capable of stimulating cytotoxic T cells (69) and others suggesting a pro-inflammatory or immunosuppressive role (70). The decline of mast cells in high-grade lesions in our study suggests that these cells may be important in the transition to advanced neoplasia. Importantly, in our models, we were able to adjust for factors that may correlate with the degree of dysplasia, such as percent villousness, area of the lesions, or personal characteristics. A comparison to early and later invasive lesions will be important to consider in future studies.

Our study had several advantages. It is first study to compare preinvasive lesions by location, villousness, and dysplasia using whole slide automated digital analysis as opposed to representative regions selected by a pathologist (18, 30). The analysis of the entire slide (vs. select regions) may better account for tumor heterogeneity (71). Further, many studies have used manual counting of the positively stained cells, which is more subject to bias than automated image analysis that our study employed (72, 73). Dominant histologic patterns of preinvasive lesions were determined by a pathologist independently with no knowledge of the clinical or personal characteristics of the patients. We were able to adjust out results for important potential confounders such as age, sex, race, and batch. Finally, our laboratory has extensive experience with optimization of antibodies, staining, imaging and analysis of immunofluorescence data (74, 75). We are also aware of several limitations of this study. We lacked investigation of important innate (macrophages, dendritic cells) and adaptive (Th1, Th2, T-regulatory, memory) immune cells. Our study had very few serrated lesions which will be important to investigate in future work. We were not able to assess immune counts in different regions of the polyps (e.g., stromal, epithelial) which are known to play different roles in growth in lesions (10, 17).

In summary, our results suggest marked heterogeneity in immune expression across innate and adaptive immune. From the higher densities, we infer more prominent immune responses in serrated lesions compared with conventional adenomas in lesions of the proximal colon compared with distal and rectal locations. A better understanding of the immune signatures of preinvasive colorectal lesions may shed light on critical immune cells involved in the transition from preinvasive to invasive lesions. A greater understanding could lead to possible preventive interventions. Future work is needed to characterize the immune milieu in preinvasive colorectal lesions fully, and to assess the associations with conventional risk factors (e.g., race, smoking status, BMI) and outcomes.

J. Brazeal reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

K. Wallace: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. G.J. El Nahhas: Software, formal analysis, methodology, writing–original draft. C. Bookhout: Conceptualization, formal analysis, investigation. J.E. Thaxton: Validation, methodology, writing–review and editing. D.N. Lewin: Conceptualization, investigation, writing–review and editing. N. Nikolaishvili-Feinberg: Supervision, validation, investigation, writing–review and editing. S.M. Cohen: Software, formal analysis, investigation, visualization, writing–review and editing. J.G. Brazeal: Data curation, project administration, writing–review and editing. E.G. Hill: Conceptualization, formal analysis, supervision, writing–review and editing. J.D. Wu: Conceptualization, funding acquisition, validation, investigation, writing–review and editing. J.A. Baron: Conceptualization, investigation, methodology, writing–review and editing. A.V. Alekseyenko: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft.

This study was partly funded by grants from National Library of Medicine (R01 LM012517). Supported in part by the Biostatistics Shared Resource, Hollings Cancer Center, Medical University of South Carolina (P30 CA138313). South Carolina Clinical & Translational Research (SCTR) Institute NIH Grant Nos. UL1 TR000062 and UL1 TR001450.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.