An integrated analysis of lymphocytic reaction, tumour molecular characteristics and patient survival in colorectal cancer

Background Histological lymphocytic reaction is regarded as an independent prognostic marker in colorectal cancer. Considering the lack of adequate statistical power, adjustment for selection bias and comprehensive tumour molecular data in most previous studies, we investigated the strengths of the prognostic associations of lymphocytic reaction in colorectal carcinoma by utilising an integrative database of two prospective cohort studies. Methods We examined Crohn’s-like reaction, intratumoural periglandular reaction, peritumoural reaction and tumour-infiltrating lymphocytes in 1465 colorectal carcinoma cases. Using covariate data of 4420 colorectal cancer cases in total, inverse probability-weighted Cox proportional hazard regression model was used to control for selection bias (due to tissue availability) and potential confounders, including stage, MSI status, LINE-1 methylation, PTGS2 and CTNNB1 expression, KRAS, BRAF and PIK3CA mutations, and tumour neoantigen load. Results Higher levels of each lymphocytic reaction component were associated with better colorectal cancer-specific survival (Ptrend < 0.002). Compared with cases with negative/low intratumoural periglandular reaction, multivariable-adjusted HRs were 0.55 (95% CI, 0.42–0.71) in cases with intermediate reaction and 0.20 (95% CI, 0.12–0.35) in cases with high reaction. These relationships were consistent in strata of MSI status or neoantigen loads (Pinteraction > 0.2). Conclusions The four lymphocytic reaction components are prognostic biomarkers in colorectal carcinoma.

the antitumour immune response, and all these factors have been associated with colorectal cancer mortality. [15][16][17][18] However, none of the studies has taken these molecular features into account in the prognostic analysis of antitumour immune response. Therefore, a comprehensive study focusing on the prognostic role of lymphocytic reaction and its relationship with the aforementioned molecular features is needed.
In this study, we utilised two large US-nationwide prospective cohort studies with covariate data of 4420 colorectal cancer cases, and a molecular pathological epidemiology database of 1465 cases, to evaluate the relationships between lymphocytic reaction patterns and patient survival. We hypothesised that more intense host lymphocytic reaction to colorectal cancer might be associated with a favourable clinical outcome, after adjusting for other potential confounders including neoantigen load. To reduce potential bias due to the availability of tumour tissue, we utilised inverse probability-weighting (IPW) method [19][20][21][22] (on the 4420 cases), which has not been used in the previous prognostic studies of immune response to tumour. In addition, we examined statistical interactions between lymphocytic reaction and MSI status or neoantigen load.

Study population
We collected data on colorectal cancer cases within two prospective cohort studies in the United States, the Nurses' Health Study (NHS, 121,701 women aged 30-55 years followed since 1976) and the Health Professionals Follow-up Study (HPFS, 51,529 men aged 40-75 years followed since 1986). 23 Every 2 years, study participants have been sent follow-up questionnaires to collect information on lifestyle factors and medical history of physician-confirmed diseases including colorectal cancer. The National Death Index was used to ascertain deaths of study participants and identify unreported lethal colorectal cancer cases. Participating physicians reviewed medical records to confirm diagnosis of colorectal cancer, and to record tumour characteristics (e.g. size, location and the American Joint Committee on Cancer tumour, node and metastases (TNM) classification), and causes of deaths for participants who were deceased. Formalinfixed paraffin-embedded (FFPE) tissue blocks were collected from hospitals where participants diagnosed with colorectal cancer had undergone tumour resection. We included 1465 patients with available data on at least one of four histopathological lymphocytic reactions. We included both colon and rectal carcinomas based on the colorectal continuum model. 24 Patients were followed until death or the end of follow-up (January 1, 2014 for HPFS; May 31 for NHS), whichever came first. Informed consent was obtained from all study participants. This study was approved by the institutional review boards at Harvard T.H. Chan School of Public Health and Brigham and Women's Hospital (Boston, MA), and those of participating registries as required.
Histopathological evaluations FFPE blocks of tumour tissues were collected from hospitals throughout the United States, where colorectal cancer patients had undergone surgical resection. A single pathologist (S.O.), who was unaware of other data, reviewed haematoxylin-and eosin-stained tissue sections, and recorded histopathological findings, including tumour differentiation and lymphocytic reaction components, as previously described. 5 Tumour differentiation was categorised as well to moderate vs. poor (>50% vs. ≤50% gland formation, respectively). Four components of lymphocytic reactions (Crohn'slike lymphoid reaction, peritumoural lymphocytic reaction, intratumoural periglandular reaction and TIL) were examined (Fig. 1). Crohn's-like lymphoid reaction was defined as transmural lymphoid reaction. Peritumoural lymphocytic reaction was defined as discrete lymphoid reaction surrounding a tumour mass. Intratumoural periglandular reaction was defined as lymphocytic reaction in tumour stroma within a tumour mass. TIL was defined as lymphocytes on top of cancer cells. For any given tumour, each of the four lymphocytic reaction components was scored as 0, 1+, 2+ and 3+, and graded as negative/low (0), intermediate (1+) and high (2+ and 3+) as previously described. 5,25 A review of 398 randomly selected cases between two independent pathologists (S.O. and J.N.G.) showed good concordance on grading of histopathological features, including lymphocytic reaction to tumour. 5 For the analyses of lymphocytic reaction and patient survival in strata of tumour neoantigen load, each of the four lymphocytic reaction components was graded as negative/low (0) and intermediate/high (1+, 2+ and 3+). The overall lymphocytic reaction score (0-12) was calculated as the sum of scores for the above four reaction components, and was graded as low (0-2), intermediate (3)(4)(5)(6) and high (7)(8)(9)(10)(11)(12).  Fig. 1 The four components of lymphocytic reaction against colorectal cancer. a Peritumoural lymphocytic reaction (arrows) and Crohn-like lymphoid reaction (asterisks). b Tumour-infiltrating lymphocytes (arrows) and intratumoural periglandular reaction (asterisks). c Peritumoural lymphocytic reaction (arrows). evaluated using ten microsatellite markers (D2S123, D5S346,  D17S250, BAT25, BAT26, BAT40, D18S55, D18S56, D18S67 and  D18S487), as previously described. 24 MSI-high status was defined as the presence of instability in ≥30% of the markers, and non-MSI-high as instability in <30% of the markers, as previously described. 26 DNA methylation was measured in eight CIMPspecific promoters (CACNA1G, CDKN2A, CRABP1, IGF2, MLH1, NEUROG1, RUNX3 and SOCS1) and in LINE-1. 27,28 CIMP-high was defined as ≥6 methylated promoters of eight promoters, and CIMP-low/negative as <6 methylated promoters. PCR and pyrosequencing were performed for KRAS (codons 12, 13, 61 and 146), BRAF (codon 600) and PIK3CA (exons 9 and 20), as previously described. 24 Neoantigen load, the number of proteins that likely give rise to immunogenic peptides in the tumour microenvironment, was predicted for 505 cases, by using a neoantigen prediction pipeline for somatic mutations based on whole-exome sequencing, and identifying peptides that bind to personal human leukocyte antigen (HLA) molecules with high affinity (<500 nM), as previously described. 29 Using NetMHCpan (version 2.4, Technical University of Denmark, DK-2800 Lyngby, Denmark), 30 we predicted the binding affinities of all possible 9and 10-mer mutant peptides to the corresponding HLA alleles inferred by the POLYSOLVER algorithm.
Immunohistochemistry for PTGS2 (cyclooxygenase-2), CTNNB1 (beta-catenin) and CD274 (PDCD1 ligand 1) We constructed tissue microarrays of colorectal cancer cases with sufficient tissue materials, including up to four tumour cores from each case. 31 Immunohistochemical analyses for PTGS2 (cyclooxygenase-2), nuclear CTNNB1 (beta-catenin) and CD274 (programmed death-ligand 1, PDCD1 ligand 1, PD-L1) were performed using an anti-PTGS2 antibody (dilution, 1:300, Cayman Chemical, Ann Arbor, MI), an anti-CTNNB1 antibody (dilution, 1:400, BD Transduction Laboratories, Franklin Lakes, NJ) and an anti-CD274 antibody (dilution, 1:50, eBioscience, San Diego, CA), respectively, as previously described. 17,31,32 Statistical analysis All statistical analyses were performed using SAS software (version 9.4, SAS Institute, Cary, NC), and all P values were two-sided. We used a two-sided α level of 0.005 for our primary hypothesis testing. 33 Our primary hypothesis testing was assessment of associations of four lymphocytic reaction components (negative/ low vs. intermediate vs. high) with colorectal cancer-specific survival in the Cox proportional hazard regression model. All other analyses, including evaluation of individual hazard ratio (HR) estimates, assessment of stratum-specific risk estimates and of interaction with MSI status and neoantigen load, represented secondary analyses.
To assess the association between ordinal categories of the level of lymphocytic reaction (negative/low, intermediate and high) and other categorical variables, the chi-square test was performed. To compare continuous variables (age and LINE-1), an analysis of variance assuming equal variances was performed.
We utilised inverse probability-weighting (IPW) method using covariate data of 4420 colorectal cancer cases with or without tumour tissue, to adjust for selection bias due to tissue availability. 19 Multivariable IPW-adjusted Cox proportional hazard regression models were used to adjust for potential confounders. The multivariable IPW-adjusted Cox proportional hazard regression models initially included sex (female vs. male), age at diagnosis (continuous), year of diagnosis (continuous), family history of colorectal cancer in any first-degree relative (present vs. absent), tumour location (proximal colon vs. distal colon vs. rectum), disease stage (I vs. II vs. III vs. IV), tumour differentiation (well/ moderate vs. poor), MSI status (MSI-high vs. non-MSI-high), CIMP (low/negative vs. high), KRAS (mutant vs. wild type), BRAF (mutant vs. wild type), PIK3CA (mutant vs. wild type), LINE-1 methylation level (continuous), PTGS2 expression (positive vs. negative) and nuclear CTNNB1 expression (positive vs. negative). A backward elimination was conducted with a threshold P of 0.05 to select variables for the final models. Cases with missing data (family history of colorectal cancer in a first-degree relative (0.3%) and tumour location (0.4%)) were included in the majority category of a given categorical covariate to limit the degrees of freedom of the models. For the cases with missing data on LINE-1 methylation (13.0%), we assigned a separate indicator variable. For cases with missing information on MSI status (11.6%), CIMP status (14.6%), KRAS mutation (14.6%), BRAF mutation (10.7%), PIK3CA mutation (17.1%), PTGS2 (15.7%) and CTNNB1 (35.8%), we assigned a separate missing indicator variable. We confirmed that excluding the cases with missing information in any of the covariates did not substantially alter the results (data not shown). For the analyses using a subset of cases with available neoantigen load data, we included neoantigen load (continuous) to the multivariable IPWadjusted Cox proportional hazard regression models in addition to the aforementioned potential confounders. The proportionality of hazards assumption in colorectal cancer survival was assessed by a time-varying covariate, which was an interaction term of survival time and the level of lymphocytic reaction (P > 0.27). We observed evidence on violation of this assumption in the hazards for four lymphocytic reaction components and the overall lymphocytic score in overall survival. However, the Schoenfeld residual plots supported the proportionality of hazards during most of the follow-up period up to 10 years (data not shown), and thus, we used Cox regression models limiting the follow-up period to 10 years. Cumulative survival probabilities were estimated using the IPW-adjusted Kaplan-Meier method, and a linear trend in survival probabilities across ordinal categories of the level of lymphocytic reaction was assessed using the weighted log-rank test for trend. For analyses of colorectal cancer-specific survival, participants were censored at the time of deaths from other causes.
In secondary analyses, we assessed the statistical interaction between levels of four lymphocytic reaction components (negative/low vs. intermediate vs. high) and each of following features: MSI status (high vs. non-high), neoantigen load (high vs. low), year of diagnosis (1995 or before vs. 1996-2000 vs. 2001-2008) and tumour location (proximal colon vs. distal colon vs. rectum), using the Wald test in the multivariable-adjusted Cox proportional hazard regression model for colorectal cancer mortality. We estimated HR for a unit increase of each lymphocytic reaction component in strata of MSI status, neoantigen load, year of diagnosis and tumour location using re-parameterisation of the interaction term in a single regression model. 27 In all survival Cox regression analyses, the IPW method was applied to reduce the potential bias due to the availability of tumour tissue. [19][20][21] Using the multivariable logistic regression model for the entire dataset of colorectal cancer cases (regardless of available tissue), we estimated the probability of the availability of tumour tissue, as previously described. 25 Each patient with complete data was weighted by the inverse probability. Weights greater than the 95th percentile were truncated and set to the value of the 95th percentile to reduce outlier effects. 21 We confirmed that the results without weight truncation did not change substantially (data not shown). The Cox regression analyses without IPW yielded similar results to the IPW-adjusted model.

RESULTS
We used covariate data of 4420 rectal and colon carcinoma cases in the two prospective cohort studies for the inverse probabilityweighting (IPW) method to adjust for selection bias due to tissue availability. 19 In 1465 cases, we examined lymphocytic reaction patterns: tumour-infiltrating lymphocytes (TIL, 1461 cases), intratumoural periglandular reaction (1462 cases), peritumoural lymphocytic reaction (1456 cases) and Crohn's-like lymphoid reaction (1195 cases) ( Table 1; Supplementary Table S1). All of the To test our primary hypothesis, we examined the relationship between each lymphocytic reaction component and patient mortality (Table 2). Higher levels of each component were associated with better cancer-specific survival (P trend < 0.002) and better overall survival (P trend < 0.009) in multivariable Cox regression analyses. Compared with cases with negative/low intratumoural periglandular reaction, multivariable-adjusted HRs for colorectal cancer-specific mortality were 0.55 (95% confidence interval (CI), 0.42-0.71) in cases with intermediate reaction, and 0.20 (95% CI, 0.12-0.35) in cases with high reaction. The Cox regression analyses without IPW yielded similar results to the IPW-adjusted model (Supplementary Table S2). When we adjusted for neoantigen load, as well as MSI, these findings remained largely unchanged (P trend < 0.1 for cancer-specific survival and P trend < 0.02 for overall survival, Supplementary  Table S3). In Kaplan-Meier survival analyses, each lymphocytic reaction component was positively associated with favourable colorectal cancer-specific survival (P < 0.0001 by the log-rank test for trend, Fig. 2).
As secondary analyses, we examined lymphocytic reaction and patient survival in strata of MSI status or neoantigen load. The prognostic associations of lymphocytic reaction were not significantly modified by either variable (P interaction > 0.2 for colorectal cancer-specific survival in strata of MSI status and neoantigen load, Tables 3 and 4).
We also examined patient survival according to the overall lymphocytic reaction score. In multivariable Cox regression analyses, a higher overall lymphocytic reaction score was associated with better colorectal cancer-specific survival and overall survival (P trend ≤ 0.0001 for both, Supplementary Table S4). The Cox regression analyses without IPW yielded similar results to the IPW-adjusted model (Supplementary Table S5). When we adjusted for neoantigen load as well as MSI, these findings remained unchanged (P trend = 0.0048 for cancer-specific survival and P trend = 0.0016 for overall survival, Supplementary Table S6). In Kaplan-Meier survival analyses, the overall lymphocytic reaction score was positively associated with favourable colorectal cancerspecific survival (P < 0.0001 by the log-rank test for trend, Supplementary Fig. S1).
As another secondary analysis, given the advance in the treatment strategy over the decades, we assessed the prognostic association of lymphocytic reaction in strata of the year of diagnosis and tumour location. The prognostic associations of lymphocytic reaction were not significantly modified by either variable (P interaction > 0.1 for colorectal cancer-specific survival in strata of year of diagnosis and tumour location, Supplementary  Tables S7 and S8).
As exploratory analyses, we assessed the prognostic interactions between the lymphocytic reaction components in relation to colorectal cancer-specific mortality. There was no prognostic interaction between the lymphocytic reaction components (P interaction > 0.1) (Supplementary Tables S9 and S10). To assess associations between the categories (negative/low, intermediate and high) of intratumoural periglandular reaction to colorectal cancer or tumourinfiltrating lymphocytes, and categorical data, the chi-square test was performed. To compare age, and LINE-1 methylation level, an analysis of variance was performed.
An integrated analysis of lymphocytic reaction, tumour molecular. . . K Haruki et al.

DISCUSSION
Utilising two US prospective cohort studies, we found that higher levels of each of four lymphocytic reaction components, and higher overall lymphocytic reaction score, were strongly associated with better colorectal cancer survival. Notably, these prognostic associations were not significantly modified by adjusting for potential confounders, including MSI, CIMP, BRAF mutation, LINE-1 methylation and neoantigen load. These findings provide strong population-based evidence for the role of host immunity in colorectal cancer prognosis. Since lymphocytic reaction can be examined by evaluating haematoxylin-and eosin-stained tissues, our study also supports the potential of lymphocytic reaction as a prognostic marker for colorectal cancer patients that could be readily implemented in clinical work.
Lymphocytic reaction has been demonstrated to reflect local immune effector response in colorectal cancer, associated with patient survival. [6][7][8][34][35][36][37] The assessment of the host immunity might also be helpful to advance current front-line immunotherapies, as immune checkpoint inhibitors aim to reactivate T-cellmediated antitumour immune response. [38][39][40] Evidence suggests that not only abundance but also spatial localisation of immune cells is prognostically relevant. [34][35][36][37] Our previous study using a population of 843 colorectal cancer patients has shown a significant positive association of lymphocytic reaction with favourable patient survival independent of tumour molecular characteristics, including CIMP, MSI status and LINE-1 hypomethylation. 5 Specifically, this association was most robust when using the overall lymphocytic score, while the four lymphocytic reaction components (Crohn's-like lymphoid reaction, peritumoural lymphocytic reaction, intratumoural periglandular reaction and TIL) had weaker associations with survival. This supported the value of grading different lymphocytic reaction components to generate a composite lymphocytic reaction score. Few other studies have evaluated the prognostic significance of such composite score, but some have reported that the individual components of the lymphocytic reaction, including TIL and Crohn's-like lymphoid reaction, are independently associated with lower colorectal cancer mortality after adjustment for MSI status. 7,8 In this study, with an expanded sample size (1465 cases) and additional important potential confounders (PIK3CA mutation, PTGS2 expression, nuclear CTNNB1 expression and neoantigen load), we identified a significant association of each of the four lymphocytic reaction components with colorectal cancer-specific survival independent of the potential confounders. In addition, only a few studies, including ours, evaluated "true" TIL that exists on top of tumour epithelium, 5,7 whereas most studies have not distinguished lymphocytes in tumour stromal regions (intratumoural periglandular reaction) from the true TIL. Thus, this study supports the robust prognostic value of both the overall lymphocytic reaction score and its four components, suggesting that the comprehensive characterisation of the lymphocytic infiltrate in different areal regions provides valuable information about the host antitumour immune response. Finally, IPW was used to minimise the potential selection bias caused by biospecimen availability. [19][20][21][22] The IPW method can utilise the information from all the incident 4420 colorectal cancer cases within the cohorts during the follow-up period in order to produce less-biased estimation of the prognostic association of lymphocytic reaction. The differences of the results between IPW-applied analysis and analysis without IPW were minor, which suggests that the selection bias may not be a major concern in this dataset, and supports the robustness of our current analyses. Colorectal cancer represents a heterogeneous group of tumours that result from not only a progressive accumulation of somatic molecular alterations, but also various host-tumour interactions, including antitumour immunity. [41][42][43] The assessment of host immunity against colorectal cancer in the tumour microenvironment is increasingly important in the translational research, and biomarkers representing tumour molecular characteristics and the immune microenvironment are likely to be more and more included in the future tumour pathology evaluation criteria. 6,35,44 Thus, integrated analyses of the immune response and tumour molecular features are necessary for the development of new immune biomarkers. In this study, we have included important confounders, including MSI, CIMP, LINE-1 methylation, KRAS, BRAF and PIK3CA mutation, PTGS2 expression, nuclear CTNNB1 expression and neoantigen load. Neoantigens are the most interesting targets for immunotherapies since neoepitopes are not subject to central tolerance in the thymus. 45 Peptides of neoantigens bound to HLA can be recognised by T cells, which initiate antitumour immune response. Our study further supports the finding that neoantigen load is positively associated with higher lymphocytic reactions in colorectal cancer patients. Importantly, the benefit associated with higher lymphocytic reaction was not significantly modified by the neoantigen load and other molecular features, confirming the independent role of lymphocytic reaction in colorectal cancer survival. To the best of our knowledge, there has been no previous study on lymphocytic reaction and patient survival, which has controlled as many molecular variables as we did in this study.
We need to point out several limitations in our study. First, there is limited data on cancer treatments in our study cohort. However, it was unlikely that clinical treatment decisions were influenced by lymphocytic reaction, because these data were not available to treating physicians. In addition, given the advances in colorectal cancer treatment, as well as differences between treatment strategies of colon and rectal carcinoma, we conducted stratified analyses according to year of diagnosis and tumour location. Second, data on cancer recurrences were not collected. However, colorectal cancer-specific mortality is considered as a reasonable colorectal cancer-specific outcome, since these two cohorts had a long follow-up duration of censored cases. Third, our study was based on evaluation of immune cells by haematoxylin-and eosin-stained tissue sections. Accumulating evidence suggests that specific immune cell types are differentially involved in host immune response. [34][35][36][37][46][47][48] Innate immune response also plays a crucial role in the tumour immune microenvironment, and may interact with adaptive immune cells. 40 Further identification of these immune cell types and immunoregulatory molecules, driving each component of the lymphocytic reaction, could contribute to better understanding of the tumour immune microenvironment. Finally, we had limited information of tumour pathological features in this study. Pathological features, such as lympho-vascular invasion, extramural vascular invasion, perineural invasion and tumour budding, represent potential unmeasured confounding factors of the current analyses. [49][50][51] The strengths of our study include utilising the two independent US prospective cohorts, which covered data on pathological findings and tumour molecular features. 12   based colorectal cancer database enabled us to rigorously examine the interactive prognostic value of lymphocytic reaction and each of lymphocytic reaction components, controlling for potential confounders. The molecular pathological epidemiology method has been utilised to assess the combined influences of exposures and immunity in cancer. In addition, compared with our previous study, an increased number of cases allow us to control for a larger group of confounders, and we utilised the IPW method to reduce the potential bias by the availability of colorectal cancer tissue.
In conclusion, a higher overall lymphocytic reaction score, along with four lymphocytic reaction components, is strongly associated with better colorectal cancer-specific survival, independent of MSI status, neoantigen load and other tumour and patient characteristics. Our population-based data support the role of host immune response as an independent prognostic indicator in colorectal cancer. The multivariable Cox regression model initially included sex, age, year of diagnosis, family history of colorectal cancer, tumour location, disease stage, tumour differentiation, microsatellite instability, CpG island methylator phenotype, KRAS mutation, BRAF mutation, PIK3CA mutation, long-interspersed nucleotide element-1 methylation level, PTGS2 (cyclooxygenase-2) expression and nuclear CTNNB1 (beta-catenin) expression. A backward elimination with a threshold P of 0.05 was used to select variables for the final models.