The Prognostic Role of Macrophage Polarization in the Colorectal Cancer Microenvironment

Globally, colorectal cancer is the third most common malignancy and the second most common cause of cancer-related deaths (1). Colorectal cancers elicit variable degrees of antitumor immune response (2), and high densities of tumor-associated T cells are well-established favorable prognostic markers (3, 4). However, the immune system may also facilitate tumor progression, as certain inflammatory pathways can enhance angiogenesis, activate tumor cell proliferation, inhibit apoptosis, promote extracellular matrix degradation and tumor invasion, or directly impair an antitumor immune response (5, 6).

Macrophages are among the most common nonneoplastic cells in human tumors, including colorectal cancer (7). Macrophages can exhibit broad functional plasticity in response to environmental stimuli, and the concept of “macrophage polarization” has been recognized as a measure of phenotypic state at any given point in space and time (8, 9). Traditionally, macrophages have been categorized as either proinflammatory M1-like macrophages or anti-inflammatory M2-like macrophages, although it is now generally recognized that a spectrum of phenotypic states exists between these two extremes of macrophage polarization (8, 10). In particular, tumor-associated macrophages, exposed to multiple polarization stimuli in the tumor microenvironment, may exhibit mixed phenotypes that are not readily classified as clearly M1 or M2 (11).

Several studies have evaluated the prognostic significance of macrophages in colorectal cancer and have generally reported an association between higher macrophage densities and better survival (12–16), although contradictory reports exist (17–19). Notably, most of these studies have either evaluated macrophages using only the general macrophage marker CD68 (12, 15), or have used single polarization markers to identify M1-like and M2-like populations (13, 14, 16, 18, 19), providing a limited view of macrophage phenotypic diversity.

To address this knowledge gap, we utilized multiplex immunofluorescence to identify and quantify macrophages in tumor epithelial and stromal regions and to estimate their polarization state across an M1:M2 spectrum in tissue microarrays (TMA) of 931 patients with colorectal cancer from two U.S.-based large prospective cohort studies. This multiplexed assay enabled us to characterize macrophage polarization by simultaneously measuring the expression of two markers associated with an M1-like antitumorigenic phenotype (CD86, IRF5) and two markers associated with an M2-like protumorigenic phenotype (MAF, MRC1). Our primary hypotheses were that higher density of M1-like polarized macrophages, lower density of M2-like polarized macrophages, and higher M1:M2 density ratio would be associated with lower colorectal cancer–specific mortality. In secondary analyses, we evaluated the relationships between the densities of macrophage subsets and tumor and patient characteristics and assessed the prognostic role of macrophage densities in strata of microsatellite instability (MSI) status. This approach allowed us to define the prognostic significance and clinicopathologic correlates of both macrophage density and polarization in colorectal cancer with greater sophistication and accuracy than prior studies.

This study used 4,420 colorectal cancer cases that occurred in two U.S. nationwide prospective cohort studies, the Nurses' Health Study (NHS; 121,701 women followed since 1976; ref. 20) and the Health Professionals Follow-up Study (HPFS; 51,529 men followed since 1986; ref. 21), including 931 cases with available follow-up data and adequate tissue specimens in TMAs (Table 1). We included both colon and rectal carcinomas in accordance with the colorectal continuum model (22). The medical records related to colorectal cancer were reviewed by study physicians, and tumor, node, metastases (TNM) stage was evaluated according to the American Joint Committee on Cancer criteria. The National Death Index was used to ascertain deaths and to identify unreported lethal colorectal cancer cases. Survival time was defined as the period from the date of colorectal cancer diagnosis to death or the end of follow-up (January 1, 2014, for HPFS; May 31, 2014, for NHS) for those who had not died.

Formalin-fixed paraffin-embedded (FFPE) tissue samples were collected, and hematoxylin and eosin (H&E)–stained tissue sections were reviewed by a single study pathologist (S. Ogino) to confirm the diagnosis and categorize tumor grade as high (≤50% gland formation) or low (>50% gland formation). Four histologic patterns of lymphocytic reactions, Crohn's-like lymphoid reaction, peritumoral lymphocytic reaction, intratumoral periglandular reaction, and tumor-infiltrating lymphocytes (TIL), were scored using a 4-tiered scale (23). DNA was extracted from FFPE tumor blocks using QIAmp DNA Mini Kit (Qiagen; ref. 24) to evaluate MSI and CpG island methylator phenotype (CIMP) status (25), KRAS, BRAF, and PIK3CA mutations (26), and neoantigen load (Supplementary Table S1). TMAs were constructed to include 1–4 cores (diameter 0.6 mm) from each tumor, selected to best represent overall morphology (27). Patients who received any type of neoadjuvant therapy prior to resection were excluded from the TMA set. The study was conducted in accordance with the U.S. Common Rule. All participants gave written informed consent for the study. The study protocol was approved by the Institutional Review Boards of the Brigham and Women's Hospital and Harvard T.H. Chan School of Public Health (Boston, MA), and those of participating registries as required.

We designed a 7-plex immunofluorescence panel to characterize macrophage polarization, incorporating a previously validated tyramide signal amplification technique to maximize sensitivity (Fig. 1; Supplementary Table S2; refs. 28, 29). On the basis of the published literature (8, 10–14, 16, 18, 19, 30, 31), a broad variety of target antigen candidates were screened with available antibody clones to evaluate suitability for panel inclusion, related to the level of evidence supporting association with macrophage polarization states in human tissue and antibody clone performance in FFPE tissue specimens. Candidate antibodies were first evaluated using standard immunohistochemistry (IHC) to assess their degree of expression pattern correspondence with prior studies and their biologically expected distribution using a bank of colorectal cancer tissue samples and normal lymphoid tissue controls. Correspondence was assessed both at a tissue level and at the level of subcellular localization. Once contextual specificity was assured, single-plex immunofluorescence was then performed, combining the selected antibody clones with Opal fluorophores (Akoya Biosciences) to further evaluate performance and to assess signal-to-noise ratio. On the basis of this testing, we compiled a multiplex immunofluorescence panel consisting of a pan-macrophage marker (CD68), two M1 phenotype markers [CD86 (32, 33), IRF5 (10, 31)], two M2 phenotype markers [MRC1 (10, 34), MAF (30, 35, 36)], cytokeratin (KRT), and DAPI (Supplementary Table S2). We optimized the sequence of the antibodies and their concentrations, as well as the fluorophore pairing and concentration, confirming similar staining patterns in standard IHC and multiplex immunofluorescence (Supplementary Fig. S1). Immunofluorescence staining was performed using an automated Leica Bond RX Research Stainer (Leica Biosystems), as described in Supplementary Fig. S2.

The immunofluorescence slides were scanned using the Vectra 3.0 Automated Quantitative Pathology Imaging System (Akoya Biosciences) equipped with a 20× objective and configured to capture images at discrete intervals from 420 to 720 nm (Supplementary Fig. S3). We used slides stained with a single fluorophore to generate a spectral library, enabling the unmixing of spectrally overlapping fluorophores, as well as a negative control slide processed with no fluorophores to profile background autofluorescence. Utilizing the captured spectral data and the fluorophore spectral library, the TMA core images were decomposed into individual fluorophore components using the inForm software package (Akoya Biosciences). Representative images were then processed with pathologist-trained machine learning algorithms implemented using the inForm software package to perform tissue category segmentation (tumor epithelium, stroma, other; using KRT expression to delineate epithelial areas), cell segmentation (using the DAPI signal to identify nuclei), and cell type classification [macrophage, tumor cell, other; using a combination of cellular morphology, CD68 expression (macrophages), and KRT expression (tumor cells) to distinguish these phenotypes; Fig. 1A–D; Supplementary Fig. S3].

Data were exported from inForm at the single-cell level for each core and further processed using the R statistical programming language version 3.5.3 (R Foundation for Statistical Computing). We inspected the consistency of the data across the TMAs using violin plots of cell densities and marker intensities. In addition, we used t-distributed stochastic neighbor embedding (tSNE) to project high-dimensional, cell-level fluorophore intensity data into two-dimensional space for visual inspection of potential heterogeneity in fluorophore intensities across all TMAs. For each CD68⁺ macrophage, an M1:M2 polarization index was calculated based on the scaled (0–1) mean intensities (CD86, MRC1: cytoplasmic; IRF5, MAF: nuclear) of the four polarization markers using the formula “(CD86 × IRF5)/(MRC1 × MAF)” (Fig. 1). In this approach, macrophages with a high index value are more strongly M1-polarized, while macrophages with a low index value are more strongly M2-polarized. As estimates of the prevalence of M1-like and M2-like macrophages in colorectal cancer vary significantly according to the markers and methods used for their detection (13, 14, 16), we opted to examine the distribution of the M1:M2 index across all macrophages within the 931 tumors and to categorize equal percentages of macrophages as M1-like or M2-like. For the main analyses, macrophages with M1:M2 index values within the highest 30% were considered M1-like, while those with M1:M2 index values within the lowest 30% were considered M2-like. These data were then combined into aggregate measures at the level of individual cores and individual tumors, with densities of M1-like and M2-like macrophages measured separately. To evaluate the robustness of this approach, we also evaluated a range (10%–50%) of M1:M2-index cutoff points, in addition to the 30% cutoff, for classifying macrophages as either M1-like or M2-like. For a secondary analysis, we defined macrophage populations (CD86⁺, IRF5⁺, MAF⁺, MRC1⁺) based on visually estimated fluorophore signal intensity cutoff points for each polarization marker. The code used to calculate the M1:M2-index and assign macrophage phenotypes is freely available (https://github.com/MPE-Lab).

The statistical analyses were conducted using SAS software (version 9.4, SAS Institute, Cary, NC). All P values were two-sided. Our primary hypothesis tested relationships between the intraepithelial and stromal densities of M1-like and M2-like macrophages, as well as the M1:M2 density ratio, and cancer-specific mortality using multivariable adjusted Cox proportional hazards regression models. We used the stringent two-sided α level of 0.005 (37). All other analyses represented secondary analyses, in which we interpreted our data cautiously, in addition to using the stringent α level of 0.005.

We used the χ² test and Spearman rank correlation test to assess the relationships between macrophage densities and clinicopathologic features, as appropriate. Cumulative survival probabilities were estimated using the Kaplan–Meier method and were compared using the log-rank test. Univariable and multivariable Cox proportional hazards regression models were used to estimate the HRs and confidence intervals (CI) of cancer-specific survival according to ordinal macrophage density or density ratio quartiles (Q1–Q4). We assessed the statistical interaction between macrophage densities or density ratio (low vs. high) and MSI status (high vs. non-high) or year of diagnosis (1995 or before vs. 1996–2000 vs. 2001–2008) in relation to cancer-specific survival using the Wald test for the cross-product in multivariable-adjusted Cox regression models. We utilized the inverse probability weighting method, using covariate data from the 4,420 incident colorectal cancer cases with or without tumor tissue, to adjust for selection bias due to tissue availability (38). A detailed description of the statistical methods is shown in Supplementary Table S3.

Multiplex immunofluorescence staining to evaluate macrophage density and polarization was completed across all cohort TMAs as a single batch, yielding data for 2,942,057 tumor cells and 367,024 macrophages in 931 colorectal carcinomas. Cell densities (Supplementary Fig. S4) and marker intensities (Supplementary Fig. S5) were consistent across TMAs. T-SNE analysis of single-cell level marker intensities did not show evidence of TMA-level batch effect (Supplementary Fig. S6), suggesting that the assay had performed in a technically uniform manner across all TMAs and that the data were suitable for aggregate analysis.

Data from 1,682 cores from 931 tumors (average 1.8 cores/tumor, range 1–4) were processed to calculate macrophage densities. Macrophages were present in all tumors, and their density was higher in tumor stromal regions (median 866/mm², IQR 485–1,414/mm²) than within tumor intraepithelial regions (median 163/mm², IQR 79–323/mm²). Intraepithelial and stromal macrophage densities showed moderate positive correlation [Spearman rank correlation coefficient (ρ) = 0.66]. Each macrophage was also assigned a position on macrophage polarization spectrum via an M1:M2 index based on joint measurement of two M1 (CD86, IRF5) and two M2 (MAF, MRC1) markers (Fig. 1). When examining the extreme 30% ends of this spectrum, the densities of M1-like and M2-like macrophages were not strongly correlated (intraepithelial: Spearman ρ = 0.09; stromal: Spearman ρ = −0.16).

The Spearman ρ between macrophage densities in two randomly chosen cores of tumors with two or more cores were 0.57 and 0.60 for intraepithelial and stromal macrophages, 0.53 and 0.59 intraepithelial and stromal M1-like macrophages, and 0.61 and 0.65 for intraepithelial and stromal M2-like macrophages (Supplementary Fig. S7), indicating moderate core-to-core correlation.

During the median follow-up time of 12.2 years (IQR 8.7–16.3 years) for censored cases, there were 567 all-cause deaths, including 288 colorectal cancer–specific deaths. Kaplan–Meier analysis demonstrated an association between higher stromal macrophage density and better cancer-specific survival (log-rank P = 0.001; Supplementary Fig. S8). However, when adjusted for potential confounders in multivariable Cox regression models, overall macrophage densities in the tumor intraepithelial region (P_trend = 0.37), tumor stroma (P_trend = 0.33), or overall region (P_trend = 0.73) were not significantly associated with cancer-specific survival (Table 2; Supplementary Table S4).

As our primary analysis, we examined the prognostic role of macrophages with either M1-like or M2-like polarization as defined by the most extreme 30% ends of the M1:M2 index (Fig. 2; Table 2). We found that higher stromal densities of M1-like macrophages were associated with a tendency toward longer cancer-specific survival (Q4 vs. Q1: HR = 0.69; 95% CI, 0.47–1.01; P_trend = 0.019), while higher stromal densities of M2-like macrophages were associated with shorter cancer-specific survival (Q4 vs. Q1: HR = 1.48; 95% CI, 1.01–2.17; P_trend = 0.005) in multivariable Cox regression models. These results suggested that the relatively weak survival associations for overall macrophage densities might be due to the fact that this population was composed of discrete subpopulations with opposing survival effects. Accordingly, higher stromal M1:M2 density ratio, reflecting M1-like macrophage polarization irrespective of overall macrophage density, was associated with longer cancer-specific survival (Q4 vs. Q1: multivariable adjusted HR = 0.58; 95% CI, 0.39–0.86; P_trend = 0.005; Table 2 and Table 3). As a secondary analysis, we examined the prognostic value of macrophage populations defined by positivity for single polarization markers. We found no significant associations between the densities of these populations and survival at the α level of 0.005 (Supplementary Table S5).

Although we initially selected a 30% cutoff point for defining macrophage polarization to balance the total number of macrophages assigned to either the M1-like or M2-like categories against the strength of macrophage polarization at a single cell level, we sought to evaluate the robustness of these results by testing a range of different cutoff points from 10% to 50% (Supplementary Figs. S9–S11). Although a 50% cutoff point assigns all identified macrophages into either an M1-like or M2-like category, it also results in both categories containing macrophages with widely varying degrees of polarization. Conversely, a 10% cutoff results in smaller populations of more uniformly polarized macrophages. These analyses indicated that the favorable prognostic tendency of stromal M1-like polarized macrophages remained quite consistent across the range of cutoff points (Supplementary Fig. S9), whereas the adverse prognostic tendency of stromal M2-like macrophages was limited to the extreme 10% to 30% most strongly polarized macrophages (Supplementary Fig. S10). High stromal M1:M2 density ratio was associated with lower colorectal cancer–specific mortality within the entire range of M1:M2 index cutoff points (P_trend ≤ 0.008 in multivariable adjusted models; Supplementary Fig. S11).

As secondary analyses, we examined the associations between macrophage densities and tumor and patient characteristics (Table 1 and Fig. 3). MSI status was a strong determinant of overall macrophage densities; tumors with high-level MSI (an MSI-high phenotype) had higher densities of both intraepithelial and stromal macrophages (P < 0.001). Moreover, higher intraepithelial and stromal macrophage densities were observed in tumors harboring other characteristics commonly associated with an MSI-high phenotype, including high tumor grade and a CIMP-high phenotype (P < 0.001 for both). Higher intraepithelial macrophage densities were also associated with proximal tumor location and higher neoantigen load (P < 0.001 for both). Tumors that were diagnosed in more recent years were associated with higher macrophage densities (P < 0.001) and were also associated with older patient age (P < 0.001), CIMP-high phenotype (P = 0.004), and high LINE-1 methylation level (P < 0.001; Supplementary Table S6).

When examining macrophages defined by the most extreme 30% cutoffs for the polarization index, higher stromal densities of M1-like macrophages (but not M2-like macrophages) were associated with an MSI-high phenotype (P < 0.001) and higher neoantigen load (P = 0.004), whereas in the tumor epithelial region, higher densities of both M1-like and M2-like macrophages were associated with an MSI-high phenotype, CIMP-high phenotype, high neoantigen load, and poor differentiation (P < 0.001 for all; Fig. 3). Taken together, these results suggested that, compared with non–MSI-high tumors, MSI-high tumors were characterized by more pronounced macrophage infiltration adjacent to tumor cells, and stronger M1-like polarization of stromal macrophages, which may reflect the higher immunogenicity of these tumors (39).

We also evaluated whether the densities of differentially polarized macrophages were associated with histologic lymphocytic reaction patterns (Supplementary Fig. S12). We found that higher intraepithelial and stromal M1-like macrophage densities were associated with more intense lymphocytic reactions of all four types (TILs, intratumoral periglandular reaction, peritumoral reaction, Crohn's-like reaction; P < 0.001 for all), whereas intraepithelial M2-like macrophage density showed significant, but relatively weaker, association with TILs (P < 0.001). Stromal M2-like macrophage density was not significantly associated with any of the four lymphocytic reaction patterns (P > 0.24).

Overall macrophage densities were not significantly associated with disease stage (Fig. 3). When macrophage polarization was evaluated, stromal M1-like macrophages tended to be more common in lower stage tumors (P = 0.015), while M2-like macrophages tended to be more common in higher stage tumors (P = 0.046), although these findings did not reach statistical significance at the stringent α level of 0.005.

Given that we identified a strong association between macrophage densities and MSI status, we performed survival analyses stratified by MSI status to further characterize the relationship between these tumor features (Supplementary Table S7). We found that high stromal densities of overall macrophages (P_interaction = 0.011) and M1-like macrophages (P_interaction = 0.057) tended to be more strongly associated with better survival in MSI-high tumors compared with non-MSI-high tumors, although these findings did not reach statistical significance at an α level of 0.005 and need to be interpreted cautiously because of low numbers of events in patients with MSI-high tumors. There were no significant differences in survival associations for macrophage densities stratified by year of diagnosis (Supplementary Table S8).

We performed quantitative, multiplexed analysis of macrophage polarization and density in colorectal cancers from patients in two large U.S. nationwide prospective cohort studies. This analysis revealed that the overall macrophage population had heterogenous prognostic significance, with M1-like and M2-like phenotypes exhibiting opposite effects. Higher M1:M2 density ratio, capturing the degree of M1-like macrophage polarization regardless of overall density, was associated with lower colorectal cancer–specific mortality, independent of potential confounding factors, including tumor stage and tumor molecular features.

During the past few decades, there have been significant conceptual and experimental advances in the understanding of macrophage biology (8). Single-cell RNA sequencing studies increasingly show that the macrophage population exhibits more heterogeneity than previously appreciated, particularly in inflammatory conditions such as cancer (40). These studies and others have demonstrated that macrophages exist along a spectrum of polarization states, with M1-like proinflammatory and M2-like anti-inflammatory phenotypes representing the extremes of this spectrum (8, 9, 33, 40). However, there is a lack of specific markers that can cleanly separate these populations (8, 10), necessitating multimarker approaches for the detection of differentially polarized macrophage subsets. Developments in multiplex immunofluorescence have enabled simultaneous visualization of multiple protein targets in a single tissue section (41). Although the number of markers in these assays is limited compared with some other methods such as bulk RNA sequencing or single-cell sequencing, a key benefit is the ability to examine macrophages in the spatial context of the tumor microenvironment across hundreds of specimens.

The multiplex immunofluorescence panel utilized in this study included two markers associated with an M1-like phenotype and two associated with an M2-like phenotype. Of these markers, IRF5 is a key transcription factor promoting inflammatory M1-like macrophage polarization, including induction of IL12A and IL12B as well as repression of IL10 in macrophages (31). Transcription factor MAF is associated with an M2-like homeostatic phenotype and is downregulated in the IFNG-driven M1-like phenotype (35). CD86 is an important costimulatory signal for antigen presentation to T cells and is commonly associated with an M1-like phenotype (32, 42, 43). MRC1 (CD206) is a scavenger receptor contributing to multiple cellular functions, including the endocytosis of mannosylated inflammatory glycoproteins (44), and is often found to be highly expressed in M2-like macrophages (10). Our four-marker approach had a balanced distribution of M1 and M2 markers, as well as transcription factors and cell surface markers that were spatially distinct at a subcellular level, enabling us to study the macrophage polarization spectrum via an M1:M2 index based on marker signal intensities within each individual macrophage.

We found that overall macrophage density was not significantly associated with survival in multivariable Cox regression models. However, this result was likely due to the fact that the overall population was comprised of M1-like cells with a favorable prognostic effect and M2-like cells with an adverse effect. Notably, high M1:M2 density ratio in tumor stroma, representing M1-like polarization of tumor stromal macrophages, was associated with lower cancer-specific mortality independent of other tumor or patient features, including tumor stage, MSI status, and KRAS, BRAF, and PIK3CA mutation status. This finding suggested that macrophage polarization state had prognostic value beyond common tumor molecular and clinical features. Interestingly, a study suggested that high macrophage density in colorectal cancer could predict benefit from 5-fluorouracil adjuvant therapy (45); further studies are warranted to assess whether this effect is dependent on macrophage polarization.

Prior studies have identified numerous potential mechanisms underlying the diverse roles for M1-like and M2-like macrophages in the tumor microenvironment (46, 47). Briefly, antitumorigenic M1-like macrophages may phagocytose tumor cells and efficiently present antigen to lymphocytes, thereby enhancing the antitumor immune response, while protumorigenic M2-like macrophages have been implicated in anti-inflammatory effects (46). Supporting the divergent roles of M1-like and M2-like macrophage populations in dictating the status of the antitumor lymphocytic response, we found that higher density of M1-like but not M2-like macrophages in tumor stroma positively correlated with the degree of four histologic lymphocytic reaction patterns (TILs, intratumoral periglandular reaction, peritumoral reaction, Crohn's-like reaction).

The progression of colorectal cancer is frequently associated with a systemic inflammatory response, characterized by the release of inflammatory mediators into the circulation, thereby supporting tumor growth and dissemination by numerous mechanisms including promotion of tumor cell proliferation and shaping premetastatic niches to promote subsequent metastasis (48, 49). We found that a lower intraepithelial M1:M2 density ratio was associated with advanced disease stage. Although a previous study did not indicate strong correlations between CD68⁺ macrophage densities in tumor tissue and serum cytokine levels (50), future studies could evaluate whether the local polarization of macrophages might reflect or contribute to a systemic inflammatory response.

Two major forms of genomic instability, namely chromosomal instability and MSI, are recognized in colorectal carcinoma (51). MSI-high tumors are characterized by a deficient mismatch repair system, leading to a hypermutated phenotype and production of immunogenic neoantigens (39). This often results in a prominent lymphocytic infiltrate in these tumors, which is associated with favorable clinical outcome (3, 4, 23, 52). We found that MSI status was an important determinant of macrophage densities in the colorectal cancer microenvironment, as MSI-high tumors had higher intraepithelial macrophage densities regardless of polarization, and higher densities of M1-like but not M2-like macrophages in tumor stroma. These results suggest that these immunogenic tumors are characterized by M1-polarization of stromal macrophages, as well as close contact between macrophages and tumor cells. However, there is also variation within the MSI-high tumors, and a case report describes high M2-like macrophage infiltration in an immune checkpoint blockade–resistant patient with an MSI-high tumor (53). Although part of the prognostic effect of macrophage infiltrates observed in our univariable analyses was attributable to associations between MSI status and macrophage densities, multivariable and MSI-stratified Cox regression models indicated that macrophage polarization harbored prognostic value independent of MSI status.

Several important limitations need to be considered in the interpretation of these results. First, the study was based on TMAs, and macrophage densities and polarization in small tissue samples may not perfectly represent the entire tumor. However, we examined an average of 1.8 cores per tumor and observed reasonable core-to-core correlation between macrophage densities, supporting the validity of this approach. Moreover, TMAs enabled us to examine a large number of samples stained uniformly as one batch. Automated computer-based image analysis ensured that all the images were processed in a uniform manner, reducing the effects of analytical imprecision. Second, we tested multiple hypotheses in this observational study. However, we had specific primary hypotheses, and we used a stringent α level of 0.005 to reduce the risk of type I statistical error (37). Third, although our 7-plex immunofluorescence panel is capable of more comprehensively characterizing macrophage polarization than single- or double-marker assays, the number of included polarization markers was still limited. We recognize the need for further validation of our approach in other cohorts as well as examination of additional markers to further define polarization stages and identify CD68⁻ macrophage populations. Fourth, as the estimates of the prevalence of M1-like and M2-like macrophages in colorectal cancer vary significantly according to the markers and methods used for their detection (13, 14, 16), we decided to examine the 30% extreme ends of the polarization spectrum in the main analyses. We acknowledge that this may not reflect the true distribution of the anti- and proinflammatory macrophage phenotypes in the colorectal cancer microenvironment. However, sensitivity analyses supported the robustness of the results across a wide range of cutoff points to define M1-like (10%–50%) and M2-like (10%–30%) phenotypes. Fifth, patients in the cohort were diagnosed and treated over several decades, a time period long enough that clinical practice, including staging accuracy and treatment strategies, may have changed. Notably, macrophage densities were lower and CIMP-high tumors were less prevalent in patients diagnosed in 1995 or before compared with those diagnosed afterwards. However, there was no significant association between year of diagnosis and tumor cell density or cytokeratin fluorophore signal intensity, suggesting that macrophage density differences over time were not due to time-dependent loss of antigenicity in tissue samples. Taken together, these results raised the possibility of evolution in the molecular epidemiology of colorectal cancer during the study period. We also adjusted for year of diagnosis in the multivariable survival models, mitigating its potential effect on the main results. In addition, there were no significant differences in survival associations for macrophage densities stratified by year of diagnosis. Finally, treatment data were unavailable for our cohorts. However, it is unlikely that macrophage densities or polarization influenced the treatment decisions, as such data were not available during treatment periods. A major strength of this study was the availability of a rich molecular pathological epidemiology database, providing comprehensive data regarding tumor molecular characteristics and clinicopathologic features. The patients were recruited from a large number of hospitals across the United States, supporting the generalizability of the findings.

In conclusion, we measured macrophage M1:M2 polarization spectrum in the colorectal cancer microenvironment using a quantitative, multiplexed immunofluorescence assay. We found that macrophage polarization rather than absolute overall density was associated with colorectal cancer mortality, with M1-like and M2-like macrophages showing opposite effects. These findings support the utility of a multimarker approach for identifying biologically and prognostically relevant macrophage subpopulations in the tumor microenvironment.

S.A. Väyrynen reports grants from Finnish Cultural Foundation and Orion Research Foundation sr during the conduct of the study. K. Fujiyoshi reports grants from Uehera Memorial Foundation (research fellowship) during the conduct of the study. J.L. Guerriero reports grants and personal fees from GlaxoSmithKline and Array BioPharma; grants from Eli Lilly; and personal fees from Codagenix, Verseau Therapeutics, and Kymera during the conduct of the study. A.T. Chan reports grants and personal fees from Bayer Pharma AG and personal fees from Pfizer Inc. and Boehringer Ingelheim outside the submitted work. R. Nishihara reports other from Pfizer (employee) outside the submitted work. C.S. Fuchs reports personal fees from Agios, Amylin Pharma, Bain Capital, CytomX Therapeutics, Daiichi Sankyo, Eli Lilly, Entrinsic Health, Evolveimmune Therapeutics, Genentech, Merck, Taiho, and Unum Therapeutics outside the submitted work, serves as a director for CytomX Therapeutics, owns unexercised stock options for CytomX and Entrinsic Health, is a cofounder of Evolveimmune Therapeutics and has equity in this private company, and has provided expert testimony for Amylin Pharmaceuticals and Eli Lilly. J.A. Meyerhardt reports personal fees from COTA Healthcare (till December 2019), Taiho Pharmaceutical (NCCN Grant Review Panel), and Ignyta (single ad board) outside the submitted work. M. Giannakis reports grants from Bristol-Myers Squibb and Merck outside the submitted work. S. Ogino reports grants from NIH (grant number R35 CA197735) during the conduct of the study. J.A. Nowak reports a patent for an algorithm for identifying signatures of tumor-immune cell interactions within the tumor microenvironment pending. No disclosures were reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of NIH. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

J.P. Väyrynen: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. K. Haruki: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M.C. Lau: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. S.A. Väyrynen: Data curation, investigation, visualization, methodology, writing–original draft, writing–review and editing. R. Zhong: Formal analysis, investigation, writing–review and editing. A. Dias Costa: Investigation, methodology, writing–review and editing. J. Borowsky: Investigation, methodology, writing–review and editing. M. Zhao: Investigation, writing–review and editing. K. Fujiyoshi: Investigation, writing–review and editing. K. Arima: Investigation, writing–review and editing. T.S. Twombly: Investigation, writing–review and editing. J. Kishikawa: Investigation, writing–review and editing. S. Gu: Investigation, writing–review and editing. S. Aminmozaffari: Investigation, writing–review and editing. S. Shi: Investigation, writing–review and editing. Y. Baba: Investigation, writing–review and editing. N. Akimoto: Investigation, writing–review and editing. T. Ugai: Investigation, writing–review and editing. A. Da Silva: Investigation, writing–review and editing. J.L. Guerriero: Methodology, writing–review and editing. M. Song: Investigation, writing–review and editing. K. Wu: Funding acquisition, investigation, writing–review and editing. A.T. Chan: Funding acquisition, investigation, writing–review and editing. R. Nishihara: Funding acquisition, investigation, writing–review and editing. C.S. Fuchs: Funding acquisition, investigation, writing–review and editing. J.A. Meyerhardt: Data curation, funding acquisition, investigation, writing–review and editing. M. Giannakis: Conceptualization, data curation, funding acquisition, investigation, writing–review and editing. S. Ogino: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing. J.A. Nowak: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing.

The authors thank the participants and staff of the Nurses' Health Study and the Health Professionals Follow-up Study for their valuable contributions as well as the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY. The authors assume full responsibility for analyses and interpretation of these data. Portions of this research were conducted on the O2 High Performance Compute Cluster, supported by the Research Computing Group, at Harvard Medical School. See http://rc.hms.harvard.edu for more information. This work was supported by NIH grants (P01 CA87969 to M.J. Stampfer; UM1 CA186107 to M.J. Stampfer; P01 CA55075 to W.C. Willett; UM1 CA167552 to W.C. Willett; U01 CA167552 to W.C. Willett and L.A. Mucci; P50 CA127003 to C.S. Fuchs; R01 CA118553 to C.S. Fuchs; R01 CA169141 to C.S. Fuchs; R01 CA137178 to A.T. Chan; K24 DK098311 to A.T. Chan; R35 CA197735 to S. Ogino; R01 CA151993 to S. Ogino; K07 CA190673 to R. Nishihara; R03 CA197879 to K. Wu; R21 CA222940 to K. Wu and M. Giannakis; and R21 CA230873 to K. Wu and S. Ogino); by the Cancer Research UK Grand Challenge Award (UK C10674/A27140 to M. Giannakis and S. Ogino); by the Nodal Award (2016–02) from the Dana-Farber Harvard Cancer Center (to S. Ogino); by grants from the Project P Fund, The Friends of the Dana-Farber Cancer Institute, the Bennett Family Fund, and the Entertainment Industry Foundation through the National Colorectal Cancer Research Alliance; and by the Stand Up To Cancer Colorectal Cancer Dream Team Translational Research Grant (SU2C-AACR-DT22-17 to C.S. Fuchs and M. Giannakis). Stand Up To Cancer is a division of the Entertainment Industry Foundation. The SU2C research grant is administered by the American Association for Cancer Research, a scientific partner of SU2C. K. Haruki was supported by fellowship grants from the Uehara Memorial Foundation and the Mitsukoshi Health and Welfare Foundation. S.A. Väyrynen was supported by grants from the Finnish Cultural Foundation and Orion Research Foundation sr. J. Borowsky was supported by a grant from the Australia Awards-Endeavour Scholarships and Fellowships Program. K. Fujiyoshi was supported by a fellowship grant from the Uehara Memorial Foundation. K. Arima was supported by grants from Overseas Research Fellowship from Japan Society for the Promotion of Science (JP201860083). K. Wu was supported by an Investigator Initiated Grant from the American Institute for Cancer Research (AICR). A.T. Chan is a Stuart and Suzanne Steele MGH Research Scholar. J.A. Meyerhardt's research is supported by the Douglas Gray Woodruff Chair fund, the Guo Shu Shi Fund, the Anonymous Family Fund for Innovations in Colorectal Cancer, the Project P fund, and the George Stone Family Foundation. M. Giannakis was supported by a Conquer Cancer Foundation of ASCO Career Development Award.

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.