Although tumor-infiltrating T cells hold a beneficial prognostic role in colorectal cancer, other lymphocytic populations are less characterized. We developed a multiplexed immunofluorescence assay coupled with digital image analysis and machine learning to identify natural killer (NK) cells (NCAM1+CD3), natural killer T-like (NKT-like) cells (NCAM1+CD3+), and T cells (NCAM1CD3+) within the PTPRC+ (CD45+) cell population and to measure their granzyme B (GZMB; cytotoxicity marker) and FCGR3A (CD16a; NK-cell maturity marker) expression. We evaluated immune cell densities and spatial configuration in 907 incident colorectal carcinoma cases within two prospective cohort studies. We found that T cells were approximately 100 times more abundant than NK and NKT-like cells. Overall, NK cells showed high GZMB expression and were located closer to tumor cells than T and NKT-like cells. In T and NKT-like cells, GZMB expression was enriched in cells in closer proximity to tumor cells. Higher densities of both T and NKT-like cells associated with longer cancer-specific survival, independent of potential confounders (Ptrend < 0.0007). Higher stromal GZMB+ and FCGR3A+ NK-cell densities associated with longer cancer-specific survival (Ptrend < 0.003). For T and NKT-like cells, greater proximity to tumor cells associated with longer cancer-specific survival (Ptrend < 0.0001). These findings indicate that cytotoxic NCAM1+CD3GZMB+ NK cells and NCAM1+CD3+ NKT-like cells are relatively rare lymphocytic populations within the colorectal cancer microenvironment and show distinct spatial configuration and associations with patient outcome. The results highlight the utility of a quantitative multimarker assay for in situ, single-cell immune biomarker evaluation and underscore the importance of spatial context for tumor microenvironment characterization.

The prognostic classification of colorectal cancer largely relies upon anatomic extent, as captured using TNM staging, which is based on the invasion depth of the primary tumor (T) and the presence of nodal metastasis (N) and distant metastasis (M). However, accumulating evidence indicates that immune microenvironment features critically modulate tumor behavior and could be used to refine prognostic categories (1–3). Lymphocytes are a heterogenous group of adaptive and innate immune cells, and include T cells, B cells, and natural killer (NK) cells (4). Numerous studies have indicated that higher T-cell density in the colorectal cancer microenvironment associates with a favorable disease outcome (3), but the significance of other lymphocytic populations is less well established.

NK cells, initially described in the 1970s based on their ability to mediate killing of certain tumors and virus-infected cells without prior antigen exposure (5), are cytotoxic lymphocytes of the innate immune system (4). The best-characterized NK cell subsets in humans include immature NK cells that express NCAM1 (CD56) but neither CD3 nor FCGR3A (Fc gamma receptor 3A, CD16; i.e., NCAM1+CD3FCGR3A cells) and mature NK cells that express NCAM1 (CD56) and FCGR3A (CD16) but not CD3 (i.e., NCAM1+CD3FCGR3A+ cells) within the total PTPRC+ (CD45+) leukocyte population (4). Although the identification of such cells has traditionally been possible using multiparameter flow cytometry (6) and, more recently, using in situ multiplexed IHC methods (7), most studies evaluating the significance of lymphocytic cells in the colorectal cancer microenvironment have been based on single-color IHC (3) and have therefore been unable to accurately identify these cell populations.

In this study, we developed a customized, multiplex immunofluorescence assay to identify NK cells (PTPRC+NCAM1+CD3), NKT-like cells (PTPRC+NCAM1+CD3+), and T cells (PTPRC+NCAM1CD3+), along with expression of FCGR3A and granzyme B (GZMB) within these cell populations. Using digital image analysis and supervised machine learning, we applied this assay to an extensively characterized cohort of 907 colorectal cancers derived from two U.S. nationwide prospective cohort studies. To study the prognostic significance of these lymphocytic populations, we controlled for potential confounding and selection bias due to tissue data availability using the inverse probability weighting (IPW) method and covariate data from 4,465 colorectal cancer cases in the cohorts. We hypothesized that higher densities of all three cell types—in particular, their cytotoxic subpopulations—would associate with longer survival. We examined the spatial organization of these populations with respect to tumor cells and evaluated spatial organization prognostic value as an exploratory investigation.

Data availability

The data underlying this article cannot be shared publicly. Further information including the procedures to obtain and access data from the Nurses' Health Studies and Health Professionals Follow-up Study are described at https://www.nurseshealthstudy.org/researchers/ and https://sites.sph.harvard.edu/hpfs/for-collaborators/.

Study population

We documented 4,465 colorectal cancer cases that had occurred during follow-up (until 2012) of two prospective cohort studies in the U.S., namely the Nurses' Health Study (NHS, 121,701 women ages 30–55 years at enrollment in 1976) and the Health Professionals Follow-up Study (HPFS, 51,529 men ages 40–75 years at enrollment in 1986). Among those 4,465 cases, 907 tumors yielded multiplex immunofluorescence data that met all quality control metrics (Table 1; Supplementary Fig. S1). Based upon the continuum of tumor characteristics across the colorectum, we included both colon and rectal carcinomas (8). 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.

Table 1.

Immune cell densities in relation to characteristics of colorectal cancer cases.

Immune cell density (cells/mm2)
Median (25th–75th percentile)
NCAM1+CD3NCAM1+CD3+NCAM1CD3+NCAM1CD3FCGR3A+
CharacteristicaTotal NNK cellsNKT-like cellsT cellsmacrophages
All cases 907 2.4 (0.7–5.9) 2.9 (0.7–6.8) 434 (206–831) 550 (308–884) 
Sex 
 Female (NHS) 503 (55%) 2.5 (0.7–6.6) 3.1 (0.7–7.5) 440 (218–875) 580 (309–904) 
 Male (HPFS) 404 (45%) 2.3 (0.7–5.4) 2.8 (0.8–6.0) 432 (193–799) 514 (305–857) 
Age (years) 
 <65 280 (31%) 2.4 (0.7–5.8) 3.3 (0.9–7.8) 425 (210–799) 536 (293–871) 
 ≥65 627 (69%) 2.4 (0.6–6.0) 2.7 (0.6–6.6) 439 (205–859) 559 (310–895) 
Year of diagnosis 
 1995 or before 296 (33%) 2.3 (0.8–5.6) 2.8 (0.6–7.5) 428 (195–811) 499 (265–930) 
 1996–2000 296 (33%) 2.0 (0–6.5) 3.0 (0.8–6.8) 430 (213–805) 573 (320–920) 
 2001–2008 315 (34%) 2.6 (0.7–5.8) 2.9 (0.7–6.7) 447 (206–892) 569 (314–840) 
Family history of colorectal cancer in first-degree relative(s) 
 Absent 713 (79%) 2.4 (0.6–5.9) 2.9 (0.7–6.8) 430 (207–805) 545 (308–895) 
 Present 190 (21%) 2.5 (0.7–6.0) 3.1 (0.7–6.9) 491 (196–1,001) 572 (306–865) 
Tumor location 
 Cecum 161 (18%) 2.7 (0.7–6.7) 4.0 (1.5–8.5) 608 (241–1,051) 653 (378–1,162) 
 Ascending to transverse colon 291 (32%) 2.6 (0.7–9.5) 3.2 (0.8–7.7) 429 (165–876) 577 (309–904) 
 Descending to sigmoid colon 270 (30%) 2.0 (0.7–5.3) 2.9 (0.6–6.2) 418 (218–769) 468 (275–776) 
 Rectum 181 (20%) 2.0 (0–4.5) 2.0 (0–4.5) 411 (209–723) 536 (309–802) 
Tumor differentiation 
 Well to moderate 823 (91%) 2.2 (0.6–5.5) 2.9 (0.7–6.8) 432 (206–809) 535 (297–849) 
 Poor 83 (9.2%) 5.3 (2.2–12.7) 2.8 (1.2–7.6) 508 (207–1,122) 848 (479–1,180) 
AJCC disease stage 
 I 196 (23%) 3.1 (0.9–8.4) 3.4 (1.0–8.2) 610 (329–1,080) 558 (296–844) 
 II 277 (33%) 2.7 (0.9–7.3) 3.3 (0.7–7.8) 445 (206–859) 604 (355–983) 
 III 241 (29%) 1.9 (0.4–5.3) 2.6 (0.8–5.0) 411 (176–732) 558 (306–908) 
 IV 131 (16%) 1.4 (0–4.0) 1.9 (0–4.8) 269 (120–505) 484 (293–752) 
MSI status 
 Non–MSI-high 727 (83%) 1.9 (0–4.5) 2.5 (0.4–5.7) 409 (193–768) 507 (281–811) 
 MSI-high 153 (17%) 7.4 (3.0–18) 5.4 (2.1–15) 694 (280–1,180) 823 (469–1,162) 
CIMP status 
 Low/negative 684 (81%) 1.9 (0–4.5) 2.5 (0.5–5.8) 412 (195–757) 497 (281–796) 
 High 156 (19%) 5.8 (2.2–16) 4.4 (1.4–13) 579 (248–1,134) 823 (457–1,117) 
Mean LINE-1 methylation level 
 ≥60% 552 (63%) 2.5 (0.7–6.3) 3.0 (0.7–7.0) 459 (212–921) 585 (318–961) 
 <60% 328 (37%) 1.8 (0.6–5.2) 2.6 (0.6–6.6) 404 (170–754) 482 (285–810) 
KRAS mutation 
 Wild-type 531 (60%) 2.6 (0.7–7.7) 3.2 (0.8–8.2) 448 (207–944) 572 (322–931) 
 Mutant 348 (40%) 2.0 (0–4.1) 2.4 (0–5.6) 420 (196–746) 509 (264–850) 
BRAF mutation 
 Wild-type 749 (85%) 2.2 (0.5–5.3) 2.7 (0.6–6.2) 429 (201–797) 521 (296–842) 
 Mutant 137 (15%) 4.5 (1.4–14) 3.9 (1.2–11) 488 (205–986) 725 (446–1,038) 
PIK3CA mutation 
 Wild-type 692 (84%) 2.4 (0.7–5.9) 2.8 (0.7–6.7) 423 (195–830) 537 (302–858) 
 Mutant 134 (16%) 2.1 (0–5.6) 2.9 (0.6–6.8) 534 (258–980) 601 (323–997) 
Neoantigen load 
 Q1 (lowest) 103 (25%) 2.0 (0–4.1) 2.8 (0.8–5.0) 432 (241–907) 514 (301–774) 
 Q2 103 (25%) 1.5 (0–3.1) 1.7 (0–4.1) 384 (156–721) 483 (264–808) 
 Q3 103 (25%) 1.8 (0.6–5.3) 3.2 (1.3–6.8) 500 (247–937) 592 (369–1,013) 
 Q4 (highest) 102 (25%) 5.5 (1.3–17) 4.5 (1.6–12) 588 (310–1,103) 723 (381–990) 
Immune cell density (cells/mm2)
Median (25th–75th percentile)
NCAM1+CD3NCAM1+CD3+NCAM1CD3+NCAM1CD3FCGR3A+
CharacteristicaTotal NNK cellsNKT-like cellsT cellsmacrophages
All cases 907 2.4 (0.7–5.9) 2.9 (0.7–6.8) 434 (206–831) 550 (308–884) 
Sex 
 Female (NHS) 503 (55%) 2.5 (0.7–6.6) 3.1 (0.7–7.5) 440 (218–875) 580 (309–904) 
 Male (HPFS) 404 (45%) 2.3 (0.7–5.4) 2.8 (0.8–6.0) 432 (193–799) 514 (305–857) 
Age (years) 
 <65 280 (31%) 2.4 (0.7–5.8) 3.3 (0.9–7.8) 425 (210–799) 536 (293–871) 
 ≥65 627 (69%) 2.4 (0.6–6.0) 2.7 (0.6–6.6) 439 (205–859) 559 (310–895) 
Year of diagnosis 
 1995 or before 296 (33%) 2.3 (0.8–5.6) 2.8 (0.6–7.5) 428 (195–811) 499 (265–930) 
 1996–2000 296 (33%) 2.0 (0–6.5) 3.0 (0.8–6.8) 430 (213–805) 573 (320–920) 
 2001–2008 315 (34%) 2.6 (0.7–5.8) 2.9 (0.7–6.7) 447 (206–892) 569 (314–840) 
Family history of colorectal cancer in first-degree relative(s) 
 Absent 713 (79%) 2.4 (0.6–5.9) 2.9 (0.7–6.8) 430 (207–805) 545 (308–895) 
 Present 190 (21%) 2.5 (0.7–6.0) 3.1 (0.7–6.9) 491 (196–1,001) 572 (306–865) 
Tumor location 
 Cecum 161 (18%) 2.7 (0.7–6.7) 4.0 (1.5–8.5) 608 (241–1,051) 653 (378–1,162) 
 Ascending to transverse colon 291 (32%) 2.6 (0.7–9.5) 3.2 (0.8–7.7) 429 (165–876) 577 (309–904) 
 Descending to sigmoid colon 270 (30%) 2.0 (0.7–5.3) 2.9 (0.6–6.2) 418 (218–769) 468 (275–776) 
 Rectum 181 (20%) 2.0 (0–4.5) 2.0 (0–4.5) 411 (209–723) 536 (309–802) 
Tumor differentiation 
 Well to moderate 823 (91%) 2.2 (0.6–5.5) 2.9 (0.7–6.8) 432 (206–809) 535 (297–849) 
 Poor 83 (9.2%) 5.3 (2.2–12.7) 2.8 (1.2–7.6) 508 (207–1,122) 848 (479–1,180) 
AJCC disease stage 
 I 196 (23%) 3.1 (0.9–8.4) 3.4 (1.0–8.2) 610 (329–1,080) 558 (296–844) 
 II 277 (33%) 2.7 (0.9–7.3) 3.3 (0.7–7.8) 445 (206–859) 604 (355–983) 
 III 241 (29%) 1.9 (0.4–5.3) 2.6 (0.8–5.0) 411 (176–732) 558 (306–908) 
 IV 131 (16%) 1.4 (0–4.0) 1.9 (0–4.8) 269 (120–505) 484 (293–752) 
MSI status 
 Non–MSI-high 727 (83%) 1.9 (0–4.5) 2.5 (0.4–5.7) 409 (193–768) 507 (281–811) 
 MSI-high 153 (17%) 7.4 (3.0–18) 5.4 (2.1–15) 694 (280–1,180) 823 (469–1,162) 
CIMP status 
 Low/negative 684 (81%) 1.9 (0–4.5) 2.5 (0.5–5.8) 412 (195–757) 497 (281–796) 
 High 156 (19%) 5.8 (2.2–16) 4.4 (1.4–13) 579 (248–1,134) 823 (457–1,117) 
Mean LINE-1 methylation level 
 ≥60% 552 (63%) 2.5 (0.7–6.3) 3.0 (0.7–7.0) 459 (212–921) 585 (318–961) 
 <60% 328 (37%) 1.8 (0.6–5.2) 2.6 (0.6–6.6) 404 (170–754) 482 (285–810) 
KRAS mutation 
 Wild-type 531 (60%) 2.6 (0.7–7.7) 3.2 (0.8–8.2) 448 (207–944) 572 (322–931) 
 Mutant 348 (40%) 2.0 (0–4.1) 2.4 (0–5.6) 420 (196–746) 509 (264–850) 
BRAF mutation 
 Wild-type 749 (85%) 2.2 (0.5–5.3) 2.7 (0.6–6.2) 429 (201–797) 521 (296–842) 
 Mutant 137 (15%) 4.5 (1.4–14) 3.9 (1.2–11) 488 (205–986) 725 (446–1,038) 
PIK3CA mutation 
 Wild-type 692 (84%) 2.4 (0.7–5.9) 2.8 (0.7–6.7) 423 (195–830) 537 (302–858) 
 Mutant 134 (16%) 2.1 (0–5.6) 2.9 (0.6–6.8) 534 (258–980) 601 (323–997) 
Neoantigen load 
 Q1 (lowest) 103 (25%) 2.0 (0–4.1) 2.8 (0.8–5.0) 432 (241–907) 514 (301–774) 
 Q2 103 (25%) 1.5 (0–3.1) 1.7 (0–4.1) 384 (156–721) 483 (264–808) 
 Q3 103 (25%) 1.8 (0.6–5.3) 3.2 (1.3–6.8) 500 (247–937) 592 (369–1,013) 
 Q4 (highest) 102 (25%) 5.5 (1.3–17) 4.5 (1.6–12) 588 (310–1,103) 723 (381–990) 

Abbreviations: AJCC, American Joint Committee on Cancer; CIMP, CpG island methylator phenotype; HPFS, Health Professionals Follow-up Study; LINE-1, long-interspersed nucleotide element-1; MSI, microsatellite instability; NHS, Nurses' Health Study; SD, standard deviation.

aPercentage indicates the proportion of patients with a specific clinical, pathologic, or molecular characteristic among all patients.

Study physicians reviewed medical records and gathered clinical information [such as tumor size, location, and the American Joint Committee on Cancer tumor, node, metastases (TNM) classification, as well as causes of deaths for deceased participants]. The National Death Index (National Center for Health Statistics, Hyattsville, MD) was utilized to confirm deaths and 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, 2016 for HPFS; June 1, 2016 for NHS). For analyses of colorectal cancer–specific survival, deaths from other causes were censored.

We collected formalin-fixed paraffin-embedded tumor blocks from 1,619 study participants with colorectal cancer from hospitals throughout the United States where participants had undergone tumor resection. Hematoxylin and eosin–stained sections were reviewed to confirm the presence of invasive cancer, evaluate tumor differentiation, and mark areas for tissue microarray construction (2–4 cores from each tumor of 0.6 mm diameter; ref. 9). DNA was extracted using QIAmp DNA Mini Kit (Qiagen), and tumor microsatellite instability (MSI) status (based on 10 microsatellites: D2S123, D5S346, D17S250, BAT25, BAT26, BAT40, D18S55, D18S56, D18S67, and D18S487), CpG island methylator phenotype (CIMP) status (based on eight CIMP-specific promoters: CACNA1G, CDKN2A, CRABP1, IGF2, MLH1, NEUROG1, RUNX3, and SOCS1), long interspersed nucleotide element-1 (LINE-1) methylation level, KRAS (codons 12, 13, 61, and 146), BRAF (codon 600), and PIK3CA (exons 9 and 20) mutation status, and neoantigen load were determined as described previously (10, 11).

Multiplex immunofluorescence

To examine NK cells and other lymphocytic populations in the tumor microenvironment, we designed a custom multiplexed immunofluorescence (mIF) assay based on the tyramide signal amplification method (12). We first tested different antibody clones with chromogenic IHC in colorectal cancer samples from 8 patients (Supplementary Table S1), to evaluate the correspondence of the staining patterns to those described in the literature. We next evaluated immunofluorescent signal-to-noise ratios for antibody clones selected on the basis of their performance in standard chromogenic IHC, using Opal fluorophores (Akoya Biosciences). We then constructed a mIF assay targeting CD3 [T-cell surface marker (13)], FCGR3A [Fc gamma receptor 3A, CD16a; a receptor expressed by mature NK cells and macrophages involved in antibody-dependent cellular cytotoxicity (4)], GZMB [ cytotoxicity marker (13)], NCAM1 [CD56; cell surface glycoprotein with a role in cell-cell adhesion; expressed by NK cells and some other cells, such as neurons (14))], PTPRC [CD45; pan-leukocyte marker (15)], KRT (keratin; epithelial cell marker), and DAPI (nuclear marker). We iteratively optimized antibody-fluorophore pairing and concentrations, as well as the sequence of the staining, and confirmed the correspondence of multiplex immunofluorescence staining patterns with those of single-marker chromogenic IHC (Supplementary Fig. S2). The final mIF protocol was automated using a Leica Bond RX Research Stainer (Leica Biosystems; Supplementary Fig. S3).

Using our validated and automated assay, we processed two complete sets of tissue microarray sections (total: 20 sections), separated by a vertical depth of >20 μm, to increase the tissue area subjected to analysis, and included MHC class I antigen presentation molecules present in all nucleated cells (16)[either B2M (section 1) or HLA-A/HLA-B/HLA-C (section 2)] as additional quality controls to monitor the staining properties of individual samples. All slides were stained in a single batch to ensure uniform processing (Supplementary Fig. S3).

Image capture and analysis

We scanned the immunofluorescence slides using the Vectra Polaris Automated Quantitative Pathology Imaging System (Akoya Biosciences), equipped with seven imaging filters and a 20× objective. We separately constructed and profiled a spectral fluorophore and autofluorescence library to enable optimal multispectral unmixing (17, 18). Spectrally unmixed images were generated using the inForm software package (v1.4.8; Akoya Biosciences) and further analyzed using pathologist-supervised machine learning algorithms built in the inForm software package (Fig. 1; refs. 17, 18). We confirmed that staining intensities were consistent across the tissue microarrays, indicating uniform assay technical performance (Supplementary Fig. S4). Tissue regions were categorized into epithelial, stromal, and “other” (such as empty space or mucin) categories. Individual cells were segmented into nuclear, cytoplasmic, and membranous regions. Cells were then phenotyped into the PTPRC+NCAM1+CD3 (NK cells), PTPRC+NCAM1+CD3+ (NKT-like cells), PTPRC+NCAM1CD3+GZMB+ (GZMB+ T cells), PTPRC+NCAM1CD3+GZMB (GZMB T cells), PTPRC+NCAM1CD3FCGR3A+ (FCGR3A+ macrophages), PTPRC+NCAM1CD3FCGR3A (other PTPRC+ immune cells), KRT+PTPRC (tumor epithelial cells), and KRTPTPRC (other cells) categories. The cell phenotype classification implemented in the inForm software package was based on multinomial logistic regression utilizing image features derived from texture analysis and cell segmentation. Single cell–level data from inForm were exported for further processing using the R statistical programming language (v. 4.0.1; R Foundation for Statistical Computing). We further classified NK and NKT-like cells into GZMB+/GZMB and FCGR3A+/FCGR3A subsets based upon fluorophore signal intensity cut-points (cytoplasmic mean intensity 0.25 for GZMB and membranous mean intensity 8 for FCGR3A). We calculated immune cell density (cells/mm2) separately for tumor intraepithelial, stromal, and overall (intraepithelial and stromal) compartments. Core-to-core correlation for cell densities was moderate to high (Supplementary Fig. S5) and was higher for cell types with higher average densities (Spearman rho = 0.65 for NCAM1CD3+ T cells; 0.64–0.65 for NCAM1CD3FCGR3A+ macrophages; 0.36–0.44 for NCAM1+CD3 NK cells, and 0.40 for NCAM1+CD3+ NKT-like cells). For each cell subset, cases were classified into quartile categories (C1–C4) if ≤25% of cases had a density of zero. If >25% of cases had a density of zero for a specific cell type, all zero-density cases were grouped together (C1 category), and the remaining (non-zero) cases were divided into tertiles (C2–C4).

Figure 1.

Spatially informed analysis of immune cell populations in 907 colorectal cancer cases. A, Example multiplex immunofluorescence image for detection of GZMB, NCAM1, KRT, CD3, FCGR3A, PTPRC, and DAPI (nuclear stain). Scale bar: 100 μm. The image was processed with pathologist-supervised machine learning algorithms to identify tumor intraepithelial and stromal regions (B) and identify and classify each cell (C). In addition to the phenotypic criteria designated in the image, PTPRC expression was required for all immune populations. The distribution of densities of various immune cell populations in tumor intraepithelial (D) and stromal (E) regions. F, Spearman rank correlation coefficients between the densities of intraepithelial (IEL) and stromal (S) immune cells.

Figure 1.

Spatially informed analysis of immune cell populations in 907 colorectal cancer cases. A, Example multiplex immunofluorescence image for detection of GZMB, NCAM1, KRT, CD3, FCGR3A, PTPRC, and DAPI (nuclear stain). Scale bar: 100 μm. The image was processed with pathologist-supervised machine learning algorithms to identify tumor intraepithelial and stromal regions (B) and identify and classify each cell (C). In addition to the phenotypic criteria designated in the image, PTPRC expression was required for all immune populations. The distribution of densities of various immune cell populations in tumor intraepithelial (D) and stromal (E) regions. F, Spearman rank correlation coefficients between the densities of intraepithelial (IEL) and stromal (S) immune cells.

Close modal

We conducted spatial analysis using the spatstat R package (v 1.63–3; ref. 19). We calculated the nearest neighbor distances between immune and tumor cells (NNDImmune cell:Tumor). For visualization, we plotted GZMB and FCGR3A intensities of individual immune cells across all cores (scaled across all immune cells by calculating Z-scores) as a function of NNDImmune cell:Tumor with the ggplot2 package (v. 3.3.0) using generalized additive model smoothing [formula y ∼ s(x)]. We evaluated colocalization of tumor and immune cells using the G-cross function [GTumor:Immune cell (r)] with Kaplan–Meier edge correction, evaluating the likelihood of any tumor cell in the sample having at least one immune cell of the specified type within radius r (20, 21). For survival analysis, we utilized function values at 20 μm [GTumor:Immune cell (20 μm)] to identify immune cell populations likely capable of direct cell-to-cell interaction with tumor cells (20, 21). This radius was preselected prior to analysis to maintain consistency with earlier studies (20–22).

Statistical analysis

We performed statistical analyses using SAS software (version 9.4, SAS Institute). As our primary analysis, we evaluated the relationship between density of each lymphocytic cell type (within intraepithelial or stromal regions, categorized into ordinal quartiles C1–C4) and colorectal cancer-specific mortality using multivariable Cox proportional hazards regression models. We used the stringent two-sided α level of 0.005, as recommended by a panel of expert statisticians (23). Other analyses were secondary, and their results were interpreted cautiously. We analyzed overall mortality as a secondary outcome measure.

Using 4,465 incident colorectal cancer cases in the cohorts, we combined the IPW method with Cox proportional hazards regression to reduce selection bias due to the availability of tumor tissue (24). Using the multivariable logistic regression model for tissue availability versus unavailability as an outcome variable in the 4,465 cases, we estimated the probability of the availability of tissue multiplex immunofluorescence data. Each patient with available tissue data was then weighted by the inverse of the availability probability. We set weights greater than the 95th percentile to the value of the 95th percentile to reduce outlier effects.

The covariates we assessed as potential confounders 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), tumor location (proximal colon vs. distal colon vs. rectum), tumor differentiation (well to moderate vs. poor), disease stage (I/II vs. III/IV), MSI status (MSI-high vs. non–MSI-high), CIMP status (high vs. low/negative), LINE-1 methylation level (continuous), KRAS mutation status (mutant vs. wild-type), BRAF mutation status (mutant vs. wild-type), and PIK3CA mutation status (mutant vs. wild-type). We conducted a backward elimination with a threshold P = 0.05 to select variables for the final models. We included cases with missing data in the majority category of a given categorical covariate to limit degrees of freedom: family history of colorectal cancer in a first-degree relative (0.4%), tumor location (0.4%), tumor differentiation (0.1%), disease stage (6.8%), MSI (3.0%), CIMP (7.4%), KRAS (3.1%), BRAF (2.3%), and PIK3CA (8.9%). For cases with missing LINE-1 methylation data (3.0%), we assigned a separate indicator variable. The proportionality of hazards assumption for colorectal cancer–specific survival was assessed by a time-varying covariate, which was an interaction term of survival time and immune cell densities (P > 0.1). Although we observed evidence of violation of this assumption in the overall survival hazard, the Schoenfeld residual plots supported the proportionality of hazards during most of the follow-up period up to 10 years. Thus, we used Cox regression models limiting the follow-up period to 10 years.

In secondary analyses, we assessed the statistical interaction between immune cell densities (low vs. high) and MSI status (high vs. non-high) in relation to cancer-specific survival. We used the Wald test for the cross-product in multivariable Cox regression models. We estimated HR for colorectal cancer mortality comparing high versus low immune cell densities in the two strata of MSI status using reparameterization of the interaction term in a single regression model (25). We also estimated cumulative survival probabilities using the Kaplan–Meier method and compared the differences between categories using the log-rank test. We evaluated relationships between immune cell densities and clinicopathologic features using the χ2 test and Spearman rank correlation test as appropriate.

Spatial organization of NK- and T-cell infiltrates in colorectal cancer

We applied our customized mIF assay to quantify NK cells, NKT-like cells, and T cells in colorectal carcinoma tissue from two prospective cohort studies. Across 3,234 core images (image per case; median 4; mean 3.6; range, 1–8) from 907 colorectal cancer cases, we identified a total of 4,759,441 tumor cells, 6,224 PTPRC+NCAM1+CD3 NK cells, 6,043 PTPRC+NCAM1+CD3+ NKT-like cells, 672,271 PTPRC+NCAM1CD3+ T cells, 700,234 PTPRC+NCAM1CD3FCGR3A+ macrophages, and 414,842 PTPRC+NCAM1CD3FCGR3A “other” immune cells. T cells were approximately as common as FCGR3A+ macrophages, but approximately 100 times more abundant than NK and NKT-like cells (Fig. 1D and E; Supplementary Fig. S6). Whereas only 30% of T cells expressed the cytotoxic marker GZMB, 76% of NK and 54% of NKT-like cells expressed this marker (Supplementary Fig. S6). Expression of FCGR3A, an NK-cell maturity marker, was relatively uncommon among the NK- and NKT-like cell populations (15% and 13%, respectively), and signal intensities were usually lower than those observed in FCGR3A+ macrophages (Fig. 2F; Supplementary Fig. S6). Overall, densities of different immune cell types showed low to moderate correlation (Fig. 1F).

Figure 2.

Immune cell–tumor cell proximity analyses with the NND function. Multiplex immunofluorescence image of an example tissue microarray core (A), corresponding cell phenotype map (B), and immune cell to nearest tumor cell distance (NNDImmune cell:tumor) plot (C). D, Distributions of NNDImmune cell:tumor for different immune cell populations. E, Distributions of NNDImmune cell:tumor for lymphocytic populations according to GZMB expression. F, Scaled intensities of GZMB and FCGR3A according to NNDImmune cell:tumor. Scale bar in A, 100 μm. The plots in DF are based on all immune cells identified in samples from 907 colorectal cancers (6,224 NCAM1+CD3; 6,043 NCAM1+CD3+; 672,271 NCAM1CD3+; 700,234 NCAM1CD3FCGR3A+; 414,842 NCAM1CD3FCGR3A). The box indicates quartile 1 to quartile 3, with the middle line denoting the median and the whiskers indicating minimum (within a distance of 1.5 times the IQR below quartile 1) to maximum (within a distance of 1.5 times the IQR above quartile 3). P values determined using the Wilcoxon rank-sum test. ****, P < 0.0001; ns, P > 0.005.

Figure 2.

Immune cell–tumor cell proximity analyses with the NND function. Multiplex immunofluorescence image of an example tissue microarray core (A), corresponding cell phenotype map (B), and immune cell to nearest tumor cell distance (NNDImmune cell:tumor) plot (C). D, Distributions of NNDImmune cell:tumor for different immune cell populations. E, Distributions of NNDImmune cell:tumor for lymphocytic populations according to GZMB expression. F, Scaled intensities of GZMB and FCGR3A according to NNDImmune cell:tumor. Scale bar in A, 100 μm. The plots in DF are based on all immune cells identified in samples from 907 colorectal cancers (6,224 NCAM1+CD3; 6,043 NCAM1+CD3+; 672,271 NCAM1CD3+; 700,234 NCAM1CD3FCGR3A+; 414,842 NCAM1CD3FCGR3A). The box indicates quartile 1 to quartile 3, with the middle line denoting the median and the whiskers indicating minimum (within a distance of 1.5 times the IQR below quartile 1) to maximum (within a distance of 1.5 times the IQR above quartile 3). P values determined using the Wilcoxon rank-sum test. ****, P < 0.0001; ns, P > 0.005.

Close modal

To study the spatial organization of lymphocytic infiltrates, we calculated distances from each immune cell to the closest tumor cell (Fig. 2AC). These analyses showed that the average distance to the nearest tumor cell from each NK cell was 30% or 15% shorter than that from each T cell or each NKT-like cell, respectively (Fig. 2D). Within T and NKT-like cell populations, the distance from each GZMB+ cell to the closest tumor cell was 36% and 27% shorter than from each GZMB cell (Fig. 2E and F), respectively, whereas FCGR3A expression showed relatively less variation according to the distance to the closest tumor cell (Fig. 2F). Taken together, these results indicate that NK cells, as well as cytotoxic GZMB+ T and NKT-like cell subpopulations, tend to be more closely colocalized with tumor cells than other lymphocytic cells or FCGR3A+ macrophages.

Associations with clinicopathologic features

Immune responses against colorectal cancer are known to associate with tumor molecular features (21, 26, 27). In particular, tumors with an MSI-high phenotype due to a defective mismatch repair (MMR) system typically accumulate larger numbers of immunogenic neoantigens, which may elicit an intense antitumor immune response driven primarily by T cells (10, 26). We therefore assessed associations between molecular features and immune cell densities (Fig. 3; Table 1). We found that an MSI-high phenotype associated with higher intraepithelial and stromal densities of numerous cell types, including NCAM1+CD3 NK cells, NCAM1+CD3+ NKT-like cells, NCAM1CD3+ T cells, as well as NCAM1CD3FCGR3A+ macrophages (P < 0.0007). Consistent with these results, higher neoantigen load associated with higher intraepithelial densities of all four cell types, as well as higher stromal densities of NK and NKT-like cells (P < 0.0002). Analysis of additional key clinicopathologic features showed that high disease stage associated with lower intraepithelial and stromal densities of all four cell types (P < 0.005), whereas high tumor grade associated with high intraepithelial and stromal densities of NK cells and FCGR3A+ macrophages, as well as high intraepithelial T-cell density (P < 0.002).

Figure 3.

Relationships between clinicopathologic features and immune cell densities in tumor intraepithelial and stromal regions. The plots show data for 907 colorectal cancer cases. P values are based on the comparison of categorical data between the ordinal categories of immune cell density by χ2 test. AJCC, American Joint Committee on Cancer; CIMP, CpG island methylator phenotype; MSI, microsatellite instability.

Figure 3.

Relationships between clinicopathologic features and immune cell densities in tumor intraepithelial and stromal regions. The plots show data for 907 colorectal cancer cases. P values are based on the comparison of categorical data between the ordinal categories of immune cell density by χ2 test. AJCC, American Joint Committee on Cancer; CIMP, CpG island methylator phenotype; MSI, microsatellite instability.

Close modal

Survival analyses

As our main analysis, we evaluated the prognostic significance of immune cell densities using 4,465 incident colorectal cancer cases (including 907 cases with available immune cell data) in two prospective cohort studies and the IPW method to adjust for selection bias. During the median follow-up time of 16.3 years [interquartile range (IQR), 12.8–20.3 years] for censored cases, there were 284 colorectal cancer–specific deaths among 640 all-cause deaths.

We first evaluated the prognostic significance of the total (PTPRC+) immune cell population (Supplementary Table S2; Supplementary Fig. S7), and found that higher intraepithelial, stromal, and overall densities of total immune cells associated with lower cancer-specific mortality, independent of other characteristics such as disease stage or MSI status (Ptrend < 0.0001). Compared with the overall tissue area, the associations were more significant for intraepithelial and stromal compartments, highlighting the value of spatially resolved analysis (Supplementary Table S2; Supplementary Fig. S7).

Analysis of PTPRC+ subpopulations (Table 2; Supplementary Table S3; Fig. 4) showed that higher intraepithelial and stromal densities of both NCAM1+CD3+ NKT-like and NCAM1CD3+ T cells, as well as stromal densities of NCAM1CD3FCGR3A+ macrophages, associated with lower cancer-specific mortality (Ptrend < 0.0007), whereas neither intraepithelial nor stromal NCAM1+CD3 NK-cell populations exhibited prognostic significance in multivariable models (Ptrend > 0.06). The most significant survival association was found for intraepithelial T cells [HR for the highest quartile (C4; vs. the lowest C1) 0.25; 95% confidence interval (CI), 0.15–0.40; Ptrend < 0.0001].

Table 2.

Immune cell densities in tumor intraepithelial and stromal regions and patient survival with IPW.

Colorectal cancer–specific survivalOverall survival
No. ofNo. ofUnivariableMultivariableNo. ofUnivariableMultivariable
caseseventsHR (95% CI)aHR (95% CI)a,beventsHR (95% CI)aHR (95% CI)a,b
NCAM1+CD3 NK cells 
Intraepithelial cell density 
 C1 535 194 1 (referent) 1 (referent) 385 1 (referent) 1 (referent) 
 C2 124 32 0.68 (0.45–1.02) 0.59 (0.38–0.91) 88 0.67 (0.48–0.95) 0.60 (0.42–0.87) 
 C3 125 36 0.91 (0.63–1.31) 1.06 (0.72–1.58) 86 0.98 (0.73–1.33) 1.06 (0.76–1.49) 
 C4 123 22 0.48 (0.30–0.76) 0.83 (0.50–1.36) 81 0.60 (0.43–0.84) 0.77 (0.52–1.14) 
Ptrendc   0.0037 0.46  0.0099 0.27 
Stromal cell density 
 C1 261 92 1 (referent) 1 (referent) 187 1 (referent) 1 (referent) 
 C2 216 90 1.26 (0.93–1.71) 1.16 (0.86–1.56) 162 1.08 (0.82–1.41) 1.01 (0.76–1.33) 
 C3 214 63 0.84 (0.60–1.19) 1.01 (0.71–1.46) 153 0.94 (0.71–1.24) 1.10 (0.83–1.47) 
 C4 216 39 0.52 (0.35–0.79) 0.64 (0.41–1.00) 138 0.63 (0.46–0.85) 0.68 (0.48–0.97) 
Ptrendc   0.0003 0.064  0.0025 0.098 
NCAM1+CD3+ NKT-like cells 
Intraepithelial cell density 
 C1 530 195 1 (referent) 1 (referent) 391 1 (referent) 1 (referent) 
 C2 126 40 0.79 (0.54–1.14) 0.81 (0.53–1.25) 88 0.88 (0.64–1.19) 0.93 (0.65–1.32) 
 C3 125 34 0.69 (0.47–1.00) 0.74 (0.51–1.08) 81 0.67 (0.49–0.93) 0.73 (0.53–1.00) 
 C4 126 15 0.30 (0.17–0.54) 0.40 (0.22–0.71) 80 0.55 (0.39–0.78) 0.63 (0.43–0.91) 
Ptrendc   <0.0001 0.0007  <0.0001 0.0041 
Stromal cell density 
 C1 250 102 1 (referent) 1 (referent) 190 1 (referent) 1 (referent) 
 C2 218 90 1.01 (0.75–1.36) 0.87 (0.65–1.16) 165 1.04 (0.80–1.36) 0.94 (0.72–1.23) 
 C3 221 51 0.49 (0.34–0.71) 0.49 (0.34–0.70) 150 0.56 (0.41–0.76) 0.56 (0.41–0.76) 
 C4 218 41 0.39 (0.27–0.58) 0.41 (0.27–0.62) 135 0.60 (0.45–0.80) 0.65 (0.48–0.88) 
Ptrendc   <0.0001 <0.0001  <0.0001 0.0002 
NCAM1CD3+ T cells 
Intraepithelial cell density 
 C1 226 117 1 (referent) 1 (referent) 178 1 (referent) 1 (referent) 
 C2 227 77 0.64 (0.47–0.87) 0.67 (0.50–0.90) 164 0.80 (0.61–1.06) 0.81 (0.61–1.07) 
 C3 227 60 0.44 (0.32–0.61) 0.52 (0.37–0.73) 154 0.55 (0.41–0.74) 0.61 (0.45–0.81) 
 C4 227 30 0.19 (0.13–0.30) 0.25 (0.15–0.40) 144 0.42 (0.31–0.56) 0.43 (0.31–0.59) 
Ptrendc   <0.0001 <0.0001  <0.0001 <0.0001 
Stromal cell density 
 C1 226 103 1 (referent) 1 (referent) 179 1 (referent) 1 (referent) 
 C2 227 94 0.87 (0.65–1.17) 0.88 (0.65–1.18) 161 0.83 (0.63–1.09) 0.85 (0.64–1.13) 
 C3 227 50 0.45 (0.31–0.64) 0.52 (0.37–0.74) 150 0.54 (0.41–0.73) 0.60 (0.45–0.80) 
 C4 227 37 0.29 (0.19–0.44) 0.31 (0.20–0.47) 150 0.50 (0.37–0.66) 0.50 (0.36–0.68) 
Ptrendc   <0.0001 <0.0001  <0.0001 <0.0001 
NCAM1CD3FCGR3A+ macrophages 
Intraepithelial cell density 
 C1 226 79 1 (referent) 1 (referent) 164 1 (referent) 1 (referent) 
 C2 227 80 1.02 (0.74–1.42) 0.98 (0.69–1.37) 161 1.06 (0.80–1.41) 0.99 (0.74–1.33) 
 C3 227 71 0.92 (0.65–1.29) 1.01 (0.72–1.42) 165 1.00 (0.75–1.34) 1.01 (0.75–1.36) 
 C4 227 54 0.69 (0.47–1.00) 0.74 (0.49–1.13) 150 0.80 (0.59–1.09) 0.82 (0.57–1.18) 
Ptrendc   0.044 0.25  0.17 0.37 
Stromal cell density 
 C1 226 95 1 (referent) 1 (referent) 177 1 (referent) 1 (referent) 
 C2 227 84 0.88 (0.65–1.21) 0.88 (0.64–1.22) 156 0.88 (0.67–1.17) 0.84 (0.63–1.13) 
 C3 227 67 0.70 (0.50–0.97) 0.78 (0.55–1.11) 163 0.82 (0.62–1.09) 0.78 (0.58–1.06) 
 C4 227 38 0.37 (0.25–0.56) 0.43 (0.28–0.64) 144 0.56 (0.41–0.76) 0.60 (0.44–0.83) 
Ptrendc   <0.0001 <0.0001  0.0002 0.0023 
Colorectal cancer–specific survivalOverall survival
No. ofNo. ofUnivariableMultivariableNo. ofUnivariableMultivariable
caseseventsHR (95% CI)aHR (95% CI)a,beventsHR (95% CI)aHR (95% CI)a,b
NCAM1+CD3 NK cells 
Intraepithelial cell density 
 C1 535 194 1 (referent) 1 (referent) 385 1 (referent) 1 (referent) 
 C2 124 32 0.68 (0.45–1.02) 0.59 (0.38–0.91) 88 0.67 (0.48–0.95) 0.60 (0.42–0.87) 
 C3 125 36 0.91 (0.63–1.31) 1.06 (0.72–1.58) 86 0.98 (0.73–1.33) 1.06 (0.76–1.49) 
 C4 123 22 0.48 (0.30–0.76) 0.83 (0.50–1.36) 81 0.60 (0.43–0.84) 0.77 (0.52–1.14) 
Ptrendc   0.0037 0.46  0.0099 0.27 
Stromal cell density 
 C1 261 92 1 (referent) 1 (referent) 187 1 (referent) 1 (referent) 
 C2 216 90 1.26 (0.93–1.71) 1.16 (0.86–1.56) 162 1.08 (0.82–1.41) 1.01 (0.76–1.33) 
 C3 214 63 0.84 (0.60–1.19) 1.01 (0.71–1.46) 153 0.94 (0.71–1.24) 1.10 (0.83–1.47) 
 C4 216 39 0.52 (0.35–0.79) 0.64 (0.41–1.00) 138 0.63 (0.46–0.85) 0.68 (0.48–0.97) 
Ptrendc   0.0003 0.064  0.0025 0.098 
NCAM1+CD3+ NKT-like cells 
Intraepithelial cell density 
 C1 530 195 1 (referent) 1 (referent) 391 1 (referent) 1 (referent) 
 C2 126 40 0.79 (0.54–1.14) 0.81 (0.53–1.25) 88 0.88 (0.64–1.19) 0.93 (0.65–1.32) 
 C3 125 34 0.69 (0.47–1.00) 0.74 (0.51–1.08) 81 0.67 (0.49–0.93) 0.73 (0.53–1.00) 
 C4 126 15 0.30 (0.17–0.54) 0.40 (0.22–0.71) 80 0.55 (0.39–0.78) 0.63 (0.43–0.91) 
Ptrendc   <0.0001 0.0007  <0.0001 0.0041 
Stromal cell density 
 C1 250 102 1 (referent) 1 (referent) 190 1 (referent) 1 (referent) 
 C2 218 90 1.01 (0.75–1.36) 0.87 (0.65–1.16) 165 1.04 (0.80–1.36) 0.94 (0.72–1.23) 
 C3 221 51 0.49 (0.34–0.71) 0.49 (0.34–0.70) 150 0.56 (0.41–0.76) 0.56 (0.41–0.76) 
 C4 218 41 0.39 (0.27–0.58) 0.41 (0.27–0.62) 135 0.60 (0.45–0.80) 0.65 (0.48–0.88) 
Ptrendc   <0.0001 <0.0001  <0.0001 0.0002 
NCAM1CD3+ T cells 
Intraepithelial cell density 
 C1 226 117 1 (referent) 1 (referent) 178 1 (referent) 1 (referent) 
 C2 227 77 0.64 (0.47–0.87) 0.67 (0.50–0.90) 164 0.80 (0.61–1.06) 0.81 (0.61–1.07) 
 C3 227 60 0.44 (0.32–0.61) 0.52 (0.37–0.73) 154 0.55 (0.41–0.74) 0.61 (0.45–0.81) 
 C4 227 30 0.19 (0.13–0.30) 0.25 (0.15–0.40) 144 0.42 (0.31–0.56) 0.43 (0.31–0.59) 
Ptrendc   <0.0001 <0.0001  <0.0001 <0.0001 
Stromal cell density 
 C1 226 103 1 (referent) 1 (referent) 179 1 (referent) 1 (referent) 
 C2 227 94 0.87 (0.65–1.17) 0.88 (0.65–1.18) 161 0.83 (0.63–1.09) 0.85 (0.64–1.13) 
 C3 227 50 0.45 (0.31–0.64) 0.52 (0.37–0.74) 150 0.54 (0.41–0.73) 0.60 (0.45–0.80) 
 C4 227 37 0.29 (0.19–0.44) 0.31 (0.20–0.47) 150 0.50 (0.37–0.66) 0.50 (0.36–0.68) 
Ptrendc   <0.0001 <0.0001  <0.0001 <0.0001 
NCAM1CD3FCGR3A+ macrophages 
Intraepithelial cell density 
 C1 226 79 1 (referent) 1 (referent) 164 1 (referent) 1 (referent) 
 C2 227 80 1.02 (0.74–1.42) 0.98 (0.69–1.37) 161 1.06 (0.80–1.41) 0.99 (0.74–1.33) 
 C3 227 71 0.92 (0.65–1.29) 1.01 (0.72–1.42) 165 1.00 (0.75–1.34) 1.01 (0.75–1.36) 
 C4 227 54 0.69 (0.47–1.00) 0.74 (0.49–1.13) 150 0.80 (0.59–1.09) 0.82 (0.57–1.18) 
Ptrendc   0.044 0.25  0.17 0.37 
Stromal cell density 
 C1 226 95 1 (referent) 1 (referent) 177 1 (referent) 1 (referent) 
 C2 227 84 0.88 (0.65–1.21) 0.88 (0.64–1.22) 156 0.88 (0.67–1.17) 0.84 (0.63–1.13) 
 C3 227 67 0.70 (0.50–0.97) 0.78 (0.55–1.11) 163 0.82 (0.62–1.09) 0.78 (0.58–1.06) 
 C4 227 38 0.37 (0.25–0.56) 0.43 (0.28–0.64) 144 0.56 (0.41–0.76) 0.60 (0.44–0.83) 
Ptrendc   <0.0001 <0.0001  0.0002 0.0023 

Abbreviations: CI, confidence interval; HR, hazard ratio; IPW, inverse probability weighting.

aIPW was applied to reduce bias due to the availability of tumor tissue after cancer diagnosis (see “Statistical Analysis” subsection for details).

bThe multivariable Cox regression model initially included sex, age, year of diagnosis, family history of colorectal cancer, tumor location, tumor differentiation, disease stage, microsatellite instability, CpG island methylator phenotype, KRAS, BRAF, and PIK3CA mutations, and long-interspersed nucleotide element-1 methylation level. A backward elimination with a threshold P of 0.05 was used to select variables for the final models.

cPtrend value was calculated across the four ordinal categories of the density of each immune cell within tumor intraepithelial and stromal regions in the IPW-adjusted Cox regression model.

Figure 4.

Kaplan–Meier estimates of colorectal cancer–specific survival. Inverse probability weighting-adjusted Kaplan–Meier survival curves for colorectal cancer–specific survival according to ordinal categories (C1–C4, from low to high) of intraepithelial (AC) and stromal (DF) densities of NCAM1+CD3, NCAM1+CD3+, and NCAM1CD3+ lymphocytic populations.

Figure 4.

Kaplan–Meier estimates of colorectal cancer–specific survival. Inverse probability weighting-adjusted Kaplan–Meier survival curves for colorectal cancer–specific survival according to ordinal categories (C1–C4, from low to high) of intraepithelial (AC) and stromal (DF) densities of NCAM1+CD3, NCAM1+CD3+, and NCAM1CD3+ lymphocytic populations.

Close modal

When lymphocytic populations were further categorized according to their GZMB expression, cytotoxic GZMB+ subpopulations of NK and NKT-like cells were more significantly associated with favorable survival than GZMB subpopulations, whereas both GZMB+ and GZMB T-cell populations showed associations with lower cancer-specific mortality (Supplementary Table S4). For stromal GZMB+ NK cells, the HR for C4 (vs. C1) was 0.56 (95% CI, 0.36–0.87; Ptrend = 0.0022). Higher stromal densities of mature (FCGR3A+) NK cells also associated with lower cancer-specific mortality [HR for C4 (vs. C1) 0.23; 95% CI, 0.09–0.56; Ptrend = 0.0002; Supplementary Table S5].

Given that MSI status is an important determinant of immune cell densities, we also performed survival analyses stratified by MSI status (Supplementary Table S6). These analyses indicated that high NK-cell density was more significantly associated with lower cancer-specific mortality in MSI-high tumors compared with non–MSI-high tumors (Pinteraction = 0.0011), although the result needs to be interpreted cautiously due to low event numbers for MSI-high cases. For other cell types, there were no statistically significant differences at the α level of 0.005.

Finally, as secondary analyses, we assessed the prognostic significance of tumor cell-immune cell colocalization, as evaluated by the G-cross [GTumor:Immune cell (r)] function, where higher function values at a given radius r indicate a greater proportion of tumor cells in the sample being colocated with at least one immune cell within that radius (Fig. 5). We found that greater colocalization between tumor cells and NCAM1+CD3+ NKT-like cells and NCAM1CD3+ T cells was associated with lower cancer-specific mortality (Ptrend < 0.0001; Fig. 5D). Because higher immune density itself can result in greater tumor and immune cell colocalization, we next evaluated whether these survival associations were independent of immune cell density (Supplementary Table S7). These analyses suggested that, for NCAM1CD3+ T cells, G-cross function values measuring colocalization harbored greater prognostic significance than density (Ptrend = 0.0009 and Ptrend of 0.35, respectively, when included in the same multivariable Cox regression model with reciprocal adjustment). Taken together, these findings not only support the prognostic significance of T and NK lineage lymphocyte densities, but also indicate that spatial proximity between tumor cells and certain immune cell types harbors independent prognostic significance.

Figure 5.

Spatial analysis of immune cell infiltrates using the tumor cell–immune cell G-cross function [GTumor:Immune cell (r)]. The function evaluates the likelihood of any tumor cell in the sample having at least one neighboring immune cell of the specified type within an r μm radius. A, Example multiplex immunofluorescence image for detection of GZMB, NCAM1, KRT, CD3, FCGR3A, PTPRC, and DAPI (nuclear stain). Scale bar: 100 μm. B, Corresponding cell map. C, Corresponding GTumor:Immune cell (r) function curves for each immune cell type. D, Univariable (black) and multivariable (red) Cox proportional hazards regression models of cancer-specific survival according to ordinal categories (C1–C4, from low to high) of GTumor:Immune cell (20 μm) in 907 patients. The forest plots show HRs along with their 95% CIs as whiskers. The multivariable Cox regression models initially included sex, age, year of diagnosis, family history of colorectal cancer, tumor location, tumor differentiation, disease stage, MSI, CpG island methylator phenotype, KRAS, BRAF, and PIK3CA mutations, and long-interspersed nucleotide element-1 methylation level. A backward elimination with a threshold P of 0.05 was used to select variables for the final models.

Figure 5.

Spatial analysis of immune cell infiltrates using the tumor cell–immune cell G-cross function [GTumor:Immune cell (r)]. The function evaluates the likelihood of any tumor cell in the sample having at least one neighboring immune cell of the specified type within an r μm radius. A, Example multiplex immunofluorescence image for detection of GZMB, NCAM1, KRT, CD3, FCGR3A, PTPRC, and DAPI (nuclear stain). Scale bar: 100 μm. B, Corresponding cell map. C, Corresponding GTumor:Immune cell (r) function curves for each immune cell type. D, Univariable (black) and multivariable (red) Cox proportional hazards regression models of cancer-specific survival according to ordinal categories (C1–C4, from low to high) of GTumor:Immune cell (20 μm) in 907 patients. The forest plots show HRs along with their 95% CIs as whiskers. The multivariable Cox regression models initially included sex, age, year of diagnosis, family history of colorectal cancer, tumor location, tumor differentiation, disease stage, MSI, CpG island methylator phenotype, KRAS, BRAF, and PIK3CA mutations, and long-interspersed nucleotide element-1 methylation level. A backward elimination with a threshold P of 0.05 was used to select variables for the final models.

Close modal

Using a quantitative and multiplexed assay, we analyzed NK- and T-cell infiltrates in more than 900 colorectal cancer cases from two U.S. nationwide prospective cohort studies. This approach enabled us to identify complex cellular phenotypes defined by combinations of multiple markers, as well as to precisely measure spatial immune cell infiltration patterns, none of which has been possible using traditional single-marker approaches. Our findings suggest that lymphocyte-subset specific spatial relationships between tumor and immune cells are associated with patient outcome independent of potential confounders such as disease stage and MSI status.

Although their antitumor potential has been appreciated for decades (5), the prognostic significance of NK cells in the colorectal cancer microenvironment has been much less studied than T cells, with only a small number of reports to date (3, 28–33). Most of these studies, utilizing conventional single-color IHC, have suggested that higher densities of NCAM1 (CD56)+ cells or B3GAT1+ [CD57+, a marker for NK-cell terminal differentiation (4)] cells are associated with a favorable outcome (29–31), although other studies did not support this association (32, 33). Notably, as single markers, neither NCAM1 nor B3GAT1 is specific for NK cells, because NCAM1 can be expressed by neurons, NKT cells, and tumor cells (34, 35), and B3GAT1 can be expressed by some terminally differentiated T and NKT cells (36). This lack of specificity may explain why prior studies reached contradictory conclusions. Using our multimarker approach, we found that the overall PTPRC+NCAM1+CD3 NK-cell population was not significantly prognostic in multivariable models, whereas both the cytotoxic (GZMB+) and mature (FCGR3A+) NK-cell subpopulations associated with a favorable outcome. Studies have suggested that NK-cell populations in gastrointestinal cancer are heterogeneous and are involved in both cytotoxic and immunomodulatory functions (37). Consistent with this observation, our evaluation of functionally distinct NK cell subsets suggests that the cytotoxic (GZMB+) and mature (FCGR3A+) subpopulations drive the association of NK cells with better survival.

NKT cells are a heterogenous population of lymphocytes with characteristics of both NK and T cells that exist at the boundary of the innate and adaptive immune systems (38). There are currently no specific markers that can identify the entirety of the NKT-cell population using IHC or flow cytometry, although the PTPRC+NCAM1+CD3+ combination has been frequently used in prior studies (38, 39). Consequently, these cells are often referred to as “NKT-like” to acknowledge the fact that not all NKT cells express both NCAM1 and CD3 (39). To our knowledge, our study is the first to examine the significance of PTPRC+NCAM1+CD3+ NKT-like cells in colorectal cancer tissue. We found that these cells are approximately as common as PTPRC+NCAM1+CD3 NK cells but approximately 100 times less common than PTPRC+NCAM1CD3+ T cells. Despite their relative rarity, higher densities of these cells associated with lower cancer-specific mortality, hinting at a potent functional role for this incompletely characterized cell type. Within the NKT-like cell population, cytotoxic GZMB+ cells, potentially capable of direct tumor cell lysis, appeared to drive prognostic associations. However, further studies using different methodologies are warranted given the heterogeneity in NKT-cell populations (38) and resulting challenges for fully capturing this heterogeneity using multiplex immunofluorescence assays.

T cells had the most significant favorable prognostic association among the cell types analyzed in this study. Even though GZMB+ and GZMB T-cell subpopulations had significantly different spatial distributions, with the cytotoxic GZMB+ subset being preferentially located closer to tumor cells, both populations showed very similar prognostic significance, suggesting that both cytotoxic and non-cytotoxic roles for T cells are important for orchestrating an effective antitumor immune response. Numerous prior studies support the prognostic significance of T cells in colorectal cancer (3), and a prognostic parameter called Immunoscore, based on measurement of overall densities of CD3+ and CD8+ T cells in the invasive margin and tumor center has been validated in an international study involving more than 2,600 patients (40). Our exploratory, secondary analyses showed that colocalization between tumor cells and T cells harbors stronger prognostic value than overall T-cell density and is also independent of T-cell density. These findings suggest that incorporation of both immune cell density and spatial configuration into future analysis methods could drive creation of tumor-immune biomarkers that provide improved personalized treatment guidance.

Our study has some important limitations. First, the measurements were based on tissue microarrays. Although we fully recognize that immune cell infiltrates exhibit spatial heterogeneity, we analyzed a median of four core images per tumor and observed good core-to-core correlation, supporting the validity of our tissue microarray approach. The tissue microarrays also enabled us to evaluate more than 900 tumors in a single batch, eliminating batch effects from staining and sectioning, thereby significantly reducing technical variability. Second, although our assay enabled detection of more detailed cell phenotypes than single-color IHC, the number of markers was still limited and did not allow for identification of some potential populations of interest, such as CD3 or NCAM1 NKT cells (38). However, our marker combinations and machine learning–based analysis algorithms did allow us to measure the key NK-cell and NKT-like cell populations with high confidence across more than 900 separate tumors. Third, we tested multiple hypotheses, which might cause false-positive findings. However, we explicitly defined our main hypotheses and interpreted the results of the secondary analyses cautiously. We used the stringent α level of 0.005 to reduce false-positive findings, as recommended by an expert panel (23). Fourth, treatment information was not available in our cohort studies. However, treatment decisions for colorectal cancer are mainly based on disease stage, for which our multivariable models were adjusted. Finally, our subjects were predominantly non-Hispanic Whites, and our findings need to be confirmed in different populations.

Our study has several notable strengths. The multiplex immunofluorescence assay enabled us to examine multiple relevant lymphocyte markers at the single-cell level in situ in a single tissue section, enabling direct comparisons between various subpopulations. Using pathologist-supervised machine learning, we were able to identify individual tumor cells and lymphocytes with high precision, facilitating accurate measurements of cell densities in different tissue compartments, as well as detailed spatial point pattern analyses that were not possible using conventional manual evaluation. Our molecular pathologic epidemiology database based on two prospective U.S.-wide cohort studies included extensive molecular and clinical information that allowed for correlative analysis with immune cells densities and detailed adjustment of multivariable survival models. The database, involving 4,465 incident colorectal cancer cases, also enabled us to use the IPW method to adjust for potential selection bias due to the availability of tumor tissue data. The patients received care at dozens of institutions across the U.S., minimizing potential bias associated with any single hospital or practice setting and increasing the generalizability of the findings.

In conclusion, higher densities of rare lymphocyte populations, NCAM1+CD3+ NKT-like cells and cytotoxic NCAM1+CD3GZMB+ NK cells, in the colorectal cancer microenvironment are associated with lower cancer-specific mortality. The spatial configuration of lymphocytic infiltrates differs according to cell population and also harbors independent prognostic significance. These results highlight the utility of a quantitative multimarker assay for in situ biomarker evaluation in a manner that may guide personalized medicine.

S.A. Väyrynen reports grants from Finnish Cultural Foundation and Orion Research Foundation sr during the conduct of the study. A.T. Chan reports grants and personal fees from Pfizer, and personal fees from Bayer Pharma AG and Boehringer Ingelheim outside the submitted work. C.S. Fuchs reports personal fees from Amylin Pharma, AstraZeneca, Bain Capital, CytomX Therapeutics, Daiichi Sankyo, Eli Lilly, Entrinsic Health, Evolveimmune Therapeutics, Genentech, Merck, Taiho, and Unum Therapeutics outside the submitted work; has served as a Director for CytomX Therapeutics and owns unexercised stock options for CytomX and Entrinsic Health; is a cofounder of Evolveimmune Therapeutics and has equity in this private company; has provided expert testimony for Amylin Pharmaceuticals and Eli Lilly; and is an employee of Genentech and Roche. J.A. Meyerhardt reports personal fees from Merck, COTA Healthcare, and Taiho Pharmaceutical outside the submitted work. M. Giannakis reports grants from Bristol Myers Squibb, Merck, Servier, and Janssen outside the submitted work. S. Ogino reports grants from NIH during the conduct of the study. J.A. Nowak reports other support from NanoString, Akoya Biosciences, and Illumina outside the submitted work. No disclosures were reported by the other authors.

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–review and editing. S.A. Väyrynen: Data curation, investigation, visualization, methodology, writing–review and editing. T. Ugai: Formal analysis, investigation, writing–review and editing. N. Akimoto: Investigation, writing–review and editing. R. Zhong: Investigation, writing–review and editing. M. Zhao: Investigation, writing–review and editing. A. Dias Costa: Investigation, methodology, writing–review and editing. J. Borowsky: Investigation, methodology, writing–review and editing. P. Bell: Investigation, writing–review and editing. Y. Takashima: Investigation, writing–review and editing. K. Fujiyoshi: Investigation, writing–review and editing. K. Arima: Investigation, writing–review and editing. J. Kishikawa: Investigation, writing–review and editing. S.-s. Shi: Investigation, writing–review and editing. T.S. Twombly: Investigation, 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. X. Zhang: Funding acquisition, investigation, writing–review and editing. C.S. Fuchs: Funding acquisition, investigation, writing–review and editing. J.A. Meyerhardt: Funding acquisition, investigation, writing–review and editing. M. Giannakis: Conceptualization, 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.

We would like to 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. This work was supported by U.S. 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; R01 CA248857, to S. Ogino, U. Peters, and A.I. Phipps; R03 CA197879, to K. Wu; R21 CA222940, to K. Wu and M. Giannakis.; R21 CA230873, to K. Wu and S. Ogino; and K07 CA188126 to X. Zhang); by Nodal Award (2016-02) from the Dana-Farber Harvard Cancer Center (to S. Ogino); by the Stand Up To Cancer Colorectal Cancer Dream Team Translational Research Grant (SU2C-AACR-DT22-17, to C.S. Fuchs and M. Giannakis), 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 National Colorectal Cancer Research Alliance. Stand Up To Cancer is a division of the Entertainment Industry Foundation. The indicated SU2C research grant is administered by the American Association for Cancer Research, 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. 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 the 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 is supported by the Douglas Gray Woodruff Chair fund, the Guo Shu Shi Fund, Anonymous Family Fund for Innovations in Colorectal Cancer, Project P fund, and the George Stone Family Foundation. M. Giannakis was supported by a Conquer Cancer Foundation of ASCO Career Development Award. 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 article.

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.

1.
Ogino
S
,
Nowak
JA
,
Hamada
T
,
Milner
DA
,
Nishihara
R
. 
Insights into pathogenic interactions among environment, host, and tumor at the crossroads of molecular pathology and epidemiology
.
Annu Rev Pathol
2019
;
14
:
83
103
.
2.
Angell
HK
,
Bruni
D
,
Barrett
JC
,
Herbst
R
,
Galon
J
. 
The immunoscore: colon cancer and beyond
.
Clin Cancer Res
2020
;
26
:
332
9
.
3.
Alexander
PG
,
McMillan
DC
,
Park
JH
. 
The local inflammatory response in colorectal cancer - type, location or density? A systematic review and meta-analysis
.
Cancer Treat Rev
2020
;
83
:
101949
.
4.
Miller
JS
,
Lanier
LL
. 
Natural killer cells in cancer immunotherapy
.
Annu Rev Cancer Biol
2019
;
3
:
77
103
.
5.
Kiessling
R
,
Klein
E
,
Wigzell
H
. 
Natural' killer cells in the mouse. I. Cytotoxic cells with specificity for mouse Moloney leukemia cells. Specificity and distribution according to genotype
.
Eur J Immunol
1975
;
5
:
112
7
.
6.
Bryceson
YT
,
Fauriat
C
,
Nunes
JM
,
Wood
SM
,
Björkström
NK
,
Long
EO
, et al
Functional analysis of human NK cells by flow cytometry
.
Methods Mol Biol
2010
;
612
:
335
52
.
7.
Francisco-Cruz
A
,
Parra
ER
,
Tetzlaff
MT
,
Wistuba
II
. 
Multiplex immunofluorescence assays
.
Methods Mol Biol
2020
;
2055
:
467
95
.
8.
Yamauchi
M
,
Morikawa
T
,
Kuchiba
A
,
Imamura
Y
,
Qian
ZR
,
Nishihara
R
, et al
Assessment of colorectal cancer molecular features along bowel subsites challenges the conception of distinct dichotomy of proximal versus distal colorectum
.
Gut
2012
;
61
:
847
54
.
9.
Chan
AT
,
Ogino
S
,
Fuchs
CS
. 
Aspirin and the risk of colorectal cancer in relation to the expression of COX-2
.
N Engl J Med
2007
;
356
:
2131
42
.
10.
Giannakis
M
,
Mu
XJ
,
Shukla
SA
,
Qian
ZR
,
Cohen
O
,
Nishihara
R
, et al
Genomic correlates of immune-cell infiltrates in colorectal carcinoma
.
Cell Rep
2016
;
15
:
857
65
.
11.
Liao
X
,
Lochhead
P
,
Nishihara
R
,
Morikawa
T
,
Kuchiba
A
,
Yamauchi
M
, et al
Aspirin use, tumor PIK3CA mutation, and colorectal-cancer survival
.
N Engl J Med
2012
;
367
:
1596
606
.
12.
Tóth
ZE
,
Mezey
E
. 
Simultaneous visualization of multiple antigens with tyramide signal amplification using antibodies from the same species
.
J Histochem Cytochem
2007
;
55
:
545
54
.
13.
Raskov
H
,
Orhan
A
,
Christensen
JP
,
Gögenur
I
. 
Cytotoxic CD8+ T cells in cancer and cancer immunotherapy
.
Br J Cancer
2021
;
124
:
359
67
.
14.
Gunesch
JT
,
Dixon
AL
,
Ebrahim
TAM
,
Berrien-Elliott
MM
,
Tatineni
S
,
Kumar
T
, et al
CD56 regulates human NK cell cytotoxicity through Pyk2
.
Elife
2020
;
9
:
e57346
.
15.
Rheinländer
A
,
Schraven
B
,
Bommhardt
U
. 
CD45 in human physiology and clinical medicine
.
Immunol Lett
2018
;
196
:
22
32
.
16.
Neefjes
J
,
Jongsma
MLM
,
Paul
P
,
Bakke
O
. 
Towards a systems understanding of MHC class I and MHC class II antigen presentation
.
Nat Rev Immunol
2011
;
11
:
823
36
.
17.
Parra
ER
,
Uraoka
N
,
Jiang
M
,
Cook
P
,
Gibbons
D
,
Forget
MA
, et al
Validation of multiplex immunofluorescence panels using multispectral microscopy for immune-profiling of formalin-fixed and paraffin-embedded human tumor tissues
.
Sci Rep
2017
;
7
:
13380
.
18.
Väyrynen
JP
,
Haruki
K
,
Lau
MC
,
Väyrynen
SA
,
Zhong
R
,
Dias Costa
A
, et al
The prognostic role of macrophage polarization in the colorectal cancer microenvironment
.
Cancer Immunol Res
2021
;
9
:
8
19
.
19.
Baddeley
A
,
Turner
R
. 
spatstat: an R package for analyzing spatial point patterns
.
J Stat Softw
2005
;
12
:
282
90
.
20.
Barua
S
,
Fang
P
,
Sharma
A
,
Fujimoto
J
,
Wistuba
I
,
Rao
AUK
, et al
Spatial interaction of tumor cells and regulatory T cells correlates with survival in non-small cell lung cancer
.
Lung Cancer
2018
;
117
:
73
9
.
21.
Väyrynen
JP
,
Lau
MC
,
Haruki
K
,
Väyrynen
SA
,
Dias Costa
A
,
Borowsky
J
, et al
Prognostic significance of immune cell populations identified by machine learning in colorectal cancer using routine hematoxylin and eosin-stained sections
.
Clin Cancer Res
2020
;
26
:
4326
38
.
22.
Väyrynen
JP
,
Haruki
K
,
Väyrynen
SA
,
Lau
MC
,
Dias Costa
A
,
Borowsky
J
, et al
Prognostic significance of myeloid immune cells and their spatial distribution in the colorectal cancer microenvironment
.
J Immunother cancer
2021
;
9
:
e002297
.
23.
Benjamin
DJ
,
Berger
JO
,
Johannesson
M
,
Nosek
BA
,
Wagenmakers
E-J
,
Berk
R
, et al
Redefine statistical significance
.
Nat Hum Behav
2018
;
2
:
6
10
.
24.
Liu
L
,
Nevo
D
,
Nishihara
R
,
Cao
Y
,
Song
M
,
Twombly
TS
, et al
Utility of inverse probability weighting in molecular pathological epidemiology
.
Eur J Epidemiol
2018
;
33
:
381
92
.
25.
Nosho
K
,
Irahara
N
,
Shima
K
,
Kure
S
,
Kirkner
GJ
,
Schernhammer
ES
, et al
Comprehensive biostatistical analysis of CpG island methylator phenotype in colorectal cancer using a large population-based sample
.
PLoS One
2008
;
3
:
e3698
.
26.
Grasso
CS
,
Giannakis
M
,
Wells
DK
,
Hamada
T
,
Mu
XJ
,
Quist
M
, et al
Genetic mechanisms of immune evasion in colorectal cancer
.
Cancer Discov
2018
;
8
:
730
49
.
27.
Haruki
K
,
Kosumi
K
,
Li
P
,
Arima
K
,
Väyrynen
JP
,
Lau
MC
, et al
An integrated analysis of lymphocytic reaction, tumour molecular characteristics and patient survival in colorectal cancer
.
Br J Cancer
2020
;
122
:
1367
77
.
28.
Zhang
S
,
Liu
W
,
Hu
B
,
Wang
P
,
Lv
X
,
Chen
S
, et al
Prognostic significance of tumor-infiltrating natural killer cells in solid tumors: a systematic review and meta-analysis
.
Front Immunol
2020
;
11
:
1242
.
29.
Coca
S
,
Perez-Piqueras
J
,
Martinez
D
,
Colmenarejo
A
,
Saez
MA
,
Vallejo
C
, et al
The prognostic significance of intratumoral natural killer cells in patients with colorectal carcinoma
.
Cancer
1997
;
79
:
2320
8
.
30.
Chaput
N
,
Svrcek
M
,
Aupérin
A
,
Locher
C
,
Drusch
F
,
Malka
D
, et al
Tumour-infiltrating CD68+ and CD57+ cells predict patient outcome in stage II-III colorectal cancer
.
Br J Cancer
2013
;
109
:
1013
22
.
31.
Chen
Y
,
Yuan
R
,
Wu
X
,
He
X
,
Zeng
Y
,
Fan
X
, et al
A novel immune marker model predicts oncological outcomes of patients with colorectal cancer
.
Ann Surg Oncol
2016
;
23
:
826
32
.
32.
Sconocchia
G
,
Zlobec
I
,
Lugli
A
,
Calabrese
D
,
Iezzi
G
,
Karamitopoulou
E
, et al
Tumor infiltration by FcγRIII (CD16)+ myeloid cells is associated with improved survival in patients with colorectal carcinoma
.
Int J Cancer
2011
;
128
:
2663
72
.
33.
Nagtegaal
ID
,
Marijnen
CA
,
Kranenbarg
EK
,
Mulder-Stapel
A
,
Hermans
J
,
van de Velde
CJ
, et al
Local and distant recurrences in rectal cancer patients are predicted by the nonspecific immune response; specific immune response has only a systemic effect–a histopathological and immunohistochemical study
.
BMC Cancer
2001
;
1
:
7
.
34.
Halama
N
,
Braun
M
,
Kahlert
C
,
Spille
A
,
Quack
C
,
Rahbari
N
, et al
Natural killer cells are scarce in colorectal carcinoma tissue despite high levels of chemokines and cytokines
.
Clin Cancer Res
2011
;
17
:
678
89
.
35.
Sconocchia
G
,
Eppenberger
S
,
Spagnoli
GC
,
Tornillo
L
,
Droeser
R
,
Caratelli
S
, et al
NK cells and T cells cooperate during the clinical course of colorectal cancer
.
Oncoimmunology
2014
;
3
:
e952197
.
36.
Kared
H
,
Martelli
S
,
Ng
TP
,
Pender
SLF
,
Larbi
A
. 
CD57 in human natural killer cells and T-lymphocytes
.
Cancer Immunol Immunother
2016
;
65
:
441
52
.
37.
Wang
F
,
Lau
JKC
,
Yu
J
. 
The role of natural killer cell in gastrointestinal cancer: killer or helper
.
Oncogene
2021
;
40
:
717
30
.
38.
Krijgsman
D
,
Hokland
M
,
Kuppen
PJK
. 
The role of natural killer T cells in cancer-a phenotypical and functional approach
.
Front Immunol
2018
;
9
:
367
.
39.
Krijgsman
D
,
de Vries
NL
,
Skovbo
A
,
Andersen
MN
,
Swets
M
,
Bastiaannet
E
, et al
Characterization of circulating T-, NK-, and NKT cell subsets in patients with colorectal cancer: the peripheral blood immune cell profile
.
Cancer Immunol Immunother
2019
;
68
:
1011
24
.
40.
Pagès
F
,
Mlecnik
B
,
Marliot
F
,
Bindea
G
,
Ou
F
,
Bifulco
C
, et al
International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study
.
Lancet
2018
;
391
:
2128
39
.