Prognostic Impact of FoxP3+ Regulatory T Cells in Relation to CD8+ T Lymphocyte Density in Human Colon Carcinomas

Background T-lymphocyte infiltration into colon carcinomas can influence clinical outcome, and interactions among T cell subsets may be more informative than either subset alone. Our objective was to examine the prognostic impact of tumor-infiltrating FoxP3+ regulatory T cells (Tregs) in relation to cytotoxic CD8+ T lymphocytes in patients with colon carcinomas characterized by DNA mismatch repair (MMR) status who participated in adjuvant chemotherapy trials. Methods FoxP3+ and CD8+ densities in tumor epithelial and stromal compartments were analyzed by immunohistochemistry and quantified in resected, stage II and III colonic carcinomas (N = 216). Immune marker density was dichotomized at the median and categorized as high vs low. MMR status was classified as MMR deficient (dMMR) or proficient (pMMR). Cox models were adjusted for age, stage, and tumor grade. Results The density of FoxP3+ infiltration was similar in tumor stroma and epithelia, whereas CD8+ was higher in stroma. The prognostic impact of FoxP3+ and CD8+ T cell infiltration was stronger in stroma vs epithelia, and the density of each marker in stroma was independently associated with improved overall survival (OS). However, the impact of FoxP3+ on survival was dependent upon CD8+ density (P interaction  = .040). Among CD8+low tumors, FoxP3+high cases had significantly improved OS compared to FoxP3+low cases after adjustment for covariates (hazard ratio 0.43; 95% confidence interval 0.19 to 0.95; P = .030). In contrast, FoxP3+ was not prognostic among CD8+high tumors. FoxP3+ remained prognostic in CD8+low tumors after further adjustment for MMR or BRAF V600E mutation status. Additionally, these immune markers identified a pMMR subgroup with a similarly favorable OS as for dMMR tumors. Conclusions The prognostic impact of FoxP3+ and CD8+ T cell density are inter-dependent, whereby FoxP3+ exerts a favorable influence on survival only in colon cancers with low CD8+ infiltration.


Introduction
Studies in patients with colorectal cancer (CRC) suggest that the density of tumor infiltrating lymphocytes (TILs) strongly influences clinical outcome [1,2,3]. High intratumoral density of CD3+ and cytotoxic CD8+ lymphocytes, as well as CD45RO+ memory T cells, has been associated with reduced incidence of tumor metastasis and favorable prognosis [1,2]. This finding suggests that clinical outcome could be governed, in large part, by the status of the local adaptive immune response. Upon antigenic stimulation, CD8+ T lymphocytes differentiate into effector cells that kill tumor cells by releasing toxic granules such as granzyme B and perforins [4,5]. We previously found that a high density of TILs in colon cancers is associated with increased tumor cell apoptosis [6]. However, not all studies have found CD8+ infiltration to be prognostic in CRC [7,8], and recent data suggests that interactions among T cell subsets is a critical factor that controls the host-tumor reaction and predicts disease outcome [9,10]. An important factor that may block the adaptive immune response are regulatory T cells (CD4+CD25+ Tregs) that express the forkhead box P3 (FoxP3) transcription factor [11]. FoxP3+ Tregs can suppress host-mediated anti-tumor immunity and tumor-specific cytotoxicity, suggesting that Treg depletion is a potential therapeutic strategy [12]. Recent evidence indicates that signaling of the T cell chemoattractant CCL5 can recruit Tregs to tumors and enhance their ability to kill CD8+ T cells, thereby providing a mechanism of immune escape [13]. A high density of infiltrating FoxP3+ Tregs has been shown to be associated with an adverse prognostic impact in some tumor types [10,11], yet a favorable impact in others, including in colon cancers [8,14,15,16,17]. Consistent with their role in immune suppression, a high density of intratumoral FoxP3+ Tregs has been associated with poor outcomes in ovarian, pancreatic, and hepatocellular carcinomas [10,11]. Increased numbers of FoxP3+ Tregs have been detected in colon cancers compared to surrounding unaffected mucosa [18,19]. However, most, but not all [7,20], studies in colon cancers have shown a paradoxical and statistically significant association with favorable prognosis [8,14,15]. This finding is consistent with the improved prognosis for FoxP3+ T cell infiltration in Hodgkin's and follicular lymphoma and in headand-neck carcinomas [16,17]. Tregs are increased in tissues with ongoing inflammation, such as in inflammatory bowel disease [21], where control of inflammation may inhibit tumor development or progression [22].
The extent of T lymphocyte infiltration in human CRC has been shown to differ based upon the status of the DNA mismatch repair (MMR) system [23,24]. CRCs with deficient MMR (dMMR) and microsatellite instability (MSI) are highly immunogenic compared to proficient MMR (pMMR) tumors, the latter of which show chromosomal instability and account for a majority of CRCs. Most dMMR tumors show hypermethylation of the MLH1 MMR gene promoter and frequent activating mutations in a mutational hotspot within exon 15 of the BRAF oncogene (V600E) [23]. An increased density of TILs, including CD3+ and CD8+ lymphocytes, are characteristic of dMMR tumors although studies analyzing the prognostic impact of immune markers have generally not accounted for MMR status [23,24]. BRAF mutation has been reported to be more frequent in stage I to IV human colorectal carcinomas with a high lymphocytic infiltration [25]. In contrast to pMMR cancers, most TILs within dMMR tumors are located within tumor epithelia or in direct contact with tumor cells [26]. Importantly, differences in the tissue location of effector T cell and Treg infiltration within tumor epithelial vs stromal compartments may influence their function and prognostic impact [27,28], yet studies have not adequately addressed this issue, especially for FoxP3+ Tregs [1,2,8].
In this study, we determined the density of FoxP3+ and CD8+ T lymphocytes in epithelial and stromal compartments of human colon cancers, and their relationship to MMR status and patient survival rates. We examined resected stage II and III colon cancers from participants in adjuvant chemotherapy trials where treatment was standardized and meticulous long-term clinical follow-up data were collected. We focused on this population, since immune markers may provide prognostic information that could enable risk stratification for adjuvant therapy. Given data indicating that interactions among T cell subsets may be more informative than either marker alone [9,10], we analyzed the intratumoral densities of FoxP3+ Tregs in relationship to effector CD8+ T cells.

Ethics Statement
The study was approved by the Mayo Clinic institutional review board.

Patient Specimens
Surgically resected TNM stage II and III primary colon adenocarcinomas (N = 216) were analyzed from three 5-fluorouracil (5-FU)-based adjuvant chemotherapy trials conducted by the Mayo Clinic and the North Central Cancer Treatment Group. Formalin-fixed, paraffin-embedded tumor blocks from a nonrandom subset of study participants with MMR data were used. Tumors at or below the splenic flexure were categorized as distal; the remainder as proximal. Of the 216 patients, 191 were randomized to systemic 5-FU, and the others received surgery alone or 5-FU by portal vein infusion. The number of patients who received each study treatment is as follows: 5-FU and leucovorin (LV) plus levamisole (N = 134; study 91-46-53), 5-FU plus levamisole vs 5-FU plus levamisole plus LV (N = 57; study 89-46-51), or portal venous 5-FU vs observation (N = 10 vs N = 15; study 79-46-04) [29,30].

Analysis and Quantification of Immune Markers
TMA slides were imaged using the Slide Scanner system (Bacus Laboratories Inc., Lombard, IL), as described [32]. For CD8+ (N = 216) and FoxP3+ (N = 156), positively-stained cells were counted within each core (206 magnification) in epithelial (within the tumor cell nests or directly contacting a tumor cell) and stromal compartments [20,33]. Density measurements were recorded as the number of positive cells per tumor core (cells per 0.28 mm 2 surface area) [20,34]. Scoring criteria were established by two pathologists (J.M.O; T.C.S.) using randomly selected cases, then all specimens were scored without knowledge of clinical information.

Statistical Analysis
Chi-square tests were used for categorical variables, and the Wilcoxon rank-sum test was used to associate continuous marker densities with 2-level categorical data. The Wilcoxon signed-rank test was used to compare stroma and epithelia densities for each immune marker. T-cell densities were divided at the median for comparisons with clinical variables, including overall survival (OS). OS, censored at 8 years, was calculated from randomization to the date of death or last contact. Within the study population, 81 patients died, and 135 are still alive (global survival rate = 62.5%). Median follow-up in the 135 alive patients was 8 years (range: 4.7-8.0), since the data was censored at 8 years.
In the overall study cohort, we found that 19 cases carried the BRAF V600E mutation (15.2%). Study and parent cohorts had similar dMMR rates.
Age, stage, and grade were significantly associated with OS ( Table 2). Five-year OS rates were similar among study and parent cohorts (70% vs 68%; P = .467). Among dMMR tumors, 5-year OS rates were also similar in the study and parent cohorts. Within the study cohort, the favorable impact of dMMR (vs pMMR) on OS was of borderline statistical significance after adjustment for covariates (HR 0.49; 95% CI 0.22 to 1.11; P = .064), but was highly significant in the parent cohort (HR 0.61; 95% CI 0.42 to 0.89; P = .0066).

Immune Cell Density
We analyzed the expression and location of infiltrating cytotoxic CD8+ T cells and FoxP3+ regulatory T cells within colon carcinomas (Figure 1). Staining was analyzed and compared within epithelial and stromal compartments. CD8+ density was significantly higher in stroma vs epithelia, whereas FoxP3+ was similar (Table S2).

Immune Marker Density in Relation to Clinicopathological Variables
For analysis of immune markers in relation to clinical variables, we dichotomized immune marker density at the median. We found that high vs low CD8+ densities were associated with highgrade histology (Table 1). Similar results were found for FoxP3+. Within tumor epithelia, high CD8+ density was significantly associated with proximal tumor site. T cell densities were not significantly associated with stage, number of malignant nodes, or T stage. CD8+ and FoxP3+ were positively associated with one another.
Analysis of immune markers in relation to MMR status revealed that CD8+ density was significantly higher in epithelia of dMMR vs pMMR cases (median 41 vs 10 cells; P,.0001), yet did not differ in tumor stroma. For FoxP3+, location did not differ by MMR status. No statistically significant associations were found between BRAF status and CD8+ density (P = .26), or between BRAF and FoxP3+ density (P = .83).

T-cell Infiltration and Prognosis
Univariately, high CD8+ T cell density was significantly associated with improved OS in tumor stroma (P = .0091), but not in epithelia (P = .0978; Table 2). Neither FoxP3+ in stroma nor epithelia was prognostic. Analysis of CD8+ density in stroma plus epithelia suggested a stronger prognostic impact compared to analyzing either tumor compartment alone (P = .0075; Table 2).
In multivariable models that included age, stage, and grade, high CD8+ density in stroma (P = .0049) and in epithelia (P = .0323) were significantly associated with OS (Table 3). For FoxP3+, high epithelial or stromal density had borderline associations with favorable OS. After adjustment for MMR status, the favorable prognostic impact of CD8+ density was weakened, whereas the impact of FoxP3+ density was strengthened. For both FoxP3+ and CD8+, MMR-adjusted models showed greater prognostic impact for stromal (vs epithelial) infiltration (Table 3). Adjustment for BRAF status did not significantly alter the findings shown in the multivariable models.
In exploratory analysis, evaluation of the immune markers in stroma plus epithelia showed greater impact on OS compared to either compartment alone. In this regard, high FoxP3+ density in both compartments was associated with substantially improved OS compared to low density in both (HR 0.29 adjusted for covariates and MMR; 95% CI 0.12 to 0.71; P = .0064), and results were similar for CD8+ (HR 0.38; 95% CI 0.19 to 0.74; P = .0048).
We also examined the converse situation where CD8+ T cell density was stratified by FoxP3+ density. We found that high CD8+ density was not prognostic among FoxP3+ high tumors (adjusted P = .530), but was associated with improved OS among FoxP3+ low tumors (HR 0.34; 95% CI 0.15 to 0.77; P = .0060). The association of CD8+ with OS became non-significant after further adjustment for MMR or BRAF.

Immune Markers in Proficient vs Deficient MMR Tumors
To remove potential confounding by dMMR, we determined the prognostic impact of the FoxP3+ and CD8+ in pMMR cases and found consistent results (Table 4). We sought to identify a favorable prognostic group using these immune markers within pMMR tumors. FoxP3+ high or CD8+ high cases that were pMMR showed similarly favorable OS as did 107 dMMR cases from the parent cohorts (Figure 3a-3b). Interestingly, pMMR cases with FoxP3+ high or CD8+ high showed differences in tumor site and histologic grade compared to phenotypic characteristics of dMMR cases [38] (Figure 3c).

Discussion
Given data suggesting that interactions among T lymphocyte subsets may regulate the host-mediated anti-tumor reaction [9,10], we examined the association of FoxP3+ and CD8+ T cells with patient prognosis in stage II and III colon cancers from participants in adjuvant chemotherapy studies. We found that FoxP3+ and CD8+ densities in tumor stroma were independently associated with patient survival in multivariable models. However, their prognostic impact differed significantly based on the relative density of the other, suggesting that these markers are interdependent and should not be directly compared within the same model. We made the novel observation that the FoxP3+ T-cell density independently predicts a favorable OS only when the density of CD8+ infiltration is low. When CD8+ density was high, patient survival was favorable regardless of the level of FoxP3+ infiltration. The converse was also true in that the favorable prognostic impact of CD8+ infiltration was most evident when FoxP3+ density was low. Of note, the best survival rates were observed in cases with high density of either marker, and lowest in cases with low density of both markers. These data suggest that CD8+ and FoxP3+ T cells can interact to regulate the anti-tumor immune response, whereby redundancy in their prognostic impact may exist when there is high-level infiltration of both. However, a microenvironment with low cytotoxic T cells enables the ability of high FoxP3+ density to exert a favorable prognostic influence. We demonstrate a paradoxically favorable effect of FoxP3+ T cells on patient survival that is supported by a study in the Apc MIN/+ mouse model of intestinal polyposis where adoptive transfer of Tregs was able to regress established tumors [39]. CRCs are heavily infiltrated by innate inflammatory immune cells (e.g., tumorassociated macrophages, neutrophils) [22], and studies in mice have shown that adoptive transfer of Tregs can inhibit bacteria-driven inflammation and subsequent colon cancer [40,41]. These data raise the possibility that FoxP3+ T lymphocytes may preferentially suppress T-cell immune responses driven by microbial rather than tumor-associated antigens [11], and could account for their observed favorable prognostic impact in tumors rich in microbial pathogens, such as cancers of the colon and oral cavity [17,42,43]. Evidence also suggests that Tregs may have differential effects depending upon the extent of tumor development/progression. In preclinical studies, depletion of FoxP3+Tregs in mice early in tumor progression enabled immune rejection whereas no therapeutic effect was seen with established tumors [44]. A recent study in human CRC and in a murine polyposis model found that the interaction of Tregs with mast cells can reverse the anti-inflammatory function of Tregs, achieving a pathologic shift in the natural role of Tregs from controlling acute inflammation to promoting inflammation [45]. This Treg skewing is consistent with the known plasticity of Tregs and was associated with their reduced expression of IL10, a cytokine that is critical for maintaining Treg suppressor function [45]. In addition, tumor-infiltrating Tregs were found to share characteristics of T H 17 proinflammatory T cells [45], and a T H 17-biased microenvironment has been commonly linked with protective effector T-cell responses in cancer [46,47]. Potential mechanisms by which Tregs can eradicate tumor cells are suggested by the observation that Tregs can express cytotoxic molecules such as granzyme and perforin, and have been shown to induce monocyte death [48]. Furthermore, adoptive immunotherapy with CD4 + CD25 + Tregs has been shown to decrease tumor multiplicity through induction of apoptosis in intestinal tumors, which supports the possibility that in certain contexts, Tregs can directly or indirectly eradicate tumor cells [39]. Human FoxP3+ Tregs have also been shown to possess an effector differentiation program resulting in the production of IL-17 [49] that was shown to inhibit tumor growth by a T cell-dependent mechanism [50].
Immune marker density in tumor stroma yielded stronger prognostic information than in epithelia. Furthermore, analyzing both compartments together was found to strengthen the association of each immune marker with prognosis compared to analysis of either alone (data not shown). We also determined whether the prognostic impact of CD8+ or FoxP3+ in epithelia and stroma were impacted by inclusion of dMMR tumors. Of note, CD8+ density in tumor epithelium, but not stroma, was higher in dMMR compared to pMMR tumors, whereas no difference was seen for FoxP3+. In survival analyses, adjustment for MMR status was shown to weaken the favorable influence of CD8+ high in both epithelial and stromal compartments, but was found to strengthen the prognostic impact of FoxP3+. While BRAF mutation has been associated with an increased immune cell infiltration in colorectal carcinomas [25], adjustment for BRAF mutational status in our study did not significantly impact the findings shown in multivariable models. Given the potential confounding influence of dMMR tumors with their pronounced T cell infiltration and favorable prognosis [23,24], we excluded them to examine the immune markers in a pMMR subset. Within pMMR cases, FoxP3+ remained favorably prognostic in CD8+ low , but not CD8+ high , tumors. However, CD8+ was not prognostic when stratified by FoxP3+ (data not shown). These findings suggest that the prognostic impact of FoxP3+ is independent of dMMR, in contrast to CD8+ T cells. Since the larger population of pMMR cases is considered poorly immunogenic which may contribute to worse clinical outcome vs dMMR [18,51], we sought to determine if high densities of our immune markers could identify a favorable prognostic group within pMMR tumors. Importantly, we found that pMMR tumors with FoxP3+ high, and to a lesser extent CD8+ high, had similarly favorable survival as did dMMR cases despite dissimilar clinicopathologic features. If confirmed in a larger cohort, these data suggest that immune markers could identify a pMMR subgroup with a favorable prognosis in whom adjuvant chemotherapy may be unnecessary, similar to the situation for stage II dMMR tumors [52].
While FoxP3 is the most specific marker for identifying Tregs [53], important caveats exist. Activated CD4 + CD25 2 effector T cells can transiently express FoxP3+ that may or may not be associated with the acquisition of suppressor functions [54]. Furthermore, there are CD8+CD25+FoxP3+ T cells that appear uncommon in CRCs, and can suppress the anti-tumor immune response [11]. Data indicate that FoxP3 demethylation identifies Tregs with stable FoxP3 expression, and is more specific than mRNA or protein expression in discriminating Tregs from activated FoxP3+ conventional T cells [55]. Further study of FOXP3 demethylation and functional studies of FoxP3+ cells in CRCs are eagerly awaited. Our finding that immune markers were independent prognostic variables in a modest-sized patient cohort underscores their robust prognostic impact and potential clinical utility. While our study population represents a subset of patients from the parent studies, our study cohort had similar OS rates as did the parent cohorts where dMMR was significantly associated with favorable outcome. A strength of our study population is the meticulous collection of long-term survival data from adjuvant trials. Since most of the patients in our study received adjuvant 5-FU-based chemotherapy, we cannot exclude the possibility that the prognostic impact of FoxP3+ and CD8+ T cell density may differ among patients who did not receive chemotherapy. In this regard, the predictive impact of our T-cell markers is unknown and awaits further study. While we examined three tumor cores per patient from tissue regions that were considered representative, we acknowledge that TMAs have limitations for analysis of tumor heterogeneity. Because our stratified analysis was not planned a priori, it will be important to confirm our findings in an independent cohort.
In conclusion, we found that the prognostic impact of FoxP3+ and CD8+ T cell density are inter-dependent, whereby FoxP3+ exerted a favorable influence on survival only in colon cancers with low CD8+ infiltration. Therefore, these data suggest a novel inter-relationship between these immune markers in colon cancers whereby the prognostic impact of FoxP3+ T cells is enhanced in a background of low cytotoxic CD8+ cells. Figure S1 Control images for CD8+ and FoxP3+ staining. Normal human liver is shown as a negative control for CD8+ (a) and FoxP3+ (b) T cell immunostaining, and tonsil tissue (c) was stained as a positive control for FoxP3+ T cells (206 magnification). (PDF)