Evaluation of a Tumor Microenvironment–Based Prognostic Score in Primary Operable Colorectal Cancer

Colorectal cancer remains the second most common cause of cancer-related death in North America and Western Europe (1). At present, prognosis and the need for adjuvant therapy is primarily based on pathologic assessment of the depth of primary tumor invasion and the presence of lymph node metastases (2); however, there is clear heterogeneity in the survival of patients with similar disease stage, particularly those with stage II (locally advanced, lymph-node negative) disease (3).

Although several high-risk pathologic characteristics, such as venous invasion or serosal involvement, are now recognized as important determinants of survival, particularly in node negative disease (4, 5), it is now clear that other host and tumor characteristics may similarly determine oncologic outcome. Indeed, alongside the intrinsic properties of tumor cells, components of the tumor microenvironment, such as tumor-infiltrating immune cells (6), and the tumor-associated stroma (7), may also determine disease progression and the outcome of patients with colorectal cancer.

Increased survival in association with peritumoral inflammatory responses was first reported in patients with colorectal cancer almost 50 years ago (8). To date, over 100 individual studies have found a reduction in disease recurrence and an increase in survival in association with a conspicuous local inflammatory cell infiltrate in patients with colorectal cancer (6). Although many studies have examined the impact on outcome of individual inflammatory cell subtypes and their density and location within the tumor microenvironment, it is now clear that semiquantitative, histopathologic assessment of the generalized inflammatory infiltrate using routine pathologic specimens, such as that offered by the Klintrup–Mäkinen (KM) grade, not only correlates with the density of individual inflammatory cell subtypes, but also provides similar prognostic detail (9–12).

Similarly, extensive characterization of the tumor-associated stroma, predominantly comprising of cancer-associated fibroblasts and extracellular matrix, has identified pertinent roles in facilitating tumor growth and invasion (13), angiogenesis (14, 15) and energy homeostasis (16). Of interest, however, it has been reported, again using routine pathologic specimens, that a high stroma to tumor ratio is associated with the presence of adverse pathologic characteristics (17). Furthermore, the presence of a high tumor stroma percentage (TSP) has been validated as a stage-independent marker of reduced survival in patients with primary operable colorectal cancer (17–19) and may also identify patients less likely to benefit from adjuvant 5-fluorouracil–based chemotherapy (17).

Of interest, although both hold independent prognostic value for patients with primary operable colorectal cancer (17), combined assessment of the tumor inflammatory cell infiltrate and tumor stroma, and subsequently the interaction and combined impact on survival, has not previously been examined. This, therefore, presents the opportunity to develop a tumor microenvironment score that may provide prognostic information complimentary to current pathologic assessment of features of the tumor itself. The aim of the present study was to evaluate a novel, tumor microenvironment–based prognostic score in patients with primary operable colorectal cancer.

Since 1997, all elective and emergency colorectal cancer resections taking place at a single surgical unit in Glasgow Royal Infirmary have been entered in to a prospectively maintained database. For the present study, patients who, on the basis of preoperative computed tomography and findings at laparotomy were considered to have undergone potentially curative, elective resection for stage I–III primary colorectal adenocarcinoma between January 1997 and May 2008 were included. Patients who had undergone neoadjuvant therapy, emergency resection or resection with palliative intent, or had died within 30 days of surgery were excluded. Local ethics committee approval was granted.

Patient demographics were collected prospectively and tumors were staged using the fifth edition of the AJCC/UICC-TNM staging system (2). Additional pathologic data were taken from reports issued following resection. Since 2003, elastica staining has been used in our institution for the detection of venous invasion, with tumors before 2003 stained retrospectively (5). Both intramural and extramural invasion were considered as evidence of venous invasion. Margin involvement was considered as involvement of any nonperitonealized surgical margin, including the circumferential resection margin. Following surgery, patients were discussed at an institutional multidisciplinary meeting; patients with stage III or stage II disease with high-risk pathologic features and without comorbid disease precluding adjuvant therapy were considered for primarily 5-fluorouracil–based chemotherapy.

Mismatch repair (MMR) status was assessed for a subset of patients who were included in a previously constructed colorectal cancer tissue microarray (TMA; ref. 20). TMA slides were placed in the ThermoFisher pH 9 pre-treatment (PT) module solution (Thermo Fisher Scientific Inc.) at room temperature. Slides were then heated in the PT module to a temperature of 96°C for 20 minutes and allowed to cool. Using the ThermoFisher autostainer, slides were incubated in peroxidase block for 5 minutes and rinsed with TBS. They were then incubated in UV protein blocker for 5 minutes and rinsed once again with TBS solution. Slides were then incubated in primary antibody for 20 minutes at a concentration of 1:100 for MLH1 and MSH6 and 1:50 for MSH2 and PMS2 (product codes: M3640, M3646, M3639, and M3647, respectively; Dako UK Ltd.). Following this incubation period, slides were rinsed with TBS and Quanto Amplifier was applied to slides for 10 minutes followed by a further wash with TBS. Quanto Polymer was then added for 10 minutes followed by a TBS wash. DAB Quanto substrate was then added for 5 minutes, slides washed in TBS, counterstained in hematoxylin, blued in Scotts' tap water, dehydrated through a series of graded alcohols and cover slips applied with distrene, plasticiser, xylene (DPX).

MMR protein expression was established by a single observer (A.G. Powell) blinded to clinical outcomes using UK NEQAS scoring guidelines. Appendix and normal colon were used as positive controls. Accuracy of expression was determined by (i) strong nuclear expression in immune cells, (ii) strong nuclear expression in the base and lower half of the crypts with fading of intensity near the top of the crypt adjacent to the luminal surface, and (iii) strong staining in lymphoid follicles. An observer blinded to clinical outcome (J.H. Park) scored 10% of cores.

Expression was reported as MMR proficient (strong nuclear staining with positive immune cells), or MMR deficient (staining intensity is either weak or patchy with normal immune cell infiltrate, or negative with complete loss of expression and normal immune cell expression). Peri-nuclear immunopositivity was not considered diagnostic for protein expression. Patchy staining of the cytoplasm with normal immune cell expression was considered to be a result of MMR protein complex destabilization and loss of binding to nuclear DNA.

Routine hematoxylin and eosin (H&E)–stained sections of the deepest point of tumor invasion were retrieved. Two investigators (C.H. Richards and C.S.D. Roxburgh), trained by a consultant colorectal pathologist and who were blinded to clinical outcomes, assessed the generalized inflammatory infiltrate semiquantitatively according to KM grade as previously described (21, 22). Briefly, using H&E-stained sections of the deepest point of tumor invasion, inflammatory cell density at the invasive margin was graded using a four-point scale and subsequently classified as low-grade (no increase or mild/patchy increase in inflammatory cells) or high-grade (prominent inflammatory reaction forming a band at the invasive margin, or florid cup-like infiltrate at the invasive edge with destruction of cancer cell islands; Fig. 1A and B).

As previously described, TSP was assessed semiquantitatively by a single investigator (J.H. Park) blinded to clinical outcomes (17). Briefly, using full sections of the deepest point of tumor invasion, the proportion of stroma was calculated as a percentage of the visible field, excluding areas of mucin deposition or necrosis. Tumors were subsequently graded as low TSP (≤50%) or high TSP (>50%; Fig. 1C and D). Coscoring of 10% of tumors was performed by a second investigator (C.S.D. Roxburgh) to ensure consistency. The interobserver intraclass correlation coefficient was 0.81 for assessment of both KM grade and the TSP group (>0.7 is considered good). Any discrepancies for the assessment of either KM grade or the TSP group were resolved by discussion between the two investigators to reach a consensus opinion.

Patients were routinely followed up for 5 years following surgery. Date and cause of death were crosschecked with the cancer registration system and the Registrar General (Scotland). Death records were complete until March 15, 2013, that acted as the censor date. Cancer-specific survival was measured from date of surgery until date of death from colorectal cancer.

The relationship between clinicopathologic and tumor microenvironment characteristics and survival was examined using univariate Cox proportional hazards regression to calculate HRs and 95% confidence intervals (CI). Variables with a P value of <0.1 on univariate regression analysis were then examined in a multivariable model using a backward conditional method. The relationship between a tumor microenvironment–based score and survival was further examined using Kaplan–Meier log-rank analysis, with 5-year survival presented as a percentage of surviving (standard error). The relationship between the tumor microenvironment score and other clinicopathologic characteristics was examined using χ² analysis for trend. A P value of <0.05 was considered statistically significant. All statistical analyses were performed using SPSS version 21.0 (IBM SPSS).

A total of 307 patients who underwent elective resection for stage I–III colorectal adenocarcinoma were included. Clinicopathologic characteristics are summarized in Table 1. Two thirds of patients were younger than 75 at time of surgery with a similar number of males and females. The majority of patients (71%) underwent colonic resection, with pathologic confirmation of lymph node negative (stage I/II) disease in just under two thirds of patients. Venous invasion was detected in 34%, surgical margin involvement in 6%, serosal involvement in 25%, and tumor perforation in 2% of patients. A low KM grade and high TSP were identified in 66% and 25% of patients, respectively. MMR status was available for 208 patients, with MMR-deficient colorectal cancer identified in 33 patients (16%). Overall, 82 patients (27%) received adjuvant chemotherapy; 59 patients (52%) with lymph node involvement received adjuvant chemotherapy compared with 23 patients (12%) with lymph node negative colorectal cancer.

The median follow-up of survivors was 126 months (range, 59–194 months), with 95 cancer-specific deaths. Five-year cancer-specific survival was 75% overall, 86% in patients with stage I/II disease, and 58% in patients with stage III disease. The relationship between clinicopathologic and tumor microenvironment characteristics and cancer-specific survival are shown in Table 1. On univariate analysis, advancing age (P < 0.05), TNM stage (P < 0.001), T stage (P < 0.01), N stage, venous invasion (both P < 0.001), margin involvement (P < 0.05), serosal involvement (P = 0.001), low KM grade and high TSP (both P = 0.001) were all associated with reduced survival. MMR deficiency showed a trend toward increased survival (P = 0.082). On multivariate analysis, age, TNM stage (P < 0.05), venous invasion (P = 0.001), KM grade (P < 0.05), and TSP (P < 0.01) were independently associated with cancer-specific survival.

The prognostic value of KM grade and TSP was further examined (Table 2). Five-year survival of patients with a low TSP was 80% (3), whereas patients with a strong KM grade had 5-year survival of 90%. The presence of a weak KM grade or high TSP was associated with 5-year survival of 68% and 62%, respectively.

A cumulative prognostic score based on these characteristics of the tumor microenvironment was subsequently derived (Table 2). As the univariate HRs and 95% confidence intervals for weak KM grade and high TSP overlapped, the presence of each characteristic scored one point, thus stratifying patients into four possible groups. Patients with a strong KM grade and low TSP comprised 27% of the study population and had a 5-year survival of 89%; conversely patients with a weak KM grade and high TSP comprised 19% of the group with a 4-fold increased risk of cancer-related death and 5-year survival of 51%. The presence of a weak KM grade and low TSP was identified in almost half of the patients studied and was associated with an intermediate 5-year survival of 75%. Only 6% of patients had a strong KM grade with a high TSP; this group had an identical 5-year survival to patients with a strong KM grade and low TSP.

As TSP was not associated with survival in patients with a strong KM grade, the cumulative prognostic score was modified to include all patients with a strong KM grade in the good prognostic group, irrespective of TSP. This modified prognostic score, termed the Glasgow Microenvironment Score (GMS), stratified patients with primary operable colorectal cancer into three distinct prognostic groups (Table 2, Fig. 2): a good prognostic group (GMS = 0 with a strong KM grade and either high or low TSP) with 5-year survival of 89%, an intermediate prognostic group (GMS = 1 with a weak KM grade and low TSP) with an almost 2-fold increased risk of cancer-related death and 5-year survival of 75%, and a poor prognostic group (GMS = 3 with a weak KM grade and high TSP) with a 4-fold increased risk of death and 5-year survival of 51%. Furthermore, on multivariate analysis, GMS was associated with reduced survival (HR, 1.93; 95% CI, 1.36–2.73; P < 0.001), independent of TNM stage (HR, 1.73; 95% CI, 1.07–2.80; P = 0.025), and venous invasion (HR, 2.37; 95% CI, 1.42–3.94; P = 0.01).

The clinical utility of the GMS was further explored in relation to lymph node involvement, venous invasion, MMR status, and use of adjuvant therapy (Table 3). GMS stratified survival of patients with both lymph node negative (stage I/II) and positive (stage III) disease (P = 0.036 and P = 0.002, respectively), and identified patients with both an excellent prognosis (stage I/II and GMS = 0, 92% 5-year survival) and a poor prognosis (stage III and GMS = 2, 37% 5-year survival). Furthermore, patients with stage III disease and GMS = 0 had 5-year survival superior to that of patients with stage I/II disease and GMS = 2 (81% vs. 69%). Similarly, GMS was able to provide further prognostic information alongside venous invasion and MMR status; 5-year survival of patients without venous invasion ranged from 93% to 70% (P = 0.025) and with venous invasion from 78% to 27% (P < 0.001), whereas the combination of MMR status and GMS stratified 5-year survival from 100% in patients with MMR-deficient colorectal cancer and GMS = 0, to 37% in patients with MMR competent colorectal cancer and GMS = 2. In addition, when patients were stratified by use of adjuvant therapy, GMS was predictive of survival of both patients who received or did not receive adjuvant chemotherapy (both P = 0.002)

The relationship between GMS and clinicopathologic characteristics was subsequently examined (Table 4). Increasing GMS was associated with use of adjuvant chemotherapy (P < 0.05), increasing T stage (P < 0.001), N stage, margin and serosal involvement (P < 0.01), and venous invasion (P < 0.05). GMS was not associated with age, sex, tumor site, differentiation, MMR status, or the presence of tumor perforation.

The present study, for the first time, examines the clinical utility of combined assessment of the tumor inflammatory cell infiltrate and tumor stroma, using the KM grade and TSP, respectively, in patients with primary operable colorectal cancer. Indeed, a simple, cumulative prognostic score based on the assessment and interaction of these characteristics using routine histopathologic specimens and termed the GMS, was able to provide improved risk stratification. Using this score, it was possible to identify a group of patients with lymph node–negative disease with 5-year survival comparable with patients with lymph node involvement. Conversely, it was also possible to identify patients with stage III disease and 5-year survival of more than 80%. Similarly, the GMS was able to stratify patients independent of their venous invasion and MMR status. Such a simple, routinely available score can be readily evaluated and validated. If this proves to be the case, then the GMS may help better inform decisions regarding the need for adjuvant therapy and surveillance for otherwise “low risk” patients, or avoid unnecessary treatment for those previously deemed “high risk.”

The results of the present study also have profound implications regarding our understanding of the nature of the tumor microenvironment. As survival of patients with a strong KM grade did not differ with TSP, it could be inferred that the presence of a strong, conspicuous inflammatory infiltrate represents the host's normal antitumor response (12). Furthermore, few patients had a high TSP in the presence of a strong KM grade. As such, it may be loss of this coordinated immune response that facilitates disease progression, allowing tumor stroma formation that in turn facilitates tumor growth and invasion. Therefore, future work must not only consider the intrinsic properties of the tumor cell itself, but also the components of the tumor microenvironment. Indeed, there is increasing evidence that the tumor microenvironment may play a role in chemoresistance (17, 23). Therefore, this new knowledge and the GMS, in particular, may be incorporated into future clinical trial design.

The results of the present study remain to be validated in an independent cohort. In particular, external validation by another research group is required before the GMS can be incorporated into routine pathologic reporting. However, given that the GMS uses routine pathologic specimens, this will facilitate external validation. Indeed, assessment of the GMS may be readily automated (24, 25), further facilitating validation and the implementation of such measures into routine clinical practice.

The present study is limited by the small number of patients with stage I disease (21 patients), and as such it was not possible to examine the clinical utility of assessment of the tumor microenvironment, or the GMS, for this subgroup of patients separately. Given that earlier, node-negative disease is likely to predominate with the introduction of screening (26), this would be an important area for further research. In addition, although the GMS stratified survival independent of MMR status, no other prognostic molecular markers were examined. To date, however, few of these markers have been recommended for use in routine clinical practice, and as such their clinical utility in the management of patients with primary operable colorectal cancer is yet to be realized (27).

In summary, the present study demonstrates the clinical utility of a novel cumulative prognostic score based on the tumor inflammatory cell infiltrate and tumor stroma in patients with primary operable colorectal cancer. This score, termed the GMS, has much to commend it because it is simple and routinely available.

No potential conflicts of interest were disclosed.

Conception and design: J.H. Park, D.C. McMillan, J. Edwards, C.S.D. Roxburgh

Development of methodology: J.H. Park, D.C. McMillan, A.G. Powell, C.S.D. Roxburgh

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.C. McMillan, A.G. Powell, C.H. Richards, C.S.D. Roxburgh

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.H. Park, D.C. McMillan, A.G. Powell, J. Edwards, C.S.D. Roxburgh

Writing, review, and/or revision of the manuscript: J.H. Park, D.C. McMillan, A.G. Powell, P.G. Horgan, J. Edwards, C.S.D. Roxburgh

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.H. Park, A.G. Powell, C.H. Richards, P.G. Horgan, C.S.D. Roxburgh

Study supervision: D.C. McMillan, A.G. Powell, P.G. Horgan, J. Edwards, C.S.D. Roxburgh

This work was supported by the Richie Trust Fellowship (grant number 90479; to A.G. Powell) and the Cancer Research: UK Centre Development Fund.

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.