An exploration of immunohistochemistry-based prognostic markers in patients undergoing curative resections for colon cancer

Immune evasion was presented as an emerging hallmark of cancer in 2011 [1]. In the tumour microenvironment, immune cells interact continuously with the cancer cells during tumorigenesis, a process that takes several years [2, 3]. Through T-cell activation the adaptive immune system has the capacity to impair tumorigenesis, when tumour-associated antigens are presented [4]. However, the cancer cells often escape immune surveillance by activation of immune checkpoint pathways, thus avoiding anticancer immunity [5]. In recent years, immune checkpoint inhibitors have been introduced [6].

As clinical outcome varies substantially among patients diagnosed within the same tumour stage this emphasizes the need for further refinement of the current classification [7]. The Immunoscore©, which is based on the expression of cluster of differentiation 3 (CD3) and CD8 on tumour-infiltrating lymphocytes (TILs) in the tumour centre and in the invasive margin, has shown superiority as a prognostic marker over Union for International Cancer Control (UICC)-TNM classification and highlighted the importance of TILs and anti-cancer immunity [7, 8].

Several other immunohistochemical (IHC) markers are under investigation as promising prognostic or predictive biomarkers. Human leukocyte antigen G (HLA-G) is a non-classical human leukocyte antigen (HLA) class Ib molecule that has immune modulatory properties [9]. The expression of HLA-G is found in both physiological and pathological conditions [10]. HLA-G can impair the function of T-cells, B-cells, and natural killer (NK) cells through several inhibitory pathways, and is a marker of immune evasion [11‐13]. Recently, HLA-G expression has been associated with a worsened prognosis in patients with colorectal cancer [14‐17].

The programmed death 1 (PD-1) pathway is involved in inhibition of the immune response and the exhaustion of T-cells [18]. Programmed death-ligand 1 (PD-L1) is expressed constitutively on T-cells, B-cells, macrophages and other hematopoietic and non-hematopoietic cells, and is inducible through cytokines and in-trans binding of the immune checkpoint PD-1 [19]. Cancer cells can express PD-L1, and several published studies have investigated the role of PD-L1 both as a prognostic marker and a predictive marker for immune checkpoint blockade [6, 20‐24].

Homeobox protein CDX2 (CDX2) is a marker of differentiation of colon cancer cells and has been proposed as a strong prognostic marker in patients with colon cancer [25].

The aim of this study was to explore the expression patterns of HLA-G, PD-L1, and CDX2 as well as CD3 and CD8 in a cohort of patients diagnosed with pT3 and pT4 colon cancers, and to investigate their value as prognostic markers individually and in a combined model.

In this study, we explored the expression of prognostic markers in patients with pT3 and pT4 colon cancers including HLA-G and PD-L1, two markers of immune evasion, as well as the expression of CDX2, a marker of differentiation, and CD3 and CD8, markers of TILs. In adjusted multivariable Cox regression models, positive HLA-G expression was associated with a shortened time to recurrence while a preserved CDX2 expression was associated with a prolonged time to recurrence. When we combined all IHC markers into a summarized score of an unfavourable expression pattern, we found an intermediate and a high combined marker score to be associated with a shortened time to recurrence.

Our results of HLA-G expression as a prognostic marker are in accordance with previously published studies on patients with colorectal cancer [14‐17]. HLA-G expression has also been shown to be associated with a shortened time to recurrence, DFS and OS in several other malignancies such as gastric cancer, breast cancer, lung cancer and malignant melanoma [30‐33]. During pregnancy, HLA-G modulates the maternal immune response to accept the semi-allogenic foetus [34, 35]. These results are all in accordance with a pathophysiological expression of HLA-G and its modulatory effects on cells of the immune system [10‐13]. We defined HLA-G-positive tumours as 10 or more positive cells in one full slide, which may be a very low cut-off. The literature is sparse and divergent on survival analyses and cut-off values for HLA-G expression. Dichotomising HLA-G expression based on positive expression (> 0% positive cells) or a 5%-cut-off has previously been used in prognostic biomarker studies on patients with colorectal cancer [15‐17, 36, 37]. We had a lower occurrence of HLA-G-positive tumours than the published studies with a > 0% cut-off with 9.0% in our cohort compared with 70.6 and 65% in two Chinese populations and 20.3% in a Dutch population, thereby in more accordance with our study [16, 17, 36]. The Dutch study utilized the same antibody (4H84) as we did while using tissue microarrays (TMAs) instead of evaluating full slides. The 4H84 mAb detects denatured HLA-G molecules. The authors included patients with colon cancer of all T-stages, although their population had primarily T3 tumours, while we in the present study only included patients with T3 and T4 colon cancer tumours [36]. The two Chinese studies both evaluated full slides; however, different anti-HLA-G antibodies were used; the MEM-G/2 mAB, which binds free heavy chain of all HLA-G isoforms, and an anti-HLA-G mAb (HGY) not available commercially that should detect both membrane and soluble HLA-G isoforms. The studies included patients with colon and rectal cancer with all T-stages. The study with the highest proportion of patients with HLA-G positive tumours did not stratify patients in colon and rectal cancer cohorts [16]. However, the other Chinese study found a lower proportion of HLA-positivity in patients with rectal cancer [17]. Direct comparison does not seem possible due to the different methods applied in these studies compared to ours e.g. TMA versus full slides and different antibodies applied. Furthermore, both inter- and intratumour heterogeneity have been reported for HLA-G in colorectal tumours [38, 39]. Thus, the expression of HLA-G varies depending on the location within the tumour. The inconsistent HLA-G findings across studies could be attributed to several factors. A number of different anti-HLA-G mAbs are used in the published studies, one (HGY) is not commercially available and staining specifities seem not to have been widely assessed. The mAbs may bind to different epitopes, which may influence the detection rate of HLA-G isoform expression in different tumours. It can be speculated that there might also be ethnic differences; the percentages of tumours expressing HLA-G are closest within the two studies including Caucasian patient groups and within the two studies including Asian patient groups, respectively. Furthermore, novel alternatively spliced HLA-G isoforms have been characterized in clear cell renal cell carcinoma specimens, which may theoretically also occur in colon cancers and influence the staining patterns [40]. Finally, even with the same population and IHC methods, formalin fixation time has been shown to affect the IHC reactions [41]. Interestingly, HLA-G may be a potential new therapeutic target for cancer immunotherapy [42]. One study utilizing chimeric antigen receptor T-cells (CAR-T cells) directed against HLA-G was recently published, while a number of patents have been filed for experimental antibodies directed against HLA-G and its receptors [43, 44].

A consensus guideline for assessment of PD-L1 has not been established for colon cancer. We used a combined positive score for cancer cells and immune cells expressing PD-L1 as a surrogate marker of immune activation. We found patients with high PD-L1 expression to have a longer DFS in unadjusted non-parametric analyses. Four studies based on TMAs have investigated the combined expression of PD-L1 in tumour and immune cells in patients with colorectal cancer [24, 45‐47]. All four studies did find an association of a high combined PD-L1 expression and longer survival, however, they used different antibodies and performed manual assessment of the PD-L1 stainings. A recent meta-analysis of PD-L1 expression and prognosis in patients with colorectal cancer did not recommend PD-L1 as a prognostic marker even though the conclusion was that immune cell expression of PD-L1 was associated with a better survival [48]. As PD-L1 expression may be a marker of good prognosis when expressed by immune cells, and may be a marker of bad prognosis when expressed by tumour cells, it might be more informative not to use a combined positive score as we did, but differentiate between the cell types [23, 24]. However, our analytic platform did not allow for this distinguishment.

CDX2 is a gastrointestinal-specific transcription factor [49]. We identified only 3.7% of our cohort with a reduced CDX2 expression. Patients with reduced CDX2 expression had significantly shortened time to recurrence. Previously, loss of CDX2 has been described as strongly associated with poor prognosis in patients with colorectal cancer [25, 50, 51]. Our results support that loss of CDX2 is a marker of poorly differentiated tumours. Furthermore, we found a reduced CDX2 expression to be associated with MSI status. Interestingly, a study reported that loss of CDX2 expression could predict survival only in patients with MSS [51]. Loss of CDX2 expression has also been suggested to identify a high-risk subgroup of patients with stage II [25].

In our study, patients with a high CD3-CD8-score had a significantly prolonged DFS as well as a prolonged OS. Thus, our results are in line with those shown for the Immunoscore© in several publications and cohorts of patients with colorectal cancer [7, 8, 52]. We did not follow the Immunoscore© protocol, as we used percentages of positive cells instead of densities, different antibodies, laboratory equipment and software for digital analysis. We did, however, adopt a similar approach when calculating a score for TILs, based on digital counts of CD3- and CD8-positive cells in two tumour compartments (the tumour centre and the invasive margin). Patients with early stage disease, UICC stage I, have been found to have a higher infiltration of TILs than patients with UICC stage II-IV [53]. We did not include patients with UICC stage I, but we did, however, find patients with a high CD3-CD8 score to have a higher occurrence of UICC stage II disease than patients with a low CD3-CD8 score. We also found patients with a high CD3-CD8 score to have a higher occurrence of MSI than patients with a low CD3-CD8 score. Accordingly, tumours with MSI are associated with a high immune system activation due to the high expression of tumour-associated antigens [7].

When we combined all our markers into a combined marker score, we identified the strongest signal in the regression analyses. Both an intermediate and a high combined marker score were significantly associated with an increased risk of recurrence and mortality. Our data confirmed that a combination of prognostic markers could provide a stronger estimate of prognosis. A previous study combining the results of HLA class I- and FoxP3-expression based on a computed immune phenotype, could identify a distinct survival pattern between three different phenotypes [36]. The width of the 95% CI in our study, reveals that our HR should be interpreted with great care. When calculating the score, all markers contributed with the same weight to the total score. However, this may not be the optimal approach as each marker may contribute differently to the risk of recurrence or death.

A strength of this study is inclusion of consecutive patients during a two-year inclusion span. We chose to focus on patients with pT3 and pT4 tumours in the colon based on the higher risk of recurrence [54]. In all tumour samples, the invasive margin was represented and assessment was performed on full slides. We investigated the expression of more than one immune checkpoint in patients with colon cancer, and each patient was analysed for TILs. Apart from the previously mentioned limitations only a low number of patients with reduced CDX2 expression (n = 7) and positive HLA-G expression (n = 17) were identified. This resulted in limited statistical power in our analyses. We saw the strongest signal in the time-to-recurrence analyses. Time-to-recurrence comes with a threat of competing risk bias in case patients die before they can develop recurrence [55]. We attempted to reduce this risk by applying the Fine-Gray method.

In conclusion, we investigated HLA-G, PD-L1, CDX2, and CD3 and CD8 as prognostic markers in patients with pT3 and pT4 colon cancers. We found positive HLA-G expression, and a high combined marker score to be independently associated with a shortened time to recurrence. Preserved expression of CDX2 was independently associated with a longer time to recurrence.