Immunosurveillance is active in colorectal cancer as downregulation but not complete loss of MHC class I expression correlates with a poor prognosis

Abstract

Many colorectal tumors lose or downregulate cell surface expression of MHC class I molecules conferring resistance to T-cell-mediated attack. It has been suggested that this phenomenon is due to in vivo immune-tumor interactions. However, evidence of the impact of MHC class I loss on outcomes from colorectal cancer is scarce. In our study of more than 450 colorectal cancers in tissue microarray format, we have shown that both high levels of MHC class I expression and absent MHC class I expression are associated with similar disease-specific survival times, possibly due to natural killer cell-mediated clearance of MHC class I-negative tumor cells. However, tumors with low level expression of MHC class I were found to confer a significantly poorer prognosis, retaining independent significance on multivariate analysis. The existence of these poor prognosis tumors, which may avoid both NK- and T-cell-mediated immune surveillance, has important implications for the design of immunotherapeutic strategies in colorectal cancer. © 2005 Wiley-Liss, Inc.

Results from both animal models and human cancer patients indicate that a functional cancer immunosurvelliance process exists in vivo, acting as an extrinsic tumor suppressor. It has also become clear that the immune system can facilitate tumor progression, at least in part by sculpting the immunogenic phenotype of tumors as they develop, a process termed “cancer immunoediting.”1 Initiation of an antitumor immune response occurs when the immune system becomes alerted to the presence of the growing tumor. This may result in tumor elimination, or alternatively, due to the inherent genetic instability of tumors, it may lead to the selection of tumor variants that are resistant to immune attack. This hypothesis may explain the observation that tumors frequently downregulate or lose cell surface expression of major histocompatibility complex (MHC) class I molecules,2 as the peptide-MHC class I complex is essential for the activation of CD8+ cytotoxic T lymphocytes bearing a cognate T-cell receptor (TCR).3 Paradoxically, tumors with absent MHC class I expression appear particularly susceptible to natural killer (NK) cell-mediated killing, due to the presence of MHC class I allele-specific killer-cell inhibitory receptors (KIRs) on the surface of NK cells. In the absence of relevant MHC class I expression, this KIR-mediated inhibitory signalling is lost, resulting in the activation of NK cytolytic effector functions in a process known as “missing-self recognition.”4 Thus, the MHC complex increasingly appears to play a central role in modulating the in vivo immune response to a developing tumor. Expression of MHC class I alleles is frequently abnormal in colorectal cancer.5 Previous immunohistochemical studies have indicated that loss of MHC class I expression in colorectal cancer is associated with poorly differentiated tumors and with those exhibiting mucinous differentiation.5, 6 No association has been found between overall MHC class I loss and either survival or metastatic spread.7, 8 Using high-throughput tissue microarray (TMA) technology,9 we have analyzed MHC class I expression in a series of more than 450 paraffin-embedded colorectal tumor specimens. Data derived from this analysis was then associated with known patient and tumor variables and with long-term patient outcome data to gain further insight into the mechanisms by which tumor cells may manipulate the immune-tumor interaction.

Material and methods

Patients and specimens

TMA construction included representative samples from 462 consecutive patients undergoing elective surgical resection of a histologically proven primary colorectal carcinoma at University Hospital Nottingham between January 1, 1994 and December 31, 2000. Clinicopathologic data from these patients, including tumor site, Duke's/TNM stage, histologic tumor type and grade and the presence of extramural vascular invasion, have been recorded prospectively. During this study period, patients with lymph node-positive disease routinely received adjuvant chemotherapy with 5-flurouracil and folinic acid. Prospective follow-up data regarding the date and cause of death were available for >99% of the cohort, from the UK Office for National Statistics. Follow-up was calculated from the date of resection of the primary tumor, with surviving cases censored for analysis at December 31, 2003. Disease-specific survival was used as the primary end point. The Local Research Ethics Committee granted approval for our study.

Tissue array preparation

Tumor samples were arrayed as described previously.10 Array blocks were constructed at a density of 80–150 cores per array, with analysis of cores from 3 different tumor regions from each individual case for both heavy- and light-chain expression.

Immunohistochemistry

Immunohistochemistry was performed using a routine streptavidin-biotin peroxidase method on freshly prepared 5 μm TMA sections, using a monoclonal antibody to the polymorphic MHC class I heavy chain (HC10, mouse monoclonal anti-human HLA class I heavy chain, gift of Prof. H. Ploegh, Harvard Medical School, Boston, MA), which preferentially recognizes an epitope in unfolded non-β₂-m-associated HLA-B and HLA-C,11, 12, 13 and a commercially available polyclonal antibody to the invariant light-chain β₂-m molecule (A0072, polyclonal rabbit anti-human β₂-microglobulin, Dako, Ely, UK). Microwave pretreatment in pH 6.0 citrate buffer was employed to retrieve antigenicity for HC10 staining. This step was omitted for β₂-m staining, based on the supplier's recommendations. Primary antibodies were incubated on the slides for 1 hr at room temperature with the optimal dilutions previously determined to be 1:2,000 (HC10) and 1:5,000 (A0072). Negative controls, consisting of NSS instead of primary antibody, confirmed the specificity of the staining. The presence of MHC class I-positive tumor-infiltrating lymphocytes, stromal cells and vascular endothelial cells provided intrinsic internal positive controls.

All sections were subsequently incubated with biotinylated goat anti-mouse/rabbit immunoglobulin (Dako), followed by incubation with preformed streptavidin-biotin/horseradish peroxidase (HRP) complex (Dako) and visualisation of antigen using 3, 3′-diaminobenzidine tetrachloride (DAB; Dako).

Evaluation of TMA immunostaining

Immunostaining was evaluated using a semiquantitative scoring system by one observer (N.W.), performed in a coded manner and without reference to clinicopathologic details. HC10 staining was scored in 3 categories: 0 for a complete absence of tumor cell membrane staining, 1 for cases with 1–50% of tumor cells exhibiting positive membrane staining and 2 for cases with 51–100% tumor cells exhibiting positive membrane staining. No attempt was made to score cytoplasmic staining with HC10, as it has been observed that the antibody defines free HLA class I heavy chain and thus may show a positive reaction in the cytoplasm in tumors with W6/32 negativity and no surface MHC class I expression.14 In the case of β₂-microglobulin staining, tumors were categorised as either β₂-m negative (complete absence of tumor cell staining) or β₂-m positive. Where discrepancies arose between the staining of cores from the same tumor, an average of the evaluable cores was taken, with confirmation by 2 observers (Z.M. and N.W.) using a double-headed microscope, with a consensus decision in all cases.

Statistical analysis

Statistical analysis of the data was performed using the SPSS package (version 11.5 for Windows; SPSS, Chicago, IL). Pearson χ² tests were used to determine the significance of associations between categorical variables. All deaths relating to colorectal cancer, including early deaths from postoperative complications, were considered in the disease-specific survival calculations. Patients who died from noncolorectal cancer-related causes and without evidence of cancer recurrence were censored at the time of death. Survival analysis was performed using Kaplan-Meier curves, and the differences in disease-specific survival between groups estimated using the log-rank test. The Cox proportional hazards model was used in multivariate analysis to determine relative risk and independent significance. p-values < 0.05 were considered as statistically significant.

Results

Patients

Of the total of 462 cases, 7 (1.5%) were not available for analysis due to a lack of cores and/or tumor tissue in all 3 samples examined. In a further 7 cases (1.5%), 1 core of the 3 was suitable for evaluation, in 29 cases (6.3%) 2 cores were suitable, and in the remaining 419 cases (90.7%), all 3 cores were evaluable. Thus, analysis of a total of 455 cases was possible. The median age of the cohort at the time of surgery was 72 years, consistent with a median age at diagnosis of colorectal cancer of 70–74 years in the UK.15 At the time of censoring for data analysis, 49% of patients had died from their disease, 14% were deceased from all other causes and 37% were alive. Median follow-up for all patients was 37 months (range 0–116 months), with a median follow-up of 74.5 months (range 36–116 months) in survivors.

MHC class I expression

Immunohistochemical staining with the anti-heavy chain HC10 antibody demonstrated consistent strong positive staining of stromal tissues and tumor-infiltrating inflammatory cells, indicating the success of the immunohistochemical technique. Within the tumor samples, a wide variation in immunoreactivity to HC10 was observed. A total of 348 tumors (76.5%) exhibited strong membrane immunoreactivity, with >50% tumor cells positive (Fig. 1a). In contrast, 45 tumors (9.9%) exhibited weak staining with 1–50% tumor cells positive (Fig. 1b) and 62 tumors (13.6%) entirely negative (Fig. 1c). With respect to the group classified as showing weak HC-10 immunoreactivity (1–50% of tumor cells positive), 6 cases exhibited complete loss of staining in 1 core, with 1–50% tumor cells positive in the remaining 2 cores evaluated. In all other cases with weak HC-10 immunoreactivity, 1–50% tumor cells were positive in all cores evaluated.

Figure 1. (a–c) HLA class I heavy chain expression (HC10). (a) High expression (51–100% cells positive). (b) Low expression (1–50% cells positive). (c) Absent expression. (d,e) β₂-microglobulin expression (A0072). (d) β₂-m positive tumor. (e) β₂-m negative tumor. (a–e) Original magnification ×20.

Staining with anti-β₂-m antibody was generally more intense than with HC10 and appeared more consistent within each tumor. A total of 379 cases (83.3%) were seen to be positive for β₂-m expression (Fig. 1d) and the remaining 76 (16.7%) entirely negative (Fig. 1e). A highly significant correlation was noted between HC10 and β₂-m staining (χ² = 42.392, p < 0.001).

Only tumours coexpressing both heavy chain and β₂-m can express functional MHC class I molecules at the cell surface. Both heavy and β₂-m can be expressed independently within the cytoplasm, and it may be possible to see β₂-m staining without HC10 staining as this antibody does not recognise all class I heavy chain alleles. However, by combining our results, it was possible to classify tumors into those theoretically unable to express functional MHC class I due to a complete absence of either heavy chain or β₂-m expression, those with reduced levels of MHC class I expression (weak HC10 immunoreactivity, β₂-m-positive tumors) and those with higher levels of class I expression (strongly HC10 immunoreactive, β₂-m positive tumors). These 3 groups comprised 112 (24.6%), 32 (7.0%) and 311 (68.4%) cases, respectively.

Comparison of MHC expression and tumor/patient characteristics

The relationship between MHC class I expression and patient/tumor characteristics as assessed by χ² test is shown in Table I. A significant relationship was noted between class I expression and tumor grade (p = 0.008). Absent or reduced class I expression was also noted to occur relatively infrequently in earlier stage tumors, although this did not reach statistical significance (p = 0.158).

Table I. Relationship Between MHC Class I Expression and Patient/Tumor Characteristics

	No. of cases (%)	Absent MHC class I n (%)	Low MHC class I n (%)	High MHC class I n (%)	p-value
Gender
Male	262 (57.6)	64 (24.4)	18 (6.9)	180 (68.7)
Female	193 (42.4)	48 (24.9)	14 (7.3)	131 (67.9)	0.979
Histologic tumor type
Adenocarcinoma	387 (85.1)	92 (23.8)	27 (7)	268 (69.3)
Mucinous	50 (11)	17 (34)	2 (4)	31 (62)
Columnar	4 (0.9)	1 (25)	0 (0)	3 (75)
Signet ring	6 (1.3)	2 (33.3)	2 (33.3)	2 (33.2)
Unknown	8 (1.8)	0 (0)	1 (12.5)	7 (87.5)	0.108
Tumor grade (differentiation)
Well	29 (6.4)	5 (17.2)	0 (0)	24 (82.8)
Moderate	348 (76.5)	79 (22.7)	28 (8)	241 (69.3)
Poor	69 (15.2)	28 (40.6)	3 (4.3)	38 (55.1)
Unknown	9 (2)	0 (0)	1 (11.1)	8 (88.9)	0.008
Tumor site
Colon	233 (51.2)	63 (27)	19 (8.2)	151 (64.8)
Rectal	179 (39.3)	36 (20.1)	12 (6.7)	131 (73.2)
Unknown	43 (9.5)	13 (30.2)	1 (2.3)	29 (67.4)	0.246
TNM stage
0	3 (0.7)	0 (0)	0 (0)	3 (100)
I	68 (14.9)	13 (19.1)	1 (1.5)	54 (79.4)
II	172 (37.8)	45 (26.2)	13 (7.6)	114 (66.3)
III	154 (33.8)	42 (27.3)	10 (6.5)	102 (66.2)
IV	51 (11.2)	11 (21.6)	8 (15.7)	32 (62.7)
Unknown	7 (1.5)	1 (14.3)	0 (0)	6 (85.7)	0.158
Vascular invasion status
Negative	221 (48.6)	56 (25.3)	15 (6.8)	150 (67.9)
Positive	126 (27.7)	31 (24.6)	12 (9.5)	83 (65.9)
Unknown	108 (23.7)	25 (23.1)	5 (4.6)	78 (72.2)	0.642

Comparison of MHC expression and prognosis

Kaplan-Meier analysis was performed to assess relationships with disease-specific survival (DSS) data. Analysis included all 455 evaluable cases (Fig. 2a). Tumors with high expression of MHC class I were associated with the longest DSS (mean DSS 68 months, 95% CI 63–74 months), followed by tumors with absent MHC class I expression (mean DSS 60 months, 95% CI 50–69 months). Tumors displaying low-level expression of MHC class I were associated with a significantly reduced DSS (mean DSS 41 months, 95% CI 26–56 months). Although this difference in survival between groups was most marked in the earlier stage patients (Fig. 2b), a similar trend was also noted in those with more advanced disease (Fig. 2c).

Figure 2. Kaplan-Meier curves showing disease-specific survival according to MHC class I heavy and light chain coexpression in (a) all patients (n = 455); (b) early stage (TNM 0–II) cancers (n = 243) and (c) late stage cancers (TNM III–IV and unknowns, n = 212).

Multivariate analysis

Multivariate analysis using the Cox proportional hazards model was performed to calculate hazard ratios (HR) and identify factors with independent prognostic significance. Analysis of all 455 cases included tumor grade, TNM stage (TNM stage III/IV vs. stage I/II tumors), vascular invasion status and MHC class I expression (low vs. high/absent). As expected, both advanced TNM stage (HR 2.929, 95% CI 2.156–3.954) and the presence of extramural vascular invasion (HR 1.842, 95% CI 1.344–2.524) displayed a high level of independent prognostic significance (both p < 0.001); however, this was not demonstrated for tumor grade (HR 1.082, 95% CI 0.567–2.063, p = 0.811). The hazard ratio for death in patients with MHC class I low tumors was 1.635 (95% CI 1.049–2.549), demonstrating independent prognostic significance (p = 0.030).

Discussion

Our study shows that colorectal tumors with high expression of MHC class I are associated with the longest DSS, followed by those with absent heavy- or light-chain expression. In contrast, tumors displaying low level coexpression of class I chains are associated with a significantly reduced DSS. Furthermore, this difference in survival between groups is most marked in earlier stage patients. Although there have been many reports of loss of MHC antigens in tumors, due to the large sample number (462 tumours) and strong clinicopathologic database, this is the first study to our knowledge that shows that downregulation of MHC class I is an independent marker of poor prognosis in colorectal cancer.

Cell surface expression of MHC-peptide complexes requires correctly folded MHC class I heavy chain to associate with β₂-m and be stabilised by the binding of antigenic peptide in the heavy-chain peptide-binding groove.16 Additional factors involved in the assembly and intracellular transport of the peptide-MHC complex include the molecular chaperones calnexin, calreticulin and ERp57,17 the transporter molecule TAP (transporter associated with antigen processing) and low-molecular-weight proteins 2 (LMP2) and LMP7.18 Defects in this antigen-processing machinery (APM) are implicated in the total loss of MHC class I surface expression in colorectal cancer.14 In genetically unstable tumors, this phenotype may arise but would have no selective advantage unless the tumor was being attacked by T cells. Under these conditions, tumors lacking MHC class I would not be recognized and would escape T-cell-mediated immune killing. However, in the absence of any surface MHC class I these tumors would be susceptible to NK cell attack, as KIRs would not be engaged. In contrast, tumor cells that downregulate specific MHC class I alleles may avoid T-cell recognition whilst retaining sufficient MHC class I expression to avoid NK cell activation. Although the theory of HLA-deficient tumor escape from T cells and NK cells has been described previously,19 this is the first large-scale analysis to demonstrate clearly a significant survival difference between tumors with low-level MHC class I loss and those with either high levels or total loss of MHC class I. Four different altered HLA class I phenotypes are commonly found in tumor tissues including: (i) Phenotype I, or total HLA loss, which results from interference with β₂-m synthesis, transporter associated with antigen processing (TAP) or structural defects in MHC genes; (ii) Phenotype II, or loss of HLA haplotype, which can result from chromosomal nondisjunction or mitotic recombination; (iii) Phenotype III or loss of HLA locus expression, which probably occurs through a mechanism of transcriptional control; and (iv) Phenotype IV or HLA allelic loss, which might result from point mutations or partial deletions of HLA class I genes.2 Without HLA typing of all of the tumors in our study, it is not possible to discriminate between these genetic events. Investigation of cryopreserved tumor samples would potentially allow a more comprehensive analysis of the molecular mechanisms responsible for alterations of HLA and β₂-m expression in individual tumors. However, few if any investigators have access to a large enough number of frozen tumor samples with sufficient follow-up data to investigate the role of MHC class I expression as a prognostic factor. In our study, rather than investigating the HLA class I phenotypes, we have instead shown that the functional consequence of MHC class I downregulation in colorectal cancer is reduced patient survival.

The observation that the influence of MHC class I expression is greatest in the earlier stages of the disease suggests that in this group of patients an immune response is occurring, which may eliminate circulating tumor cells and prevent the development of metastases. Corroborating evidence for this hypothesis is provided by the finding that primary colorectal tumors display sparse NK-cell infiltration compared to the presence of tumor-infiltrating CD8+ T cells, suggesting that the survival benefit observed in tumors with total loss of MHC class I expression may be due to clearance of metastasizing tumor cells by NK cells in the systemic circulation and the liver rather than the primary tumor itself.20 Despite this, if early stage tumors become relatively resistant to immune attack due to the selective downregulation of MHC class I, then recurrence of the disease after surgical treatment will be more likely. These results are therefore particularly pertinent to the development of immunotherapeutic strategies for colorectal cancer, suggesting that early stage patients who have downregulated MHC class I expression may derive greater benefit from adjuvant chemotherapy, as their immune response is unlikely to remove any circulating tumor cells after surgery. In contrast, patients with normal MHC class I levels would appear to be good candidates for T-cell-based vaccine strategies.

Acknowledgements

We thank Prof. H.L. Ploegh for providing the antibody HC10, and Mr. J. Ronan for technical advice.