HLA-E and HLA-G Expression in Classical HLA Class I-Negative Tumors Is of Prognostic Value for Clinical Outcome of Early Breast Cancer Patients

There has been strong evidence that tumor progression is controlled by the host’s immune system (1). However, because of their intrinsic genetic unstable nature, tumor cells may acquire properties to escape from immune recognition (2). These poorly immunogenic clones frequently have lost expression of classical HLA class I (HLA-A, HLA-B, and HLA-C), which enables them to escape CTL attack. However, in that case, they may be vulnerable to NK cell elimination. Expression of nonclassical HLA class I molecules (HLA-E and HLA-G), which play a pivotal role in immune surveillance by NK cells, may therefore also determine outcome of tumor–immune interaction (3). Under normal circumstances, expression of the HLA-E molecule is found in most tissues that express HLA-A, -B, -C, or -G molecules and is thought to provide an important “self-signal” to the immune system by accommodating and presenting peptide fragments from leader sequences of these molecules (3, 4). HLA-G expression, in contrast, has very restricted tissue expression and has been mostly found in extravillous trophoblastic cells, where it mediates semiallograft immunotolerance during pregnancy (5). Expression of HLA-E and HLA-G on the cell surface can respectively bind with the inhibitory receptors CD94/NKG2A and KIR2DL4/p49 of NK cells and thereby cause inhibition of their proliferation and cytotoxic effector functions (6, 7). HLA-E also binds activating CD94/NKG2C receptors, present on T and NK cells, however, with a 6-fold lower affinity (8).

Tumors may acquire or upregulate expression of HLA-E and HLA-G as protective property against immune recognition and elimination of tumors (3). HLA-E is regularly expressed in various healthy tissues and correlates with expression of classical HLA class I molecules. This physiological correlation with classical HLA class I molecules has been found to be disturbed in tumors, suggesting that malignant cells which escape T cell immune recognition by downregulation of classical HLA class I expression, may further escape immune recognition by upregulation of HLA-E (9). In addition, expression of HLA-G protects against “missing self” recognition of NK. Expression of this molecule, which is rarely found in healthy tissues, is frequently observed in pathological conditions such as in tumors (10, 11). Previous studies showed that both HLA-E and HLA-G had increased expression in different types of tumor (12–15). Studies on the prognostic value of HLA-E expression in colorectal and cervix cancer showed that expression of this molecule correlated to tumor progression and had a trend toward a worse clinical outcome. The prognostic value of HLA-G expression has been investigated in colorectal, gastric, esophageal squamous cell carcinoma, and non-small cell lung cancer and revealed it to be an independent prognostic factor for poor clinical outcome (16–19). In addition, expression of HLA-G has also been found in breast cancer; however, no statistically significant associations were found with outcome of patients (20–22).

The prognostic effect of HLA-E and HLA-G expression in breast cancer is unknown. The purpose of this study was to analyze the prognostic relevance of expression of HLA-E and HLA-G in a large cohort of early breast cancer patients. Previously, we determined classical HLA class I expression in the same patient cohort. Therefore, we were able to stratify patients based on classical HLA class I expression of tumors and to analyze the impact of HLA-E and HLA-G expression on clinical outcome of early breast cancer patients.

The patient population comprised all nonmetastasized breast cancer patients primarily treated with surgery in the Leiden University Medical Center between 1985 and 1994 (n = 677). Patients with bilateral tumors or a prior history of cancer (other than basal cell carcinoma or cervical carcinoma in situ) were excluded. The following data were known: age, tumor grade, histological type, Tumor-Node-Metastasis stage, local and systemic therapy, locoregional/distant tumor recurrence, secondary tumor, survival, and expression of estrogen receptor (ER), progesterone receptor, and human epidermal growth factor receptor 2 (HER2) (23). All tumors were graded according to current pathological standards by a pathologist (V.T.H.B.M.S). In addition, for about half the cohort of patients (n = 266) a tissue microarray (TMA) of paired histologically normal breast tissue was available. Normal breast tissue originated from the cancer-affected breast but localized more distal from the tumor tissue.

MEM-E/02 (Abcam, Cambridge, U.K.) and 4H84 (Nuclilab, Ede, The Netherlands) Abs were used to recognize HLA-E and HLA-G, respectively. MEM-E/02 reacts specifically with the denatured H chain of human HLA-E24. The 4H84 Ab recognizes denatured HLA-G molecules and has been described to react with classical HLA class I molecules (25–27). Tissue sections of 4 μm were cut from a previously constructed TMA of formalin-fixed paraffin-embedded tumors (23). Tissue sections were deparaffinized and rehydrated. For Ag retrieval, 0.01 M Trizma EDTA buffer (pH 6) was used for 10 min at maximum power in a microwave oven. Endogenous peroxidase was blocked for 20 min in 0.3% hydrogen-peroxide methanol. Sections were incubated overnight with primary mAbs using predetermined optimal concentrations. After 30 min of incubation with secondary Ab Envision anti-mouse (K4001; DakoCytomation, Glostrup, Denmark), sections were visualized using diaminobenzidine solution. Tissue section were counterstained with hematoxylin and then dehydrated and finally mounted in malinol. For each primary Ab, all slides were stained simultaneously to avoid interassay variation. For each staining, placenta tissue slides served as positive control. Negative controls were placenta tissue slides that did undergo the whole immunohistochemical staining without primary Abs. Sections of paired normal tissue TMA were stained with MEM-E/02 and 4H84 to assess frequency of staining in normal breast tissue samples.

Tumor staining for classical HLA class I using the mouse mAbs HCA2 and HC10 (anti–HLA-A and anti–HLA-B/C, respectively) was described previously (28).

Microscopic analysis of HLA-E and HLA-G was assessed by two independent observers (A.S., E.M.d.K.) in a blinded manner. Both markers were scored in a binary manner, considering any specific staining of tumor cells as positive expression and no staining as no expression. A combined variable of HLA-E and HLA-G scores was created: HLA-EG. HLA-EG expression was considered positive when both HLA-E and HLA-G were expressed and negative when either HLA-E or HLA-G was not expressed.

Statistical analyses were performed using the statistical package SPSS (version 16.0 for Windows; SPPS, Chicago, IL). Cohen’s κ coefficient was used to assess interobserver agreement in quantification. This revealed a substantial agreement in classification for HLA-E (κ = 0.72) and a very good agreement in classification for HLA-G (κ = 0.90). The χ² test was used to evaluate associations between various clinicopathological parameters and HLA-E and HLA-G expression. Relapse-free period (RFP) was the time from date of surgery until an event (locoregional recurrence and/or a distance recurrence, whichever came first). Overall survival (OS) was defined as date of surgery until death. The Kaplan-Meier method was used for survival plotting and log-rank test for comparison of survival curves. RFP is reported as cumulative incidence function, after accounting for death as competing risk (29). Cox regression was used for univariate and multivariate analysis for RFP and OS. Significant variables (p < 0.1) in univariate analysis were included in multivariate analysis. To analyze the independent prognostic effect of HLA-E and HLA-G on clinical outcome, tumors were stratified based on a previously determined expression characteristics of classical HLA class I molecules.

We finally analyzed whether the specificity of the anti–HLA-G Ab would interfere with the results of our survival analyses by separately analyzing the set of patients in which those who stained positive for this Ab on normal breast tissue were excluded.

Tumor material was available and incorporated in the TMA of 86% (574 of 677) of the patients. Paired normal breast tissue was available on TMA in 46% (266 of 574) of the patients. Median age of patients was 57 y (range, 23–96 y). Median follow-up of patients alive was 19 y (range, 14–23 y). Clinicopathological and treatment characteristics are shown in Table I.

Microscopical quantification was successful in 86% (493 of 574) of tumors for HLA-E and in 87% (501 of 574) for HLA-G. Respectively, 14 and 13% of tumors were damaged or lost on the TMA slides, a problem associated with preparation, staining, and mounting of TMA slides. Two groups, expression versus no expression, were defined for HLA-E and HLA-G (Fig. 1A–D). Expression was found in 50% (247 of 493) and in 60% (299 of 501) of tumors for HLA-E and HLA-G, respectively (Table I). Expression of HLA-EG was found in 23% (100 of 428) of tumors. HLA-G stained positive in 1% (3 of 266) of normal tissue samples (Fig. 1E, 1F), whereas HLA-E showed positive staining in all normal tissue samples (Fig. 1G).

In the whole cohort of patients, HLA-E, HLA-G, and HLA-EG showed no statistically significant difference in outcome between expression versus no expression for RFP (log-rank p values, respectively: 0.52, 0.95, and 0.72) or OS (log-rank p values, respectively: 0.86, 0.74, and 0.27) (Figs. 2A, 2D, 2G, 3A, 3D, 3G).

Next, we stratified patients based on classical HLA class I tumor expression, classified as expression versus loss. Among the subgroup of classical HLA class I expression results were similar as in the whole cohort of patients: for HLA-E, HLA-G, or HLA-EG, a statistically significant difference was not found for different expression levels in outcome for RFP (log-rank p values, respectively: 0.73, 0.69, and 0.51) or OS (log-rank p values, respectively: 0.64, 0.74, and 0.22) (Figs. 2B, 2E, 2H, 3B, 3E, 3H). Interestingly, among the subgroup of patients with loss of tumor expression of classical HLA class I, HLA-E and HLA-G expression showed significant differences for RFP (log-rank p values, respectively: 0.03 and 0.04) and OS (log-rank p values, respectively: 0.03 and 0.12) between both expression groups (Figs. 2C, 2F, 3C, 3F). Of the patients with no tumor expression of HLA-E or HLA-G, respectively, 60 and 56% of patients were relapse free after 10 y, whereas of the patients with tumor expression of HLA-E or HLA-G, respectively, 35 and 39% of patients were relapse free after 10 y. The combination variable HLA-EG showed, similarly to HLA-E and HLA-G separately, differences in outcome between expression and no expression among the subgroup of classical HLA class I loss, but at a much higher level of significance than each separately (log-rank p values: RFP, 0.001; OS, 0.007) (Figs. 2I, 3I). Among the patients with no expression of HLA-EG, 55% were relapse free after 10 y, compared with 17% for expression of HLA-EG. Cox proportional multivariate analysis was performed for relapses over time including the following factors: tumor stage, lymph node status, ER status, HER2 expression, local therapy, endocrine therapy, and HLA-EG. This analysis revealed that lymph node status and HLA-EG (p = 0.011; hazard ratio [HR], 2.87; 95% CI, 1.28–6.43) were independent factors for RFP among the subgroup of classical HLA class I loss patients (Table II). These data showed that HLA-EG possesses a specific prognostic effect but only among classical HLA class I loss patients. To prove that classical HLA class I and HLA-EG were significantly cooperating variables, an interaction term was introduced in Cox regression analysis. This analysis showed a statistically significant interaction (p = 0.002) between the two markers, suggesting that there is a biological connection between classical HLA class I and HLA-EG.

The 4H84 Ab has been described to occasionally cross-react with classical HLA class I molecules (27). Therefore, we performed additional immunohistochemical analyses to examine whether this cross-reaction would interfere with our survival results.

Expression on paired normal breast tissue of half the cohort was found in 1% (3 of 266) for HLA-G. These three patients who showed weakly positive staining for HLA-G on normal breast tissue, also stained positive for classical HLA class I on normal and tumor tissue, indicating that the 4H48 Ab possibly occasionally cross-reacted with these classical HLA class I molecules. It should be noted, however, that the staining on normal tissue was only modest when compared with tumor staining with the 4H84 Ab (compare Fig. 1F with 1D). To examine whether the occasional cross-reaction of the 4H84 Ab would interfere with our results, we performed a subanalysis by selecting only the tumors of the 266 patients of whom paired normal tissue was available. In this analysis, we excluded the three cases that showed positive staining for HLA-G on normal breast tissue (the presumed cases which showed cross-reaction for the 4H84 Ab) and examined whether survival analyses would reveal similar results as to when these cases would not be excluded. When excluding these three cases, no survival analyses reached statistical significance (log-rank p ≥ 0.426) in neither the total population of patients nor the patient population with expression of classical HLA class I. This was concordant with the results found without exclusion of these cases (log-rank p ≥ 0.693). Importantly, no expression was seen of HLA-G in normal breast tissue of patients whose tumor showed no classical HLA class I expression, but resulted positive for HLA-G expression. Taken together, these results suggests that the occasional cross-reaction of 4H84 with classical HLA class I molecules did not interfere with our results.

Tumor–immune interaction may be of great importance for clinical outcome (2). In this study we showed that in tumors devoid of classical HLA class I expression, HLA-E and HLA-G expression were of statistically significant influence on outcome of breast cancer patients independently of known clinicopathological parameters, with an almost three times higher risk of relapse over time for patients with expression of HLA-EG compared with patients with no expression of HLA-EG. This is the first study, to our knowledge, providing evidence for a prognostic value of nonclassical HLA class I molecule expression in a large cohort of breast cancer patients. In addition, to our knowledge, we are the first to report that such an effect on outcome of patients interplays with expression of classical HLA class I molecules. Importantly, these results can be explained by underlying biology and support and add to previous studies on tumor–immune interaction in breast cancer (3, 12–19).

Previous studies have found elevated expression levels of the non-classical HLA class I molecules, HLA-E and HLA-G, in tumor tissues (3, 12–19). Normally, HLA-G is not expressed on nonmalignant cells. Corresponding to this fact, we found in our study that 4H84 HLA-G Ab did stain in a considerable number of tumor tissues but in a negligible number of normal mammary tissues. Under normal circumstances, HLA-E surface expression is dependent on the availability of HLA class I signal sequence-derived peptides. Therefore, HLA-E surface expression is usually found to be coexpressed with classical HLA class I, which comes to expression in almost all healthy tissues (3, 4). Corresponding to this fact, we did not find any normal mammary tissue that did not express HLA-E molecules. In some tumor tissue, however, HLA-E expression seems to be independent of the availability of classical HLA class I sequence-derived peptides and can be expressed in cells that lack classical HLA class I expression (9, 30). Indeed, we found cytoplasmic expression of HLA-E in classical HLA class I-negative tumors in our study. The disturbed balances of expression of classical HLA class I, HLA-E and HLA-G, as found in our study, suggest a cooperation between these molecules in evading immune recognition. According to the immunoediting hypothesis, tumors may become shaped through interaction with the immune system, leading to the selective outgrowth of highly tumorigenic clones that escape from immune recognition and elimination (31). Downregulation of classical HLA class I expression in tumors, with simultaneous loss of cell surface expression of HLA-E because of lack of peptide fragments that it can bind, is believed to reflect CTL immune escape (3). However, these tumor cells become highly vulnerable to NK cells, which recognize these “missing self” cells (14). Through a variety of factors, such as epigenetic control, hypoxia, stress, and cytokines, expression of HLA-G and HLA-E may be upregulated and counteract this susceptibility to NK cells (3, 10, 32, 33). Supportive for a specific NK cell inhibition of the nonclassical HLA class I molecules, for both HLA-E and HLA-G an inverse correlation was found with NK cell infiltrate in a colorectal cancer and gastric cancer study, respectively (34, 35). In addition, in various studies using colon cancer and melanoma cell lines it was demonstrated that overexpression of HLA-E and HLA-G respectively directly inhibited NK-mediated cell lysis (35–39). The statistical interaction between HLA-E and HLA-G with classical HLA class I molecules, as found in our study, adds to this evidence, suggesting that specifically in tumors devoid of classical HLA class I expression, upregulation of HLA-E and HLA-G expression counteracts the resulting NK cell susceptibility, leading to immune escape of tumor cells. Our study supports and adds to previous findings, suggesting that HLA-E and HLA-G contribute to tumor immune escape, specifically NK cells, a phenomenon that is likely to have impact on clinical outcome of patients.

Prognostic associations of HLA-E and HLA-G have been studied in various types of tumors (16–19, 30, 35). In cervical cancer, HLA-E expression increased with the progression of the lesion. One study analyzed the prognostic effect of HLA-E expression in colorectal cancer. A statistically significant association with outcome was noticed where high expression of HLA-E resulted in a worse disease-free survival of patients (35). HLA-G expression showed a positive correlation with higher histological grade and clinical stage in colorectal cancer, gastric cancer, epithelial squamous cell carcinomas, and cutaneaous T cell lymphoma. In addition, expression of HLA-G was an independent prognostic factor for a worse outcome of patients in colorectal cancer, epithelial squamous cell carcinomas, and non-small cell lung cancers (16, 18, 19). We described that tumor expression of HLA-E and HLA-G has an independent prognostic influence in breast cancer patients, resulting in a worse patient outcome. Previously, similar results for disease-free survival were found for breast cancer, albeit that these results did not reach statistical significance (20). This study was similar to ours in terms of patients selection criteria and immunohistochemical staining methods but was probably limited by the small number of breast cancer patients studied (n = 43). The results of our study demonstrate for the first time, to our knowledge, a statistically significant association of HLA-E and HLA-G expression with clinical outcome in a large cohort of breast cancer patients, which is particularly revealed in patients with tumors lacking expression of classical HLA class I molecules. Moreover, patients with tumors with simultaneous expression of HLA-E and HLA-G had an increased risk of relapses compared with patients with tumors expressing either HLA-E or HLA-G, a phenomenon that has been previously described as well (13). In addition, we were able to demonstrate a statistical interaction in outcome analyses, indicating that the effect on outcome of HLA-E and HLA-G expression and the effect on outcome of HLA class I expression do not only operate simultaneously, but that the combined effect on outcome of these molecules is more than additive. These data correspond to the hypothesis that tumor expression of the non-classical HLA class I molecules E and G may indeed serve to protect tumor cells from NK-cell attack, but this is mostly relevant in a situation that NK cells are activated (i.e., in case classical HLA class I molecule expression is downregulated [10]).

Taken together, these results provide insights in breast cancer tumorigenesis and provide further evidence that the immune system is able to recognize and eliminate breast cancer cells. However, it is also evident that breast cancer cells are capable of escaping immune attack. A better understanding of the various phases of tumor–immune interactions in breast cancer (i.e., elimination, equilibrium and finally escape) may lead to a better prediction of clinical outcome of patients. Furthermore, this knowledge may be used for the development of tailored immunotherapeutic treatment modalities.

We thank J. Molenaar for help with the database and colleagues at the research laboratory of the surgery department at the Leiden University Medical Center for help and advice.

Disclosures The authors have no financial conflicts of interest.