Purpose: Despite improvements in cancer treatment, the prognosis of ovarian cancer remains low and imperfectly predicted by traditional pathologic criteria. Biomarkers that predict prognosis independently of such criteria shed light on important molecular variations, aiding in the development and targeting of novel therapies. Previous work has shown human leukocyte antigen (HLA) class I antigen expression to be independently predictive of prognosis in colorectal and breast cancer. We investigated the prognostic potential of HLA class I antigen expression by studying a large series of ovarian cancers.

Experimental Design: A tissue microarray of 339 ovarian cancer cases linked to prospectively recorded clinicopathologic and follow-up data was constructed. This was stained following a standard immunohistochemical protocol for HLA class I heavy chain (HC-10) and β2-microglobulin (β2-m). HLA class I antigen expression was compared with clinicopathologic factors and overall disease-specific survival using the Pearson χ2 test, Kaplan-Meier curves, and the log-rank test. Cox regression was used to test for the independence and magnitude of effects.

Results: There were no univariate correlations between HLA class I antigen expression and clinicopathologic factors. Deviation from an HC-10+2-m+ phenotype correlated with reduced survival in univariate analysis (log-rank, 5.69; P = 0.017); a retained HC-10+2-m+ phenotype predicted improved prognosis independently of age, stage, level of cytoreduction, and chemotherapy usage on multivariate analysis (hazard ratio, 0.587; 95% confidence interval, 0.442-0.781; P < 0.001).

Conclusions: HLA class I antigen expression is an independent prognostic marker in ovarian cancer, its loss correlating with a poor prognostic outcome.

Ovarian cancer is the most common gynecologic cancer in the United Kingdom and represents the fourth most common cancer site in women. Its low 5-year survival of 29% has altered little over the last 40 years despite advances in treatment (1). Improving our understanding of the molecular events that underlie the divergent outcomes of apparently identical cases through the search for biomarkers that predict prognosis independently of traditional criteria is fundamental to the development and subsequent targeting of novel treatments.

Down-regulation of human leukocyte antigen (HLA) class I antigen expression has been documented in a variety of tumors, including ovarian cancer, and is often, although not exclusively, associated with features of aggressive disease and a poorer prognosis (28). HLA class I antigen down-regulation as a biomarker has also been shown to have prognostic powers in breast and colon cancer independent of the traditional prognostic markers for these diseases (9, 10). In ovarian cancer, the presence of tumor-specific cytotoxic T cells (CTL) within the tumor-infiltrating lymphocytes of advanced ovarian cancer suggests an important role for HLA class I antigens (11). However, whereas HLA class I antigen down-regulation has been shown to predict poor prognosis in aneuploid types (12), the independence of this prognostic power is as yet unreported in unselected ovarian neoplasms.

The aim of this study was thus to investigate whether HLA class I antigen down-regulation predicts prognosis in unselected ovarian cancer, the hypothesis being that this would be associated with poorer survival. We used established tissue microarray techniques and immunohistochemistry to study 339 sequential cases of ovarian cancer (1315).

Patients and tissues. Patients with ovarian cancer undergoing primary surgical management at the Derby City General Hospital have been followed up prospectively since 1982. Data recorded includes age at diagnosis, International Federation of Gynecology and Obstetrics stage, degree of cytoreduction (optimal being the absence of macroscopic disease postoperatively), adjuvant treatment, and disease-specific survival (calculated from the surgery date up until November 31, 2005 when survivors were censored). The database was updated and audited to ensure accuracy of data entry. There were no major inaccuracies, and >97% of data were available. It was estimated that 300 cases would be required to give an 80% chance of detecting hazard ratios of ≥1.4 and ≤0.7 (EGRET, Cytel Statistical Software); thus, 395 sequential cases of ovarian cancer treated between January 1, 1982 and December 31, 1997 were considered for inclusion. Patients received adjuvant chemotherapy unless they had low-grade stage I disease. The type of chemotherapy provided followed the best practice for the day and was platinum based in later years.

Following resection, the tumors were fixed immediately in 10% neutral buffered formalin and subsequently embedded in paraffin wax before long-term storage in an ambient environment. New sections were cut, stained with H&E, and reviewed by a gyne-pathologist (S.D.) blinded to the original clinical information, and tumors were typed and graded (gyne-oncology group system) with representative areas to be included on the arrays marked. Cases were excluded if not found to be ovarian cancer on review, or if no archived tissue was available (Fig. 1). Ultimately, 0.6-mm-diameter cores from 339 tumors were precisely arrayed into four blocks using a manual tissue arrayer (Beecher Instruments) with 76 to 133 tumors per block, as described previously (13). The tissue microarrays were assembled in five copies, each biopsy being from a different, representative area of the tumor.

Fig. 1.

Flow diagram illustrating the derivation of a 298-case subgroup of the original series of ovarian cancers that was analyzed immunohistochemically for functional HLA class I antigen expression.

Fig. 1.

Flow diagram illustrating the derivation of a 298-case subgroup of the original series of ovarian cancers that was analyzed immunohistochemically for functional HLA class I antigen expression.

Close modal

Ethical approval to carry out the study was granted by the Derbyshire Local Research Ethics Committee.

Immunohistochemistry and evaluation of staining. To stain for the widest range of HLA class I heavy chains with a single antibody, the mouse anti-human monoclonal antibody to HLA class I heavy chain (HC-10-Gift; Prof. H Ploegh, Harvard Medical School) was selected. This antibody was raised to denatured HLA class I heavy chains freed from the β2-microglobulin light-chain (β2-m) element of denatured native HLA class I antigen and preferentially recognizes HLA-B and HLA-C on formalin-fixed, paraffin-embedded tissue (16). HLA class I antigens require the presence of β2-m to function, and co-staining these tumors for β2-m using a commercial polyclonal rabbit anti-human antibody (A0072, DAKO) allowed the generation of HC-10/β2-m phenotypes. Any variation from the expected HC-10+2-m+ phenotype was considered a down-regulation of the HLA class I antigen. An HC-10+2-m phenotype is likely to represent a state of genuinely free, non-functional HLA class I heavy chain and an HC-102-m phenotype complete loss of HLA class I antigen. An HC-102-m+ phenotype may represent down-regulation of HLA class I antigen and/or the presence of other HLA class I heavy-chain locus products not identified by HC-10.

Freshly cut 4-μm sections of tissue microarray, random whole ovarian cancer sections, and positive controls (tonsil) were deparaffinized and rehydrated, and endogenous peroxidase activity was blocked by immersion in 0.3% hydrogen peroxide for 15 min. Following antigen retrieval by microwave in citrate buffer (pH 6.0) for 10 min at high power and 10 min at low power (for HC-10 only), the sections were incubated in normal swine serum for 10 min and then incubated with the primary antibody for 60 min. The optimal dilutions of the primary antibodies were ascertained by titration on whole ovarian cancer sections previously found to express HC-10 and β2-m. Negative controls were ovarian tumors known to stain positively with the primary antibodies that were incubated instead with normal swine serum at this point. The sections were then incubated with a biotinylated goat anti-mouse/rabbit secondary antibody at 1:100 dilution (DAKO) for 30 min and then by a streptavidin-biotinylated horseradish peroxidase complex at 1:100 dilution for 45 min (DAKO). Visualization was by the application of 3,3′-diaminobenzidine (DAKO).

Following the review of the whole section staining by two authors (P.R and S.D), a semiquantitative scoring system was adopted. Only cell membrane staining was considered, and for each core, the intensity of staining was scored in four categories (0, absent; 1, weak; 2, moderate; 3, strong), and the percentage of tumor cell membranes staining was recorded (Fig. 2A-D). The cores were scored independently by the same two authors blinded to the clinical information, with a consensus being reached in difficult cases. To allow differential levels of expression to be analyzed within the individual biomarker groups, the percentage of cells staining for β2-m was used to stratify the distribution of this marker effectively, whereas the intensity of the cell membrane more effectively stratified the HC-10 staining. Average values for % cells β2-m positive were determined over ×200 power fields: 0 field (24 of 339, 7.1% cases), 1 field (102 of 339, 37.2% cases), and 2 fields (213 of 339, 62.8% cases). Average values for intensity of membranes staining HC-10 positive were determined over ×200 power fields: 0 field (28 of 339, 8.3% cases), 1 field (82 of 339, 24.2% cases), and 2 fields (229 of 339, 67.6% cases). In defining positivity (to allow for coexpression analysis of both markers), any degree of positive staining was considered to indicate positive expression, thereby minimizing false negatives while also adopting a reproducible system. This equated to a cutoff of ≥10% for β2-m staining (cores scored to within the nearest 10%) and ≥1 intensity for HC-10; ≥10% of cell membranes staining with ≥1 intensity effectively defined positivity for both markers.

Fig. 2.

A, ovarian cancer tissue core staining weakly (1) with HC-10 in the cell membrane. B, ovarian cancer tissue core staining moderately (2) with HC-10 in the cell membrane. C, ovarian cancer tissue core staining strongly (3) with HC-10 in the cell membrane. D, ovarian cancer tissue core staining positive for β2-m in the cell membrane of 100% of the cells.

Fig. 2.

A, ovarian cancer tissue core staining weakly (1) with HC-10 in the cell membrane. B, ovarian cancer tissue core staining moderately (2) with HC-10 in the cell membrane. C, ovarian cancer tissue core staining strongly (3) with HC-10 in the cell membrane. D, ovarian cancer tissue core staining positive for β2-m in the cell membrane of 100% of the cells.

Close modal

Statistical analysis. The data were analyzed using SPSS version 12 (SPSS). Univariate associations between categorical variables were calculated using the Pearson χ2 test, and the survival analysis was done using Kaplan-Meier curves; differences between the groups are estimated using the log-rank test. Cox regression was used in a multivariate analysis to determine hazard ratios and the independence of effects. P < 0.05 defined statistical significance.

Neither the degree of HC-10 nor the degree β2-m expression showed any correlations with clinicopathologic characteristics (data not shown) or prognosis (log rank, 1.18; P = 0.28 and log rank, 1.89; P = 0.17, respectively) and, as intended, subsequent analyses focused on the analysis of their combined phenotypes. Of the series of 395 patients, 298 had expression data complete enough to allow for co-analysis (Fig. 1); 32.8% (102 of 298) of cases were negative for HC-10, and 55.6% (175 of 298) of cases were negative for β2-m. With only 34.6% (103 of 298) of cases HC-10+2-m+, 65.4% (195 of 298) of cases had phenotypes suggestive of HLA class I antigen down-regulation, revealing this to be a common event in this series (Table 1).

Table 1.

The immunohistochemical expression of HLA class I heavy chain (HC-10), β2-m light chain, and their combined phenotypes

No. cases% Cases
HLA class I heavy chain (HC-10) intensity (n = 311)   
    <1 102 32.8 
    1-2 85 27.3 
    2-3 124 39.9 
HLA class I heavy chain (HC-10) positive (≥1 intensity) 209 67.2 
β2-m light chain % cells positive (n = 315)   
    0-25 242 76.8 
    26-50 42 13.3 
    51-75 19 
    76-100 12 3.8 
β2 microglobulin light chain positive (>10% cells) 140 44.4 
HLA Class I antigen expression phenotypes (n = 298)   
    HC-10+2-m+ 103 34.6 
Down-regulated 195 65.4 
    HC-10+2-m 98 32.9 
    HC-102-m+ 68 23.8 
    HC-102-m 29 9.7 
No. cases% Cases
HLA class I heavy chain (HC-10) intensity (n = 311)   
    <1 102 32.8 
    1-2 85 27.3 
    2-3 124 39.9 
HLA class I heavy chain (HC-10) positive (≥1 intensity) 209 67.2 
β2-m light chain % cells positive (n = 315)   
    0-25 242 76.8 
    26-50 42 13.3 
    51-75 19 
    76-100 12 3.8 
β2 microglobulin light chain positive (>10% cells) 140 44.4 
HLA Class I antigen expression phenotypes (n = 298)   
    HC-10+2-m+ 103 34.6 
Down-regulated 195 65.4 
    HC-10+2-m 98 32.9 
    HC-102-m+ 68 23.8 
    HC-102-m 29 9.7 

There were no striking differences between the clinicopathologic characteristics of the entire series and this apparently randomly selected 298 strong subgroup of cases (Table 2). The median age of diagnosis for these cases was 62 years; the median follow-up for this group was 21.7 months (range, 0-271 months); and the 5-year survival was 27.1%. This group's characteristics are similar to those of large-scale reports (1, 1720). On univariate analysis using the χ2 method, the HC-10+2-m+ phenotype was not found to be associated with age (using the Surveillance, Epidemiology, and End Results program of the National Cancer Institute subdivisions; ref. 17), International Federation of Gynecology and Obstetrics stage, degree of cytoreduction, grade, histologic type, or the use of chemotherapy (Table 2). However, as seen in the Kaplan-Meier plot (Fig. 3), those patients without this phenotype did have a worse prognosis (log rank, 5.69; P = 0.017). These cases with phenotypes indicating down-regulation of HLA class I antigen had a 5-year survival of 23.2% compared with 34.7% when both HC-10 and β2-m were expressed. Factors in this series found to predict prognosis independently of each other were age (as a continuous variable), International Federation of Gynecology and Obstetrics stage, the absence of residual macroscopic disease following primary surgery, and whether the patient received chemotherapy of any form. These factors were included in the Cox model along with the HLA class I antigen expression phenotype. The presence of the HC-10+2-m+ phenotype was found to retain its power to predict an improved prognosis independently of the other factors (hazard ratio, 0.587; 95% confidence interval, 0.442-0.781; P < 0.001; Table 3). When the other phenotypes were analyzed in a similar fashion, no correlations were found in either univariate or multivariate analysis (data not shown), indicating that it is the lack of an expected HC-10+2-m+ phenotype that is deleterious in ovarian cancer.

Table 2.

Clinicopathological characteristics of the entire series

Whole series (n = 395), no. cases (%)Analyzed cases (n = 298), no. cases (%)HC-10+2-m+ (n = 103), no. cases (%)Down-regulated phenotypes (n = 195), no. cases (%)χ2 (p)
Age n = 394 n = 297 n = 103 n = 194  
    <30 1 (0.3) 1 (0.3) 1 (0.5)  
    30-59 167 (42.4) 119 (40.1) 42 (40.8) 77 (39.7) 0.554 
    >60 226 (57.4) 177 (59.6) 61 (59.2) 116 (59.8) (0.758) 
FIGO stage n = 375 n = 290 n = 99 n = 191  
    1 99 (26.4) 77 (26.6) 27 (27.3) 50 (26.2)  
    2 46 (12.3) 33 (11.4) 9 (9.1) 24 (12.6)  
    3 188 (50.1) 144 (49.7) 51 (51.5) 93 (48.7) 0.836 
    4 42 (11.2) 36 (12.4) 12 (12.1) 24 (12.6) (0.841) 
Optimally debulked n = 376 n = 287 n = 99 n = 188 0.057 
    Yes 157 (41.8) 119 (41.5) 42 (42.4) 77 (41.0) (0.811) 
Grade n = 376 n = 296 n = 99 n = 195  
    1 50(13.3) 32 (10.8) 11 (10.9) 21 (10.8)  
    2 93 (24.7) 66 (22.3) 19 (18.8) 47 (24.1) 1.102 
    3 233 (62) 198 (66.9) 71 (70.3) 127 (65.1) (0.576) 
Histologic type n = 395 n = 298 n = 103 n = 195  
    Serous 203 (51.4) 162 (54.4) 52 (50.5) 110 (56.4)  
    Endometrioid 46 (11.7) 37 (12.4) 11 (10.7) 26 (13.3)  
    Mucinous 50 (12.7) 26 (8.7) 12 (11.7) 14 (7.2)  
    Undifferentiated 65 (16.5) 45 (15.1) 17 (16.5) 28 (14.4)  
    Clear cell 26 (6.6) 24 (8.1) 10 (9.7) 14 (7.2) 3.264 
    Other 5 (1.3) 4 (1.4) 1 (1) 3 (1.5) (0.659) 
Chemotherapy n = 388 n = 292 n = 100 n = 192 0.228 
    Yes 283 (73.2) 214 (73.3) 75 (75) 140 (72.9) (0.633) 
Whole series (n = 395), no. cases (%)Analyzed cases (n = 298), no. cases (%)HC-10+2-m+ (n = 103), no. cases (%)Down-regulated phenotypes (n = 195), no. cases (%)χ2 (p)
Age n = 394 n = 297 n = 103 n = 194  
    <30 1 (0.3) 1 (0.3) 1 (0.5)  
    30-59 167 (42.4) 119 (40.1) 42 (40.8) 77 (39.7) 0.554 
    >60 226 (57.4) 177 (59.6) 61 (59.2) 116 (59.8) (0.758) 
FIGO stage n = 375 n = 290 n = 99 n = 191  
    1 99 (26.4) 77 (26.6) 27 (27.3) 50 (26.2)  
    2 46 (12.3) 33 (11.4) 9 (9.1) 24 (12.6)  
    3 188 (50.1) 144 (49.7) 51 (51.5) 93 (48.7) 0.836 
    4 42 (11.2) 36 (12.4) 12 (12.1) 24 (12.6) (0.841) 
Optimally debulked n = 376 n = 287 n = 99 n = 188 0.057 
    Yes 157 (41.8) 119 (41.5) 42 (42.4) 77 (41.0) (0.811) 
Grade n = 376 n = 296 n = 99 n = 195  
    1 50(13.3) 32 (10.8) 11 (10.9) 21 (10.8)  
    2 93 (24.7) 66 (22.3) 19 (18.8) 47 (24.1) 1.102 
    3 233 (62) 198 (66.9) 71 (70.3) 127 (65.1) (0.576) 
Histologic type n = 395 n = 298 n = 103 n = 195  
    Serous 203 (51.4) 162 (54.4) 52 (50.5) 110 (56.4)  
    Endometrioid 46 (11.7) 37 (12.4) 11 (10.7) 26 (13.3)  
    Mucinous 50 (12.7) 26 (8.7) 12 (11.7) 14 (7.2)  
    Undifferentiated 65 (16.5) 45 (15.1) 17 (16.5) 28 (14.4)  
    Clear cell 26 (6.6) 24 (8.1) 10 (9.7) 14 (7.2) 3.264 
    Other 5 (1.3) 4 (1.4) 1 (1) 3 (1.5) (0.659) 
Chemotherapy n = 388 n = 292 n = 100 n = 192 0.228 
    Yes 283 (73.2) 214 (73.3) 75 (75) 140 (72.9) (0.633) 

NOTE: Univariate analysis by χ2 of phenotype status versus clinicopathologic criteria is also displayed; showing no associations.

Abbreviation: FIGO, International Federation of Gynecology and Obstetrics.

Fig. 3.

Kaplan-Meier analysis showing the overall survival of ovarian cancer patients according to the presence of a HC-10+/β2-m+ phenotype representing retained HLA class I antigen expression.

Fig. 3.

Kaplan-Meier analysis showing the overall survival of ovarian cancer patients according to the presence of a HC-10+/β2-m+ phenotype representing retained HLA class I antigen expression.

Close modal
Table 3.

Multivariate analysis by Cox regression

Hazard ratios (95% confidence intervals)Significance (P)
Age at diagnosis (y) 1.027 (1.015-1.038) <0.001 
FIGO Stage   
    1  <0.001 
    2 2.934 (1.591-5.414) 0.001 
    3 5.796 (3.311-10.146) <0.001 
    4 8.071 (4.247-15.338) <0.001 
Residual macroscopic disease 2.026 (1.405-2.921) <0.001 
Patient received chemotherapy 0.430 (0.287-0.644) <0.001 
HC-10+2-m+ phenotype 0.587 (0.442-0.781) <0.001 
Hazard ratios (95% confidence intervals)Significance (P)
Age at diagnosis (y) 1.027 (1.015-1.038) <0.001 
FIGO Stage   
    1  <0.001 
    2 2.934 (1.591-5.414) 0.001 
    3 5.796 (3.311-10.146) <0.001 
    4 8.071 (4.247-15.338) <0.001 
Residual macroscopic disease 2.026 (1.405-2.921) <0.001 
Patient received chemotherapy 0.430 (0.287-0.644) <0.001 
HC-10+2-m+ phenotype 0.587 (0.442-0.781) <0.001 

As with all tissue microarray work, the final group analyzed (n = 298) is smaller than the original series identified (n = 395) due to incomplete data and missing tissue. Studying the coexpression of two markers exacerbated this. Significant selection bias within the final analyzed group could potentially have been introduced by non-random loss of tissue, but as the final group varies little from the original series regarding its clinicopathologic characteristics, this is unlikely to have occurred. With 298 cases available, the power of the study was retained.

In this study, we show, along with other work, that down-regulation of HLA class I antigen is a common event in ovarian cancer (2, 8). The 34.6% of tumors that retained a complete HC-10+2-m+ phenotype had a better prognosis than other cases; furthermore, the expression/loss of this phenotype is an independent prognostic factor in ovarian cancer of a similar magnitude to the presence/absence of residual macroscopic disease (Table 3). The absence of complete HLA class I antigen expression leading to a worse prognosis supports work by Vitale et al. (2), which concluded that down-regulation of HLA type I loss was associated with worse disease, in this case, stage III cancers. Differences in our methodologies, regarding powering, biomarker co-analysis, and the use of Union Internationale Contra Cancer independent prognostic factors (21) in statistical analyses, meant that while we should not duplicate this specific finding, we were better positioned to show a novel and important independent association with prognosis.

Cancer cells may be eliminated by, be in equilibrium with, or escape innate or adaptive immunologic control mechanisms via immunediting (22, 23), and an important mechanism by which the immune system eliminates cancer cells is via a CTL-based response to abnormal peptide presented in conjunction with HLA class I antigen on tumor cells (24). Given the genetic instability of cancer, such a selective pressure might result in the evolution and ultimate protection of tumors from this process via down-regulation of HLA class I antigen expression, as either an early or multistep process (25). This is likely to be important in the clinical course of ovarian cancer (26), and evidence for such tumor sculpting comes from the observation that highly tumor-specific CTL exist within the tumor-infiltrating lymphocyte population of advanced disease whose function can be inhibited by anti–HLA class I monoclonal antibodies but which fail to elicit a clinically adequate immune response (11, 27). That down-regulation of HLA class I antigen should associate with worse survival, as we found, is therefore logical.

When HLA class I antigen expression is lost, killer cell–inhibitory receptors on the surface of natural killer cells (which produce an inhibitory signal when bound to HLA class I antigen) no longer function. The “missing self” hypothesis concludes that a resultant increase in natural killer cell killing activity will compensate for diminished T-cell killing in these circumstances (28). This may occur clinically in breast and colorectal cancer and explain the improvement in prognosis of those tumors with total loss of HLA class I antigen expression (9, 10). However, our data suggest that this mechanism must be less effective in ovarian cancers. This may be due to a concurrent acquired resistance to natural killer cell killing (29), a lack of natural killer cell presence within the tumor infiltrating lymphocytes (7, 30), or the rapid exhaustion of this mechanism by the large volume nature ovarian disease (22).

There are seven reported mechanisms by which a state of down-regulated HLA class I antigen might occur: total HLA class I antigen loss due to the disrupted synthesis of antigen-processing machinery components such as β2-m and transporter associated with antigen processing; haplotypic loss of either one or two haplotypes due to chromosomal events; HLA A, B, or C locus down-regulation due to altered HLA class I antigen gene transcription; individual HLA allelic loss due to gene point mutations or deletions; a compound state of two or more of these events; an unresponsiveness to IFNs due to down-regulation of the IFN response sequence element and subsequent pathway disruption; and finally down-regulation of HLA A-B-C molecules with a concurrent up-regulation of HLA-E. The most frequent event observed in this series was loss of β2-m. This contributed to over 55% of the loss of HLA class I antigen. The mechanisms for β2-m loss in cancer are, like those for heavy chain, varied but commonly include loss of mRNA stability (31) and mutational silencing (32). However, genomic losses due to chromosomal defects, particularly in chromosome 15, are not frequently observed in ovarian cancer. In melanoma, β2-m mutations are primarily related to extensive loss of chromosome 15 material (33); although alterations in chromosome 15 in ovarian cancer have been well documented, there are no reports of frequent alterations in the β2-m locus (15q21). These mechanisms may relate to ovarian cancer where we have shown a significant loss and clear clinical correlation with poor outcome. However, other reports suggest that HLA class I antigen down-regulation in ovarian cancer is a post-translational event that can be restored upon IFN-γ treatment (3437). The exact mechanism of HLA-class I antigen loss in ovarian cancer warrants further investigation and would be most successfully achieved on freshly isolated tumor samples.

Our work shows that ovarian cancers not expressing the HC-10+2-m+ phenotype have a poorer prognosis irrespective of their gross appearance. Future work could also test whether foreknowledge of this phenotype might predict those cases that will respond best to CTL- or IFN-γ–based immunotherapies.

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.

We thank Claire Paisch, Rob Moss, John Rohan, Ian Ellis, Zahra Madjd, and Nick Watson for their technical advice; Sam Crockett, Claire Coveney, and Denise Wingate for their help updating and auditing the final database; Steve Kyte and Andrea Gooding for their help identifying the archived tissue blocks and manufacturing new sections; Irshad Soomro and Zia Choudry for reviewing the histology of the early cases; Sarah Lewis and Tricia McKeever for their statistical guidance; Sheila O'Malley for her governance input; and David Liu for his support.

1
Quinn M, Great Britain. Office for National Statistics. Cancer trends in England and Wales, 1950-1999. London: The Stationery Office; 2001.
2
Vitale M, Pelusi G, Taroni B, et al. HLA class I antigen down-regulation in primary ovary carcinoma lesions: association with disease stage.
Clin Cancer Res
2005
;
11
:
67
–72.
3
Hicklin DJ, Marincola FM, Ferrone S. HLA class I antigen downregulation in human cancers: T-cell immunotherapy revives an old story.
Mol Med Today
1999
;
5
:
178
–86.
4
Concha A, Esteban F, Cabrera T, Ruiz-Cabello F, Garrido F. Tumor aggressiveness and MHC class I and II antigens in laryngeal and breast cancer.
Semin Cancer Biol
1991
;
2
:
47
–54.
5
Stein B, Momburg F, Schwarz V, et al. Reduction or loss of HLA-A,B,C antigens in colorectal carcinoma appears not to influence survival.
Br J Cancer
1988
;
57
:
364
–8.
6
Levin I, Klein T, Goldstein J, et al. Expression of class I histocompatibility antigens in transitional cell carcinoma of the urinary bladder in relation to survival.
Cancer
1991
;
68
:
2591
–4.
7
Kabawat SE, Bast RC, Jr., Welch WR, Knapp RC, Bhan AK. Expression of major histocompatibility antigens and nature of inflammatory cellular infiltrate in ovarian neoplasms.
Int J Cancer
1983
;
32
:
547
–54.
8
Norell H, Carlsten M, Ohlum T, et al. Frequent loss of HLA-A2 expression in metastasizing ovarian carcinomas associated with genomic haplotype loss and HLA-A2-restricted HER-2/neu-specific immunity.
Cancer Res
2006
;
66
:
6387
–94.
9
Madjd Z, Spendlove I, Pinder SE, Ellis IO, Durrant LG. Total loss of MHC class I is an independent indicator of good prognosis in breast cancer.
Int J Cancer
2005
;
117
:
248
–55.
10
Watson NF, Ramage JM, Madjd Z, et al. Immunosurveillance is active in colorectal cancer as downregulation but not complete loss of MHC class I expression correlates with a poor prognosis.
Int J Cancer
2006
;
118
:
6
–10.
11
Santin AD, Bellone S, Palmieri M, et al. Restoration of tumor specific human leukocyte antigens class I-restricted cytotoxicity by dendritic cell stimulation of tumor infiltrating lymphocytes in patients with advanced ovarian cancer.
Int J Gynecol Cancer
2004
;
14
:
64
–75.
12
Moore DH, Fowler WC, Jr., Olafsson K. Class I histocompatibility antigen expression: a prognostic factor for aneuploid ovarian cancers.
Gynecol Oncol
1990
;
38
:
458
–61.
13
Kononen J, Bubendorf L, Kallioniemi A, et al. Tissue microarrays for high-throughput molecular profiling of tumor specimens.
Nat Med
1998
;
4
:
844
–7.
14
Hoos A, Urist MJ, Stojadinovic A, et al. Validation of tissue microarrays for immunohistochemical profiling of cancer specimens using the example of human fibroblastic tumors.
Am J Pathol
2001
;
158
:
1245
–51.
15
Rosen DG, Huang X, Deavers MT, et al. Validation of tissue microarray technology in ovarian carcinoma.
Mod Pathol
2004
;
17
:
790
–7.
16
Stam NJ, Spits H, Ploegh HL. Monoclonal antibodies raised against denatured HLA-B locus heavy chains permit biochemical characterization of certain HLA-C locus products.
J Immunol
1986
;
137
:
2299
–306.
17
Quirk JT, Natarajan N. Ovarian cancer incidence in the United States, 1992-1999.
Gynecol Oncol
2005
;
97
:
519
–23.
18
Brun JL, Feyler A, Chene G, et al. Long-term results and prognostic factors in patients with epithelial ovarian cancer.
Gynecol Oncol
2000
;
78
:
21
–7.
19
Yancik R. Ovarian cancer. Age contrasts in incidence, histology, disease stage at diagnosis, and mortality.
Cancer
1993
;
71
:
517
–23.
20
ICON2: randomised trial of single-agent carboplatin against three-drug combination of CAP (cyclophosphamide, doxorubicin, and cisplatin) in women with ovarian cancer. ICON Collaborators. International Collaborative Ovarian Neoplasm Study.
Lancet
1998
;
352
:
1571
–6.
21
Gospodarowicz MK, Henson DE, Hutter RVP, O'Sullivan B, Sobin LH, Wittekind CH, eds. Prognostic factors in cancer. 2nd ed. New York: Chichester: Wiley-Liss; 2001.
22
Ochsenbein AF, Klenerman P, Karrer U, et al. Immune surveillance against a solid tumor fails because of immunological ignorance.
Proc Natl Acad Sci U S A
1999
;
96
:
2233
–8.
23
Dunn GP, Bruce AT, Ikeda H, Old LJ, Schreiber RD. Cancer immunoediting: from immunosurveillance to tumor escape.
Nat Immunol
2002
;
3
:
991
–8.
24
Seliger B, Maeurer MJ, Ferrone S. Antigen-processing machinery breakdown and tumor growth.
Immunol Today
2000
;
21
:
455
–64.
25
Klein G, Klein E. Evaluation of tumours and the impact of molecular oncology.
Nature (Lond)
1985
;
315
:
190
–5.
26
Algarra I, Cabrera T, Garrido F. The HLA crossroad in tumor immunology.
Hum Immunol
2000
;
61
:
65
–73.
27
Santin AD, Hermonat PL, Ravaggi A, et al. In vitro induction of tumor-specific human lymphocyte antigen class I-restricted CD8 cytotoxic T lymphocytes by ovarian tumor antigen-pulsed autologous dendritic cells from patients with advanced ovarian cancer.
Am J Obstet Gynecol
2000
;
183
:
601
–9.
28
Ljunggren HG, Karre K. In search of the ‘missing self’: MHC molecules and NK cell recognition.
Immunol Today
1990
;
11
:
237
–44.
29
Garrido F, Ruiz-Cabello F, Cabrera T, et al. Implications for immunosurveillance of altered HLA class I phenotypes in human tumours.
Immunol Today
1997
;
18
:
89
–95.
30
Ferguson A, Moore M, Fox H. Expression of MHC products and leucocyte differentiation antigens in gynaecological neoplasms: an immunohistological analysis of the tumour cells and infiltrating leucocytes.
Br J Cancer
1985
;
52
:
551
–63.
31
Hicklin DJ, Wang Z, Arienti F, et al. beta2-Microglobulin mutations, HLA class I antigen loss, and tumor progression in melanoma.
J Clin Invest
1998
;
101
:
2720
–9.
32
Bicknell DC, Kaklamanis L, Hampson R, Bodmer WF, Karran P. Selection for beta 2-microglobulin mutation in mismatch repair-defective colorectal carcinomas.
Curr Biol
1996
;
6
:
1695
–7.
33
Paschen A, Arens N, Sucker A, et al. The coincidence of chromosome 15 aberrations and beta2-microglobulin gene mutations is causative for the total loss of human leukocyte antigen class I expression in melanoma.
Clin Cancer Res
2006
;
12
:
3297
–305.
34
Boyer CM, Dawson DV, Neal SE, et al. Differential induction by interferons of major histocompatibility complex-encoded and non-major histocompatibility complex-encoded antigens in human breast and ovarian carcinoma cell lines.
Cancer Res
1989
;
49
:
2928
–34.
35
Santin AD, Rose GS, Hiserodt JC, et al. Effects of cytokines combined with high-dose gamma irradiation on the expression of major histocompatibility complex molecules and intercellular adhesion molecule-1 in human ovarian cancers.
Int J Cancer
1996
;
65
:
688
–94.
36
Mobus VJ, Asphal W, Knapstein PG, Kreienberg R. Effects of interferon gamma on the proliferation and modulation of cell-surface structures of human ovarian carcinoma cell lines.
J Cancer Res Clin Oncol
1993
;
120
:
27
–34.
37
Freedman RS, Kudelka AP, Kavanagh JJ, et al. Clinical and biological effects of intraperitoneal injections of recombinant interferon-gamma and recombinant interleukin 2 with or without tumor-infiltrating lymphocytes in patients with ovarian or peritoneal carcinoma.
Clin Cancer Res
2000
;
6
:
2268
–78.