HLA class II alleles associated with infection by HPV16 in cervical cancer in situ
Article first published online: 16 JUN 2001
Copyright © 2001 Wiley-Liss, Inc.
International Journal of Cancer
Volume 93, Issue 6, pages 817–822, 15 September 2001
How to Cite
Beskow, A. H., Josefsson, A. M. and Gyllensten, U. B. (2001), HLA class II alleles associated with infection by HPV16 in cervical cancer in situ. Int. J. Cancer, 93: 817–822. doi: 10.1002/ijc.1412
- Issue published online: 13 AUG 2001
- Article first published online: 16 JUN 2001
- Manuscript Accepted: 10 APR 2001
- Manuscript Revised: 5 APR 2001
- Manuscript Received: 22 JAN 2001
- Swedish Cancer Foundation. Grant Number: 3436-B97-04XAA
- National Institutes of Health (USA). Grant Number: NR 1 RO1 CA61197-01A3
- cervical cancer
HLA class II alleles have been associated with an increased risk of developing cervical cancer through infection with oncogenic forms of human papilloma virus (HPV). We have examined the association of variation at the DRB1 and DQB1 loci with HPV16 infection and risk of development of cervical cancer by analysis of 440 cases diagnosed with cervical cancer in situ and 476 age-matched controls in a retrospective case-control study. The infection history of a woman was studied by analysis of cervical smears taken at multiple times during a period of up to 27 years (1969–95). The frequency of a number of alleles are either increased (DRB1*0801, DRB1*1501, DQB1*0402 and DQB1*0602) or decreased (DRB1*0101, DRB1*1301, DQB1*0501 and DQB1*0603) in the cancer patients compared to the controls. After correction for multiple testing, only the DQB1*0602 and the DRB1*1501 alleles remain associated with cancer and only in HPV16-infected patients (DQB1*0602: 102/264 (39%) vs. 130/476 (28%), p = 0.028 and DRB1*1501: 104/259 (40%) vs. 132/469 (28%), p = 0.027). These alleles are associated primarily with infection by HPV and only indirectly affect the risk of developing cervical cancer in situ. To study the impact of these alleles on persistence of infection, women with short-term infections were compared to those with long-term infections. Carriers of DQB1*0602 and DRB1*1501 were more frequent in the group with long-term HPV infections, indicating that these class II alleles contribute to the inability to clear an HPV infection. © 2001 Wiley-Liss, Inc.
Infection with oncogenic forms of human papillomavirus (HPV) is the major etiologic factor for the development of cervical cancer. Using sensitive detection methods, oncogenic HPVs have been found in almost every cervical tumor.1, 2 However, infection with oncogenic HPVs is common among women having normal cytology,3 mild lesions regress without intervention and most infections are cleared spontaneously. This implies that factors determining the course of an HPV infection contribute to the risk of developing cervical cancer. Such factors may include the inherent ability to respond to HPV infection.
It has been suggested that differences in the immune response to HPV affect whether an HPV infection is cleared or becomes persistent.4 Given the pivotal role of HLA molecules in the recognition of foreign peptides, several studies have been performed to examine the association of specific HLA alleles with HPV infection status and development of cervical cancer. In 1991, the HLA-DQw3 antigen was reported to be associated with cervical cancer in a German cohort,4 and later studies in Spanish,5 British6 and African-American cohorts7 showed similar results. Other HLA alleles have also been proposed to increase the risk of cervical cancer: DRB1*1501,5, 8 DRB1*04 and DRB1*11.9 A number of alleles that are less common in patients relative to controls have also been reported, e.g., DRB1*1301,8 DQB1*05019 and DQB1*0603.5 An increased risk for HPV16-associated cervical cancer has been reported for the DRB1*1501-DQB1*0602 haplotype,10 the DQA1*0102-DQB1*0602 haplotype11 and for the DRB1*0701 allele.12, 13 Finally, an HPV16-specific protective effect for DR13 alleles has been reported.10 While a direct comparison between these studies is somewhat difficult due to differences in cancer stage studied, laboratory methods and study population, they indicate an effect of HLA alleles on risk of cervical cancer. In our study, we have used a large case-control material to investigate the association of HLA class II alleles with infection by HPV16 and to the risk of developing cervical cancer in situ. Our results show that HLA class II alleles are associated primarily with infection of HPV and only secondarily with cervical cancer.
MATERIAL AND METHODS
The participants in our study were selected from a cohort of women residing in Uppsala County, Sweden, between 1969–95.14 The participants had to fulfill the following entry criteria: (i) born in Sweden; (ii) less than 50 years of age at entry and; (iii) their first smear classified as normal by cytology on squamous cell (PAP=1). Histologic specimens (either a small biopsy or a complete “cone”) from all eligible cases were reviewed by an experienced pathologist (Jan Pontén). All cases with invasive cancer or adenocarcinoma were excluded from the study, and only cases with squamous-cell cancer in situ, CIS, were included. For each case, 5 separate controls, individually matched by date of entry to the cohort (±90 days) and by year of birth, were randomly selected from the study cohort. The women included as controls had to be alive without having developed cancer in situ or invasive cancer before the date of diagnosis for their corresponding matched case. Also, the first smear of the cases and controls had to be classified as normal by cytology on squamous cell (PAP=1).15, 16
Collection of smears
The archival smears included in our study were collected during the period of 1969–95. A total of 3,480 smears with a mean of 3.1 smears per control and 4.4 smears per case were included.
Taqman assays for detection of HPV16 and β-actin were performed as previously described.18
One smear, with the lowest β-actin Ct-value (highest amount of genomic DNA) based on the Taqman assay, was selected from each woman for HLA typing. Two microliters of the DNA from each smear was used for PCR amplification of the DQB1 and DRB1 loci, using biotinylated PCR primers. The PCR products were hybridised to reverse-dotblots containing sequence-specific oligonucleotides, as provided by the manufacturer (Dynal REL™ SSO; Dynal, Sweden). These procedures have been described previously.19, 20 For subtyping, group-specific amplifications of the DR*03, *04, *11, *13 and *14 alleles were performed as previously described.21, 22
Due to the large number of alleles at individual HLA loci and the consequence of a large number of independent tests, we estimated p-values corrected for multiple testing by randomising the data according to the following procedure. A program that uses χ2 statistics was created to compare observed differences in carrier frequency between cases and controls. Prior to the randomising of genotypes, χ2 values were calculated for each allele and each comparison. The HLA genotypes were then randomised 5,000 times, while keeping case-control status and HPV status constant. χ2 values were then calculated for each allele from each randomised set of data. The χ2 value for the original, nonrandomised data was then compared to those of the randomised data sets. The uncorrected p-value for an allele was set as the proportion of χ2 values that gave a higher value than the χ2 value of the original data set. To correct for multiple testing, p-values were calculated for every allele in all permutations. From each of the 5,000 permutations, the lowest p-value among all alleles was selected. The uncorrected p-value of each allele was then compared to the set of lowest p-values, and the corrected p-value was calculated as the proportion of permutations with a p-value that is lower than the uncorrected. A program for randomising the data and calculating the corrected p-values is available upon request.
To study the effect of specific HLA alleles on HPV16 infection status, p-values and odds ratios (OR) with 95% confidence intervals (CI) were estimated using the SAS procedure FREQ. The probability of interactions was studied by using the GENMOD procedure in SAS.
The cohort studied was composed of 478 women diagnosed with cervical cancer in situ and 608 age-matched controls.18 HLA-DRB1 typing results were obtained for 433 cases and 469 controls, while HLA-DQB1 typing results were obtained for 440 cases and 476 controls. Due to poor DNA quality, incomplete typing results (only 1 of the 2 alleles could be detected) were obtained for 6 cases and 12 controls for DRB1 and 9 cases and 17 controls for DQB1. Since the statistical analysis was based on HLA carrier frequency, these individuals were included in the data set. Also, for 3 cases and 1 control, HPV16 typing data was lacking and these were included only in analysis of all women. A woman was considered as HPV16-positive if 1 or more of her cervical smear samples tested positive. In total, 60% (288/478) of the cases and 16% (99/608) of the controls tested HPV16-positive.
The distribution of DRB1 alleles in cases and controls and in HPV16-positive individuals is shown in Table I. Based on tests for single alleles, the DRB1 alleles *0101 (71/433 [16%] vs. 106/469 [23%]; p = 0.018), *0801 (51/433 [12%] vs.34/469 [7%]; p = 0.022), *1301 (41/433 [9%] vs. 69/469 [15%]; p = 0.011) and *1501 (155/433 [36%] vs. 132/469 [28%]; p = 0.019) show frequency differences between cases and controls. For HPV16-positive women, the alleles *0701 (52/259 [20%] vs. 59/469 [13%]; p = 0.007), *1301 (23/259 [9%] vs. 69/469 [15%]; p = 0.023) and *1501 (104/259 [40%] vs. 132/469 [28%]; p = 0.002) show differences between cases and controls. However, when the correction for multiple tests is applied, using randomisation of data, only the comparison for the *1501 allele is significant and only for HPV16-positive cases compared to controls (p = 0.027). For the *1501 allele, the OR = 1.7 (95% CI, 1.2–2.3).
|All n = 469||16 positive n = 76||16 negative n = 392||All n = 433||16 positive n = 259||16 negative n = 171|
The distribution of DQB1 alleles in cases and controls and in HPV16-positive women is shown in Table II. In the tests of individual alleles, the following DQB1 alleles show frequency differences between cases and controls *0402 (51/440 [12%] vs. 34/476 [7%], p = 0.023), *0501 (84/440 [19%] vs. 122/476 [26%], p = 0.021), *0602 (154/440 [35%] vs. 130/476 [27%], p = 0.007) and *0603 (43/440 [10%] vs. 69/476 [14%], p = 0.026). For HPV16-positive women, the *0402 (30/264 [11%] vs. 34/476 [7%], p = 0.037) and *0602 (102/264 [39%] vs. 130/476 [27%], p = 0.002) alleles show differences between patients and controls. When correcting for multiple tests, only the comparison between HPV16-infected women and controls for the *0602 allele (p = 0.028) remains statistically significant. For the *0602, OR = 1.7 (95% CI, 1.2–2.3). A summary of the associations found between HLA alleles and the development of cervical cancer in situ is shown in Table III.
|All n = 476||16 positive n = 79||16 negative n = 396||All n = 440||16 positive n = 264||16 negative n = 174|
|Cases vs. controls||HPV16-positive cases vs. controls|
|Uncorrected p-values||Corrected p-values||Uncorrected p-values||Corrected p-values|
Given that HPV is found in the majority of cervical cancer tumours, it is important to determine whether the DQB1*0602 and the DRB1*1501 alleles are associated primarily with infection or with development of cervical cancer. These 2 class II alleles both occur at a significantly higher frequency among infected compared to uninfected individuals (DQB1*0602 p = 0.004 and DRB1*1501 p = 0.001, Table IV). In these analyses no interaction was found with cancer in situ status (DQB1*0602 p = 0.56 and DRB1*1501 p = 0.95, Table IV); the magnitude of the association with infection is similar in both cases and controls when analysed separately. Therefore, carriers of these class II alleles have an increased risk of becoming infected by HPV16, and this indirectly increases the risk of development of cervical cancer in situ.
|HLA allele||Study group||No. of carriers among infected/ total no. of infected||No. of carriers among uninfected/total no. of uninfected||OR||Effect modified by cancer status p-value||95% CI||p-value|
Next, we examined whether the infection history of carriers of the DRB1*1501 and DQB1*0602 alleles differs from that of carriers of other alleles, making use of the study design that allowed us to assess the infection status of individual women at multiple times. To study the relationship between HLA carrier frequency and infection history, HPV16-positive women, regardless of case-control status, were divided into 2 groups: women with short-term infections and women with long-term infections. The smears used in our study are derived from women participating in the organised screening program, where women are invited to attend the clinic every 3–4 years. This introduces a limitation in the time period between samples from individual women, in that from many women smears were not sampled at shorter intervals than 3 years. Due to the length of the intervals between sampling occasions, the group with short-term infections included women with at least 1 HPV16-positive smear and a maximum of 3 years from the first HPV16-positive smear to the following negative smear. A total of 74 women qualified for these criteria (39 cases and 35 controls). The group with long-term infections included those with at least 3 consecutive positive smears and with a minimum of 3 years from the first positive smear to the last positive smear. Since the smears were not taken at completely regular time intervals, a maximum of 4 years was allowed between HPV16-positive smears in order for the woman to be included in the group with long-term infections. A total of 89 individuals qualified for these criteria (88 cases and 1 control). Long-term infections are found at a significantly higher frequency among cases than controls (p < 0.001 for a test of independence). This comparison is somewhat biased by the small number of observations in 1 cell and the difference in mean number of smears between cases and controls (4.4 and 3.1, respectively). The DRB1*1501 and DQB1*0602 carrier frequencies were then compared between women with short- and long-term infections (Table V). While the results are not statistically significant at the 5% level, carriers of the DRB1*1501 (p = 0.09) and DQB1*0602 (p = 0.12) allele both show a tendency of increased risk for long-term infections. Our results support the hypothesis that carriers of these alleles have a higher risk of developing long-term HPV16 infections, which indirectly increases the risk of developing cervical cancer. As the time for spontaneous clearance of an HPV infection has been shown to be considerably shorter than the mean interval between smears in our present study,3, 23 it is likely that some women in the group with long-term infections represent reinfections rather than persistent infections. Such a bias is likely to reduce the power of the test. Thus, a study design that would have allowed for more frequent sampling of smears may have increased the ability to detect a difference between short- and long-term infections.
|Study group||Short-term infection (n = 74)||Long-term infection (n = 89)||p-value|
|DRB1*1501 carriers||22 (30%)||38 (43%)|
|DRB1*1501 noncarriers||52 (70%)||51 (57%)||0.09|
|DQB1*0602 carriers||22 (30%)||37 (42%)|
|DQB1*0602 noncarriers||52 (70%)||52 (58%)||0.12|
We studied the contribution of the DRB1 and DQB1 loci to the risk of developing cervical cancer in situ. The comparison between cases and controls indicate an association of several alleles at each of these loci with risk of cancer (Table III). Since the HLA loci have multiple alleles and tests are performed without having an a priori hypothesis for the particular alleles involved, the level of significance for an individual test has to be corrected for multiple testing. A common means of correcting for multiple testing is to multiply the p-value for the individual test with the number of tests conducted (number of alleles). This Bonferroni correction provides a simple means of evaluating test results but is a conservative approach resulting in that some true associations will remain undetected. An alternative method is to calculate p-values from the actual data set and compare these to the expected p-values under the hypothesis of no association. Since this approach uses the actual data set, it is likely to present a more relevant correction for multiple testing. In using this method, we randomise the observed genotypes between individuals and determine the minimum p-value obtained in each randomisation. The corrected significance levels are then estimated as the proportion of randomisations that yields a smaller p-value than the original uncorrected p-value. This approach reduces the number of true associations that are discarded, while applying an appropriate correction for multiple testing. When applied to our data, both the Bonferroni correction and the randomisation method yield consistent information with regard to the HLA alleles showing a significant association with cervical cancer in situ.
Several of the DRB1 and DQB1 alleles found by us to be associated with cervical cancer in situ have also been reported by others, such as DR*15 in patients with CINI/II and invasive cervical cancer in a Dutch population,8 DR*2 (DR*15 and DR*16) with squamous-cell carcinoma of the cervix (SCCC) and DQB1*0402 in patients with CINIII/CIS in a Spanish population.5 None of the previous studies have presented positive associations of DRB1*0801 and DQB1*0602 with cervical cancer. Several alleles that show a negative association with cancer in situ in our study have shown a similar effect in other patient populations. For instance, DRB1*0101 and DQB1*0501 with CINI and CINIII in British patients,9 DR*6 (DR*13 and DR*14) with SCCC in German patients4 and DQB*0603 with CINI/II and invasive SCCC (ISCCC) in Spanish patients.5 We did not detect an increased risk of cervical cancer in situ for DQB1*03 carriers, as has been reported for cases with CINI/II,5 CINI and CINIII6, 9 and SCCC.4 Previous studies of Swedish patients with CINII/III24 or invasive cervical cancer25 did not detect this association. Although it is tempting to interpret a similarity in the alleles reported to be associated with the pre-stages or cervical cancer in different studies as support for true associations, it is noteworthy that most of these studies did not apply any correction for multiple testing. Also, differences in the populations studied and the stages of the disease examined between studies makes direct comparisons difficult. In our study, none of the associations mentioned above remained significant after a correction for multiple testing. Thus, comparisons between patients and controls, even using the relatively large sample size of our study, appear insufficient to unambiguously establish a contribution of HLA to the development of cervical cancer in situ.
Since infection by HPV is considered a necessary etiological factor in cervical cancer, it is natural to stratify the patients and controls on HPV infection status. When comparing HPV16-positive cases and controls, both DRB1*0701 and DQB1*0402 occur at a higher frequency among cases, although this difference does not hold after correction for multiple testing. A similar trend has been seen for DRB1*0701 in studies of Dutch CIN patients.12, 13 However, for the DRB1*1501 and DQB1*0602 alleles, we observe a significant association that is maintained after the correction for multiple testing. These 2 alleles are in strong linkage disequilibrium; in our data 95% of DRB1*1501 carriers also have the DQB1*0602 allele and 93.5% of the DQB1*0602 carriers have the DRB1*1501 allele. Carriers of the DRB1*1501/DQB1*0602 alleles have been reported to confer an increased risk of developing invasive cervical cancer in Hispanic10 and Norwegian patients11 and in Swedish patients to develop CINII/III.24 Additional studies have shown a similar trend but with higher p-values (around 0.05 or higher).8, 13, 26–28 Only one study has shown a contrary result; a decreased risk for high-grade squamous intraepithelial lesions (HGSIL) for DRB1*1501/DQB1*0602 carriers was found in a U.S. population.29 The same study also reported an increased (although not statistically significant) frequency of these 2 alleles in women with low-grade squamous intraepithelial lesions (LGSIL) and/or HPV16 infection. Since the p-values in their study were not corrected for multiple testing, it is difficult to evaluate these findings. Taken together, alleles on the DRB1*1501/DQB1*0602 haplotype appear to affect the risk of developing both HPV16-positive, low-grade dysplasias, dysplasias with increasing grades of CIN, cervical cancer in situ and invasive cervical cancer.
To determine whether the association of the DQB1*0602/DRB1*1501 haplotype with cancer in situ is primarily due to an increased risk of infection by HPV16, we compared HPV16-infected women with the group of uninfected women, irrespective of case-control status. A significant difference was found in carrier frequencies between these 2 groups. Since there was no interaction with cancer status (the effect was similar in both cases and controls), we conclude that DQB1*0602/DRB1*1501 carriers have an increased risk of becoming HPV16 infected. These results imply that the HLA alleles are associated primarily with infection by HPV. Since the DQB1*0602/DRB1*1501 haplotype is associated with different stages in the development of cancer, from low-grade dysplasia to invasive cervical cancer, it appears likely that the HLA haplotype affects whether an infection becomes persistent or clears spontaneously. To study the importance of these HLA alleles on infection persistence, we compared women with a capacity to clear an HPV infection within a more limited time to women carrying more long-term infections, irrespective of case-control status. We applied a simple model for classification conferred by the limitations of study design, as described in Results. However, the model used appears to describe the difference in infection history between cases and controls well, since only 1 control is present among the women with long-term infections. The increased frequency of the DRB1*1501 and DQB1*0602 alleles among women with long-term infections supports the notion that these alleles, or alleles at nearby linked loci, affect whether a woman is able to clear an infection or is prone to carry the virus for more extended time periods. Persistent infection has previously been shown to be associated with a higher risk of cervical cancer development,30, 31 presumably by increasing the risk of cell transformation.
As the MHC region contain a number of closely linked genes, associations with individual genes may be secondary to alleles at neighbouring loci in genetic disequilibrium. For instance, the class I HLA-B7 allele, which is in linkage disequilibrium with the DRB1*1501 and DQB1*0602 alleles, has shown an association with cervical cancer and been proposed to be involved in the ability of certain HPV16 strains to evade immune surveillance.32 Also, a combination of a tumour necrosis factor allele (TNFa-11) and the DQB1*0602 allele has been reported to be associated with the development of CINII/III in HPV16-positive patients from Northern Sweden.33 Further, different combinations of alleles at several HLA loci may contribute to the association. For instance, certain susceptibility profiles, including different HLA alleles, have been identified that increase the risk of developing AIDS.34–36 If a similar situation exists for cervical cancer, the DRB1*1501 and DQB1*0602 alleles may represent only 1 of several susceptibility profiles. Clearly, complete HLA haplotype data from large studies is required in order to assign the observed association with cervical cancer in situ to specific loci and alleles.
The mechanism by which HLA alleles affect infection persistence and development of cervical cancer is not known. An HPV infection is known to cause an immunologic response, as illustrated by the presence of HPV-specific antibodies in the sera of individuals with known history of infection. It is not known whether viral clearance is dependent on the cell-mediated immune response, but the high frequency of cervical HPV infections and lesions in immune-deficient women (such as HIV patients) suggests that the cellular immune response is important in viral clearance. It appears likely that individuals vary in their ability to present viral epitopes to the T-cell system. Carriers of certain HLA alleles, such as those found to be associated with cervical cancer in situ in our study, may be less capable of displaying viral peptides to circulating T cells. Such a mechanism would have consequences on the potential to produce prophylactic and therapeutic vaccines against HPV. The therapeutic vaccines often include synthetic parts of E6 and/or E7 peptides.37 Our results suggest that carriers of some HLA alleles have a reduced efficiency in presenting viral peptides to the immune system. This could imply that vaccines might not be as effective for women carrying the particular HLA alleles most strongly associated with risk of HPV infection.
In conclusion, our results show that HLA alleles affect the risk of an HPV infection and probably also the risk of developing a long-term HPV infection. As a consequence, carriers of these alleles are at higher risk of developing cervical cancer in situ. Although a number of alleles have been reported to be associated with an increased risk of developing cervical cancer, only the combination of DRB1*1501/DQB1*0602 alleles have an effect that can be unambiguously demonstrated. We have previously shown that women developing cervical cancer in situ have higher titer of HPV16 already a number of years prior to the development of cytological changes.17 This indicates that the initial immune response to an HPV infection can affect the number of viral copies and that this may be crucial for the risk of cell transformation and later the development of cervical cancer. Further studies will have to demonstrate whether the association between these HLA alleles and HPV16 infection in cervical cancer in situ is related to a reduced ability to control replication of the HPV genome.
We thank Dr. J. Terwilliger for constructive ideas about the statistics, Dr. P. Magnusson and B.Sc. M. Ingman for comments on the manuscript and Dr. H. Erlich for generous support during the HLA typing.
- 11An increased risk of cervical intra-epithelial neoplasia grade II-III among human papillomavirus positive patients with the HLA-DQA1*0102-DQB1*0602 haplotype: a population-based case-control study of Norwegian women. Int J Cancer 1998;76: 19–24., , , et al.