Association of HLA diversity with the risk of 25 cancers in the UK Biobank

Summary

Background

The human leukocyte antigen (HLA) is a highly polymorphic region, and HLA diversity may play a role in presenting tumour-associated peptides and inducing immune responses. However, the effect of HLA diversity on cancers has not been fully assessed. We aimed to explore the role of HLA diversity on cancer development.

Methods

A pan-cancer analysis was performed to evaluate the effect of HLA diversity, measured by HLA heterozygosity and HLA evolutionary divergence (HED), on the susceptibility of 25 cancers in the UK Biobank.

Findings

We observed that the diversity of HLA class II locus was associated with a lower risk of lung cancer (OR_hetero = 0.94, 95% CI = 0.90–0.97, P = 1.29 × 10⁻⁴) and head and neck cancer (OR_hetero = 0.91, 95% CI = 0.86–0.96, P = 1.56 × 10⁻³). Besides, a lower risk of non-Hodgkin lymphoma was associated with an increased diversity of HLA class I (OR_hetero = 0.92, 95% CI = 0.87–0.98, P = 8.38 × 10⁻³) and class II locus (OR_hetero = 0.89, 95% CI = 0.86–0.92, P = 1.65 × 10⁻¹⁰). A lower risk of Hodgkin lymphoma was associated with the HLA class I diversity (OR_hetero = 0.85, 95% CI = 0.75–0.96, P = 0.011). The protective effect of HLA diversity was mainly observed in pathological subtypes with higher tumour mutation burden, such as lung squamous cell carcinoma (P = 9.39 × 10⁻³) and diffuse large B cell lymphoma (P_{class I} = 4.12 × 10⁻⁴; P_{class Ⅱ} = 4.71 × 10⁻⁵), as well as the smoking subgroups of lung cancer (P = 7.45 × 10⁻⁵) and head and neck cancer (P = 4.55 × 10⁻³).

Interpretation

We provided a systematic insight into the effect of HLA diversity on cancers, which might help to understand the etiological role of HLA on cancer development.

Funding

This study was supported by grants from the National Natural Science Foundation of China (82273705, 82003520); the Basic and Applied Basic Research Foundation of Guangdong Province, China (2021B1515420007); the Science and Technology Planning Project of Guangzhou, China (201804020094); Sino-Sweden Joint Research Programme (81861138006); the National Natural Science Foundation of China (81973131, 81903395, 81803319, 81802708).

Research in context.

Evidence before this study

The human leukocyte antigen (HLA) plays a critical role in antigen recognition and immune response, which is supported by the structural basis of an impressive degree of polymorphisms. The diversity of HLA, measured by HLA heterozygosity and HLA evolutionary divergence (HED), has been associated with the susceptibility to many diseases, especially those with infectious or autoimmune etiologies. The HLA diversity has also been associated with the susceptibility to cancer with infectious or autoimmune etiologies. Individuals who are heterozygous at the HLA locus have a lower risk of non-Hodgkin lymphoma and hepatitis B virus-associated hepatocellular carcinoma. However, associations between HLA diversity and the susceptibility of many other cancers remain unclear.

Added value of this study

This study included 25 types of cancers from the UK Biobank. A protective effect of increased HLA diversity was observed in the development of lung cancer, non-Hodgkin lymphoma and Hodgkin lymphoma. Subgroup analysis showed that the protective effect of HLA diversity was mainly observed in pathological subtypes with higher tumour mutation burden, such as lung squamous cell carcinoma and diffuse large B cell lymphoma, as well as the smoking subgroups of lung cancer.

Implications of all the available evidence

Our findings suggested an etiological role of HLA diversity in cancer development, especially in cancers with a high mutational load and infectious or autoimmune etiologies, which might add complementary evidence for understanding the etiological role of HLA on cancer development.

Introduction

The human leukocyte antigen (HLA) plays a critical role in antigen recognition and immune response,¹ which is supported by the structural basis of an impressive degree of polymorphisms.² Encoded by different alleles, HLA molecules are characterized by variable amino acid sequences of peptide-binding grooves, which express diverse binding affinity of antigen peptides.^3,4

The diversity of HLA, usually measured qualitatively by HLA heterozygosity, has been associated with susceptibility to diseases with infectious or autoimmune etiologies.^5,6 For example, HLA heterozygotes who carry two different alleles at the HLA locus were reported to have a lower risk of virus infection and infectious disease progression, such as human immunodeficiency virus (HIV) and hepatitis B virus (HBV).7, 8, 9 HLA heterozygosity has also been associated with cancer susceptibility. Heterozygous individuals at the HLA locus have a lower risk of non-Hodgkin lymphoma and HBV-associated hepatocellular carcinoma.^7,10,11 In addition to HLA heterozygosity, the diversity of HLA could also be measured quantitatively by HLA evolutionary divergence (HED).¹² HED is calculated for heterozygous individuals carrying two different alleles at the same HLA locus according to the magnitude of sequence divergence between alleles. HED was reported to be positively correlated with the number of peptides bound by HLA alleles, which may contribute to the heterogeneity of immune response.12, 13, 14

The heterozygote advantage and the divergent allele advantage were proposed as important mechanisms in maintaining HLA diversity.^15,16 Compared with inherited homozygous HLA alleles or low HED, individuals with more diverse HLA alleles were hypothesized to present a broader range of tumour-associated peptides to T cells, which may increase the probability of eliciting an enhanced immune response and thus play roles in cancer development.^17,18 However, the associations between HLA diversity and the susceptibility of many cancers remain unclear. To better understand the role of HLA diversity on cancer development, a pan-cancer analysis using the data from the UK Biobank was performed to systematically investigate the associations of HLA diversity with the susceptibility of 25 types of cancers.

Methods

Study populations

The UK Biobank is a biomedical database with over 500,000 individuals recruited between 2006 and 2010. In the baseline assessment, a large amount of phenotypic and genotypic information was collected, and long-term follow-up was carried out for various health statuses.¹⁹

In our study, 411,615 individuals were included after the following steps of genotype quality control. Firstly, 15,223 individuals were excluded for missing genotyping data, 837 individuals were excluded for discordance between genetic and reported gender or aneuploidy of the sex chromosome, and 378 individuals were excluded for the poor quality of genotyping data (high missing rate or abnormal heterozygosity). For each pair of subjects implied as third-degree or closer relatives with a kinship coefficient exceeding 0.0442, only the one with the highest genotype call rate was included, and 74,452 individuals were removed from further analysis. A detailed subject selection flowchart is shown in Supplementary Fig. S1.

Cancer data

Cancer diagnosis data from the UK Biobank was acquired from the hospital, national cancer registries, and death records. Both prevalent cases diagnosed before recruitment and incident cases diagnosed after recruitment were included. The cancer classification was based on the 9th Revision of International Classification of Diseases (ICD-9) and the 10th Revision of International Classification of Diseases (ICD-10).

The 25 most common site-specific cancers in the UK Biobank were investigated. 69,671 individuals diagnosed with benign tumours, carcinoma in situ, or non-melanoma skin cancer, and 3557 individuals diagnosed with other cancers were excluded. Finally, 48,574 cancer cases and 289,813 controls were investigated. Meanwhile, the 25 cancers were grouped into seven categories according to their original organs or systems. Detailed information about ICD codes, case numbers, and classification for each cancer is listed in Supplementary Table S1. In the subsequent analysis, non-Hodgkin lymphoma and Hodgkin lymphoma were further classified into different pathological subtypes according to the third edition International Classification of Diseases for Oncology (ICD-O-3), and lung cancer was classified according to the World Health Organization Classification of Lung Tumors.²⁰

HLA genotyping

Genotyping of samples was performed using two similar SNP genotyping arrays sharing 95% of marker content, the Affymetrix UK Biobank Axiom array for about 450,000 samples and the Affymetrix UK BiLEVE Axiom array for about 50,000 samples. HLA alleles (at four-digit resolution) for the classical class I and class II locus were imputed by the HLA∗IMP:02 algorithm, and a multi-population reference panel was used.²¹ A total of 211 HLA class I alleles, of which 53 are in the HLA-A locus, 126 are in the HLA-B locus, 32 are in the HLA-C locus, and a total of 135 HLA class II alleles, of which 59 are in the HLA-DRB1 locus, 18 are in the HLA-DQB1 locus, 14 are in the HLA-DQA1 locus, 36 are in the HLA-DPB1 locus, 8 are in the HLA-DPA1 locus were included. A posterior probability cutoff of 0.5 was used to filter the imputed HLA alleles with poor quality. For each HLA locus, the successful genotyping rate is over 97.7%. Individuals with two different alleles at the four-digit resolution of a specific locus were defined as heterozygotes. Assuming each HLA class I and HLA class II locus has the same contribution to antigen presentation,¹³ the total number of heterozygotes at the HLA class I and class II locus was counted and analysed in this study.

HED calculation

To evaluate the evolutionary divergence of alleles, we used the method described in the previous study to calculate HED (12). We acquired the allele sequences in the IMGT/HLA database.² The exon 2 and exon 3 of the HLA class I alleles and exon 2 of the HLA class II alleles, which encode the peptide-binding groove of HLA molecules, were extracted according to the annotation information released from the Ensembl.²² The HED was calculated according to the sequence divergence of peptide-binding domains between two alleles using the Grantham distance metric, as described by a previous study.¹² The Grantham distance is a comprehensive index for measuring the differences of paired amino acid sequences, taking into account the biochemical composition, polarity, and volume. One unit of HED is estimated to represent a difference of 2.2 amino acids.⁹

Statistics analysis

The effects of HLA heterozygosity and HED on cancer risk were estimated using a logistic regression model. Sex, age at recruitment, and genotyping array were adjusted in the model. Additionally, considering that the top two PCs in the UK Biobank could separate individuals with sub-Saharan African ancestry, European, and east-Asian ancestry,²¹ and usually the first 5–20 PCs would be adjusted for stratification,²³ we selected the top five PCs in the adjustment. For female cancers (breast cancer, cervical cancer, endometrium, ovary cancer), covariates including hormone replacement therapy use (never, ever), oral contraceptive pill use (never, ever), parity (nulliparous, 1–2, ≥3), age at first birth (<25, 25–29, ≥30), age at menarche (<12, 12–13, ≥14), menopause status (premenopausal, postmenopausal) were additionally adjusted. We additionally estimated the associations using generalized estimating equations (GEEs) with non-robust SEs to check the stability of the results.^24,25 The dose–response effect of the number of heterozygotes on cancer risks was evaluated by the Cochran-Armitage trend test. Due to the limitation of sample size, participants with 0 or 1 heterozygote were merged into the reference group. For multiple tests, the adjusted p-value (P_adj) was calculated using the Benjamini-Hochberg method controlling the false discovery rate (FDR). P_adj < 0.05 was defined as a significant association, and P_adj < 0.1 was defined as a suggestive association to improve statistical power.

Subgroup analysis was performed to evaluate the associations between HLA diversity and the risk of cancers among different pathological subtypes and smoking status (never, ever). A subgroup analysis of participants with their ethnic background as white was performed.

All statistical analyses were performed using R version 4.0.4. The Strengthening the Reporting of Genetic Association Studies (STREGA) checklist was followed.²⁶

Ethics statement

Our data resource was the UK Biobank, which obtained ethics approval from the North West Multi-Centre Research Ethics Committee (MREC). We have received ethics approval from the UK Biobank (ID: 58,450). All participants had written consent to participate and agreed to have their health information followed.

Role of funders

The funders of this study had no role in study design, data collection, data analyses, interpretation, or in the drafting of the manuscript.

Results

Association between HLA diversity and risk of solid tumours

We investigated the association between HLA diversity and the risk of 21 types of solid tumours, which were divided into six major cancer categories. Individuals carrying two different alleles for each HLA locus were defined as heterozygotes. HED was calculated as a quantitative description of the sequence divergence between the two alleles.^12,13 The characteristics of the heterozygosity and HED of each HLA locus are shown in Supplementary Table S2. The number of heterozygotes was summed up, and the average of the HED was calculated for the HLA class I and class II locus, respectively. Considering the similar result between the logistic regression analysis and GEE with non-robust SEs (Supplementary Table S3), the results of logistic regression analysis were shown.

Lower risk of lung cancer was observed with the increased number of heterozygotes at the HLA class II locus with an OR of 0.94 (95% CI = 0.90–0.97, P = 1.29 × 10⁻⁴, P_adj = 1.99 × 10⁻³) and a significant effect trend was shown (P_trend = 4.10 × 10⁻³), whereas no significant association was observed in the HLA class I locus (Fig. 1, Fig. 2; Table 1). Compared with individuals carrying homozygotes or one heterozygote at the HLA class II locus, the ORs are 0.85 (95% CI = 0.69–1.05), 0.70 (95% CI = 0.59–0.83), 0.71 (95% CI = 0.60–0.83), 0.71 (95% CI = 0.60–0.84) for individuals carrying two to five heterozygotes respectively. The heterozygosity of HLA-DRB1 (OR = 0.79, 95% CI = 0.69–0.89, P = 2.34 × 10⁻⁴), HLA-DQB1 (OR = 0.77, 95% CI = 0.69–0.87, P = 1.18 × 10⁻⁵), and HLA-DQA1 (OR = 0.85, 95% CI = 0.77–0.95, P = 2.58 × 10⁻³) (Supplementary Table S5) was significantly associated with a lower risk of lung cancer. A lower risk of lung cancer was also observed for individuals with higher HED at the HLA class II locus (OR = 0.98, 95% CI = 0.97–1.00, P = 5.84 × 10⁻³, P_adj = 0.072) (Fig. 2).

Fig. 1.

Associations between HLA heterozygosity (left panel) and HED (right panel) at class I locus and the risk of 25 cancers.

Fig. 2.

Associations between HLA heterozygosity (left panel) and HED (right panel) at class II locus and the risk of 25 cancers.

Table 1.

The trend test for the associations between the number of heterozygotes at the HLA locus and the risk of cancers.^a

No. of heterozygote	Control	Lung cancer			Cervical cancer			Non-Hodgkin lymphoma			Hodgkin lymphoma
	N (%)	N (%)	OR (95% CI)	P	N (%)	OR (95% CI)	P	N (%)	OR (95% CI)	P	N (%)	OR (95% CI)	P
Class I locus
0 + 1	20,147 (7.1%)	182 (8.0%)	1.00 (ref)	–	49 (6.0%)	1.00 (ref)	–	168 (8.4%)	1.00 (ref)	–	34 (8.8%)	1.00 (ref)	–
2	43,550 (15.5%)	365 (16.1%)	0.96 (0.80–1.15)	0.649	109 (13.2%)	1.82 (0.92–3.60)	0.083	334 (16.6%)	0.92 (0.76–1.11)	0.384	76 (19.6%)	1.04 (0.69–1.55)	0.866
3	217,952 (77.4%)	1723 (75.9%)	0.90 (0.77–1.05)	0.189	664 (80.8%)	2.38 (1.29–4.41)	6.00 × 10−3	1504 (75.0%)	0.83 (0.71–0.97)	0.021	277 (71.6%)	0.75 (0.53–1.08)	0.122
Ptrendb	–	–	–	0.061	–	–	0.027	–	–	6.30 × 10−3	–	–	0.014
Class II locus
0 + 1	17,108 (6.2%)	191 (8.6%)	1.00 (ref)	–	59 (7.3%)	1.00 (ref)		178 (9.0%)	1.00 (ref)	–	36 (9.4%)	1.00 (ref)	–
2	16,710 (6.0%)	156 (7.0%)	0.85 (0.69–1.05)	0.135	45 (5.5%)	1.03 (0.53–2.00)	0.919	140 (7.1%)	0.81 (0.65–1.01)	0.059	30 (7.8%)	0.86 (0.53–1.39)	0.528
3	64,196 (23.3%)	500 (22.5%)	0.70 (0.59–0.83)	3.64 × 10−5	201 (24.7%)	1.00 (0.60–1.67)	0.998	501 (25.3%)	0.75 (0.63–0.89)	8.73 × 10−4	81 (21.1%)	0.60 (0.40–0.89)	0.011
4	105,090 (38.1%)	818 (36.8%)	0.71 (0.60–0.83)	2.11 × 10−5	300 (36.9%)	0.95 (0.58–1.57)	0.852	705 (35.7%)	0.64 (0.55–0.76)	1.77 × 10−7	146 (38.0%)	0.66 (0.46–0.95)	0.025
5	72,807 (26.4%)	558 (25.1%)	0.71 (0.60–0.84)	5.29 × 10−5	208 (25.6%)	0.99 (0.59–1.65)	0.970	453 (22.9%)	0.61 (0.51–0.72)	2.52 × 10−8	91 (23.7%)	0.61 (0.41–0.89)	0.011
Ptrend	–	–	–	4.10 × 10−3	–	–	0.427	–	–	2.20 × 10−16	–	–	0.091

The significant associations (P-value < 0.05) were indicated in bold.

^aOdds ratios (ORs), and 95% confidence intervals (95% CIs) were calculated using logistic regression models. All models were adjusted for sex, age at recruitment, genotyping array, and the top five principal components. Additional covariates were adjusted for cervical cancer, including hormone replacement therapy use, oral contraceptive pill use, parity, age at first birth, age at menarche, and menopause status.

^bP values for trend were assessed using the Cochran-Armitage trend test.

Interestingly, a risk effect of heterozygotes was observed in cervical cancer (OR = 1.40, 95% CI = 1.13–1.73, P = 1.72 × 10⁻³, P_adj = 0.018) (Fig. 1), and individuals carrying two or three heterozygotes exhibit higher risks of cervical cancer with ORs of 1.82 (95% CI = 0.92–3.60) and 2.38 (95% CI = 1.29–4.41, P_trend = 0.027) respectively (Table 1). A consistent risk effect of higher HED was observed in cervical cancer (OR = 1.08, 95% CI = 1.02–1.15, P = 4.71 × 10⁻³, P_adj = 0.072). Further analysis of each HLA class I locus identified a significant association between the risk of cervical cancer and the diversity of HLA-A and HLA-B (Supplementary Tables S5 and S6).

We found a protective effect of heterozygote at the HLA class II locus (OR = 0.91, 95% CI = 0.86–0.96, P = 1.56 × 10⁻³, P_adj = 0.018) with a significant trend in effect sizes (P_trend = 0.011) on head and neck cancer. A consistent but not significant protective effect of higher HED (OR = 0.98, 95% CI = 0.96–1.00, P = 0.029, P_adj = 0.140) was also observed (Fig. 2; Supplementary Table S4). Among the two cancers in the head and neck cancer, a decreased oral cancer risk (OR = 0.91, 95% CI = 0.84–0.98, P = 8.21 × 10⁻³, P_adj = 0.065) was observed with the number of heterozygotes of HLA class II locus increased.

In addition, a decreased prostate cancer risk (OR = 0.99, 95% CI = 0.99–1.00, P = 8.87 × 10⁻³, P_adj = 0.075) was observed in the HED analysis of the HLA class II locus (Fig. 2). No significant association of HLA heterozygosity and HED with the risk of other solid tumours was observed.

Association between HLA diversity and risk of lymphatic and hematopoietic cancers

We observed that an increasing number of heterozygotes at the HLA class I locus (OR = 0.92, 95% CI = 0.88–0.96, P = 8.71 × 10⁻⁵, P_adj = 1.80 × 10⁻³) (Fig. 1) and HLA class II locus (OR = 0.93, 95% CI = 0.91–0.95, P = 1.33 × 10⁻⁸, P_adj = 4.12 × 10⁻⁷) (Fig. 2) was associated with lower risk of lymphatic and hematopoietic cancers. Meanwhile, the increase of HED at the HLA class I locus (OR = 0.98, 95% CI = 0.96–0.99, P = 1.63 × 10⁻³, P_adj = 0.034) (Fig. 1) and class II locus (OR = 0.98, 95% CI = 0.98–0.99, P = 1.06 × 10⁻⁴, P_adj = 6.56 × 10⁻³) (Fig. 2) were associated with lower risk of lymphatic and hematopoietic cancers.

Among the four types of cancers in the lymphatic and hematopoietic cancer category, an increased number of heterozygotes at the HLA class I locus was associated with a lower risk of non-Hodgkin lymphoma with OR of 0.92 (95% CI = 0.87–0.98, P = 8.38 × 10⁻³, P_adj = 0.065) (Fig. 1). This finding is in line with a previous study.¹⁰ At the HLA class II locus, an increased number of heterozygotes was also associated with a lower risk of non-Hodgkin lymphoma with an OR of 0.89 (95% CI = 0.86–0.92, P = 1.65 × 10⁻¹⁰, P_adj = 1.05 × 10⁻⁸) (Fig. 2). Significant trends in effect sizes were shown in the HLA class I locus (P_trend = 6.30 × 10⁻³) and HLA class II locus (P_trend = 2.20 × 10⁻¹⁶) (Table 1). Consistently, higher HED at class I and class II locus were associated with a reduced risk of non-Hodgkin lymphoma with ORs of 0.97 (95% CI = 0.96–0.99, P = 0.010, P_adj = 0.075) (Fig. 1) and 0.98 (95% CI = 0.97–0.99, P = 5.63 × 10⁻⁴, P_adj = 0.017) (Fig. 2). Analysis of the heterozygosity of each HLA locus identified significant associations between the risk of non-Hodgkin lymphoma and the heterozygosity of HLA-A, HLA-C, HLA-DRB1, HLA-DQB1, HLA-DQA1, HLA-DPB1, HLA-DPA1. Similar associations between HED of these HLA locus and the risk of non-Hodgkin lymphoma were identified except for the HLA-C locus (Supplementary Tables S5 and S6).

In addition to non-Hodgkin lymphoma, the heterozygosity (OR = 0.85, 95% CI = 0.75–0.96, P = 0.011, P_adj = 0.079) and HED (OR = 0.94, 95% CI = 0.90–0.98, P = 7.49 × 10⁻³, P_adj = 0.075) of HLA class I locus (Fig. 1) were also associated with a lower risk of Hodgkin lymphoma. A significant trend in effect size was shown with the increased number of heterozygotes at the HLA class I locus (P_trend = 0.014) (Table 1). The subsequent analysis at each HLA class I locus showed a statistically significant association for the heterozygosity and HED of the HLA-B locus (Supplementary Tables S5 and S6). The heterozygosity of the HLA class II locus was also associated with a lower risk of Hodgkin lymphoma (OR = 0.91, 95% CI = 0.84–0.98, P = 0.015, P_adj = 0.092) (Fig. 2), while no significant or suggestive association was observed in the HED analysis. Protective effects of heterozygosity and HED at HLA class I and class II locus were also observed for leukaemia. However, these associations did not reach the significance level after FDR correction (Fig. 1, Fig. 2).

The effect of HLA diversity among different pathological subtypes

Subgroup analysis was conducted to explore the effect of HLA diversity among pathological subtypes. For lung cancer, a decreased risk of lung squamous cell carcinoma was associated with the heterozygosity of HLA class II locus (OR = 0.90, 95% CI = 0.83–0.97, P = 9.39 × 10⁻³) (Table 2), with ORs of 0.84 (95% CI = 0.52–1.35), 0.58 (95% CI = 0.40–0.86), 0.59 (95% CI = 0.42–0.85), and 0.60 (95% CI = 0.41–0.88) for carriers of two and five heterozygotes (P_trend = 0.048) (Supplementary Table S7). The effect of HED at the HLA class II locus was consistent with HLA heterozygosity in lung squamous cell carcinoma (Table 2). In addition, a lower risk of lung adenocarcinoma was also found with the increased number of heterozygotes at the HLA class II locus (OR = 0.93, 95% CI = 0.88–0.99, P = 0.022), while no statistical significance of HED analysis was observed. Besides, no significant association was found between HLA heterozygosity or HED and the risk of lung neuroendocrine carcinoma.

Table 2.

Associations between the HLA diversity and the risk of different tumour pathological subtypes.^a

Cancer	Pathological subtype	N	HLA class I locus				HLA class II locus
			No. of heterozygoteb		HED		No. of heterozygote		HED
			OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P
Lung cancer	Squamous cell carcinoma	400	1.00 (0.87–1.15)	0.997	0.97 (0.93–1.02)	0.200	0.90 (0.83–0.97)	9.39 × 10−3	0.96 (0.94–0.99)	5.60 × 10−3
	Adenocarcinoma	732	0.96 (0.87–1.07)	0.495	0.99 (0.96–1.02)	0.596	0.93 (0.88–0.99)	0.022	1.00 (0.98–1.02)	0.759
	Neuroendocrine carcinoma	343	0.93 (0.80–1.07)	0.308	0.99 (0.94–1.04)	0.621	0.93 (0.85–1.01)	0.104	0.97 (0.94–1.00)	0.071
non-Hodgkin lymphoma	Diffuse large B cell lymphoma	540	0.82 (0.74–0.92)	4.12 × 10−4	0.95 (0.92–0.99)	0.014	0.87 (0.81–0.93)	4.71 × 10−5	0.97 (0.95–1.00)	0.022
	Follicular lymphoma	385	1.08 (0.92–1.26)	0.345	1.02 (0.97–1.07)	0.369	0.89 (0.82–0.97)	5.52 × 10−3	0.98 (0.96–1.01)	0.228
	T-cell lymphoma	142	0.96 (0.76–1.21)	0.721	1.01 (0.94–1.09)	0.762	0.99 (0.86–1.14)	0.841	0.99 (0.95–1.04)	0.811
Hodgkin lymphoma	Mixed cellularity	62	0.74 (0.55–1.01)	0.061	0.92 (0.82–1.03)	0.153	0.78 (0.64–0.95)	0.014	0.96 (0.89–1.03)	0.290
	Nodular sclerosis	119	1.03 (0.76–1.39)	0.867	1.01 (0.92–1.11)	0.856	1.02 (0.85–1.22)	0.828	1.01 (0.96–1.07)	0.675

The significant associations (P-value < 0.05) were indicated in bold.

^aOdds ratios (ORs), and 95% confidence intervals (95% CIs) were calculated using logistic regression models. All models were adjusted for sex, age at recruitment, genotyping array, and the top five principal components.

^bThe number of heterozygotes at HLA locus was taken as a continuous variable in the logistic regression models.

Among the three pathological subtypes of non-Hodgkin lymphoma, the tumour mutation burden was highest in diffuse large B cell lymphoma (DLBCL), followed by follicular lymphoma (FL) and T-cell lymphoma (TCL).²⁷ Lower risk of DLBCL was observed with the increased number of heterozygotes at the HLA class I (OR = 0.82, 95% CI = 0.74–0.92, P = 4.12 × 10⁻⁴) and class II locus (OR = 0.87, 95% CI = 0.81–0.93, P = 4.71 × 10⁻⁵) (Table 2), and significant trends in effect sizes were shown in the HLA class I locus (P_trend = 7.00 × 10⁻⁴) and HLA class II locus (P_trend = 4.50 × 10⁻³) (Supplementary Table S8). A similar protective effect was observed in the analysis between DLBCL and HED. The heterozygosity of the HLA class II locus showed protective effects on FL (OR = 0.89, 95% CI = 0.82–0.97, P = 5.52 × 10⁻³), but no significant association was observed between HED and the risk of FL. In TCL, no significantly statistical association was found between HLA heterozygosity or HED and cancer risk.

In the analysis of Hodgkin lymphoma, we only observed a significant association of the heterozygosity of the HLA class II locus with the risk of mixed cellularity Hodgkin lymphoma (OR = 0.78, 95% CI = 0.64–0.95, P = 0.014) (Table 2; Supplementary Table S9), which was reported to have a high prevalence of Epstein–Barr virus infection.²⁸ No significant association was identified between HLA heterozygosity and nodular sclerosis Hodgkin lymphoma.

The effect of HLA diversity among different smoking status

Smoking is considered a strong oncogenic driver in the progression of cancer and is the leading risk factor for lung cancer and head and neck cancer. Compared with non-smokers, the somatic mutation burden was significantly elevated in smokers,²⁹ which might create antigenic peptides with immunogenicity. To explore the HLA diversity effect on the cancer risk among individuals with different smoking statuses, we performed a subgroup analysis among smokers and non-smokers (Table 3; Supplementary Table S10). In smoker subgroup, individuals with an increased number of heterozygotes or higher HED at the HLA class II locus had a significantly lower risk of lung cancer with ORs of 0.93 (95% CI = 0.90–0.96, P = 7.45 × 10⁻⁵) and 0.98 (95% CI = 0.97–0.99, P = 2.17 × 10⁻³) respectively. Similarly, individuals with an increased number of heterozygotes (OR = 0.91, 95% CI = 0.85–0.97, P = 4.55 × 10⁻³) and higher HED (OR = 0.98, 95% CI = 0.95–1.00, P = 0.039) at HLA class II locus also had a lower risk of head and neck cancers. However, no significant association was observed between HLA heterozygosity and HED and the risk of lung cancer and head and neck cancer in the non-smoker subgroup.

Table 3.

Associations between the HLA diversity and the risk of cancer in subgroups stratified by smoking status.^a

Cancer	Smoking status	N	HLA class I locus				HLA class II locus
			No. of heterozygoteb		HED		No. of heterozygote		HED
			OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P
Lung cancer	Smokers	2061	0.94 (0.89–1.00)	0.067	0.98 (0.96–1.00)	0.025	0.93 (0.90–0.96)	7.45 × 10−5	0.98 (0.97–0.99)	2.17 × 10−3
	Non-smokers	235	0.97 (0.81–1.17)	0.753	0.99 (0.94–1.05)	0.828	0.97 (0.87–1.08)	0.549	1.01 (0.97–1.04)	0.756
Head and neck cancer	Smokers	601	0.93 (0.83–1.04)	0.204	1.01 (0.97–1.04)	0.790	0.91 (0.85–0.97)	4.55 × 10−3	0.98 (0.95–1.00)	0.039
	Non-smokers	156	0.91 (0.74–1.12)	0.377	0.95 (0.89–1.02)	0.165	0.89 (0.79–1.02)	0.084	0.98 (0.94–1.02)	0.326

The significant associations (P-value < 0.05) were indicated in bold.

^aOdds ratios (ORs), and 95% confidence intervals (95% CIs) were calculated using logistic regression models. All models were adjusted for sex, age at recruitment, genotyping array, and the top five principal components.

^bThe number of heterozygotes at HLA locus was taken as a continuous variable in the logistic regression models.

Discussion

In this study, we systematically investigated the effect of HLA diversity, measured by heterozygosity and HED, on the risk of 25 cancers in the UK Biobank, which included 48,574 cases and 289,813 controls. We demonstrated that HLA heterozygosity and higher HED were associated with reduced risks of lung cancer, non-Hodgkin lymphoma, Hodgkin lymphoma, and a higher risk of cervical cancer. The heterogeneity test results (I²)³⁰ showed that the heterogeneity between the All-population test and the White-specific test was 0% (Supplementary Tables S11 and S12). These results suggested that HLA diversity might be important in the development of cancers with autoimmune or infectious etiologies and a high mutation burden.

Among solid tumours, we observed a significant protective effect of HLA class II diversity on lung cancer and head and neck cancer. A previous study showed no association between HLA heterozygosity and the risk of 12 non-virus-associated solid tumours.³¹ The heterogeneity in study design might contribute to the different results. The result comparison of previous studies and this study was compared in Supplementary Table S13. Our results suggested that the breadth of the antigen-binding repertoire of HLA II molecules may be important to the progression of some solid tumours. In the subsequent analysis, we found that the protective effect of HLA diversity for those two cancers was observed in subgroups with higher tumour mutation burden, such as lung squamous cell carcinoma and smoking-associated cancers. With exposure to mutagens of cigarette smoking, a high mutation burden might be accumulated in smoking-associated cancers.^29,32,33 An increased abundance of mutations may promote the production of immunogenic antigenic peptides.³⁴ Our findings suggested that smokers with increased HLA diversity at the HLA class II locus might render broader mutant peptides and thus clear mutated cells more effectively than smokers with low HLA diversity.

For solid tumours with infectious etiology, the heterozygosity of HLA class II was reported to associate with a lower risk of HBV-associated hepatocellular carcinoma and chronic HBV infection.⁷ In our study, no significant protective effect of HLA diversity was observed on the risk of liver cancer. It was speculated that the protective effect of HLA heterozygosity on liver cancer might be mediated by tumour-associated virus infection and disease progression. Persistent infection with high-risk HPV is an essential factor in the development of cervical cancer. Interestingly, the diversity of HLA has a risk effect on cervical cancer. Wang et al. have also observed that individuals heterozygous for HLA-B or HLA-C showed a higher risk of cervical neoplasia, but this result was not statistically significant.³⁵ Considering the role of HLA diversity in the infection of tumour-associated virus, it appears necessary for further studies to investigate the relationship between HLA heterozygosity and HPV infection to decipher the risk effect of heterozygotes on cervical cancer.

A lower risk of non-Hodgkin lymphoma and Hodgkin lymphoma was associated with heterozygosity and higher HED of HLA locus. The heterozygosity of the HLA class I locus showed a protective effect on the risk of non-Hodgkin lymphoma, which has been observed in the previous study.¹⁰ In addition, we found that the HLA class II heterozygosity also confers a lower risk of non-Hodgkin lymphoma. The HLA class I molecules typically bind endogenous peptides to CD8+ T lymphocytes. The HLA class II molecules bind exogenous peptides to CD4+ T lymphocytes. These molecules are responsible for infection and autoimmune diseases.³⁶ These findings suggested that the protective effect of heterozygosity at HLA class I and II locus on non-Hodgkin lymphoma may be modulated by the recognition of broader antigens associated with pathogen infection and autoimmune disorders, which are in line with our current understanding of the risk factors for non-Hodgkin lymphoma, including the autoimmune conditions and virus infection.³⁷ Furthermore, we investigated the HLA diversity effect among different pathological subtypes. Our findings were consistent with previous research conducted on the European population.¹¹ The heterozygosity of the HLA class I and II locus showed protective effect on DLBCL. In contrast, only the heterozygosity of HLA class II locus was associated with the reduced risk of FL. The heterogeneity of tumour mutation burden between these subtypes might be related to the different association patterns.

We observed a protective effect of diversity at HLA class I locus on Hodgkin lymphoma. Although the association between HLA diversity and susceptibility to Hodgkin lymphoma has not been reported before, several studies have found a strong association of the risk of Hodgkin lymphoma with the HLA region at the single nucleotide (SNP) or allele level.³⁸ The HLA class I region has been mainly reported to be associated with Epstein–Barr virus (EBV) positive Hodgkin lymphoma. About 40% of Hodgkin lymphoma are EBV positive, and EBV infection in Hodgkin lymphoma is highly correlated with the pathological subtypes.²⁸ The EBV infection is present in 68% of mixed cellularity Hodgkin lymphoma, while only in 10%–20% of nodular sclerosis Hodgkin lymphoma. Coincidentally, the protective effect of heterozygotes was only observed in mixed cellularity Hodgkin lymphoma. We speculated that the protective association of HLA diversity with Hodgkin lymphoma might be attributed to a broader range of recognized EBV antigens and a more efficiently activated T-cell response to improve the control of EBV infection. It seems necessary to conduct analysis stratified by tumour EBV infection status. However, there is no information about tumour EBV status in the UK Biobank. Further study should be performed to investigate the effect of HLA diversity on Hodgkin's lymphoma risk stratified by EBV infection status.

For cancers that lack significant association, further studies are needed, especially for cancers with evidence associated with the HLA region or cancers with the heterogeneity of HLA associations in different subtypes. Such as breast cancer and liver cancer, further studies could be performed to investigate the effect of HLA diversity on the risk of cancer in subgroup stratified by different characteristics. The difference in virus infection status, tumour mutation burden, and tumour mutation status might lead to differences in the association of HLA diversity.

To the best of our knowledge, this study is the most systematic report to evaluate the effect of HLA diversity on cancer risk. An advantage of the study was the large sample size, including 48,574 cases and 289,813 controls, which involved most of the common cancers. The large sample size could ensure relatively well-powered association estimates. In addition, we comprehensively assessed the results of HLA heterozygosity and HED from the same population to estimate the effect of HLA diversity on cancer, which is essential for the stability and reliability of associations. However, our study lacked of the information of pathogen infection associated with cancer, which restricted us from exploring more deeply the HLA diversity effect on those infection-related cancers. The participants in our study were mainly comprised of European ancestry individuals. Further research is necessary to validate the HLA diversity effect in other ethnic groups. For cancers with small sample size, the results should be interpreted with caution and should be validated in studies with larger sample size.

Conclusion

In conclusion, the protective effect of increased HLA diversity was observed in cancers typical of autoimmune or infectious etiologies and high mutation burden, including non-Hodgkin lymphoma, Hodgkin lymphoma, and lung cancer. Our results provided a systematic insight into the effect of HLA diversity on cancers, which might add complementary evidence for understanding the etiological role of HLA on cancer development.

Contributors

J.W.H, conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing. W.Q.L, conceptualization, software, formal analysis, visualization, validation, methodology, writing-original draft, writing-review and editing. W.T.M, conceptualization, supervision, validation, methodology, writing-original draft, writing-review and editing. D.C.M, formal analysis, methodology. Z.W.L, formal analysis, methodology. H.Y.Q, data curation. X.W.Q, data curation. L. Y, data curation. Y.D.W, data curation. Z.M.Q, data curation. Jia W.H., Wang Q.L. and Wang T.M. directly accessed and verified the underlying data reported in the manuscript. All authors read and approved the final version of the manuscript.

Data sharing statement

Due to the privacy and sensitive information contained in the data, the data used for this study will not be made publicly available. Data is accessible for authors upon reasonable request and an approval of the UK Biobank (https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access).

Declaration of interests

The authors declare no potential conflicts of interest.