Classical celiac disease is more frequent with a double dose of HLA-DQB1*02: A systematic review with meta-analysis

Background and aims Experimental data suggest that the HLA-DQ2 gene dose has a strong quantitative effect on clinical outcomes and severity of celiac disease (CD). We aimed to conduct a meta-analysis with systematic review to investigate the association between HLA-DQB1*02 gene doses and the characteristics of CD. Methods We searched seven medical databases for studies discussing HLA-DQB1 gene dose in CD and various disease characteristics, such as clinical presentation, histology, age at diagnosis, and comorbidities. Odds ratios (OR, for categorical variables) and weighted mean differences (for age) were calculated to compare patients with a double dose of HLA-DQB1*02 versus those with single and zero doses. Heterogeneity was tested with I2-statistics and explored by study subgroups (children and adults). Results Twenty-four publications were eligible for meta-analysis. Classical CD was more frequent with a double versus single dose of the HLA-DQB1*02 allele (OR = 1.758, 95%CI: 1.148–2.692, I2 = 0.0%). In pediatric studies, gene dose effect was more prominent (OR = 2.082, 95%CI: 1.189–3.646, I2 = 0.0% and OR = 3.139, 95%CI: 1.142–8.630, I2 = 0.0% for the comparisons of double versus single and double versus zero dose, respectively). Atrophic histology was more prevalent with a double versus zero dose (OR = 2.626, CI: 1.060–6.505, I2 = 21.3%). We observed no gene dose effect regarding diarrhea, age at diagnosis, the severity of villous atrophy, and the association with type 1 diabetes mellitus. Conclusion A double dose of HLA-DQB1*02 gene seems to predispose patients to developing classical CD and villous atrophy. Risk stratification by HLA-DQB1*02 gene dose requires further clarification due to the limited available evidence.


Methods
We searched seven medical databases for studies discussing HLA-DQB1 gene dose in CD and various disease characteristics, such as clinical presentation, histology, age at diagnosis, and comorbidities. Odds ratios (OR, for categorical variables) and weighted mean differences (for age) were calculated to compare patients with a double dose of HLA-DQB1*02 versus those with single and zero doses. Heterogeneity was tested with I 2 -statistics and explored by study subgroups (children and adults).

Introduction
Celiac disease (CD) is an immune-mediated systemic disorder triggered by gluten that occurs in genetically susceptible individuals [1,2]. CD is characterized histologically by small intestinal mucosal damage, clinically by various intestinal and extraintestinal manifestations.
The presence of HLA-DQ2 or DQ8 is essential in the disease pathogenesis. T-lymphocytes recognize gliadin peptides presented by antigen presenting cells expressing DQ2 or DQ8 on cell surface, exclusively. Therefore, theoretically, either haplotype must be present in all CDpatients [3].
Since the DQ2 molecule plays a crucial role in CD pathogenesis, the number of HLA DQB1 � 0201 copies might have important consequences in CD patients: 4 αβ-chain combinations can be synthesized in heterozygotes but all HLA-DQ molecules are identical in homozygotes [5,9,10]. Experimental data support this connection: HLA-DQ2.5 homozygotes can present gluten peptides on antigen-presenting cells more effectively than HLA-DQ2.5 heterozygotes [4,5]. HLA-DQ2.5 homozygotes are at fivefold risk of CD as compared to HLA-DQ2.5 heterozygotes [6,[9][10][11]. The presence of a second β chain seems decisive in determining the risk, whereas the role of a second α chain appears less important [10,12,13]. The magnitude of immune response depends on gene dose: HLA-DQ2.5 homozygotes show maximal T-cell activation and proinflammatory response, whereas heterozygotes exhibit less prominent responses. The more DQ2.5 molecules are expressed on antigen presenting cells, the stronger the immune activation is [4][5][6]. Based on these in vitro immunological studies, one might assume that HLA-DQ2 homozygosity alters the course of CD; furthermore, the gene dose effect is manifested clinically with an increased risk of complications [14].
To our best knowledge, no meta-analyses have been conducted on the clinical effects of HLA-DQB1 � 02 gene dose in CD. It is assumed that patients with a double dose of HLA-DQB1 � 02 exhibit worse clinical outcomes, though some papers reported on how HLA-DQ2 gene dose (i.e., the number of DQB1 � 02 alleles) is associated with CD phenotype. This metaanalysis aims to investigate the association between HLA-DQB1 � 02 gene dose and the disease characteristics.

Materials and methods
We conducted our meta-analysis observing the rules of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) Statement (S1 Appendix) [15].

Search
PubMed (MEDLINE), Embase, Cochrane Controlled Register of Trials (CENTRAL), Web of Science, WHO Global Health Library, ClinicalTrials.gov, and Scopus were searched for articles discussing gene dose effect from inception until 22 nd October 2018. Our query was "celiac AND (homozyg � OR heterozyg � OR "gene dose" OR "gene dosage" OR "double dose" OR "double dosage" OR "single dose" OR "single dosage" OR "zero dose" OR "zero dosage" OR DQB1 OR DQ2.5 OR DQ2.2 OR DQ7 OR DQ8 OR "DQ2 in trans")". Reference lists and citing articles of the relevant studies were hand-searched for further papers. No filters were imposed upon the search. Draft of search in Embase is presented in S2 Appendix.

Eligibility, selection, and data extraction
We included original papers and conference abstracts, with the exclusion of case reports; and excluded comments, letters, editorials, and review articles. Eligible study design included both observational and experimental studies with adequate description of the genetic background and disease characteristics.
Eligible studies discussed celiac patients diagnosed in accordance with the current guidelines with known gene dose of HLA-DQB1 � 02 (double, single, and zero doses) and reported at least one of our predefined outcomes of patients by genotypes, separately. On inclusion, only PCR-based HLA-typing (sequence-specific primer or oligonucleotide probes) was acceptable.
Outcomes included clinical presentation dichotomized into classical and non-classical phenotype according to the Oslo criteria. Classical and non-classical CD are defined by the presence and absence of signs and symptoms of malabsorption (i.e., diarrhea, steatorrhea, weight loss, or growth failure), respectively [16], or into groups with and without diarrhea. Age at diagnosis and at disease onset were assessed separately. On one hand, diagnostic histology was divided into atrophic (Marsh 3 grade) and non-atrophic (Marsh 0-2 grades) mucosal damage [17]. On the other hand, we analyzed the severity of villous atrophy graded by the Marsh-Oberhuber classification (Marsh 3c vs. Marsh 3a-b) [18]. Tissue transglutaminase antibodies and endomysial antibodies were in the focus when discussing diagnostic serology. Other outcomes included disease complications and comorbid conditions (i.e., anemia, osteoporosis, autoimmunity, dental complications, dermatitis herpetiformis, malignant tumors, and type 1 diabetes mellitus).
We combined the yield of search in a reference manager software (EndNote X7.4, Clarivate Analytics, Philadelphia, PA, US), followed by the removal of overlaps between database content and duplicate records. The duplicate-free pool was searched first by title, then by abstracts, and full-texts against our eligibility criteria. Each phase of selection was carried out by two independent review authors (PV and KS) in duplicate, discrepancies were resolved by third party (MS) arbitration. Investigators had no contact with the authors of the original papers.
The following data were collected by two investigators onto pre-constructed Excel sheets: publication data, study design, population (numbers and characteristics of participants), HLA-DQB1 � 02 gene dose, age at onset; age, histology, serology, clinical presentation, anemia at diagnosis; concomitant immune-mediated disorders and complications: type 1 diabetes mellitus, dermatitis herpetiformis, dental enamel defect, recurrent aphthous stomatitis, enteropathy-associated T-cell lymphoma (EATL), small bowel carcinoma (SBC).

Risk of bias assessment
Two authors (MB and EP), unblinded to publication data, assessed the methodological quality of each study by using a tool developed a priori by our review team based on the Newcastle-Ottawa Scale (S1 Table). Results of risk of bias assessment were taken into account when assessing the limitations of the individual studies.

Statistical analysis
A biostatistician (NF) carried out the statistical analysis by using Comprehensive Meta-Analysis software (Version 3, Biostat, Englewood, NJ). The random effect model with DerSimonian-Laird estimation was used for analysis [29]. For dichotomous outcomes (i.e., histology, clinical presentation, and diarrhea), we calculated odds ratios (ORs) and 95% confidence interval (CIs). For age at diagnosis, weighted mean difference (MD) and 95% CIs were calculated. Statistical significance was attained when p<0.05.

Search and selection
Five studies [24,26,28,40,44] reported on the presence of diarrhea at diagnosis, all were included in the analysis of double vs. single dose of HLA-DQB1 � 02 and that of double vs. zero dose of the allele, as well. We failed to detect a significant gene dose effect, nor in the subgroups of children and adults in a homogeneous dataset (I 2 = 0.0%) ( Table 2, S1 and S2 Figs).

Age at diagnosis
Four [28,36,41,43] and five studies [28,36,41,43,47] were included in the comparison of double vs. zero dose of HLA-DQB1 � 02 and double vs. single dose of the allele concerning age at diagnosis, respectively. Patients with a double dose of HLA-DQB1 � 02 were similar in age at

Type 1 diabetes mellitus
Five [20,21,27,39,46] studies were included in the analysis of double vs. single dose of HLA-DQB1 � 02; of them, four [20,21,27,46] were eligible for inclusion in the analysis of double vs. zero dose of the allele. Analyses revealed no significant gene dose effect concerning the coexistence of type 1 diabetes mellitus and CD, which remained unchanged in the subgroups of children and adults. All analyses suffered from significant heterogeneity (S7 and S8 Figs, Table 2).

Other patient and disease characteristics
There were several reports on the association between HLA-DQ2 gene dose effect and the clinical phenotype of CD (age at onset, anemia, serology, autoimmunity, body mass index, osteoporosis, oral manifestations, complicated CD, and dermatitis herpetiformis). Data reported were insufficient for quantitative synthesis. We report the studies and the direction of associations with HLA-DQ gene dose in Table 3.

Publication bias
Although funnel plots seem symmetric, the low number of studies raised concerns about an uncertain assessment of symmetry (S3 Appendix).

Sensitivity analysis
When we omitted studies one-by-one, there was no change in the direction of the main association, except for two outcomes.

Risk of bias assessment
Results of risk of bias assessment are presented in Fig 6. Ten out of 24 studies (41.7%) aimed to analyze gene dose effect primarily. Twenty-two out of 24 (91.7%) and 18 out of 24 (75.0%) studies used a low-risk classification for diagnosing CD and rating diagnostic histology, respectively. However, the average CD population was clearly represented only in 62.5% of the studies included. Appropriate blinding was applied in only one study (4.2%).

Discussion
We aimed to review the current knowledge about the influence of HLA-DQB1 � 02 gene dose on the phenotype of celiac disease in our study.
Although reasonable molecular mechanisms have been proposed by in vitro experiments [5,6], gene dose effect seemed to influence only the clinical phenotype defined by Oslo criteria and histology. Patients with a double dose of HLA-DQB1 � 02 exhibited more often classical phenotype and villous atrophy, as compared to those with a single dose of the allele, whereas no evidence of gene dose effect was collected on diarrhea, age at diagnosis, the severity of villous atrophy, and the frequency of type 1 diabetes mellitus ( Table 2).
Lack of clinically significant gene dose effect is also supported by experimental results, as well: (1) an equal magnitude of specific T-cell responses characterizes homo-and heterozygotes, (2) the amount of DQA1 � 05 and DQB1 � 02 mRNS in heterozygotes exceeded the expected 50% of that measured in homozygotes [80]. Other factors should be taken into consideration, e.g., complementing HLA risk stratification with ten non-HLA loci changed the allocation of 10% of study population from the moderate-to the high-risk group [81]. The effect of non-HLA loci may outweigh that of HLA. In addition, non-genetic (environmental) factors appear to contribute to CD phenotype [23,28,40].
Concerning comorbid conditions, data allowed us to perform analysis only on type 1 diabetes mellitus, which is the most frequent co-existing immune-mediated disorder in CD. We failed to observe a significant gene dose effect in the frequency of type 1 diabetes mellitus. This corroborates with previous studies confirming that the combined occurence of DQ2 and DQ8 is high in patients with both CD and type 1 diabetes mellitus [27,46,54].  Table 3. Summary of studies reporting on gene dose effect.

Characteristics
Association between HLA-DQ2 gene dose and the clinical phenotype
Although we were unable to meta-analyze other disease complications due to discrepancies in reporting, the importance of HLA-DQ2 gene dose might be highlighted by these studies (Table 3). HLA status appears to be important in the development of malignant complications. Lymphomatous complications were more frequent in patients with a double dose of HLA-DQ2 in a study [64]. In a prospective study conducted by Al-Toma et al., HLA-DQ2.5 homozygosity was associated with serious CD complications, namely, type 2 refractory celiac disease and enteropathy-associated T-cell lymphoma [14]. Prevalence of DQ2 homozygosity was significantly higher in patients with a complicated disease (i.e., patients with type 1 and type 2 refractory CD, EATL, or SBC), as compared to those with non-complicated CD in a retrospective Italian and a multicentre study [49,50]. However, it is noteworthy that data of Howell et al. did not confirm these observations [58]. Further studies aiming to resolve these controversies are awaited to test whether determining HLA gene dose is appropriate for risk stratification for severe complications.

Strengths and limitations
To date, no meta-analyses have investigated HLA-DQB1 � 02 gene dose effect in CD. Our main strength is the transparent and comprehensive search and the rigorous selection process; however, we must acknowledge that the evidence is limited due to a number of reasons.
We did not contact the original authors of the included or excluded papers to acquire further information; only published material was used to preserve reproducibility.
Data on the clinical phenotype of celiac disease were collected from patients' files retrospectively. Regarding HLA status, HLA-typing was performed (1) at the time of enrolment to the study (at or after the diagnosis of celiac disease) or (2) before enrolment (retrospectively collected from charts). In relation to the outcomes, both ways are retrospective. However, HLA is a genetic marker, an unchangeable feature of the patients. Therefore, in our opinion, timing of HLA-typing does not affect data quality.  Table. https://doi.org/10.1371/journal.pone.0212329.g006 HLA-DQB1*02 gene dose effect in celiac disease Design of the studies included were (1) controlled studies including celiac and non-celiac subjects (case-control or cross-sectional studies) or (2) uncontrolled case series reports of celiac patients. In this meta-analysis, data on celiac patients were desirable, exclusively (while the non-celiac groups were irrelevant for us). As a consequence, there were studies included not aiming to analyze gene dose effect, as a primary objective, which may limit the data reported from the studies.
The number of eligible studies was low, not allowing us to draw reliable conclusions from the symmetry of the funnel plots to assess small-study effect (S3 Appendix).
Relevant clinical questions as to whether gene dose effects anemia, complications of CD, or serology have not been meta-analyzed due to incoherent data reporting across studies (S2 Table).
Self-reported clinical symptoms (e.g., bloating, diarrhea, and abdominal discomfort) are difficult to quantify objectively. To minimize the distortion, we used the widely accepted Oslo classification (i.e., dichotomization of patients into classical [with malabsorption] and nonclassical [without malabsorption] phenotypes) [16].
Some analyses suffer from statistical heterogeneity ( Table 2) which can be explained by methodological heterogeneity (i.e., histological assessment, HLA-typing, inclusion and exclusion of subjects) and clinical heterogeneity (i.e., geography, diverse environmental factors [e.g., timing of gluten introduction], and genetics [i.e., HLA and non-HLA loci]). A part of heterogeneity may originate from the age of participants: heterogeneity reduced when analyzing data from children and adults separately in some analysis, while persisted in others ( Table 2).

Implications for clinical practice
Our results suggest a significant gene dose effect regarding clinical presentation: classical clinical presentation and villous atrophy are more frequent in patients with a double dose of HLA-DQB1. We were unable to prove a similar effect in terms of diarrhea at diagnosis, age at diagnosis, the degree of atrophy, and type 1 diabetes mellitus. Recent guidelines do not require HLA-typing to set up the diagnosis of CD apart from pediatric cases diagnosed without intestinal biopsy. The role of it is mainly restricted to the exclusion of CD [82,83]. Patients with high-risk HLA status may be at higher risk of severe disease course, raising concerns about the need for a stricter gluten-free diet and follow-up. However, these results should be treated with caution due to the limitations of the data available for our study.

Implications for research
Studies validating our results and investigating the association between gene dose effect and disease complications (e.g., malignant tumors [EATL, SBC], autoimmune disorders, and RCD) are awaited.