No association between gluten sensitivity and amyotrophic lateral sclerosis

All participants in the serological and questionnaire studies were included in the Prospective ALS study the Netherlands (PAN) between 1 January 2006 and 31 December 2015 [18]. Patients were diagnosed according to the revised El Escorial criteria [19]. The population-based design of the PAN study was ensured by inclusion of patients through multiple sources: neurologists, rehabilitation physicians, the Dutch neuromuscular Patient Association and our ALS website. Age- (±5 years) and gender-matched controls were included through the general practitioner of the participating patient. In the Netherlands, every resident is registered at a general practitioner, which makes the sample representative of the general population.

For genetic correlation analyses, we used data from a GWAS of ALS, which included 12,577 sporadic ALS cases and 23,475 unaffected controls, and a GWAS of CD, which included 4533 patients and 10,750 controls. Both studies comprised cohorts from Europe and the USA [7, 20].

Separately, ALS cases and population-matched controls are being collected worldwide as part of Project MinE [21]. A subset of these samples is drawn from the neuromuscular clinic at the University Medical Center Utrecht and controls are selected from the PAN study to match cases by geography, age and gender. Data collected on the samples include disease status and genotyping on the Illumina 2.5M array. As part of the Project MinE, samples and variants have undergone quality control (detailed description in Online Resource 1) [7].

In 359 population-based patients and 359 matched controls, we measured TG6 antibody concentrations using the ELISA IgA kit (Zedira, Darmstadt, Germany). TG6 positivity was calculated using the mean of the control group (17.33 U/mL) plus 2 standard deviations (2 × 19.59 U/mL), and was set at 56.51 U/mL. In a subset of 199 patients and 199 controls, antibodies to TG2 were measured using the ELiA Celikey IgA kit (Phadia AB, Uppsala, Sweden). Serum samples containing a TG2 antibody titer of more than 10 U/mL were considered positive, as recommended by the manufacturer. In the same subset, we determined the prevalence of IgA class endomysial antibodies (EMA) using indirect immunofluorescence (Biognost, Heule, Belgium). Total IgA was measured, and a serum IgA concentration below 0.05 g/L was regarded as IgA deficiency. Gluten-associated antibodies were compared between groups using Student’s t test for antibody concentrations and the Chi-square test for comparison of the seroprevalence, or Fisher’s exact test in case of cell counts below five. Due to low numbers, we were unable to test for interaction with gender using an interaction term in a logistic model; instead, we stratified according to gender. We compared antibody levels and positivity between bulbar and spinal patients, and antibody measures in patients with a time to diagnosis within the first quartile (representing patients with a rapid disease course) with controls. Clinical characteristics between groups were compared using the Mann–Whitney U test for continuous characteristics; the Chi square for categorical characteristics (or Fisher’s exact test as appropriate); the Cox proportional hazard model for survival.

We examined a structured questionnaire collected from 1829 population-based ALS patients and 3920 matched controls to collect medical histories [18]. For patients, we analyzed data referring to the period before symptom onset. In addition, in the 199-item food frequency questionnaire, participants were asked about their adherence to a specific diet [22]. For patients, this was defined as the 1-month period before symptom onset. All questionnaires were checked for missing data or inconsistencies, and participants were contacted by telephone to complete or correct the data. The survival status of patients was obtained by checking the municipal population register and/or contacting the general practitioner every 3 months. Proportions of participants with CD or a GFD were compared using the Fisher’s exact test because of the small numbers.

Correlations in genetic architectures between two traits can provide insight into potential overlapping etiology between diseases and may help disentangle shared potential causal mechanisms [17]. To estimate the genome-wide genetic correlation between ALS and CD, we used Linkage Disequilibrium Score Regression (LDSC). In short, LDSC considers the fact that, for polygenic traits, the association signal at a single-nucleotide polymorphism (SNP) will also capture information for SNPs nearby in linkage disequilibrium. This relationship between linkage disequilibrium and association signal can also be used to test for relationships between two traits at all SNPs across the genome. A GWAS of ALS had been performed using a linear mixed model (LMM) implemented in the Genome-Wide Complex Trait Analysis (GCTA) software [7, 23], from which the summary statistics for 1,217,311 SNPs were available [7]. Summary statistics for 499,513 SNPs from the CD GWAS were obtained via personal communication with the authors. Details of the statistical LDSC analysis are provided in the Online Resource 2.

We constructed a genetic risk score from associated CD SNPs, to see if such a score could help predict ALS disease status. We downloaded all SNPs associated by GWAS to CD at p < 1 × 10⁻⁶ from the NHGRI GWAS Database [24]; data downloaded included SNP rsID, chromosome, position, risk allele, P value, and reported odds ratio from the GWAS. A total of 31 SNPs had complete information. For duplicate SNPs, we kept the SNP results from the study with the largest sample size. We extracted the genotypes for these SNPs from 2833 unrelated Dutch samples (1823 cases and 1010 controls) that had been genotyped on the Illumina 2.5 M array as part of Project MinE. A total of 21 SNPs were found in the genotyping data. We LD-pruned the SNPs (r ² < 0.05) to ensure that all SNPs represented independent associations to CD; LD pruning removed two SNPs. We used the genotypes for the remaining 19 SNPs to construct a genetic risk score by weighting the genotypes by the odds ratio reported in the CD GWAS and summing across all weighted genotypes (performed in Plink). We used logistic regression to test for association between ALS disease status and CD-based genetic risk score, correcting for ten principal components and gender.

The same 2833 unrelated Dutch samples from Project MinE were used to perform imputation of the MHC. MHC genotypes (build hg19, chromosome 6, positions 24,092,021–38,892,022) for these samples were extracted from genotypes generated using the Illumina 2.5 M genotyping array. Details on sample quality checks and imputation of the MHC genotypes are provided in the Online Resource 3. When imputation was complete, we performed QC, removing any sample with >2.5 alleles at any HLA allele (Online Resource 3). We used the resulting genotype dosages in 1786 ALS cases and 999 controls to perform association testing between ALS status and four CD-associated HLA alleles: HLA-DQA1*0501, HLA-DQB1*0201, HLA-DQB1*0202, and HLA-DQB1*0302. We adjusted for ten principal components and gender. The fifth HLA allele of interest, HLA-DQA1*0505, could not be tested as it was absent from the SNP2HLA imputation reference panel.

We tested TG6 antibody levels in 359 patients and 359 controls, matched for baseline characteristics (Table 1). The subgroup of 199 patients and 199 controls with measured TG2 antibody and EMA levels were also characteristic-matched (data not shown). None of the patients was IgA deficient. TG2 positivity (>10 U/mL) was found in one patient and zero controls. This patient was also the only participant who tested positive for EMA, and was one of the eleven patients with positive antibodies to TG6. Despite having TG2, EMA and TG6 antibodies, this patient had clinical characteristics (probable ALS, bulbar onset at the age of 66 years) and a survival (1.33 years from onset) compatible with typical ALS. Furthermore, routine blood tests were normal, indicated no anemia, and magnetic resonance imaging (MRI) revealed no white matter lesions.

Patients and controls had comparable TG6 antibody concentrations (Student’s t test p = 0.66, Table 2) and positivity for TG6 antibodies [11 patients (3.1%) and 12 controls (3.3%), Chi-square test p > 0.99]. Stratifying by gender did not reveal a difference in antibody concentrations or prevalence (Student’s/Fisher’s test all p > 0.48, data not shown). Clinical characteristics of the 11 seropositive patients were similar to those of the 348 seronegative patients (Table 3). Eighty-one patients with a short time to diagnosis (defined as <6 months and used as a surrogate variable for a rapid disease course) had TG6 antibody levels similar to controls [mean (SD): 21.16 (31.56) vs. 17.33 (19.59), Student’s p = 0.30]. Two of the 81 patients (2.5%) tested positive for TG6 antibodies (Fisher’s p > 0.99; data not shown). Similarly, bulbar and spinal patients had comparable TG6 antibody levels [mean (SD): 17.53 (27.37) vs. 18.30 (20.20), Student’s p = 0.79] and positivity [n (%): 3 (2.5%) vs. 8 (3.4%), Fisher’s p = 0.76; data not shown].

Questionnaire-derived data of 1829 ALS patients and 3920 controls showed they had similar characteristics (Table 1). According to the data, 3 of 1829 ALS patients (0.16%) and 4 of 3920 controls (0.10%) had been diagnosed with CD (Fisher’s p = 0.69). 1470 of these 1829 patients and 3683 of the 3920 controls filled in a food frequency questionnaire; in addition to the 3 patients and 4 controls with a medical history of CD, we identified 2 more patients (n = 5 in total; 0.3%) and 8 controls (n = 12 in total; 0.3%) who were following a GFD prior to disease onset (Fisher’s p > 0.99). Table 4 shows the clinical characteristics and disease course of the 5 patients who adhered to a GFD. On average, their survival after disease onset was 3.3 years, consistent with standard ALS disease progression.

We applied LDSC analysis using summary-level results from a GWAS of ALS (12,577 cases and 23,475 controls) and of CD (4533 CD cases and 10,750 controls). The genetic correlation between ALS and CD was 0.06 (standard error = 0.07; p = 0.37), indicating that, genome-wide, there was no evidence for genetic correlation between the traits. While LDSC looks for trait correlation, genetic risk scores can be used to test for association between two traits at specific loci. We tested a genetic risk score based on CD loci in the Dutch Project MinE samples, including 1823 ALS cases and 1010 matched controls. The score was not associated with ALS disease status (p = 0.47).

We performed HLA imputation and association analysis in the same Project MinE Dutch ALS patients and controls. After additional QC of the data, we found no association at any of the four HLA alleles associated with CD in 1786 ALS cases and 999 controls (all p > 0.28, Table 5).