Gut Microbiota-Dependent Trimethylamine N-Oxide Associates With Inflammation in Common Variable Immunodeficiency

The Regional Committee for Medical and Research Ethics approved the study protocol. All study participants signed a written, informed consent. The work described has been carried out in accordance with the Declaration of Helsinki.

CVID patients were recruited from the outpatient clinics at the Section of Clinical Immunology and Infectious Diseases, Oslo University Hospital, Rikshospitalet. CVID was defined as decreased serum levels of IgG, in addition to IgA and/or IgM by a minimum of two standard deviations below the mean for age, and exclusion of other causes of hypogammaglobulinemia (Primary immunodeficiency diseases. Report of an IUIS Scientific Committee. International Union of Immunological Societies. Clin Exp Immunol. 1999) (). CVID subgroups were classified as “Infection only” or “Complications” based on criteria previously defined (), however, CVID enteropathy was defined as persistent diarrhea after exclusion of gastrointestinal infection (). ^{15 3 8}

Peripheral venous blood was collected into sterile blood collection tubes with EDTA as an anticoagulant. The tubes were immediately immersed in melting ice and subsequently centrifuged within 15 min at 2,000 g for 20 min to obtain platelet-poor plasma. Plasma was stored at −80°C in multiple aliquots.

Plasma levels of tumor necrosis factor (TNFα) and interleukin (IL)-6, IL-8, and IL-12 were analyzed using a multiplex cytokine assay (Bio-Plex Human Cytokine Plex Panel; Bio-Rad Laboratories Inc., Hercules, CA). The samples were analyzed on a Multiplex Analyzer (Bio-Rad Laboratories) according to instructions from the manufacturer.

Free carnitine and γBB were analyzed in plasma using MS/MS as described previously () with some modifications of the high-performance liquid chromatography conditions (). Stable isotope dilution liquid chromatography–tandem mass spectrometry (LC/MS/MS) was used for the quantification of TMAO, choline, and betaine in the same plasma samples (). ¹⁶ Supplementary Methods Supplementary Methods

LPS was analyzed by Limulus Amebocyte Lysate chromogenic assay (Lonza, Walkersville, MD) according to the manufacturer's instructions, with the following modifications: In the 104 CVID patients and 30 controls, samples were diluted 10-fold to avoid interference with background color, and preheated to 68°C for 10 min prior to analyses to dissolve immune complexes as previously described (,). ^{8 17}

Participants collected stool samples at home within 24 h prior to their hospital visit, or alternatively at the hospital, with a standardized collection device (). The stool samples were then transferred by the participants to stool collection tubes with Stool DNA Stabilizer (Stratec Biomedical, Birkenfeld, Germany) (). Samples were stored at minimum −20°C according to the manufacturer's recommendations until DNA extraction. Bacterial DNA was extracted using the PSP Spin Stool DNA Plus Kit (Stratec) before being subjected to amplification of the 16S ribosomal RNA gene with dual-indexed barcodes according to an established protocol (), followed by sequencing on an Illumina MiSeq (San Diego, CA;). ^{18 19 20} Supplementary Methods

To assess bacteria with TMA producing potential we applied qPCR DNA assays targetingandusing modified primers described by Rath et al. (). Briefly, 30 ng DNA was analyzed using HOT FIREPol EvaGreen qPCR Mix Plus (Solis, BioDyne, Brumath Cedex, France) and Applied Biosystems 7900HT, primer concentration and thermal cycles as described (). All products in the melting curve with a melting temperature above 78°C were considered aproduct.levels were normalized to 16S rRNA gene levels and presented as ddCt. CntA CutC CntA CutC ^{12 12}

CVID patients were asked to complete a self-administrated, validated Norwegian food frequency questionnaire designed to reflect dietary habits over the past year (,). The questionnaire offers multiple choice alternatives and also the opportunity to provide supplementary information regarding specific dietary restrictions or habits. It covers 180 food items and has serving size alternatives specified in households units, which is then converted to grams per day using a software developed at the Institute for Nutrition Research, University of Oslo (). ^{21 22 21}

The datasets for TMAO pathway metabolites were not normally distributed and so the data was log transformed. Where normal distribution was achieved, we continued with multivariate testing and where normal distribution was not achieved, we applied Mann–Whitney non-parametric testing. Fisher's exact test was used for categorical variables. Depending on the data distribution, we performed correlation analyses with either Spearman's rank correlation test (correlation coefficient rho) or Pearson's correlation test (correlation coefficient). Additionally, stepwise linear regression analysis was performed to adjust for age and sex. For the longitudinal data, we compared datasets from three different time points using the Friedman test. Univariate Repeated measures ANOVA (UNIANOVA) was used to assess the effect of treatment (rifaximin), focusing on the interaction between time and treatment group for TMAO. Calculations were performed in SPSS (version 24, IBM, NY). r

We consecutively collected plasma samples from 104 CVID patients over 13 months from the outpatient clinic at Section of Clinical Immunology and Infectious Diseases, Oslo University Hospital Rikshospitalet, and included 30 healthy controls as previously described (). In addition, stool samples, blood samples, and food frequency questionnaire (FFQ) from 40 CVID patients (FFQ = 38 patients) from a preceding intervention trial () were included for sub-analysis. In CVID patients included for gut microbiota analyses, the use of antibiotics in the last 12 weeks was an exclusion criterion. Patients' characteristics are given for both cohorts in,, and. Thirty-two CVID patients were included in both cohorts, but at different time points (,). The 86 controls used in the analyses ofandin stool samples did not have corresponding blood samples (). ^{8 23} Table 1 Supplementary Table 1 Supplementary Methods ^{8 23} Supplementary Methods CutC CntA

CVID patients (= 104) had significantly elevated plasma concentrations of TMAO compared to healthy controls (). The TMA precursors carnitine, choline, and betaine, as well as the intermediate metabolite y-BB, were also significantly increased in CVID patients compared to healthy controls (). The seven outliers in the CVID group (with the highest plasma concentration of TMAO in) were tested with a gene panel for monogenic disease in relation to their primary immunodeficiency (), but were all negative. n Figure 2A Figures 2B–E Figure 2A ²⁴

To study the temporal course and stability of carnitine-TMAO pathway metabolites, we measured the concentrations in 20 CVID patients (mean age 51.2 years ± 11.8 [SD], 13 [65%] females) three times over 8 weeks. Sixteen patients completed all three measurements. Applying the Friedman test, we discovered no significant changes in concentrations of TMAO (= 0.444), carnitine (= 0.144), betaine (= 0.174), choline (= 0.472), or y-BB (= 0.646) during temporal testing (). p p p p p Supplementary Figure 1

We and others have previously demonstrated increased inflammatory cytokines such as IL-6, IL-8, IL-12, and TNFα in CVID (–). These cytokines were all markedly elevated in the 104 CVID patients compared to controls () (). We therefore selected these inflammatory markers to explore the association between systemic inflammation and TMAO. Plasma TMAO was positively associated with plasma levels of TNFα and IL-12 (), but not with IL-6 (= 0.263, rho = 0.11) or IL-8 (= 0.214, rho = 0.12). Data on other measured cytokines and their association to TMAO are shown in. ^{25 27} Supplementary Table 2 ⁸ Figures 3A,B Supplementary Table 3 p p

Using existing data on gut microbiota composition from a subset of the CVID patients (= 40) (), we explored if specific bacterial taxa previously found to differentiate between CVID patients and healthy controls correlated with TMAO pathway metabolites in plasma sampled at the same time point. These included the ten taxa making up the CVID dysbiosis index found to capture CVID dysbiosis in the gut:(increased in CVID), and(reduced in CVID) (). In addition, we included five taxa identified to differ between CVID and healthy controls in another publication using a different statistical approach (ANCOM):, and(increased in CVID) and(reduced in CVID) (). The relative abundance ofin stool samples from CVID patients correlated positively with concentrations of TMAO in plasma (). On genus level, we found positive correlations between abundance ofin stool samples and plasma concentrations of TMAO () and y-BB (= 0.046,= 0.32) in CVID patients. Both of these taxa were increased in CVID patients compared to controls in previous publications from our group ()., a Gram positive bacterial genus previously shown to be reduced in CVID patients, was negatively correlated with the TMA precursor betaine (= 0.037,= −0.33). n Bacilli, Dorea, Roseburia, Gammaproteobacteria Bifidobacterium, Odoribacteracea, Christensenellaceae, Blautia, Sutterella, Desulfovibrionacea Hungatella, Flavonifractor, Veillonella Escherichia-Shigella Christensenellaceae R-7 group Gammaproteobacteria Escherichia-Shigella p r Blautia p r ^{23 8 28} Figure 4A Figure 4B Supplementary Table 4

Bacterial genes that have been suggested to encode enzymes responsible for TMA production (),and, were measured in stool samples from 40 CVID patients and 85 age-, sex, and body mass index (BMI)-matched controls (). These bacterial genes have also been reported to be associated with increased abundance of, demonstrated to be enriched in the gut microbiota from CVID patients (). In a previous study,was detected in only 26.0% of human samples whereaswas found in all human samples (). Due to the anticipated low number of individuals that would havepresent, we therefore compared the number of individuals that hadpresent in the two groups. Using qPCR on DNA stool samples, we found thatwas present in significantly more CVID samples (= 30, 75%), than controls (= 45, 53%),= 0.020 (Fisher's exact test,). As expected,was present in all samples. Forwe went on to measuring gene expression (ddCT) and found an increase ingene expression in CVID patients compared to controls,= 0.003 (). To conclude, both TMA producing bacterial genes;and, seem to be more prevalent in stool samples from CVID patients compared to controls. ²⁹ Supplementary Methods Supplementary Table 4 ¹² Figure 5A Figure 5B CutC CntA Gammaproteobacteria CntA CutC CntA CntA CntA n n p CutC CutC CutC p CutC CntA

We went on to exploring if there was any correlation between the expression ofin stool samples and the taxa making up the CVID dysbiosis index or the other the five taxa differing between CVID patients and controls (as defined above). We found thatlevels were negatively correlated with(= 0.04, rho = −0.33) and(= 0.028, rho = −0.35), but positively correlated with(= 0.003, rho = 0.46).is thought to have health-promoting properties and is used in probiotic supplements (). We have previously found very low abundance ofin CVID patients compared to controls (), and the negative correlation withcould suggest that in CVID patients, TMA producing bacteria are increased at the expense of beneficial gut bacteria such as. CutC CutC Bifidobacterium p Bacilli p Hungatella p Bifidobacterium Bifidobacterium CutC Bifidobacterium ^{30 8}

We have previously shown that there are increased amounts of the microbial product LPS in the plasma of patients with CVID (= 104,) (), potentially contributing to systemic inflammation in these patients. Herein, we found that plasma levels of LPS were significantly correlated with TMAO and the TMA precursors carnitine and y-BB in CVID patients (), indicating a relation between gut leakage and TMAO pathway metabolites. n Supplementary Table 2 ⁸ Figures 6A–C

Similar to CVID, patients with selective IgA deficiency also have lower gut microbial diversity (), thus it is of interest to explore whether undetectable serum IgA levels could influence TMAO concentrations in CVID patients. We compared CVID patients with undetectable serum IgA levels (IgA < 0.1 g/L) to CVID patients with detectable serum IgA levels (IgA ≥ 0.1g /L) and found no significant difference in plasma TMAO concentration between these two groups (= 0.844,). This may be due to a cut-off level for serum IgA of <0.1 g/L being too strict (), and that low, but detectable, serum levels of IgA could still affect gut microbial composition in CVID, and thereby TMAO levels. ²⁸ Supplementary Table 5 ²⁸ p

Furthermore, LPS and low serum IgA concentrations have been found to correlate with development of autoimmune cytopenias in CVID patients (,). We found no significant difference in LPS levels between CVID patients with serum IgA < 0.1 g/L compared to CVID patients with IgA ≥ 0.1 g/L, or CVID patients with or without autoimmune cytopenia, organ specific autoimmunity or enteropathy (). To further look into the potential role of LPS and undetectable serum IgA levels in the immunopathogenesis of CVID, we investigated certain inflammatory cytokines in relation to these markers. LPS correlated positively with IL-6 (= 0.028,= 0.216), a central cytokine in the pathogenesis of several autoimmune disorders as well as of inflammation in CVID (), but not did not correlate significantly with IL-8 (= 0.192, rho = 0.129), IL-12 (= 0.064, rho = 0.182), or TNFα (= 0.143, rho = 0.145). Additional correlation analyses between LPS and cytokines are given in. Cytokine analyses of CVID patients with serum IgA < 0.1 g/L compared to CVID patients with IgA ≥ 0.1 g/L showed increased concentrations of two chemokines involved in inflammatory processes (i.e., IP-10 and MCP-1) in patients with IgA < 0.1 g/L, but did not display an overall different pattern in cytokine levels than those with IgA ≥ 0.1 g/L (). To conclude, the link between LPS and inflammation in CVID could appear to occur on an overall group level. However, low IgA levels may be of greater importance than demonstrated here, conditional upon the cut-off used for low/undetectable levels of serum IgA. ^{31 32} Supplementary Table 6 ⁴ Supplementary Table 7 Supplementary Table 8 p r p p p

A detailed self-reported food frequency questionnaire was obtained from 38 CVID patients. Food items that have been described to be associated with metabolites in the carnitine-TMAO pathway include meat, fish, eggs, and dairy products. CVID patients did not appear to consume more of these food items compared to a reference population () (). Hence, an increased consumption by CVID patients of egg, dairy products, meat, or fish did not appear to be responsible for the increased TMAO concentrations seen in these patients. Additionally, we investigated other food items relevant to gut microbiota and inflammation such as fiber, protein, wine, and sugar (–) for association analyses with TMAO. Wine consumption correlated negatively with TMAO (= 0.023, rho = −0.37) in CVID patients, and when separating wine consumption into red wine and white wine, the negative correlation with TMAO, corrected for age and sex, remained significant only for red wine (= 0.018, β = −0.332). We also found that intake of table sugar, but not total sugar consumption, was associated with increased plasma concentrations of TMAO (= 0.006, rho = 0.44). In contrast, dietary intake of meat, fish, egg, dairy products, and fiber were not found significantly associated with TMAO in plasma (), further supporting that other mechanisms than diet are responsible for the increased TMAO in CVID. ³³ Supplementary Figure 2 ^{34 36} Supplementary Table 9 p p p

It has previously been shown that a course of short-term antibiotics such as ciprofloxacin and metronidazole can reduce TMAO concentrations in plasma of healthy humans, as well as in mouse models of cardiovascular disease (administering vancomycin, neomycin sulfate, metronidazole, or ampicillin for 3 weeks) (,). We have previously shown that a 2-week course of the broad spectrum oral antibiotic rifaximin alters gut microbial composition, but not systemic inflammation in CVID (). Based on the knowledge that rifaximin acts purely locally in the gut and that several antibiotics have previously been demonstrated to influence TMAO concentrations, it was intriguing to investigate whether rifaximin (550 mg twice-daily) could have a similar impact. We therefore measured the effect of 2 week administration of rifaximin on plasma concentrations of TMAO in 40 CVID patients and found that there was no difference in TMAO concentrations between the rifaximin intervention group (= 20) (baseline 5.0 [3.5–6.2] μmol/L, 2 weeks 4.3 [2.7–7.0] μmol/L, and 6 weeks 4.4 [3.1–6.9] μmol/L) and the no intervention group (= 20) (baseline 3.2 [2.0–5.5] μmol/L, 2 weeks 5.5 [2.5–9.0] μmol/L, and 6 weeks 3.6 [1.9–11.2] μmol/L),= 0.975 (UNIANOVA), data are given in median (25–75 percentile) (). ^{37 38 23} Supplementary Figure 3 n n p

When dividing the CVID patients (= 104) into two main subgroups by phenotype (i.e., infection only [= 25] and those with non-infectious inflammatory and/or autoimmune complications [= 79]), we found no differences in TMAO concentration between “infection only” (6.2 ± 4.0 μmol/L) vs. “non-infectious complications” (7.3 ± 7.3 μmol/L, mean ± SD),= 0.922. Specific analyses of CVID subgroups including autoimmune cytopenia, organ-specific autoimmunity, and enteropathy did not show significant differences in TMAO concentration between patients with and without these features (). Only two of the 104 CVID patients were not on Ig-replacement therapy and notably, these two were not outliers in the plasma TMAO concentration data (both had plasma TMAO concentration <15 μmol/L). n n n p Supplementary Table 9

Both age and sex correlated significantly with metabolites in the carnitine-TMAO pathway in CVID patients and controls (< 0.001, rho = 0.36 and= 0.02, rho = 0.24, respectively), and were therefore adjusted for in statistical analyses (see Methods). BMI has previously been found to correlate positively with TMAO in some studies (,), but we did not find this in our dataset (= 0.802, rho = −0.03). Finally, kidney function could potentially influence TMAO concentrations. We therefore explored if there was any difference in estimated glomerular filtration rate (eGFR) between CVID patients and controls, which could potentially have been a confounding factor. We found no significant difference in eGFR between the patients and controls (59.4 ± 4.2 vs. 60.0 ± 0.0, mean ± SD),= 0.277. Interestingly, within the CVID group, we found no significant difference in plasma TMAO concentration between those with eGFR < 60 vs. those with eGFR > 60 (11.1 ± 6.4 vs. 6.9 ± 6.7 μmol/L, mean ± SD,= 0.082). p p p p p ^{39 40}