The NHS is a prospective cohort study comprised of nearly 121,700 U.S. women, ages 30–55 years, who completed a mailed health questionnaire starting in 1976. Information on health information has since been updated every 2 years through follow-up questionnaires. NHSII is a similar cohort of 116,686 U.S. women between the ages of 25–42 years that was started in 1989. These women have been followed with similar biennial questionnaires. The Human Research Committee at Partners Healthcare approved this study.

Dietary assessment was performed using a 161-item semi-quantitative food frequency questionnaire (SFFQ) starting in 1984 in NHS and updated in 1986 and every 4 years since. Similarly in NHSII, dietary intake was assessed using SFFQ staring in 1991 and updated every 4 years. We calculated nutrient intake from the number of servings of food items and U.S. Department of Agriculture data on the content of nutrients in specified portions. The reproducibility and validity (against dietary records) of the FFQs have been extensively documented (13, 14). All nutrient values were adjusted for total caloric intake by the residual method to estimate intakes independent of the amount of total intake. The correlation between energy-adjusted potassium obtained from FFQ and diet records was 0.53 (15). In addition, the correlation between dietary potassium obtained from FFQ and urinary secretion of potassium was 0.56 in men and 0.39 in women and that of sodium was 0.66 in men and 0.56 in women (16). For questionnaire cycles with no assessment of dietary intake, we carried forward data from the prior questionnaire cycle when this detailed assessment was done. However, we did not carry forward dietary information. Specifically, women with missing dietary data in a 2-year cycle when dietary data was specifically queried did not contribute person-year to the analyses.

Detailed information on lifestyle factors, including body weight, smoking status, use of non-steroidal anti-inflammatory drugs (NSAIDs), menopausal hormone therapy, and oral contraceptives were collected and updated in each questionnaire cycle. Information on physical activity was also estimated using an instrument administered every 2–4 years that has been described in detail (17). Participants’ self-report of body weight, height, physical activity, and use of oral contraceptives have been previously validated (18–20). In 1992, in NHS and 1991 in NHSII, women were also asked about their latitude of residence at age 30, which we have previously shown to be associated with risk of UC and CD (21). Information about history of appendectomy was collected in 1992 in NHS and 1995 in NHSII.

Information on ancestry was collected in the 1992 questionnaire in NHS and baseline questionnaire in NHSII. Specifically, women were asked whether they had the following ancestries: African, Asian, Hispanic, Scandinavian, southern European/Mediterranean, other white, or other ancestry. Most of women (93% in the NHS I, 91% in the NHSII) reported white ancestries, consistent with underrepresentation of minorities in healthcare at the inception of both cohorts. More than 90% of the women reported a single ancestry. Ancestry data were available in more than 95% of women in NHS and 99% of women in NHSII.

Detailed information on confirmation of self-reported cases of CD and UC has been previously reported (17, 21, 22). Briefly, since 1976, participants have reported diagnoses of UC or CD through an open-ended response on biennial surveys. Diagnoses of UC and CD have also been specifically queried since 1982 and 1992, respectively. Similarly in NHSII, both diagnoses were specifically queried starting in 1993. Participants who reported diagnosis of CD or UC on any biennial questionnaire received a supplementary questionnaire. In addition, medical records from these participants were requested and reviewed by two gastroenterologists blinded to exposure information. We used standardized criteria to confirm cases of CD and UC (23–26).

For analysis of dietary sodium, potassium, and food items, person-time for each participant was calculated from the date of return of their baseline questionnaires to the date of the diagnosis of UC or CD, date of last returned questionnaire, or June 1, 2010 for NHS and June 1, 2011 for NHSII, whichever came first. At baseline, we excluded participants with missing dietary data and history of IBD or cancer (with the exception of non-melanoma skin). We calculated adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) using Cox proportional hazards modeling incorporating time-varying covariates. Consistent with prior analyses, information on BMI was derived from baseline questionnaire to avoid potential bias from changes in weight by preclinical disease (27, 28). Dietary sodium and potassium were modeled as quintiles while food items were modeled based on number of servings consistent with prior dietary analysis (28, 29). We observed no heterogeneity in the association of dietary sodium, potassium, and food items with CD or UC in separate analyses of NHS and NHSII (P for heterogeneity > 0.50 for both UC and CD). Thus, we pooled individual-level data from NHS and NHSII and adjusted for cohort in all analyses. We used the first quintile as the reference category and calculated the risk of CD and UC for each quintile of dietary sodium and potassium by comparing higher quintiles to the first quintile. Using previously described methods, we performed mediation analysis to determine whether the reported association of fiber intake on risk of CD was explained by intake of potassium (30). We used SAS version 9.3 (Cary, NC, USA) for all analyses.

In 1989–1990, 32,826 NHS participants (aged 43–69 years) returned a blood sample on ice packs by overnight courier and completed a short questionnaire (31). Between 1996 and 1999, 29,611 NHSII participants (aged 32–54 years) provided blood samples and completed a short questionnaire in a similar protocol (32). Similarly in 2001–2004, 29,684 participants in NHS and 29,859 participants in NHSII mailed in a sample of buccal cells collected using a “swish-and-spit” method. Participants who provided the buccal cells had not previously provided a blood specimen. Among participants who provided a blood or saliva specimen, we matched 169 CD cases to 740 controls and 202 UC cases to 740 controls based on age, menopausal status, month of blood collection, and fasting status. Genomic DNA was isolated from buccal cells or blood samples using conventional methods (33).

We used the most recent meta-analysis of GWA studies to identify six single nucleotide polymorphisms (SNPs) in genes involved in T_H17 pathway that have previously been associated with risk of CD and UC: rs10758669 (JAK2), rs12942547 (STAT3), rs1819333 (CCR6), rs7657746 (IL21), rs3024505 (IL10), and rs11209026 (IL23R) (3). We directly genotyped these SNPs by 5′ nuclease assay (TaqMan^®), using the ABI PRISM 7900HT Sequence Detection System (Applied Biosystems, Foster City, CA, USA). TaqMan^® primers and probes were designed using the Primer Express^® Oligo Design software v2.0 (ABI PRISM). The status of individual case–control status was not known to laboratory personnel. Genotyping procedures were validated using duplicates from nearly 10% of the samples; concordance was 100% for these samples. Primers, probes, and conditions for the genotyping assay are available upon request. We confirmed that all SNPs were in Hardy–Weinberg equilibrium among the controls using the Chi-square test (all P > 0.30).

We examined associations according to each of the T_H17 pathway SNPs except rs11209026 (IL23R) due to low minor allele frequency (~7%). However, we also constructed two summary risk scores (one for each outcome) incorporating all six T_H17 pathway SNPs associated with CD and UC risk using a previously described weighting method (34, 35). Briefly, each SNP was recoded as 0, 1, or 2 according to the number of alleles associated with increasing risk of CD or UC; each SNP was weighted by its relative effect size (β coefficient) derived from the previously reported meta-analysis data (separate β coefficient for UC and CD) (3). We constructed the genetic risk score using the equation: genetic risk score = (β1 × SNP1 + β2 × SNP2 + … + βn × SNPn)/(sum of the β coefficients).

We used conditional logistic regression for risk of CD or UC with a multiplicative interaction term for dietary potassium and genotype in T_H17 pathways defined according to the number of risk alleles while adjusting for other potential risk factors (see Assessment of Other Covariates). Since CD and UC are rare outcomes, we used odds ratios as estimates of relative risks. To minimize the potential influence of reverse causality, we analyzed prospectively collected data on diet from the questionnaires administered 4 years prior to diagnosis of CD or UC for cases and their matched controls. Although, there were no significant variations in allele frequency of genotyped risk variants according to European ancestry, we adjusted all models of gene–environment interaction for ancestry (see Assessment of Other Covariates). We used SAS version 9.3 (Cary, NC, USA) for these analyses. All P-values were two-sided and <0.05 was considered statistically significant. As the selection of the SNPs was done a priori and interaction with only five SNPs was tested, we did not adjust for multiple testing in our analyses.

After obtaining consent, we collected 10 cc of peripheral blood from 25 healthy volunteers, ages 18–35 years without history of HIV, tuberculosis, or any autoimmune diseases (e.g., rheumatoid arthritis, systemic lupus, etc.) and who were not taking steroids or other immunosuppressive medications. All experiments were performed in accordance to the approved guidelines of Human Ethics Committee of Translational Health Science and Technology Institute (THSTI), Faridabad, India.

Peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll Paque density gradient centrifugation (GE Healthcare). Naïve CD4⁺ CD45RO⁻CD25⁻ or memory CD4⁺CD45RO⁺ T cells were sorted as previously described (36). Sorted cells were cultured in 96-well U-bottom plates at a concentration of 0.1 × 10⁶ cells/ml in X-VIVO 15 medium (Lonza) supplemented with 100 units/ml Penicillin/Streptomycin, l-glutamine, NEAA, and Sodium Pyruvate. Cells were stimulated with plate bound anti-CD3 (UCHT1; 10 μg/ml), soluble anti-CD28 (28.2; 3.0 μg/ml), and were differentiated either with TGFβ1 (1 ng/ml) or with T_H17 condition with addition of following cocktails: TGFβ1 (5.0 ng/ml; R&D), IL-6 (25 ng/ml; Peprotech), IL-1β (12.5 ng/ml; Peprotech), IL-21 (20 ng/ml; Peprotech), and IL-23 (20 ng/ml; Peprotech) for 6 days. Cells were supplemented with or without potassium chloride (Sigma Aldrich). On day 6, cells were stained with LIVE/DEAD Fixable Dead Cell Stain Kit (Invitrogen) according to manufacturer’s protocol. Cells were then fixed and permeabilized with BD Cytofix/Cytoperm and stained with anti-human Foxp3-PE antibody (BD Bioscience; 560082). Labeled cells were acquired on BD FACS Verse and analyzed by FlowJo (Treestar, USA). All FACS plots were gated on live cells.

For proliferation assay, sorted naïve human CD4⁺ T cells (1 × 10⁶/ml) were labeled with 5.0 μM of CFSE in PBS at 37°C for 5 min. Cells were shaken intermittently in order to get homogenously labeled cell population. Cells were thoroughly washed with glutamine rich media (RPMI-1640) to wash off unlabeled dye. Cells were stimulated with plate bound anti-CD3 and soluble anti-CD28 in round bottom 96-well plate at a concentration of 0.1 × 10⁶/ml with and without KCl (40mM). On day 5, cells were washed extensively with 1× PBS and fixed and permeabilized with Foxp3 staining kit (eBiosciences; 00-5523-00) and were labeled with fluorescent-conjugated Foxp3 antibody (Biolegend, 320014) for 30–40 min on ice. Cells were also stained with live/dead marker to exclude dead population during analysis. Prism 5.0 was used for statistical analysis and differences were considered statistically significant with a P value of less than 0.05.

RNA was isolated using RNAeasy mini kit [Qiagen (#74134)], and converted to cDNA using iScript cDNA synthesis kit [BioRad (#1708891)]. The Taqman primers used for this study were purchased from Applied Biosystems (Smad3 Hs00969210_m1, Smad2 Hs00183425_m1, Smad7 Hs00998193_m1, and SGK1 Hs00985033_g1 Taqman primers). The values are represented as the difference in Ct values normalized to GAPDH for each sample.

Peripheral blood mononuclear cells were isolated from healthy volunteers and naïve and memory CD4⁺ T cells were sorted as described previously. Cells were stimulated for 1 h with plate bound anti-CD3 and soluble anti-CD28. Thereafter, cells were washed extensively with ice cold 1× PBS and stained with live dead stain (Thermo fisher) for exclusion of any dead cell population during analysis. Cells were fixed with 2% paraformaldehyde at 37°C for 20 min in order to cross link the phospho proteins. Cells were again washed twice with ice cold 1× PBS and permeabilized with 90% ice cold methanol at 4°C for 60 min. Cells were washed again and incubated with FACS buffer (PBS + 2% FCS) to prevent non-specific binding. Cells were stained with phospho Smad specific antibodies at 4°C for 40′ according to the manufacture’s recommendations. Finally, cells were stained with appropriate fluorescent-conjugated secondary antibodies, data were acquired by BD FACSVerse and analyzed by FlowJo. Following antibodies (phosphorylation sites included in parentheses) from Abcam were used: ab53100 (pSMAD2, pS467) and ab52903 (pSMAD3, pS423 + S425) for these experiments.