Limited new NAD+ production contributes to gut lining inflammation in inflammatory bowel disease

Clinical studies at University Hospital Schleswig-Holstein, Campus Kiel, were approved by the ethics committee of Kiel University (D 490/20, D 489/20, A 124/14, AZ 156/03-2/13, EA 1/300/15) following the Declaration of Helsinki. Enrolled subjects provided written informed consent. Procedures were carried out in accordance with national guidelines. Clinical characteristics are recorded in Tables S1–S3.

Samples were measured at Biocrates Life Sciences AG (Innsbruck, Austria) with their ‘Tryptophan Metabolism Assay’ via ultra-high-performance liquid chromatography mass spectrometry (HPLC-MS) with multiple reaction monitoring in positive mode using a SCIEX API 5500 QTRAP^® (AB SCIEX, Germany) and electrospray ionization. Half of measurements were performed after phenyl isothiocyanate derivatization. Analytes were quantified with external seven-point calibration.

Metabolites below the limit of quantitation and detection were considered missing; those with ≥80% non-missing values were retained (80% rule22). Following filtering, 3-OH-anthranilic acid (3OHAnth), quinaldic acid, Xanth and 3-IPA had missing values (1.2%, 2.9%, 2.3%, 6.1%). These values were imputed as half of each metabolite’s minimum value. Metabolites were log₁₀ transformed and auto-scaled prior to statistical analysis.

The software MetaboINDICATOR^TM (Biocrates) was used to further reveal the biological significance of metabolic changes observed. In addition to the pre-calculated ratios provided, we included two ratios (QA:sumIDO and QA:3OHAnth) based on reported increases to these ratios in disease flares.8 QA:sumIDO was calculated as: QA/(3OHAnth + 3-OH-kynurenine [3OHKyn] + KA + kynurenine [Kyn] + picolinic acid [PA] + QA + Xanth). The QA:Trp ratio reflects a measure of overall QA synthesis relative to the entire Trp pool.

For cohort #1, serum Trp was assessed by HPLC for n = 180 patients with n = 812 observations.

Targeted metabolomics was conducted in subcohort #1a of cohort #1 (n = 52 Crohn’s didease [CD], n = 82 ulcerative colitis [UC], n = 343 observations) with available samples at baseline, week 2, and week 14 after advanced therapy induction.

We included the following potential confounding variables to linear mixed models (LMMs) used to assess Trp changes over the first 52 weeks of biologic therapy: biologic therapy, diagnosis (UC/CD), body mass index (BMI), age at baseline, recruitment cohort, and sex assigned at birth. For targeted metabolomics, LMMs included the covariates: sex at birth, age, BMI, biologic status (at time of observation), and, where appropriate, diagnosis. Spearman correlations were performed using week 14 observations, where, owing to treatment, we expected the widest variance in disease activity. All analyses were performed using R v.4.4.0.

DSS colitis was performed at Pennsylvania State University, and approved by IACUC (PROTO202202188). Male 8- to 12-week-old C57BL/6J mice (Jackson Laboratory) were acclimatized for 7 days in specific-pathogen-free (SPF) conditions, receiving laboratory diet 5010. Mice were single-housed and supplied with autoclaved water with/without 2.5% DSS (MP Biomedicals) for 5 days (n = 8 control/n = 12 DSS), followed by regular water. DSS-water was exchanged every other day, and consumption was monitored. Disease activity indices (DAIs; weight loss, stool consistency, rectal bleeding, behavior, activity, fur) were assessed daily.

For NAM supplementation, male C57BL/6J mice (>12 weeks, Charles River France) were held under SPF conditions (Institut Pasteur de Lille, France; #B59‐35009) according to governmental guidelines (#2010/63/UE; Décret 2013‐118) and animal ethics (protocol #05273.01). While acclimatizing for 1.5 months, animals received diet A4 (Scientific Animal Food and Engineering, France) and water ad libitum. Animals were exposed to 12-h light–dark cycles (7 am to 7 pm). For NAM supplementation, a Trp-free diet (Ssniff, Germany; powdered to prepare pellets) supplemented with control granules or a controlled NAM release formulation (Solural Pharma, France) in three different concentrations was provided for 12 days prior to DSS treatment (1.5%, TDB Consultancy, Sweden) for 5 days (n = 10 per group). The diet was continued throughout the experiment. DAIs (diarrhea, fecal blood, macroscopic appearance of inflammation) were obtained daily.

Experiments followed ARRIVE guidelines.

Blood and biopsies were obtained from two longitudinal IBD intervention cohorts from northern Germany (n = 32 UC, n = 30 CD) over 14 weeks. Patients who were treatment-naive for biologics were introduced to anti-TNFα (n = 22), anti-α4β7-integrin (n = 21), anti-IL6-trans-signaling (n = 16), or anti-IL6-R (n = 3) and were clinically and endoscopically monitored. In total, 120 biopsies and 123 whole blood samples were used for RNA sequencing (RNA-seq), and 128 serum samples for metabolomics (Biocrates MxP Quant 500 kit, Biocrates Life Sciences AG, Austria) as previously described.3^,23^,24 Analysis of differentially expressed genes (DEGs) was performed with variance stabilized read counts (DESeq2) and applying LMMs (lme4) (vst ∼ sex * HBI/Mayo|Kyn: Trp + [1 | Patient ID]). Enrichment analyses were based on means of gene set enrichment (GSEA) or hypergeometric tests for overrepresentation (ClusterProfiler). For GSEA, a ranking vector was created summing t-values for the main effect of HBI (Harvey–Bradshaw Index)/Mayo or serum Kyn:Trp and interaction with sex. Analyses were performed in R. All P-values were adjusted with Benjamini–Hochberg correction.

Trp degrading pathways were extracted with the human metabolic model (recon3D). Expression values (as transcripts per million [TPM]) were mapped to Trp pathways applying gene–protein–reaction relations of recon3D (github.com/Porthmeus/CORPSE) to estimate reaction activity scores (RAS). RAS were associated with HBI/Mayo scores in LMMs (rxnExpr ∼ sex * HBI/Mayo + [1 | Patient ID]). Linear modeling was used to impute missing HBI/Mayo according to individual time trajectories assuming a logarithmic decline in HBI/Mayo over time. All P-values were adjusted with Benjamini–Hochberg correction.

Mucosal expression of IDO1 and QPRT were extracted from transcriptomics datasets GSE73661↗ (microarray of UC) and GSE109142↗ (RNA-seq TPM counts of pediatric UC) and processed using R v.4.2.2. Expression levels were correlated with (1) the Mayo endoscopic subscore for 166 biopsies from n = 67 UC (treated with infliximab or vedolizumab) and n = 12 HC (GSE7366125) patients, and (2) the histology severity score of n = 206 new-onset UC and n = 20 HC (GSE10914226) patients. IDO1 and QPRT expression was furthermore correlated with IFNγ, TNFα, and IL23 scores using the R package singscore,25–28 the IFNγ gene set (IFITM1, MX1, OAS3, IFIT1, IFI44L, IFI16),29 and the TNF and IL23 gene sets described by Martínez et al.28.

Normalized IDO1 and QPRT expression in UC patient biopsies was assessed in published microarray and RNA-seq data. Endoscopic improvement was defined as endoscopic Mayo (eMayo) score ≤1. Biopsies were collected within three longitudinal cohorts. (1) UC patients were treated with tofacitinib (baseline = 27, week 8 = 25).30 Raw sequencing reads were assessed with FastQC (v.0.11.9)31 and aligned to the human reference genome (GRCh38) using STAR aligner (v.2.7.10a).32 Gene-level read counts were obtained with featureCounts from the Subread package (v.2.0.1) using Gencode (v.38) for gene models.33 DEG analysis was performed with DESeq2 (v.1.44.0) in R (v.4.4.1)34 correcting for age, sex, prednisone, anti-TNF, vedolizumab, eMayo, response, and visit. (2-3) A prospective study of UC subjects with active colonic disease received either vedolizumab or infliximab24^,35 (vedolizumab-treated: baseline = 9; week 2 = 7; week 6 = 8; week 14 = 8; infliximab-treated: baseline = 8; week 2 = 7; week 6 = 7; week 14 = 6). RNA-seq was performed. Raw gene expression counts were normalized using the DESeq2 R package34 v.1.40.2.

Statistical tests were selected considering data distribution and variance characteristics: for in vitro experiments, unpaired Student’s t-tests (two groups) and one-way ANOVA with Šidák correction (multiple groups) were used. For animal experiments, Mann–Whitney U (two groups) or Kruskal–Wallis and Dunn’s correction (multiple groups) were employed. A threshold of P < .05 was considered statistically significant. Statistical tests were performed with GraphPad PRISM software 10.

Patients achieving long-term disease control (n = 120) exhibited a stable and continuous rise of serum Trp over the entire 52 weeks (Figure 1A), independent of disease entity (UC vs. CD, Figure 1B; cohort #1, Table S1). By contrast, patients without therapy persistence after 1 year (n = 60) did not exhibit a rise in serum Trp concentration (Figure 1A, B). We conducted targeted metabolomics in a subcohort of cohort #1 to examine longitudinal changes of Trp and corresponding KP, indolic, and serotonin metabolites (Fig. S1) in IBD patients (n = 52 CD, n = 82 UC) with available samples at baseline, week 2, and week 14 after advanced therapy induction (cohort #1a; Figure 1C, Table S2). Of these, 91 patients with available endoscopic scores at week 14 were classified into remission and non-remission based on eMayo (<2) or simple endoscopic score of CD (SES-CD; <3; n = 34 remission, n = 57 non-remission). We found significant differences in the second principal component between remission and non-remission (Figure S2A), whereas baseline inflammation levels were comparable between remitters and non-remitters (Figure S2B).

We assessed individual associations between metabolites or metabolite ratios and clinical (Crohn’s disease activity index: CDAI, CD; partial Mayo: pMayo, UC), endoscopic (SES-CD, CD; eMayo, UC), and biochemical (C-reactive protein: CRP) disease activity indices (Figure 1D, E), to identify serum metabolites with temporal fluctuations coinciding with inflammation resolution (defined as change from active disease to endoscopic remission). In line with previous studies, the Kyn to Trp (Kyn:Trp) ratio was higher in individuals not achieving remission at week 14 at weeks 0, 2 and 14 (Figure 1F).3^,7^,8 The ratio of PA to QA (PA:QA) was higher in remitters at each time point tested (Figure 1G). Additionally, the correlation between QA and Kyn:Trp was stronger for QA (Spearman’s rho, ρ = 0.77, false discovery rate [FDR]-corrected P < .0001) compared to PA (ρ = −0.11, FDR-corrected P = .38) (Figure 1H). These data show that Trp degradation along the KP resulting in QA accumulation (Figure 1I) is associated with disease activity and an unfavorable disease course in IBD.

We aimed to understand the underlying transcriptional changes in the inflamed mucosa driving QA accumulation in IBD patients’ serum using a second longitudinal IBD cohort (UC, n = 30; CD, n = 32) with serial biosampling before and after advanced therapy induction (cohort #2; Figure 2A, Table S3).23^,24^,35 By estimating reaction activity scores of enzymatic reactions catabolizing Trp in bulk transcriptomics from blood and intestinal mucosa, we correlated reaction abundances of Trp-degrading pathways with clinical disease activity indices (HBI, CD; total Mayo score, UC).36^,37 We identified significant correlations of Trp-degrading enzymes with disease activity in the mucosa, but not in the blood (Figure 2B–D). Inferring mucosal flux of Trp metabolism, we observed decreased reaction abundance for genes converting QA into nicotinic acid mononucleotide (NAMN) (reaction ID R24a) upon high disease activity, indicating serum QA accumulation might result from insufficient mucosal conversion into NAD⁺ (Figure 2B, D). To confirm this hypothesis, we assessed IDO1 and QPRT expression in the inflamed colonic mucosa. IDO1 and QPRT in transcriptomics from sigmoid UC biopsies significantly correlated with endoscopic (Figure 2E) and histological (Figure 2F) disease activity (cohort #3).38^,39 Whereas IDO1 expression increased with inflammatory severity, QPRT decreased (Figure 2E, F). Using intestinal UC biopsies (eMayo 0 vs. 3; cohort #1a), we confirmed heightened protein expression of IDO1 and reduced QPRT with enhanced mucosal inflammation (Figure 2G).

We next assessed whether mucosal IDO1 and QPRT expression is also informative of the response towards advanced therapies. We therefore analyzed mucosal transcriptomics of UC patients treated with the JAK inhibitor (JAKi) tofacitinib (cohort #4),30 infliximab or vedolizumab (Figure 3; cohort #5/6). Regardless of the therapeutic class, IDO1 was downregulated and QPRT upregulated between baseline and follow-up in endoscopic responders, resulting in overall lower mucosal IDO1 and heightened QPRT levels in responders vs. non-responders at follow-up. This pattern suggested that resolution of inflammation restores mucosal IDO1 and QPRT expression.

To confirm our findings experimentally and to show that mucosal QA accumulation coincides with NAD⁺ depletion, we used DSS colitis for targeted metabolomics of serum and tissues from different inflammation states (flare-up [day 5; d5], flare [d8/11]) (Figure 4A). Starting from d3, serum Kyn increased, with heightened Kyn:Trp ratios from d5 to d11 (Figure S5). When analyzing Trp and derivatives in all anatomical locations post mortem, we found that Trp degradation almost exclusively occurred in the inflamed colon (Figure 4B, D). Trp catabolism coincided with mucosal QA accumulation and reduced NAD(H), linking KP activation with NAD⁺ depletion in inflammation (Figure 4F–H). Therefore, our data support mucosal QA build-up and decreased NAD(H) as key metabolic changes during mucosal inflammation.

Having shown that Trp degradation in the inflamed mucosa leads to accumulation of QA and depletion of NAD⁺, we aimed to pinpoint upstream transcriptional signatures driving Trp catabolism in IBD that feed into the metabolic constraint at QPRT. Therefore, we combined bulk transcriptomics of blood and intestinal biopsies with serum metabolomics (cohort #2, longitudinal sampling from UC, n = 30; CD, n = 32; Figure 2A) to integrate expression of Trp-degrading enzymes with serum Trp catabolism. We identified commonly regulated signal transduction pathways correlating with (1) serum Kyn:Trp ratios as a proxy for Trp degradation and (2) disease activity scores (HBI/total Mayo) using LMMs. Among the pathways found for both approaches, nuclear factor ‘kappa-light-chain-enhancer’ of activated B-cells (NF-κB) and JAK/STAT were upregulated in inflamed tissue (Figure 5A, B). Assessing overlapping signaling transduction pathways between blood and tissue in association with high disease activity and elevated serum Kyn:Trp, we validated upregulation of mucosal JAK/STAT signaling (Figure S6A, B).

We confirmed the molecular link between JAK/STAT signaling and mucosal expression of the key KP enzyme IDO1, a known JAK/STAT downstream target gene, by inferring the putative upstream cytokine network (cohort #3).40^,41 Mucosal IDO1 induction significantly correlated with IFNγ, IL23, and TNF, implying mucosal JAK/STAT-dependent cytokines drive KP-mediated Trp breakdown (Figure 5C–H). Interestingly, we noted a similar association between JAK/STAT signaling and QPRT downregulation, suggesting that IDO1 upregulation and QPRT suppression coincide upon inflammation (Figure 5C–H).

We hypothesized that JAK/STAT-mediated activation of the KP results in QA accumulation. Given the known robust expression of IDO1 in immune cells, peripheral blood mononuclear cells (PBMCs) were chosen as an in vitro model to investigate Trp catabolism along the KP.42 After identifying IFNγ as the strongest IDO-inductor in human PBMCs (Figure 6B, Figure S7A), we co-treated PBMCs with IFNγ and JAKi (pan-JAKi: tofacitinib; selective JAKi: upadacitinib, filgotinib). IFNγ-induced IDO gene and protein expression was completely blocked upon co-treatment with JAKi (Figure 6C–E, Figure S7B). JAK/STAT-activated IDO1 upon IFNγ stimulation enhanced Trp-to-Kyn turnover in PBMC lysates and supernatants, which was reverted by JAKi (Figure 6F–H, Figure S7C–E). To show elevated Kyn results from Trp degradation and no other potential sources, we employed isotope-labeled Trp (¹³C₁₁ Trp) in PBMCs. Again, IFNγ induced conversion of ¹³C₁₁ Trp into ¹³C₁₀ Kyn, which was abrogated by JAKi (Figure S7F–K). To investigate whether JAK/STAT-driven activation of the KP led to QA accumulation, we profiled downstream Trp metabolites by subjecting PBMCs to IFNγ ± tofacitinib or epacadostat (IDO1 inhibitor). IFNγ induced increases of Kyn and the KP metabolites anthranilic acid (Anth), 3OHAnth, and QA (Figure 6I–N, Figure S8B–F). Conversely, coadministration of tofacitinib or epacadostat abrogated QA accumulation. QA was undetectable in supernatants, suggesting intracellular accumulation (Figure S8B–F).

We aimed to further validate that IFNγ-induced Trp catabolism along the KP resulted in QA accumulation in PBMC lysates due to a KP blockade at QPRT. Indeed, QPRT expression was suppressed following KP activation with IFNγ treatment (Figure S8G). We speculated that reduced QPRT not only caused QA accumulation, but also decreased QA-to-NAD⁺ conversion. NAD(H) production as assessed by an NAD/NADH-Glo^TM assay in human PBMCs was indeed diminished by IFNγ (Figure 6O), which was further aggravated in Trp-deficient medium, confirming Trp as the major source of NAD⁺ synthesis upon JAK/STAT activation (Figure S8H). Together, our data show that JAK/STAT-mediated Trp degradation via the KP feeds into QA accumulation due to a blockade of de novo NAD⁺ synthesis at QPRT.

Lastly, we aimed to understand the direct impact of QPRT on modulating immune responses in the three most relevant intestinal cell types, namely PBMCs, fibroblasts, and intestinal epithelial cells (IECs).

To silence QPRT, we utilized siRNA in fibroblasts and IECs, whereas for PBMCs, we embarked on our previous finding of downregulated QPRT following IFNγ treatment (Figure S8G). QPRT suppression significantly increased pro-inflammatory cytokines upon stimulation with IFNγ and LPS (Figure 7A–H, Figure S9A–F).

To clarify whether NAD⁺ depletion or QA buildup drives cytokine expression in QPRT-silenced cells, we tested QA’s cytotoxic effects. QA neither amplified cytokine responses nor significantly reduced viability of PBMCs (Figure S10A–D), IECs (Figure S10E–H), or fibroblasts (Figure S10I–K), arguing against direct pro-inflammatory effects of QA.

We thus hypothesized that QPRT knockdown drives pro-inflammatory cytokine expression via impaired de novo NAD⁺ synthesis and assessed cytokine expression and NAD(H) levels following treatment with IFNγ and LPS in combination with the NAD⁺ precursor NR.43 NR supplementation ameliorated increased cytokine induction in QPRT-silenced fibroblasts, pointing towards a protective role of restoring cellular NAD(H) levels (Figure 7A–C). Indeed, NAD(H) levels were significantly reduced upon QPRT knockdown, which was further aggravated by concomitant IFNγ and LPS stimulation and rescued by NR supplementation (Figure 7D, Figure S9G). Notably, the IFNγ- and LPS-induced increase of pro-inflammatory cytokines in QPRT-silenced cells was not ameliorated by epacadostat, again arguing against QA as the main inflammation driver (Figure 7A–C). Further confirming that NR can bypass insufficient de novo NAD⁺ synthesis from Trp, the anti-inflammatory property of NR was validated in IFNγ- and LPS-treated PBMCs under Trp-free conditions (Figure 7E–H). Similarly, NR treatment of human organoids derived from moderately inflamed colon of UC patients significantly boosted their growth (Figure 7I, J).

Lastly, we assessed if supplementation with NAD⁺ precursors would similarly lessen intestinal inflammation in vivo. To induce robust abrogation of de novo NAD⁺ synthesis from Trp via QPRT, DSS colitis was conducted in C57BL/6J mice that were fed a Trp-free diet and treated with both control and NAM-containing granules prior to and throughout the DSS colitis (Figure 7K, L). Although it must be considered that the effects of NAM may be amplified due to pronounced mucosal shortage of NAD⁺ upon both inflammatory and artificial Trp-deficient conditions, NAM supplementation significantly reduced disease activity indices in a dose-dependent manner.

Data and analytic methods will be available to other researchers either in the materials and methods section or via publicly available online databases.