Tryptophan degradation as a systems phenomenon in inflammation – an analysis across 13 chronic inflammatory diseases

Summary

Background

Chronic inflammatory diseases (CIDs) are systems disorders that affect diverse organs including the intestine, joints and skin. The essential amino acid tryptophan (Trp) can be broken down to various bioactive derivatives important for immune regulation. Increased Trp catabolism has been observed in some CIDs, so we aimed to characterise the specificity and extent of Trp degradation as a systems phenomenon across CIDs.

Methods

We used high performance liquid chromatography and targeted mass spectrometry to assess the serum and stool levels of Trp and Trp derivatives. Our retrospective study incorporates both cross-sectional and longitudinal components, as we have included a healthy population as a reference and there are also multiple observations per patient over time.

Findings

We found reduced serum Trp levels across the majority of CIDs, and a prevailing negative relationship between Trp and systemic inflammatory marker C-reactive protein (CRP). Notably, serum Trp was low in several CIDs even in the absence of measurable systemic inflammation. Increases in the kynurenine-to-Trp ratio (Kyn:Trp) suggest that these changes result from increased degradation along the kynurenine pathway.

Interpretation

Increases in Kyn:Trp indicate the kynurenine pathway as a major route for CID-related Trp metabolism disruption and the specificity of the network changes indicates excessive Trp degradation relative to other proteogenic amino acids. Our results suggest that increased Trp catabolism is a common metabolic occurrence in CIDs that may directly affect systemic immunity.

Funding

This work was supported by the DFG Cluster of Excellence 2167 “Precision medicine in chronic inflammation” (KA, SSchr, PR, BH, SWa), the BMBF (e:Med Juniorverbund “Try-IBD” 01ZX1915A and 01ZX2215, the e:Med Network iTREAT 01ZX2202A, and GUIDE-IBD 031L0188A), EKFS (2020_EKCS.11, KA), DFG RU5042 (PR, KA), and Innovative Medicines Initiative 2 Joint Undertakings (“Taxonomy, Treatments, Targets and Remission”, 831434, “ImmUniverse”, 853995, “BIOMAP”, 821511).

Research in context.

Evidence before this study

The amino acid tryptophan can be broken down into several bioactive substances. The kynurenine pathway is one important route through which tryptophan breakdown occurs, and pathway activity is increased in some inflammatory diseases, like inflammatory bowel disease and systemic lupus erythematosus, but also in some cancers, where pathway activity appears to play an immunosuppressive role. The bioactivity of the compounds produced via this pathway varies: some possess immunosuppressive effects, while others appear to be pro-inflammatory. Emerging evidence suggests that kynurenine pathway activity may play a pro-inflammatory role in inflammatory bowel disease, yet systematic analysis of the presence of tryptophan degradation as metabolic hallmark across chronic inflammatory diseases is lacking.

Added value of this study

This study provides a comprehensive analysis of tryptophan metabolism across 13 different chronic inflammatory disorders. We show that catabolic turnover of Trp in response to inflammation is particularly disruptive to amino acid metabolism, while other essential amino acids may be affected to a lesser extent. In addition, degradation of tryptophan is the result of enzymatic activity and not a result of overall reduction in protein intake/uptake, microbial influence or excessive protein synthesis rates.

Implications of all the available evidence

Disruption of tryptophan metabolism along the kynurenine pathway is a unifying feature of chronic inflammatory diseases. Despite this, there is still significant variation in tryptophan metabolism across disease entities. These findings suggest that targeted approaches designed to specifically intervene or restore tryptophan metabolism in chronic inflammatory diseases may be a viable route to disease management.

Introduction

Chronic inflammatory diseases (CIDs) affect various organs including the intestine, skin and joints and represent a common disease group of the immune system with rising incidence in industrialized countries. Targeted interception of cytokine signalling, and intracellular signal transduction events provide a unifying therapeutic concept in CIDs, suggesting common underlying molecular events in CID pathophysiology.¹

As an essential amino acid, tryptophan (Trp) needs to be taken in through exogenous sources. Trp can be further broken down (catabolised) to various bioactive derivatives. There are three main routes of Trp catabolism: the kynurenine pathway, the serotonin pathway, and indolic compound production.² The so-called kynurenine pathway not only produces kynurenine, but a variety of additional metabolites, sometimes referred to as kynurenines, which can have pro- and anti-inflammatory effects.³ CIDs with Trp metabolism alterations include inflammatory bowel disease (IBD), systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA).2, 3, 4, 5, 6 Indeed, the kynurenine-to-Trp ratio (Kyn:Trp), reflecting kynurenine pathway activity, increases in a disease-dependent manner in IBD and SLE, among others.^4,6,7 However, a comprehensive analysis on the manifestation of Trp degradation across different CIDs is missing.

We set out to understand the overall distribution of Trp alteration across CIDs, with a focus on 13 CIDs affecting the gastrointestinal tract (i.e., IBD), the musculoskeletal system, and the skin. By integrating high performance liquid chromatography (HPLC) based measurements into standard clinical practice, we were able to assess serum Trp levels longitudinally in 1949 CID patients, leading to 29,012 clinical observations. Complimenting this approach, we use targeted mass spectrometry to further probe Trp-derived serum metabolites in 198 patients whose HPLC-based Trp measurements indicated an extreme Trp phenotype (i.e., extremely high or low), using 179 healthy individuals as a reference. By selecting the extreme ends of the distribution, our aim was to understand whether the kynurenine pathway plays a significant role in disturbed Trp metabolism. Thus, we present a thorough and comprehensive examination of CID-associated Trp metabolism disruption.

Methods

Patient recruitment

Cohort 1

To investigate the role of Trp metabolism in CIDs, we implemented serum Trp measurements into clinical laboratory assessments at our CID outpatient clinic (UKSH, Kiel, Germany). In cohort 1, serum Trp was measured via HPLC (detailed below), and CRP was measured according to standard clinical diagnostics and the patients had diseases affecting the gastrointestinal tract (GI), musculoskeletal (MSK) system, and skin. GI CIDs (i.e., IBD) included patients with Crohn's disease (CD) and ulcerative colitis (UC). MSK CIDs included diseases affecting joints (seronegative RA: RAneg, seropositive RA: RApos, axial spondyloarthritis: AxSpA, psoriatic arthritis: PsA), connective tissue (SLE, systemic sclerosis: SSc, Sjögren's syndrome: SjS, unspecified connective tissue disease: CTD), and blood vessels (giant cell arteritis: GCA, polymyalgia rheumatica: PMR). Finally, cohort 1 includes psoriasis (Pso), which primarily affects the skin. Cohort 1 includes longitudinal measurements of 1949 patients with a total number of 29,012 observations, and an average number of 14.7 ± 16.6 (mean ± standard deviation) follow-ups. Additionally, single observations of 291 healthy individuals are included as a reference (Supplementary Table S1). Patients were recruited in the IBD outpatient clinic of the Department of Internal Medicine I at the University Medical Center Schleswig–Holstein (Campus Kiel, Germany) between January 2015 to June 2023. This cohort consists of all available patient data for which Trp measurements were available at the time of data export (June 2023).

Cohort 2

We selected patients with extreme Trp levels for inclusion into cohort 2; these patients were selected based on their HPLC-measured Trp values (below 45 μM or above 60 μM). We analysed Trp and selected Trp derivatives plus canonical proteogenic amino acids via targeted metabolomics with serum samples from 198 patients (58 GI, 88 MSK and 52 Skin) and 179 healthy individuals (Supplementary Table S2). In total, 407 serum samples were included for measurement and analysis. In addition, we assessed faecal Trp levels through targeted metabolomics from a sub-cohort of 164 patients (51 GI, 67 MSK and 46 Skin). Patient recruitment occurred at the CID outpatient clinic of the Department of Internal Medicine I at the University Medical Center Schleswig–Holstein Kiel Campus, and the Comprehensive Center for Inflammation Medicine and Clinic of Rheumatology at the University Medical Center Schleswig–Holstein Lübeck Campus in Germany. The healthy individuals were recruited at the University Hospital Schleswig Holstein, Campus Kiel in 2016, and none of these participants had received any antibiotics or other medication 2 months prior to inclusion. Individuals with diabetes were excluded from the analysis. Total stool and serum samples analysed reflects sample availability.

Cohort 3

To assess the relationship between serum and stool Trp levels, we included 72 additional serum and stool samples from 37 patients with GI and MSK CIDs (Supplementary Table S3). In this longitudinal cohort, patients were sampled at baseline and 6- and 14-weeks following administration of one of 11 targeted immunomodulatory therapies(infliximab, ustekinumab, cyclophosphamide, vedolizumab, adalimumab, tofacitinib, golimumab, etanercept, tocilizumab, secukinumab, and upadacitinib). Serum Trp was measured by HPLC, and stool Trp via targeted metabolomics. The total number of samples analysed reflects the total number of cases for which a patient had matching stool and serum samples available.

Ethics

Approval was granted by the ethics committee of the medical faculty of Kiel University before the start of the study (AZ D489/14, A 124/14 and AZ 156/03-2/13). In addition, approval was granted for the analysis of the healthy individuals in cohort 2 by the local ethic committee in Kiel (D441). Informed consent was obtained for all study participants. This study was conducted in compliance with ethical approval.

Measurement of serum metabolites

HPLC

Serum levels of Trp were determined using an In Vitro Diagnostic Conformité Européenne (CE) certified HPLC kit. Lower limits of quantification and lower limits of detection were calculated according to DIN 32645 guidelines.

Targeted Mass Spectrometry of Trp derivatives

Trp derivatives were measured in the patient serum and stool of cohort 2 via tandem mass spectrometry (LC-MS-MS) using the MxP Quant 500 kit (Biocrates Life Sciences AG, Innsbruck, Austria) according to the manufacturer's instructions. Serum samples were collected using serum s-monovette (9 mL, Sarstedt, Germany) and incubated upright at room temperature for 30 min prior to a 10-min centrifugation at 2000g. Serum was aliquoted in 500 μL tubes and stored at −80 °C until sample preparation for MS. The MxP Quant 500 kit simultaneously measures 630 metabolites covering 14 small molecule and 12 different lipid classes. It combines flow injection analysis tandem mass spectrometry (FIA-MS/MS) using SCIEX 5500 Q-Trap mass spectrometer (SCIEX, Darmstadt, Germany) for lipids and liquid chromatography tandem mass spectrometry (LC-MS/MS) using Agilent 1290 Infinity II liquid chromatography (Santa Clara, CA, USA) coupled with a SCIEX 5500 Q-Trap mass spectrometer for small molecules using multiple reaction monitoring (MRM) to detect the analytes. Data evaluation for serum metabolite concentrations and quality assessment was performed with the software SCIEX Analyst software (Version 1.7.2) and the MetIDQ™ software package (Oxygen-DB110-3023), which is an integral part of the MxP Quant 500 kit.

Data analysis

Pre-processing of metabolomics data

Metabolite concentrations below the lower limit of quantitation were removed from the analysis (i.e., considered missing values), and metabolites were further filtered based on the 80% rule⁸: only metabolites with 80% non-missing values were included for further analysis. Just 4 metabolites required missing value imputation. The proportions of missing values were 1.1% (serotonin), 3.8% (IAA and IndSO4), and 7.2% (IPA). Various factors can influence missing value rates, including unique metabolite quantitation limits and variability in metabolite levels between samples. The remaining missing values were imputed using GSimp,⁹ which uses iterative Gibbs sampling for missing value imputation. This algorithm was designed for imputation of metabolomics data, which tends to be left-censored. The imputation was performed using default parameters. Further details can be found in the GSimp publication.⁹ Only Trp, Trp derivatives, and proteogenic amino acids were considered for further analysis. For the stool Trp levels of cohort 3, 5.4% missing values were imputed.

Statistical analysis

To assess the difference between the serum Trp (derivative) concentrations in controls and different CID entities as well as in patients with high and low serum Trp, we used linear mixed models with Trp as the dependent variable, and CID entity and sex assigned at birth (hereafter referred to as sex) as fixed effects, with patient IDs as random intercepts. Because sex influences Trp levels,¹⁰ it was included as a fixed effect in all models. We additionally performed analyses on males and females separately (i.e., sex-stratified) to detect cryptic patterns related to Trp and CRP/disease activity. To assess the significance of the observed differences in the magnitude and direction of the relationship between Trp and disease activity, we additionally used linear mixed models with an interaction term specified between sex and the disease activity indices (e.g., CDAI and DAS28).

Assessment of differences in serum amino acids between control and high/low Trp groups included the primary affected organ or organ system (i.e., GI, MSK and Skin) as an additional fixed effect, i.e., the metabolite as the dependent variable, with Trp status (high/low), affected organ, and sex as fixed effects, and patient-specific random intercepts. The reference group in the aforementioned cases were the healthy individuals. To assess the predictive power of Trp on CRP and disease activity indices (e.g., BASDAI or DAS28), Trp was modelled as a fixed effect (predictor variable) with the disease activity metric as the dependent variables, with patient-specific random intercepts. The Gaussian distribution is assumed for the random effects terms in the linear mixed models. Models were fit using maximum likelihood (REML) and the degrees of freedom were estimated using Satterthwaite's method. Because the within patient samples of cohort 1 and 2 were collected in an unstructured manner, we did not assume within-group correlations when modelling covariance structures. Linear models were employed to estimate the relationship between Trp and PASI, without an adjustment for sex as there were only 3 female patients with available Trp and PASI. The metabolites, CRP, as well as DAS28 scores were all log₁₀ transformed to ensure the linearity of predictors assumption was met for linear (mixed) models. Previous reports have demonstrated low Trp levels in some CIDs, so we specifically assessed whether the Trp levels were lower in the disease groups by calculating left-tailed p-values; all other p-value were calculations were calculated two-sided.

In cohort 3, autoregression was appropriate as the samples were taken at specific points relative to therapy induction (baseline, 6 and 14 weeks). We specified serum Trp, sex and disease cohort as fixed effects in the model, with stool Trp levels as the predictor. Finally, we included patient-specific intercepts and slopes for the effect of time (specified in weeks from baseline) on stool Trp. In case of disease-specific relationships, we further stratified the data into GI and MSK entities and ran the models independently. For MSK, which had fewer observations, we dropped the random slopes term due to convergence issues. Thus, to assess the relationship between stool and serum Trp for MSK, patient-specified random intercepts were used instead. Stool and serum Trp were log₁₀ transformed. Each linear (mixed) model was checked for deviations from normality and homoscedasticity of residuals and linearity of quantitative predictors via visual inspection of fitted versus residual plots, assessment of residual distribution and by plotting residuals versus the response variable. We have included detailed overviews of the 95% confidence intervals (CI), beta coefficients, and raw/adjusted p-values for all linear (mixed) models in the supplementary files (Supplementary Tables S4–S15).

Partial correlations included an adjustment for sex, and in order to meet the assumption of sample independence, only the first observation per patient was used in this analysis. A Student's t-test with Welch approximation was used to assess differences between faecal amino acid levels in the high and low serum Trp groups. Normality was achieved following log₁₀ transformation; this was verified via visual inspection of quantile-quantile plots. Linear models were employed to estimate the relationship between Trp and PASI, without an adjustment for sex as there were only three female patients with available Trp and PASI.

In all cases where a multiple comparison adjustment was made to p-values, we used the Benjamini-Hochberg procedure. In cohort 1, the false discovery rate (FDR) correction was made across the different disease entities. In cohort 2, it was made across the metabolites and affected area for Trp high and Trp low groups together (i.e., across 36 p-values). To compare variation in Trp values between patients undergoing therapy escalation, the standard deviation of Trp and CRP was calculated at the patient level for patients with 3 or greater observations. Differences between groups were assessed using the Wilcoxon Rank Sum test. Analyses were performed with the statistical software R (version 4.2.1).¹¹ R packages lme4 (version 3.1-3)¹² and lmerTest (version 1.1-30)¹³ were used to fit the linear mixed models, and ppcor for partial Spearman correlations (version 1.1).¹⁴ For cohort 3, linear mixed models were built using R package nlme¹⁵ so that an autoregressive covariance structure could be specified, since lme4 lacks this functionality.

Role of funders

The funders did not contribute to study design, data analysis, interpretation or writing of this article.

Results

Model selection

We tested for an association between Trp, age, and sex in the healthy population using linear models, where log₁₀-transformed Trp was the independent variable, and sex and age were the dependent variables. We found no relationship for age (95% CI: −0.00076 to 0.00054, p = 0.74) but, in agreement with existing literature,¹⁰ a significant increase in Trp for males relative to females (95% CI: 0.029–0.067, p < 0.0001). Thus, age was not included as a covariate in our models.

Widespread Trp reduction in CIDs in response to systemic inflammation

In cohort 1 (Fig. 1a), we observed that serum Trp was reduced in 9 of the 13 CIDs tested, with no detectable reduction in serum Trp in patients with CTD, AxSpA, PsA or Pso (Fig. 1b, Supplementary Table S4). Naturally, we anticipated a potential influence of disease activity on the observed Trp patterns, so we used the established systemic inflammation marker CRP as a proxy for disease activity in order to identify the disease entities that have a persistent Trp catabolic footprint even during disease quiescence. Using a cut-off value of 5 mg/L to designate inactive disease, we identified serum Trp reduction in a smaller selection of CIDs even during periods of biochemical disease inactivity: specifically, in CD, UC, RApos, SjS, and SLE (Fig. 1c, Supplementary Table S5). To assess the relationship between disease activity and Trp levels, we analysed the association between CRP and Trp using linear mixed models for each CID. We observed significant negative relationships between serum CRP and Trp in most of the CIDs studied, except for CTD and SjS, which showed non-significant negative trends (Fig. 1d, Supplementary Table S6). In summary, our findings suggest that decreased Trp levels are associated with systemic inflammatory activity across various CIDs, which is consistent with the previously described role of Trp in inflammatory processes.⁵ Furthermore, excessive serum Trp reduction in the absence of overt systemic inflammation is a feature of a smaller subset of CIDs. These findings suggest that monitoring of Trp metabolism might be a sensitive approach to assess subtle levels of inflammation in CIDs.

Fig. 1.

Reduced serum Trp across CIDs. a, Cohort consists of 13 different CIDs plus a healthy reference group. b, Assessment of serum Trp levels across all available observations. Here and in the next panel, asterisks and p-values indicate significance in comparison with controls. c, Serum Trp levels of patients with low disease activity, defined as CRP <5 mg/L. d, Linear mixed models were used to assess the relationship between CRP and Trp for each CID. Estimates depict the estimated change to CRP associated with a one-unit change in Trp (both variables were log₁₀ transformed). e, Exemplary Trp trajectories for two patients; one requiring ADT escalation and one who did not. f, g Within-patient standard deviations in Trp and CRP levels are higher for patients undergoing therapy escalation compared to those who did not. Significance annotation: ∗ FDR <0.05, ∗∗ FDR <0.01, ∗∗∗ FDR <0.001. For box-and-whisker plots, whiskers extend to 1.5 times the interquartile range; all other points are plotted individually as outliers.

Trp fluctuations correlate with long-term disease course

We further hypothesised that disrupted Trp metabolism in CID is associated with the long-term disease course. Critical transition theory posits that increased variance is an indicator of a shift in previously stable systems.¹⁶ This holds true for diabetes, where long-term glycaemic variability is associated with cardiovascular complications.¹⁷ We adapted this concept for Trp metabolism to investigate whether Trp variability indicates complicated disease course in CIDs. Targeted immunomodulatory therapies, such as antibodies against cytokines or Janus kinase (JAK) inhibitors, hereafter referred to as advanced drug therapy (ADT), are commonly used as second-line treatment upon failure of conventional medications (e.g., disease-modifying antirheumatic drugs, topical agents, corticosteroids, and aminosalicylates). Therefore, patients not requiring ADT during the observation period are expected to experience less aggressive disease courses compared to patients who undergo therapy escalation. In line with our hypothesis, we observed that patients requiring ADT escalation had higher variation in Trp and CRP levels compared to patients not requiring escalation (Fig.1 e–f). Thus, in line with critical transition theory, CRP and Trp variability may indicate phase shifts toward more complicated diseases courses, suggesting possible utility for these metrics in a predictive context.

Trp levels track with select disease activity indices

Next, we further leveraged cohort 1 by assessing links between serum Trp levels and available clinical disease activity indices. Consistent with previous reports,⁷ we confirmed a negative association between Crohn's disease activity index (CDAI) and total Mayo scores in CD and UC (n = 87 and 96, respectively, Fig. 2a and b, Supplementary Table S7). We further analysed the association between serum Trp and the Bath Ankylosing Spondylitis Disease Activity Index (BASDAI, n = 57 AxSpA patients), Disease Activity Score 28-CRP (DAS28, n = 59 RAneg and 134 RApos patients) and Psoriasis Area and Severity Index (PASI, n = 22). We identified significant negative associations between BASDAI and Trp (Fig. 2c), as well as between DAS28 and Trp for RAneg (Fig. 2d) but, most surprisingly, not for RApos (Fig. 2e). There was no clear relationship between PASI and Trp (Fig. 2f). These differences in the strength of the association between Trp and disease activity indices could potentially indicate systemic differences in immune network perturbation. It is also possible that the total inflammatory burden was not high enough to reveal an association between PASI, which solely evaluates skin symptoms, and Trp.

Fig. 2.

Association between Trp and different disease activity indices in cohort 1. a-c, Negative relationships between Trp and CDAI, total Mayo, and BASDAI for CD, UC, and AxSpA, respectively. d and e, The relationship between DAS28 and Trp in RA varies by disease entity, with a detectable negative association between DAS28 and Trp in patients with RAneg but not RApos. f, No significant association between Trp and PASI was detected in patients with Pso. All available observations are included in each plot, and the regression line is depicted in black. LMM: Linear mixed model, n = total number of patients. Shaded areas represent the 95% confidence intervals of the plotted data. Confidence intervals from LMM: CDAI: −220 to −55, Total Mayo: −13 to −6.1, BASDAI: −5.5 to −0.79, DAS28 (RAneg): −1.2 to −0.25, DAS28 (RApos): −0.49 to 0.18, PASI: −26 to 28. DAS28 was log₁₀-transformed in the LMM.

Sex-specific Trp differences in CIDs

To delve more deeply into differences between males and females, we performed sex-stratified analysis for serum Trp and CRP in cohort 1. While female CID patients have lower serum Trp than males (p < 0.0001, 95% CI: 0.024–0.040, Supplementary Fig. S1A), we did not find differences in CRP levels between the sexes (p = 0.21, 95% CI: −0.017 to 0.077, Fig. 1b). We observed the most differences between males and females for overall Trp values within the disease entities (Supplemental Fig. S1C and D, Supplementary Tables S8 and S9). Strikingly, these differences tended to disappear when examining the relationship between Trp and CRP (Supplementary Fig. S1E and F, Supplementary Tables S10 and S11). Another notable difference in the sex-stratified analysis is the different relationships between the disease activity indices and Trp (Supplementary Fig. S2A–E, Supplementary Tables S12–S14). We initially performed these tests by stratifying the sexes into two separate analyses and noted that the negative relationship between DAS28 and Trp for RAneg seems to be driven primarily by female patients, as there was no obvious trend for male patients (Supplementary Fig. S3C). Conversely, the relationship between CDAI and Trp was apparent for male patients with CD, but not for female patients (Supplementary Fig. S3D). We directly compared the slopes of these Trp-disease activity relationships between males and females by building another model where we assess influence of the interaction between sex and disease activity on Trp levels and found evidence of significant differences only for CDAI (95% CI: −340 to −17 for males relative to females, p = 0.032, Supplementary Table S14).

Breakdown of correlations between Trp and other amino acids

To identify the primary driver of Trp reduction, we introduced a second cohort (cohort 2, Fig. 3a) for targeted metabolomics of stool and serum in CID patients. We considered four potential contributors to serum Trp levels: (1) dietary protein uptake, (2) protein synthesis, (3) targeted Trp degradation (e.g., via the kynurenine pathway), and (4) microbial influences via dietary Trp degradation or de novo Trp production. We found no evidence of differing faecal Trp levels in patients with high versus low serum Trp levels, neither at the level of the affected organ group, nor when the diseases were combined (Fig. 3b), suggesting no longstanding microbial influence on Trp levels. However, it must be acknowledged that these stool samples were not obtained on the date of serum sampling (mean difference −31.3 days) so we cannot rule out short term changes to faecal Trp levels. Thus, we recruited a third cohort, (cohort 3), for which the stool and serum samples were sampled with only a mean difference of −0.98 days. Here, we found no evidence for a relationship between serum and stool Trp levels, further suggesting a strong influence of catabolic processes on serum Trp levels (Supplementary Fig. S3). Considering other potential drivers of Trp depletion, in cases of dietary protein restriction, reduced uptake or increased protein synthesis, Trp should be reduced along with all or specific proteogenic amino acids. Conversely, under excessive Trp catabolism, we expect Trp levels to be largely independent from other amino acid levels since they are not required for Trp-specific catabolic processes. Thus, we correlated Trp with 18 other proteogenic amino acids within three groups: healthy individuals, and CID patients with low and high serum Trp. The underlying concept is that amino acid correlation networks are expected to exhibit a high level of connectivity, except under the pressure of extreme catabolic activity. In such cases, a breakdown in the network is anticipated to occur at the node corresponding to the responsible amino acid.

Fig. 3.

Lower serum Trp levels accompany kynurenine-pathway activation. a, Cohort 2 overview: serum samples from 198 patients with CIDs plus 179 controls were selected for mass spectrometry-based targeted metabolomics. Additionally, 164 stool samples from patients with CIDs were sent for metabolomics analysis. Patients were selected for cohort 2 based on available HPLC-based Trp measurements (below 45 μM and above 60 μM). Mass spectrometry-based serum Trp measurements are included here as an overview. b, Faecal Trp levels in high and low serum Trp groups. c, Partial Spearman's correlations between Trp and the other measured proteogenic amino acids. Correlations with FDR <0.05 are circled in black, with increasing circle size denoting decreasing FDR via calculation of -log₁₀(FDR). d-i, Serum levels of selected Trp derivatives and the ratio of kynurenine-to-Trp (Kyn:Trp). Diamonds indicate means and comparisons are made against the control group. Significance annotation: ∗ FDR <0.05, ∗∗ FDR <0.01, ∗∗∗ FDR <0.001.

We observed strong, positive correlations between Trp and 16 of 18 amino acids in healthy individuals, suggesting that under physiological conditions, serum Trp concentration fluctuates with dietary levels, general amino acid uptake and/or protein synthesis (Fig. 2c). CID patients with high serum Trp had notably intact correlation networks compared to CID patients with low serum Trp (15 versus 0 amino acid correlations with Trp, respectively) (Fig. 2c). Increased correlations in patients with high Trp compared to patients with low Trp may imply that elevated Trp levels resulted from increased intestinal amino acid intake/uptake or decreased protein synthesis. Moreover, none of the amino acids correlated with Trp in CID patients with low serum Trp. This implies an increased influence of Trp catabolism on serum Trp levels in low Trp CID patients relative to the influence of diet or protein synthesis. It also underlines a unique role of the catabolism of Trp, among other amino acids, in the context of systemic inflammation. Consistent with this theory, patients with high Trp values had increased levels of 13 amino acids relative to the healthy population, suggesting that in CIDs, extremely high serum Trp arises due to differences in protein intake or use. Conversely, patients with low serum Trp have more variation in amino acid profiles, with half of amino acids either higher or not significantly different compared to healthy individuals (n = 2 and 7 amino acids, respectively, Supplementary Fig. S4). Extending the correlation networks to all measured proteogenic amino acids reflects this volatility, with notably intact correlation networks for healthy individuals and patients with high Trp, and less overall correlations between amino acids for patients with low Trp (Supplementary Fig. S5).

Next, we quantified representative metabolites from the three main Trp catabolic routes: the kynurenine pathway, serotonin pathway and indolic compound production.^2,18 We focus our discussion on Trp derivatives that are higher in patients with low Trp relative to healthy individuals since they are more likely to convey information about the fate of Trp under extreme catabolism rather than diet-specific information. The implicit assumption here is that metabolites accumulate when Trp is catabolized along the respective pathways, providing information about which pathways may actively degrade Trp.

Trp degradation via the kynurenine pathway results in the production of a variety of immunomodulatory metabolites.³ The rate-limiting step of this pathway is the production of kynurenine itself, thus, Kyn:Trp is often used as a proxy for the pathway's activity.¹⁹ Our analysis revealed that Kyn:Trp is highest in low Trp patients compared to healthy individuals across all CID groups (Fig. 3d). We further assessed the levels of kynurenine alone, which were elevated in all disease entities when Trp was high and, strikingly, even when Trp levels were low in MSK CID (Fig. 3e). Hence, our data strongly indicate excessive kynurenine pathway activity as a prevailing feature of CIDs.

Out of the four other Trp derivatives measured (indoxylsulfate (IndSO4), indole-3-acetic acid (IAA), indole propionic acid (IPA), and serotonin), IndSO4 was the only derivative with increased levels in patients with low Trp CID relative to healthy individuals (Fig. 3f–i). On the contrary, relative to healthy individuals, IPA and serotonin were reduced in across CID in patients with low and high serum Trp (Fig. 3g and i). We considered this is a possible indication that there is a general decrease in Trp metabolism along these catabolic routes, potentially because of increased kynurenine pathway demand, however, we did not find evidence for a negative correlation between serotonin and Trp or IPA and Trp in patients with CIDs or in the healthy reference population (Supplementary Table S16). While the correlation coefficients were close to zero in all other models, a trend towards a positive correlation between Trp and IPA was observed for patients with low serum Trp (ρ = 0.23), but it was not significant after multiple hypothesis correction adjustment (p_raw = 0.042, FDR = 0.13).

Discussion

CIDs share several pathophysiological features beyond mere tissue inflammation and destruction at sites of anatomical manifestation, e.g., chronic kidney damage or fatigue.^20,21 In this comprehensive analysis, we tested the hypotheses that Trp reduction is common in CIDs and that this reduction arises primarily due to increased kynurenine pathway activity. In doing so, we identified a consistent negative association between serum Trp and CRP levels. In light of our targeted metabolomics analysis, these findings suggest that diversion of Trp metabolism towards the kynurenine pathway is a pervasive feature of acute flares of chronic inflammation. Increased Trp degradation via this pathway has been observed in numerous cell types as a result of inflammatory cytokine signalling (e.g., interferon-γ and transforming growth factor-β).^22,23 In several cancer types, upregulation of the kynurenine pathway is generally considered a hallmark of immune evasion as it may suppress anti-tumour immunity.²² This suggests that Trp catabolism through kynurenine could be upregulated in CIDs due to aberrant cytokine signalling without direct proinflammatory consequences or even as a futile attempt to down-regulate excessive immune signalling. Therefore, consideration of possible pathological connections between excessive kynurenine pathway activity in CIDs needs to be weighed against the fact that this pathway is also active in immunosuppressive environments (i.e., cancer) and that some Trp derivatives have demonstrated anti-inflammatory properties.²⁴ Harden et al. proposed a model that would explain this apparent dichotomy.⁵ The model contrasts the enzymes involved in the first step of the kynurenine pathway (indoleamine 2,3-dioxigenase (IDO) and tryptophan dioxygenase (TDO)) with downstream kynureninase expression. Kynureninase outweighs combined expression of IDO and TDO in CIDs, while the opposite pattern occurs in cancer. Hence, CIDs should have an enrichment of downstream proinflammatory metabolites with a relative paucity of upstream anti-inflammatory metabolites. We have extended our findings to include more CIDs, but the predominance of kynureninase activity over IDO-TDO could explain the generalizability of low Trp across the CIDs observed here and is suggestive that under the right conditions, excessive kynurenine pathway activity may contribute to proinflammatory environments.

We have additionally unveiled sex-specific patterns in Trp levels when comparing CIDs to the general population. We found low Trp levels for female patients in only 4 of the 13 CIDs tested. For male CID patients, this number rose to 8. Discerning whether the underlying cause stems from biological factors, social influences, or a combination of both, is challenging with our existing data. However, the average CRP levels do not suggest that the systemic inflammatory load was higher between male and female patients in the cohort. Intriguingly, the overall patterns in the relationships between Trp and CRP were relatively consistent between the sexes, albeit with differing effect sizes. Taken together, we might infer that Trp drops more quickly in males than in females upon increases in systemic inflammation. This suggests a biological underpinning to the observed patterns and would be interesting to investigate in longitudinal studies. Another notable difference in the sex-stratified analysis is the difference between the association between Trp and DAS28 and CDAI. While a detectable, negative relationship between DAS28 and Trp was found for female patients with RAneg, this was not the case for male patients. Conversely, the negative relationship with CDAI was only detected for male patients, not females. Notably, unlike the DAS28, we found statistically significant differences in the interaction between sex and CDAI, which is suggestive that these sex-based differences are particularly robust. The DAS28 calculation includes CRP, number of swollen and tender joints as well as a score to indicate the patient's global health, and the CDAI is also calculated with a mix of subjective and more objective measures of disease activity. Overall, our results suggest that there are gender and/or sex-specific influences on the relationship between Trp and disease activity. This underscores the significance of incorporating gender and sex considerations for investigations that aim to understand the mechanistic underpinnings of these associations.

We were further able to identify a subgroup of disease entities (CD, UC, RApos, SjS, SLE) with reduced Trp levels even in absence of systemic inflammation (CRP <5 mg/L). As our data indicate that Trp reduction results, in large part, due to enzymatic degradation, these disease entities may have distinct inflammatory metabotypes in which subtle Trp degradation is an initial causal pathophysiological event. Interestingly, IDO is expressed in a Janus kinase-signal transducer and activator of transcription (JAK-STAT) dependent manner.²⁵ Overactivation of the JAK-STAT pathway is present in some CIDs26, 27, 28 and inhibition of this pathway is therapeutically employed using pan-JAK or JAK1-selective inhibitors in CD,²⁹ UC,³⁰ RA,^26,31 and SLE.²⁸ Hence, our data suggest that monitoring of Trp metabolism might not only be a sensitive and precise biomarker of subtle inflammation but could also facilitate a more targeted therapeutic approach in CIDs.

We also found increased IndSO4 in patients with GI and MSK CIDs with low Trp. Inflammatory effects of IndSO4 have been investigated in chronic kidney disease, where it disrupts the intestinal barrier and contributes to inflammation via activation of the nuclear factor kappa B signalling pathway.³² Of note, decreased kidney function is a potential complication of GI and MSK CIDs.^33,34 Two other Trp derivatives displayed notable trends relative to the healthy population tested: serotonin and IPA. Studies on serum serotonin in CIDs have reported conflicting results, e.g., in one study, patients with RA had lower serum serotonin relative to the healthy population,³⁵ while another reported higher levels.³⁶ Administration of anti-tumour necrosis factor drugs do reportedly lower serum serotonin levels,³⁷ which may be contributing to the reported differences.^35,38,39 IPA has anti-inflammatory effects that are mediated via aryl hydrocarbon receptor (AHR) signalling and its serum levels are reportedly reduced in UC in a disease activity-dependent manner.^40,41 We identified IPA reduction across CIDs, indicating a paucity of AHR signalling as a potential joint feature of CIDs, which, based on the correlation results between Trp and these two metabolites, does not appear to be strongly linked to the low Trp phenotype.

The disrupted correlation network between Trp and other amino acids suggests that reduced Trp levels are driven in large part by Trp catabolism. However, this does not rule out contributions by alternative mechanisms. For example, decreased Trp uptake could also play a role. Though we did not find specific evidence for increased levels of Trp in the stool of patients with low serum Trp, it is difficult to rule out this possibility owing to the compounding issues of individual differences in Trp uptake rates, intestinal barrier permeability, diet, and stool and serum sampling times. In fact, we have previously found reduced expression of the neutral amino acid transporter SLC6A19 in intestinal biopsies of IBD patients with active disease in comparison to healthy controls.⁷ This transporter has 12 reported substrates, including Trp.⁴² Because the Trp correlation network was specifically impacted in patients with CIDs and low Trp and did not seem to influence the other 11 substrates, we do not believe that reduced intake alone is responsible for the findings reported here. In addition to SLC6A19, transport of Trp from the gut lumen can also be accomplished by the peptide transporter SLC15A1, and serum levels could be influenced further if there are changes to basolateral transport (i.e., SLC7A5/SLC3A2, SLC7A8/SLC3A2, and SLC16A10).⁴³ Finally, microbial Trp metabolism may influence human Trp metabolites, and direct influences of microbially Trp-derived metabolites on host physiology has already been described in the literature. For instance, indole-3-ethanol, indole-3-pyruvate and indole-3-aldehyde levels have been demonstrated to influence intestinal barrier function,⁴⁴ and tryptamine seems to influence intestinal secretion, thereby accelerating intestinal transit time,⁴⁵ which could, in turn, influence amino acid uptake.

While the Trp correlation networks indicate a particularly severe dysregulation of Trp metabolism in patients with CIDs and low Trp, our findings also suggest that low serum Trp levels in these patients are accompanied by low levels of 9 other amino acids (asparagine, histidine, isoleucine, leucine, lysine, methionine, threonine, tyrosine and valine). This observation raises the intriguing possibility that inflammation-related amino acid metabolic dysregulation extends beyond Trp. The extent to which amino acid catabolism, protein synthesis and turnover, or amino acid uptake influences these levels will require further study. However, it is notable that 7 of the 9 decreased amino acids are essential to the human diet and can be metabolically converted to non-essential amino acids. Decreased serum essential amino acid levels may have arisen because of increased protein synthesis requirements, which have previously been noted for certain organs upon experimentally induced colitis in rats.⁴⁶ Coupling amino acid conversion with increased protein synthesis, the overall demand for the essential amino acids is likely quite high under inflammatory conditions. This theory is consistent with the amino acid correlation networks in patients with low Trp (Supplementary Fig. S5), where correlations between the proteogenic amino acids are notably less connected than those for the reference population and patients with high Trp.

Important limitations of our study include the fact that serum Trp measurement was conducted in a cross-sectional cohort, so we cannot draw conclusions about metabolic dynamics that arise before or after successful treatment in individual patients. Furthermore, we cannot exclude the possibility that unmeasured variables could have confounded our analysis. For example, race/ethnicity, body mass index, socioeconomic status, stress and other psychological states, or certain medications may influence Trp intake and/or metabolism. In fact, stress has previously been linked to increased kynurenine pathway activity,⁴⁷ and we might expect that patients with CIDs are more likely experience psychological stress. Lastly, we did not include age as a covariate in our models because some CIDs are enriched for patients of older age than the reference population and this obscures the relationship between some of the CIDs tested and Trp. We'd also like to recognise that the two sex categories analysed here do not adequately represent the lived experiences of the broader population.

Despite the evidence from our targeted metabolomics analyses indicating Trp degradation along the kynurenine pathway, the mere accumulation of kynurenine relative to Trp (Kyn:Trp) does not provide a global overview of kynurenine pathway activity. Hence, dynamic measurement of Trp metabolism (e.g., in-vivo Trp fluxomics⁴⁸) should be considered as a diagnostic tool in patients with CIDs and to confirm the contribution of the kynurenine pathway to the low Trp phenotype. This approach would not only enable the assessment of individual levels of Trp degradation, but also shed light on the metabolic fate of Trp, which could ultimately provide support for future studies to disentangle correlation from causation. The approach would additionally help overcome the lack of detailed dietary information, which also critically impacts host Trp levels and cannot be completely disentangled within cross-sectional cohorts. Beyond the Trp derivatives included in this study, there are additional metabolites within the kynurenine pathway, as well as indolic compounds that could offer valuable insights into the pathology and comorbidities of CIDs. For instance, Trp metabolism is disrupted in chronic fatigue syndrome and depression,^49,50 and fatigue and depression are common comorbidities of CIDs.⁵¹ As mentioned above, interferon-γ induces IDO expression and thereby Trp degradation. Another inflammatory marker induced by the same cytokine is neopterin.⁵⁰ It would be interesting to examine the overall predictive power of a combination of Trp, CRP and neopterin on disease prognosis in CIDs.

Taken together, our results indicate that excessive kynurenine pathway activity occurs across CIDs, but the extent and circumstances leading to activation of the pathway may have disease-specific aspects. We therefore suggest that a detailed exploration of Trp turnover in patients with CIDs from diverse demographic backgrounds might provide deeper insights into the molecular origins of disease pathophysiology and could also lead to therapeutic concepts for targeted therapies.

Contributors

DH, SWa, KA, PR, SSchr contributed the study concept.

DH, SSz, JL, SWa designed the analysis approaches.

DH, SSz, SWa performed the statistical analyses.

DH prepared the visualisations and drafted the manuscript.

SSchu, NE, RJ, CC, DT, MO, AF generated metabolomics data.

FT, LW, HG, MS, JL, HZ, SWe, AF, RF, BH contributed to data curation and verification.

SSchr, PR, SWa, KA, DH acquired study funding.

SSz, BH, SSchr, PR, SWa, KA revised the manuscript.

All authors read and approved the final manuscript.

Data sharing statement

The data and code used to generate these findings are available upon request and following approval of a data transfer agreement. Requests should be made to the corresponding author.

Declaration of interests

The authors declare no competing interests related to the conception and results of this study. Financial disclosures: F.T. received speaker's fees from Abbvie, Bristol-Myers-Squibb, Celltrion Healthcare, Dr Falk Pharma, Eli Lilly, Ferring Pharmaceuticals, Janssen, and funding from Sanofi/Regeneron. K.A. received speaker's fees from Abbvie, Dr Falk Pharma, Eli Lilly, Janssen, Takeda and Galapagos. S.S. received consulting fees from AbbVie, Amgen, Arena, Boehringer Ingelheim, Bristol-Myers Squibb, Celltrion Healthcare, Dr Falk Pharma, Eli Lilly, Ferring Pharmaceuticals, Galapagos/Gilead Sciences, Genentech/Roche, GlaxoSmithKline, IMAB Biopharma, MSD, Pfizer, Shire, and Takeda. M.S. received speaker's fees from AbbVie, Affibody AB, Akari Therapeutics Plc, Almirall-Hermal, Amgen, Argenx BV, AstraZeneca AB, Biogen Idec, Bioskin, Bristol-Myers Squibb, Boehringer-Ingelheim, Celgene, Dermira, Eli Lilly, Foamix, Forward Pharma, Galderma, Hexal AG, Incyte Inc., Janssen-Cilag, Johnson & Johnson, Klinge Pharma, Kymab, Leo Pharma, Medac, MSD, Novartis, Pfizer, Regeneron Pharmaceutical, Sandoz Biopharmaceuticals, Sanofi-Aventis, Trevi Therapeutics, UCB Pharma. S.We. received research grants from Leo Pharma, Pfizer, and Sanofi and consulting fees from AbbVie, Almirall, Boehringer, Eli Lilly, Galderma, Pfizer, Sanofi, and Regeneron, honoraria for lectures or presentations from AbbVie, Almirall, Eli Lilly, Pfizer, Sanofi, Regeneron, and support for attending meetings from AbbVie and Sanofi. M.S. received speaker's fees from AbbVie, Almirall-Hermal, Eli Lilly, Sanofi-Aventis and travel support from AbbVie, Almirall-Hermal, Eli Lilly, Sanofi-Aventis and Amgen. D.H. received a travel grant from United European Gastroenterology. H.G. received conference travel support from Abbvie. D.T. received research grants from AbbVie, and Novartis, consulting fees from AbbVie, Almirall, Celltrion, Eli Lilly, Janssen-Cilag, Novartis, Samsung and UCB, honoraria for lectures and presentations from AbbVie, Amgen, Biogen, Eli Lilly, New-Bridge, Novartis, Samsung, Sandoz and UCB, and participates on data safety monitoring or advisory boards for AbbVie, Eli Lilly, Novartis and UCB. All other authors have no conflicts of interest to declare.