Development of a tryptophan metabolomics-based diagnostic model for hepatic encephalopathy in liver cirrhosis

Abstract

Background

Tryptophan (TRP) metabolism is implicated in the pathogenesis of hepatic encephalopathy (HE). Abnormal TRP metabolites correlate with HE severity and may represent potential biomarkers. This study employed targeted metabolomics to characterize serum TRP metabolic profiles in patients with liver cirrhosis (LC) at different stages. Furthermore, a diagnostic model integrating clinical indicators was developed to facilitate the early detection of hepatic encephalopathy in LC patients.

Method

Ninety LC patients (25 without HE, 30 with covert HE, and 35 with overt HE) and 50 healthy controls were enrolled from January 2023 to December 2024. Serum TRP metabolites were quantified using targeted liquid chromatography–mass spectrometry (LC–MS). A diagnostic model was developed using logistic regression by integrating differential metabolites with clinical indicators. Diagnostic accuracy and utility were evaluated through the area under the curve (AUC), bootstrap validation, calibration curves, and decision curve analysis (DCA).

Result

Compared with healthy controls, cirrhotic patients exhibited significant alterations in TRP metabolites. TRP and serotonin (SER) were consistently reduced (P < 0.001), whereas quinolinic acid, indole-3-lactic acid, 3-hydroxykynurenine, indole-3-acetic acid, and indole-3-carboxaldehyde were markedly elevated (P < 0.01). Between Non-HE LC and HE patients, only SER differed significantly, with lower levels in the HE group (P < 0.001). No significant metabolic differences were observed between the CHE and OHE groups. The diagnostic model integrating blood ammonia and SER achieved strong discriminatory performance, with an AUC of 0.902, sensitivity of 76.9%, and specificity of 88.0%. Bootstrap validation confirmed model robustness (AUC = 0.902, 95% CI: 0.839–0.965). Calibration curves demonstrated excellent agreement between predicted and observed outcomes, while decision curve analysis indicated substantial net clinical benefit.

Conclusion

Targeted TRP metabolomics distinguishes disease stages in cirrhotic patients, while a diagnostic model integrating blood ammonia and SER provides a reliable tool for early detection and monitoring of hepatic encephalopathy.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40001-026-03863-6.

Introduction

Hepatic encephalopathy (HE) is a serious neuropsychiatric complication of end-stage liver disease [1, 2]. Research shows that at least 30% of patients with cirrhosis may experience HE, and once HE occurs, the 1-year survival rate is 42% and the 3-year survival rate is 23% [3, 4]. Based on clinical severity, HE is classified into covert hepatic encephalopathy (CHE) and overt hepatic encephalopathy (OHE) [5, 6]. It is reported that among patients with liver cirrhosis who are tested, the incidence rate of CHE is 30–85% [7], the incidence rate of OHE is estimated to be as high as 30–40%, and the risk of OHE occurring for the first time within 5 years is as high as 5% ~ 25% [8]. Due to the lack of specific clinical manifestations of CHE, its diagnosis can often only be detected through psychological measurement evaluation or electrophysiological and other functional brain tests, and the results may not be accurate [9, 10]. Electroencephalography examination is also difficult to diagnose and misdiagnosed due to its complexity, high cost, and low sensitivity [11]. OHE often has triggering factors, and after excluding other neurological disorders such as acute cerebrovascular accidents, alcohol-related problems, and other forms of metabolic encephalopathy, combined with medical history and obvious clinical manifestations, it can be diagnosed [12]. However, there is currently no universally accepted “gold standard” laboratory marker for the diagnosis of OHE [6]. Although hyperammonemia is frequently observed in population-level studies of OHE, its diagnostic value in individual patients is limited due to confounding factors such as age, sex, diet, physical activity, and comorbidities like renal failure that can also elevate ammonia levels [13, 14]. Therefore, blood ammonia is not routinely used as a definitive diagnostic marker [13].

The progression of chronic liver disease to HE involves complex and multifactorial pathological mechanisms [15–18]. Studies have shown that metabolic disorders of tryptophan (TRP) play a key role in the pathogenesis of HE [19, 20]. As an essential amino acid, TRP and its metabolites—such as kynurenine and serotonin—may contribute directly to the neurological dysfunction in HE by modulating neurotransmitter balance, blood–brain barrier integrity, and neuroinflammatory responses [19, 21]. This study aims to use targeted metabolomics technology to systematically analyze the serum TRP metabolic profile characteristics of patients with different degrees of LC and concurrent HE, and screen for differential metabolites related to disease progression. Further combining clinical indicators such as blood ammonia (BA) and prothrombin activity, a multi-factor risk prediction model is constructed to address the limitations of single biomarkers and insufficient sensitivity in traditional HE diagnosis, and to provide a new metabolic clinical combined prediction tool for early identification and personalized intervention of HE.

Materials and methods

2.1 Research object.

Patients with chronic liver disease, including varying degrees of liver cirrhosis (LC) with or without concomitant hepatic encephalopathy (HE), who were hospitalized in the Departments of Gastroenterology, Infectious Diseases, and Hepatobiliary Surgery at our institution between January 2023 and December 2024 were enrolled in this study. Exclusion criteria were as follows: severe comorbidities affecting the heart, brain, or lungs; a history of psychiatric disorders; neurological or psychiatric conditions secondary to metabolic disturbances, toxins, or intracranial infections; and incomplete medical records. A total of 140 blood samples were collected, comprising 50 healthy controls (NC), 25 patients with liver cirrhosis without HE (Non-HE LC), 30 patients with covert HE (CHE), and 35 patients with overt HE (OHE). Healthy controls underwent routine health examinations at our hospital during the same period. This study was reviewed and approved by the Ethics Committee of our institution (approval number: 2024-Ky-233). The diagnostic criteria for LC and HE were based on the latest clinical guidelines for the diagnosis and treatment of liver cirrhosis and hepatic encephalopathy issued by the Hepatology Branch of the Chinese Medical Association [22]. The diagnosis and classification of hepatic encephalopathy were performed according to the following procedure. Patients were first evaluated using the West Haven criteria. Those with a West Haven grade ≥ II were diagnosed with overt hepatic encephalopathy. Patients with a West Haven grade of 0–I underwent further neuropsychological assessment using the Psychometric Hepatic Encephalopathy Scale (PHES). A PHES score ≤  − 4 was defined as covert hepatic encephalopathy.

Collection of clinical indicators

Clinical laboratory indicators were collected on the day of admission and mainly included alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bile acid (TBA), total bilirubin (TBIL), direct bilirubin (DBIL), indirect bilirubin (IBIL), serum albumin (ALB), alkaline phosphatase (ALP), blood ammonia (BA), gamma-glutamyl transpeptidase (GGT), white blood cell count (WBC), platelet count (PLT), neutrophil percentage (NEUT%), C-reactive protein (CRP), and prothrombin activity (PTA).Within 24 h of hospital admission, 5 mL of fasting peripheral venous blood was collected from each subject. A portion of the collected blood was centrifuged at 2000–3000 rpm for 15–20 min at room temperature, and the supernatant (serum) was collected. Immediately the frozen tube was stored in a −80 ℃ freezer for TRP metabolomics testing.

LC–MS detection of serum tryptophan and its metabolites

Serum tryptophan and its related metabolites were analyzed using a targeted metabolomics approach, including metabolites involved in the kynurenine and serotonin pathways. Targeted tryptophan metabolomic analysis was performed using an ultra-high-performance liquid chromatography system (ExionLC™, AB Sciex, Foster City, CA, USA) coupled with a triple quadrupole mass spectrometer (QTRAP 6500 Plus, AB Sciex, Foster City, CA, USA). Serum samples were precipitated with 80% methanol and spiked with the internal standard Trp-d5. After centrifugation, the resulting supernatant was subjected to LC–MS analysis. Chromatographic separation was performed on an ACQUITY UPLC HSS T3 column (2.1 × 150 mm, 1.8 μm). The mobile phase consisted of 0.1% formic acid in water (A) and 0.1% formic acid in methanol (B), delivered using a gradient elution program. Column temperature was maintained at 40 °C, the flow rate at 0.25 mL/min, and the injection volume was 5 μL. Mass spectrometric detection was conducted in positive electrospray ionization (ESI) mode with multiple reaction monitoring (MRM) for quantitative analysis. Raw data were processed using XCMS software for peak identification, alignment, and filtering. Metabolite annotation was performed using AMDIS and NIST/Wiley databases, and normalization was conducted using the internal standard. Method validation demonstrated strong linearity with correlation coefficients (r) > 0.99 for standard curves, and relative standard deviations (RSD) < 15% for quality control samples, confirming the method’s stability and reliability. Plasma metabolite concentrations are quantitatively measured and expressed in μg/mL.

Heatmap construction and hierarchical clustering

Metabolite concentrations were log-transformed when appropriate and Z-score standardized before hierarchical clustering and heatmap visualization using pheatmap in R (version 3.6.3). No apparent batch-related clustering was detected during quality control, and thus, no batch correction was applied.

Statistical analysis

Targeted metabolomics data analysis: Multivariate data including sample name, compound name, and concentration details were analyzed using R (v3.3.2) software. To reduce noise and variance between variables, the data were normalized and log-transformed. After these adjustments, the overall heatmap was drawn, and principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were used to identify inter group or intra group differences and evaluate the reliability of the model. T-test in R (v3.3.2) software was used to screen for differential metabolites of tryptophan. Clinical parameters: SPSS version 27.0 statistical software was used. For continuous variables, an independent samples t-test was applied when assumptions of normality and homogeneity of variance were met. When data were normally distributed but variances were unequal, Welch’s t-test was used, and the results are expressed as mean ± standard deviation. If the normality assumption was violated, the Mann–Whitney U test was applied, and the results are expressed as median (P25, P75). Categorical variables are expressed as frequencies (percentages), and between-group differences were analyzed using the chi-square test. All tests were two sided, with p < 0.05 considered significant, and FDR correction was applied using the Benjamini–Hochberg method.

Predictors selected by the Lasso method were first tested with univariate logistic regression to identify clinically significant variables. To evaluate the robustness of feature selection, bootstrap resampling was performed with 1,000 iterations. Within each bootstrap sample, LASSO regression with tenfold cross-validation was applied. The selection frequency of each predictor was calculated as the proportion of bootstrap samples in which the variable retained a non-zero coefficient at the optimal penalty parameter (lambda.min). The final model was constructed using multivariate logistic regression. Model performance was assessed using the area under the ROC curve (AUC). Bootstrap internal validation with 1,000 resamples was implemented using the caret R package to assess model development and predictive performance. ROC curves were generated with the "pROC" package, and a nomogram was built with the "rms" package. Calibration curves were plotted with the "rms" and "ResourceSelection" packages. Decision curve analysis (DCA) was conducted with the "rmda" package to evaluate the model’s clinical utility.

result

Characteristics of participants

A total of 140 participants were enrolled in this study based on strict inclusion and exclusion criteria, comprising 50 healthy controls (NC) and 90 patients with liver cirrhosis (LC) at various stages. The LC cohort was further stratified into 25 patients without hepatic encephalopathy (Non-HE LC), 30 with covert hepatic encephalopathy (CHE), and 35 with overt hepatic encephalopathy (OHE) (Fig. 1). Clinical parameters were compared among the Non-HE LC, HE, and NC groups (Table 1). Results demonstrated that, relative to the NC group, the Non-HE LC group exhibited significantly elevated serum levels of AST, ALT, TBA, TBIL, DBIL, IBIL, γ-GT, ALP, BA, and CRP, accompanied by markedly reduced ALB, PLT, and PTA levels. Compared with the NC group, patients in the HE group had significantly higher levels of AST, TBA, TBIL, DBIL, IBIL, ALP, BA, CRP, and NEUT%, and markedly lower levels of WBC, ALB, PLT, and PTA. Further comparison between the HE and Non-HE LC groups revealed that patients in the HE group exhibited significantly elevated TBIL, DBIL, IBIL, TBA, γ-GT, and BA levels, along with significantly reduced ALB, WBC, PLT, and PTA levels, all of which were statistically significant.

Fig. 1.

Flow diagram of participant enrollment and study cohort selection

Table 1.

Characteristics of participants

Variables	NC group (n = 50)	Non-HE LC group (n = 25)	HE group (n = 65)	Pa	Pb	Pc
Age	59 (52.75–65.25)	59 (55–62)	61 (55–67.5)	0.783	0.06	0.215
Sex (male)	35 (70%)	19 (76%)	45 (69.2%)	0.585	0.929	0.595
ALT	15.75 (11.05–24.75)	25.2 (15.7–42.6)	19.6 (14.8–30.6)	0.013	0.045	0.229
AST	19.85 (17.6–23.25)	37 (25.15–58.15)	34.9 (27.5–58.15)	< 0.001	< 0.001	0.896
TBIL	13.8 (9.8–17.48)	24.9 (15.35–46.7)	38.9 (27.9–71.3)	< 0.001	< 0.001	0.009
DBIL	3.25 (2.78–4.03)	10.8 (4.45–24.3)	19 (11.8–34.8)	< 0.001	< 0.001	0.016
IBIL	10.65 (6.98–13.48)	12.5 (9.25–23.15)	19.78 (14–37.9)	0.018	< 0.001	0.007
TBA	2.6 (1.88–5.45)	32.2 (8.35–53.25)	78 (38.3–146.05)	< 0.001	< 0.001	< 0.001
γ-GT	23.5 (13–34)	65 (45–108.5)	30 (15–66)	< 0.001	0.036	0.001
ALP	71.95 (57.53–84.23)	98 (79–152.8)	114 (86.55–148)	< 0.001	< 0.001	0.525
ALB	47.2 (44.38–48.33)	35.5 (30.55–40.7)	31 (27.2–34.85)	< 0.001	< 0.001	0.003
BA	37.05 (32.53–47.35)	60.3 (49.15–73)	86.7 (62.45–132.05)	< 0.001	< 0.001	< 0.001
WBC	5.72 (4.14–6.98)	4.93 (3.12–6.3)	2.97 (1.98–4.22)	0.136	< 0.001	0.002
PLT	226.5 (184.75–280.5)	95 (56–132.5)	52.5 (35.5–69)	< 0.001	< 0.001	< 0.001
NEUT%	62.8 (55.05–69.43)	68.9 (60.4–75.6)	68.3 (54.7–75.85)	0.055	0.048	0.85
CRP	0.62 (0.33–1.58)	4.99 (0.9–17.32)	3.26 (0.75–11.97)	< 0.001	< 0.001	0.452
PTA	116.5 (102–123)	72 (60.5–87)	53.5 (42–70.5)	< 0.001	< 0.001	< 0.001

ALT (alanine aminotransferase), AST (aspartate aminotransferase), TBIL (total bilirubin), DBIL (direct bilirubin), IBIL (indirect bilirubin), TBA (total bile acids), γ-GT (gamma glutamyl transpeptidase), ALP (alkaline phosphatase), ALB (serum albumin), BA (blood ammonia), WBC (white blood cells), PLT (platelets), NEUT% (percentage of neutrophils), CRP (C-reactive protein), PTA (prothrombin activity). P ^a: P values of Non-HE LC group and NC group; P ^b: P values of HE group and NC group; P value for HE group and Non-HE LC group

Analysis of TRP metabolites in patients with different stages of liver cirrhosis

Serum TRP and its metabolites were quantitatively analyzed in the case and control groups using liquid chromatography–mass spectrometry (LC–MS). A total of 23 TRP metabolites were identified, including nicotinamide (NAm), nicotinic (NA), pyridine carboxylic acid (PA), indole-3-carboxaldehyde (IALd), 3-hydroxyanthranilic acid (3-HAA), tryptamine (TRM), indole-3-acetic acid (IAA), indole-3-propionic acid (IPA), kynurenic acid (KYNA), 5-hydroxyindole-3-acetic acid (5-HIAA), tryptophan (TRP), indole-3-lactic acid (ILA), kynurenine (KYN), 5-hydroxytryptophan (5-HTP), serotonin (SER), 3-hydroxykynurenine (HK), quinolinic acid (QA), xanthurenic acid (XA), indole-2-carboxylic acid (IET), α-methyltryptophan (MT), melatonin (MEL), indole-3-acetamide (IAM), and indole-3-acrylic acid (IArA). Compared with the NC group, significant alterations in the concentrations of several TRP metabolites were observed in the LC group. Specifically, TRP and SER concentrations were decreased, whereas QA, ILA, 5-HTP, HK, Indole-3-carboxaldehyde (IALd), and IAA were generally elevated. XA, NA, and PA showed elevation in some samples from both groups. MT, MEL, IAM, and IArA were elevated only in a subset of LC samples, whereas in the NC group they were predominantly below the detection threshold (Fig. 2A). Principal component analysis (PCA) demonstrated a trend toward separation between the LC and NC groups along the first and second principal components, indicating differences in the serum TRP metabolite profiles between the two groups. Orthogonal partial least squares discriminant analysis (OPLS-DA) score plots revealed a more distinct separation between groups, indicating that the model effectively differentiates patients with cirrhosis from healthy controls (Fig. 2B, C).

Fig. 2.

Cluster and multivariate statistical analyses of targeted TRP metabolites were performed between NC and LC groups. A Heatmap illustrating the overall distribution of TRP metabolites. B PCA plots comparing NC and LC groups. C OPLS-DA plots comparing NC and LC groups

Statistical significance was determined using fold change (FC ≥ 1.2 or ≤ 0.8) and p values, with a false discovery rate (FDR) ≤ 0.05. Eight differentially expressed metabolites were identified between the LC and NC groups. Among these, IALd, IAA, ILA, KYN, HK, and QA were significantly elevated in the LC group, whereas TRP and SER were significantly reduced. No significant differences were observed for NAm, NA, PA, 3-HAA, IPA, KYNA, 5-HIAA, 5-HTP, or XA (Table 2).

Table 2.

Analysis of serum TRP metabolite concentrations in LC and NC groups at different levels

Metabolite abbreviation	NC group (n = 50)	LC group (n = 90)	FC	P	FDR
NAm	0.036 ± 0.063	0.045 ± 0.041	1.250	0.362	0.473
NA	0.004 ± 0.005	0.006 ± 0.005	1.500	0.882	0.963
PA	0.005 ± 0.006	0.005 ± 0.004	1.000	0.486	0.589
IALd	0.0006 ± 0.001	0.0013 ± 0.001	2.167	0.004	0.010
3-HAA	0.003 ± 0.002	0.003 ± 0.004	1.000	0.943	0.963
IAA	0.024 ± 0.040	0.074 ± 0.103	3.083	0.001	0.004
IPA	0.118 ± 0.445	0.057 ± 0.133	0.483	0.222	0.314
KYNA	0.006 ± 0.002	0.015 ± 0.039	2.500	0.146	0.248
5-HIAA	0.008 ± 0.003	0.013 ± 0.018	1.625	0.074	0.139
TRP	10.975 ± 2.601	8.153 ± 3.869	0.743	< 0.001	< 0.001
ILA	0.096 ± 0.032	0.140 ± 0.114	1.458	0.008	0.018
KYN	0.821 ± 0.341	1.082 ± 0.616	1.318	0.006	0.016
5-HTP	0.004 ± 0.001	0.006 ± 0.006	1.500	0.202	0.311
SER	0.092 ± 0.04	0.016 ± 0.022	0.174	< 0.001	< 0.001
HK	0.005 ± 0.003	0.016 ± 0.013	3.200	< 0.001	< 0.001
QA	0.005 ± 0.012	0.098 ± 0.190	19.600	0.001	0.004
XA	0.004 ± 0.007	0.004 ± 0.006	1.000	0.963	0.963

NAm (nicotinamide), NA (niacin), PA (picolinic acid or pyridine-2-carboxylic acid*), IALD (indole-3-carboxaldehyde), 3-HAA (3-hydroxyanthranilic acid), IAA (indole-3-acetic acid), IPA (indole-3-propionic acid), KYNA (kynurenic acid), 5-HIAA (5-hydroxyindole-3-acetic acid), TRP (tryptophan), ILA (indole-3-lactic acid), KYN (kynurenine), 5-HTP (5-hydroxytryptophan), SER (serotonin), HK (3-hydroxykynurenine), QA (quinolinic acid), XA (xanthurenic acid). FC: fold change; FDR: false discovery rate

Analysis of TRP metabolites among NC, Non-HE LC, and HE groups

Compared with the NC group, significant alterations in the concentrations of several TRP metabolites were observed in both the Non-HE LC and HE groups. TRP and SER levels were decreased in both groups, whereas QA, ILA, 5-HTP, KYN, HK, IALd, and IAA were elevated in a subset of samples. In addition, TRM and IAM exhibited more pronounced increases in certain HE patients. However, the overall TRP metabolite profile in the HE group did not significantly differ from that in the Non-HE LC group (Fig. 3A). PCA revealed a trend toward separation among the Non-HE LC, HE, and NC groups along the first and second principal components, whereas no distinct separation was observed between the Non-HE LC and HE groups. PCA score plots demonstrated substantial overlap among several groups, particularly between the NC and Non-HE LC groups, as well as between the CHE and OHE groups, suggesting highly similar global metabolic profiles. These findings suggest differences in serum TRP metabolite profiles among the Non-HE LC, HE, and NC groups, whereas the differences between the Non-HE LC and HE groups were relatively minor. OPLS-DA score plots revealed a more distinct separation between the Non-HE LC and HE groups versus the NC group, but no clear separation between the Non-HE LC and HE groups. These results indicate that the OPLS-DA model effectively differentiates LC patients from healthy controls, but has limited ability to discriminate between Non-HE LC and HE patients (Fig. 3B–D). For OPLS-DA, model performance and robustness were evaluated using cumulative R2X, R2Y, and Q2 values, together with permutation testing (200 iterations), as shown in the Supplementary Materials (Supplementary File 1: Table S1 and Figures S1–S7).

Fig. 3.

Cluster and multivariate statistical analyses of targeted TRP metabolites performed across NC, Non-HE LC, and HE groups. A Heatmap depicting the overall metabolite distribution. B PCA and OPLS-DA plots comparing NC and Non-HE LC groups. C PCA and OPLS-DA plots comparing NC and HE groups. D PCA and OPLS-DA plots comparing Non-HE LC and HE groups

The results of serum TRP metabolite concentration analyses in different cirrhosis groups and healthy controls are summarized in Supplementary File 1. Nine differentially expressed metabolites were identified between the Non-HE LC and NC groups, whereas eight were identified between the HE and NC groups. Among these, IAA, 5-HIAA, ILA, HK, and QA were significantly elevated in both the Non-HE LC and HE groups compared with healthy controls, whereas TRP and SER were significantly reduced. TRM and KYN exhibited more pronounced increases in the Non-HE LC group, whereas IALd showed a more prominent elevation in the HE group. All these differences were statistically significant (Supplementary File 1: Tables S2 and S3). Furthermore, only one differential metabolite was detected between the Non-HE LC and HE groups: SER, which was significantly lower in the HE group than in the Non-HE LC group (Supplementary File 1: Table S4).

Analysis of TRP metabolites among Non-HE LC, CHE, and OHE groups

In selected samples from the Non-HE LC, CHE, and OHE groups, levels of IAM, 3-HAA, QA, ILA, KYN, 5-HTP, HK, and IAA were elevated, whereas overall TRP concentrations remained unchanged. No significant differences in TRP metabolite concentrations were observed between the CHE and OHE groups (Fig. 4A). PCA results showed no clear separation among the CHE, OHE, and Non-HE LC groups on the first and second principal component planes, suggesting minimal differences in serum TRP metabolite composition. OPLS-DA analysis revealed similarly weak separation trends among the groups. Overall, the OPLS-DA model failed to effectively distinguish Non-HE LC from HE patients or to differentiate CHE from OHE patients (Fig. 4B–D).

Fig. 4.

Cluster and multivariate statistical analyses of targeted TRP metabolites were performed across Non-HE LC, CHE and OHE groups. A Heatmap depicting the overall metabolite distribution. B PCA and OPLS-DA plots comparing Non-HE LC and CHE groups. C PCA and OPLS-DA plots comparing Non-HE LC and OHE groups. D PCA and OPLS-DA plots comparing CHE and OHE groups

The detailed results of TRP metabolite concentrations in the Non-HE LC, CHE, and OHE groups are provided in Supplementary File 1. Compared with the Non-HE LC group, only one differential metabolite, SER, was identified in both the CHE and OHE groups, with significantly lower concentrations (Supplementary File 1: Tables S5 and S6). SER levels in the OHE group were lower than in the CHE group, but the difference was not statistically significant. The remaining 16 metabolites (e.g., 3-HAA, IPA, KYNA, 5-HIAA, TRP, ILA, KYN, and HK) showed no significant differences among the three groups (Supplementary File 1: Tables S5 and S6). Furthermore, no significant differences in TRP or its metabolites were found between the CHE and OHE groups (Supplementary File 1: Table S7).

Construction and validation of diagnostic models

To minimize the impact of multicollinearity among candidate predictors, LASSO regression was applied. The LASSO coefficient path (Fig. 5A) and cross-validation curve (Fig. 5B) were generated. Using tenfold cross-validation, five variables (BA, SER, PLT, XA, and PTA) were retained at the optimal penalty parameter (lambda.min). Bootstrap stability analysis showed that BA and SER exhibited extremely high selection frequencies (99.9% and 99.3%, respectively), indicating robust and stable associations. PLT and XA demonstrated moderate selection frequencies (73.8% and 65.7%), whereas PTA exhibited relatively low stability (53.0%) across bootstrap samples (Supplementary File 1 Table S8). Univariate logistic regression showed that BA, PLT, PTA, and SER were significantly associated with HE in cirrhosis patients (P < 0.05) (Table 3). These variables were incorporated into a multivariate logistic regression model, and stepwise backward selection was applied to derive the final model. The final model retained two predictors: BA (OR = 1.063, 95% CI: 1.023–1.103) and SER (OR = 0.926, 95% CI: 0.888–0.967) (Table 3).

Fig. 5.

Model Construction. A LASSO coefficient path diagram. B LASSO regularization path diagram. C ROC curve for the BA model. D ROC curve for the SER model. E ROC curve for the diagnostic model

Table 3.

Univariate and multivariate logistic regression analyses performed to assess the risk of HE in patients with LC

Variable	Univariable analysis		Multivariable analysis
	OR (95%CI)	P value	OR (95%CI)	P value
BA	1.042 (1.017–1.068)	0.001	1.063 (1.023–1.103)	0.002
PLT	0.979 (0.967–0.991)	0.001
PTA	0.962 (0.941–0.985)	0.001
SER	0.933 (0.899–0.968)	< 0.001	0.926 (0.888–0.967)	< 0.001
XA	0.939 (0.875–1.007)	0.078

BA (blood ammonia), PLT (platelets), PTA (prothrombin activity), SER (serotonin), XA (xanthurenic acid). OR: ratio of ratios; Statistical significance is shown by bold values (P < 0.05)

The model’s discriminatory ability was assessed using the area under the ROC curve (AUC). ROC analysis of the independent predictors showed strong ability to distinguish HE patients (Figs. 5C, D). The diagnostic model based on BA and SER achieved an area under the curve (AUC) of 0.902. The optimal cutoff value, determined using the Youden index, yielded a sensitivity of 76.9% and specificity of 88.0% for predicting HE in LC patients, outperforming univariate models (Fig. 5E).

Validation with 1,000 bootstrap resamples yielded an AUC of 0.902 (95% CI: 0.839–0.965) (Fig. 6A), confirming the model’s robustness. For clinical application, an individualized diagnostic model was developed and presented as a nomogram (Fig. 6B). Model calibration was assessed using a calibration curve, which compared predicted risks with observed outcomes. The calibration curve showed good agreement between predicted and observed probabilities, indicating satisfactory calibration (Fig. 6C). Decision curve analysis (DCA) was then conducted to evaluate the model’s clinical utility. The DCA curve outperformed the two extreme strategies (“treat none” and “treat all”), indicating that LC patients could derive substantial clinical net benefit from this model (Fig. 6D).

Fig. 6.

Validation of the diagnostic model. A ROC curves for internal validation using the bootstrap method. B Nomogram for risk prediction of HE occurrence in LC patients. C Calibration curve of the diagnostic model. D Decision curve analysis (DCA) for clinical utility

Discussion

The three primary catabolic pathways of tryptophan (TRP) include the kynurenine (KYN) pathway, the 5-hydroxytryptamine (5-HT) pathway, and the indole pathway. In this study, LC–MS was employed to quantify TRP and its 23 metabolites in the serum of patients with varying degrees of liver cirrhosis (LC), with or without concurrent hepatic encephalopathy (HE). Key findings are as follows: Compared with healthy controls, significant alterations were observed across all three TRP catabolic pathways in LC patients, irrespective of HE status. Specifically, the indole and KYN pathways were upregulated, whereas the 5-HT (serotonin, SER) pathway was downregulated, resulting in corresponding changes in metabolite levels and their derivatives. These results further corroborate the close association between TRP metabolic dysregulation and the onset and progression of LC and HE.

Differences between Non-HE LC and HE patients: Comparative analysis of TRP metabolic profiles between Non-HE LC and HE patients (including CHE and OHE) revealed that only serotonin (SER) levels differed significantly (P < 0.05), whereas no metabolites showed significant differences between CHE and OHE patients. These findings contrast with some previous studies. For instance, Qiang Wang et al. [23] reported that, based on metabolomic analysis, serum levels of 3-hydroxykynurenine (HK), 5-hydroxytryptophan (5-HTP), 5-hydroxyindole-3-acetic acid (5-HIAA), indole-3-lactic acid (ILA), kynurenine (KYN), and melatonin were elevated in HE patients compared with LC patients, while 5-HT levels were decreased. Discrepancies between these results may be attributed to factors such as sample size, lifestyle differences (e.g., diet), and comorbidities (e.g., diabetes), which warrant further investigation in future studies.

Indole pathway: Tryptamine (TRM), indole-3-carboxaldehyde (IALd), indole-3-acetic acid (IAA), and ILA are key metabolites of the tryptophan indole pathway, predominantly generated through the catabolic activity of gut microbiota. The composition and metabolic activity of the gut microbiota change with the progression of liver fibrosis and play a critical role in the development of HE in patients with decompensated liver cirrhosis [24]. In cirrhosis, impaired hepatic function reduces the liver’s capacity to metabolize gut-derived substances. As a result, indole and its derivatives, which are normally metabolized and cleared by the liver, accumulate in systemic circulation and are oxidized into compounds such as indoxyl sulfate. The absorption of indole derivatives into the bloodstream can affect brain function and behavior; studies have shown that these compounds can significantly decrease neuronal excitability, and their accumulation may contribute to neurological symptoms associated with liver dysfunction [25]. Furthermore, IAA and indole-3-propionic acid (IPA) have been shown to influence intestinal permeability and immune function [26]. In the present study, serum levels of indole pathway metabolites (including TRM, IAA, ILA, and IALd) were significantly elevated in both Non-HE LC and HE patients compared with healthy controls (P < 0.05), indirectly suggesting alterations in gut microbiota. These findings are consistent with recent studies [27–29], which indicate a strong correlation between gut microbiota dysbiosis and HE development in LC patients, with significant changes in microbiota composition observed in HE patients.

Kynurenine (KYN) pathway: KYN is a central metabolite of tryptophan (TRP) metabolism [20] with distinct antioxidant properties. Under conditions of reactive oxygen species (ROS) generation, KYN can efficiently scavenge hydrogen peroxide and superoxide, thereby protecting cells from oxidative stress-induced damage. Elevated KYN levels can also suppress ROS production by activated neutrophils [30] and mitigate free radical-mediated DNA and protein degradation. Additionally, KYN has been reported to reduce ROS production and lipid peroxidation induced by pro-oxidants, such as iron (II) sulfate, peroxynitrite, and 3-nitropropionic acid, in rat brain homogenates [31]. Although KYN has been used as a neuroprotective agent in various neurotoxic models, its effects are often attributed to downstream metabolites, such as kynurenic acid (KYNA) [32]. However, emerging evidence indicates that KYN itself contributes, at least partially, to these neuroprotective effects [33]. KYN also plays a critical role in the progression of liver cirrhosis (LC) to hepatic encephalopathy (HE) through its conversion to quinolinic acid (QA), an endogenous excitatory neurotoxin. QA overactivates N-methyl-D-aspartate (NMDA) receptors, causing neuronal excitotoxicity, disruption of neural signaling, and ultimately brain dysfunction. In this study, serum KYN levels in LC and HE patients were significantly higher than those in healthy controls (P < 0.05), consistent with previous reports [34]. Notably, there were no statistically significant differences in KYN concentrations between LC and HE patients (with slightly lower levels in HE) or between CHE and OHE patients. This suggests that the KYN pathway is generally activated in LC and HE; however, as the disease progresses—particularly toward HE—KYN may be more rapidly metabolized into downstream products, including neuroprotective KYNA and neurotoxic QA, resulting in stable or slightly decreased serum KYN levels. Supporting this notion, key intermediate metabolites of the KYN pathway, QA and 3-hydroxykynurenine (HK), were significantly elevated in Non-HE LC and HE patients compared with healthy controls (P < 0.05), whereas no significant differences were observed between CHE and OHE patients.

5-Hydroxytryptamine (5-HT, serotonin, SER) pathway: 5-HT is a key neurotransmitter produced via tryptophan catabolism. In conditions of severe liver dysfunction, metabolic disturbances increase blood–brain barrier permeability, facilitating the entry of circulating 5-HT into the central nervous system and disrupting neurotransmitter homeostasis. Theoretical models suggest that abnormally elevated 5-HT levels may excessively inhibit central nervous system activity, contributing to classical HE symptoms, including altered consciousness and coma. Additionally, alterations in 5-HT may affect cerebral vasomotor function, thereby influencing cerebral blood flow and perfusion, exacerbating hypoxia and metabolic imbalance in brain tissue, and promoting HE development. Contrary to this expectation, in the present study, serum 5-HT levels were significantly lower in LC and HE patients compared with healthy controls (P < 0.05) and were further reduced in HE patients relative to LC patients (P < 0.05). Although OHE patients exhibited lower levels than CHE patients, this difference did not reach statistical significance (P > 0.05). This finding—decreased 5-HT in HE—appears inconsistent with the inhibitory neurotransmitter hypothesis. In HE patients, TRP metabolism shifts toward the KYN and indole pathways, with concurrent suppression of the 5-HT pathway. Notably, previous studies report conflicting directions of change in circulating 5-HT levels in LC patients. For example, Beaudry et al. [35] observed a significant decrease in total 5-HT levels, whereas Culafic et al. [36] reported increased plasma 5-HT levels in Child–Pugh A/B patients compared with C-grade patients, highlighting the complexity of 5-HT regulation in liver disease. 5-Hydroxyindoleacetic acid (5-HIAA), the main metabolic end product of 5-HT, functions as an inhibitory neurotransmitter [37]. In this study, serum 5-HIAA levels were significantly elevated in both LC and HE patients compared with healthy controls (P < 0.05); however, no significant differences were observed between Non-HE LC and HE patients or between CHE and OHE patients (P > 0.05). These results suggest that while 5-HIAA accumulates during liver dysfunction, its levels do not distinguish HE severity.

TRP metabolites, which directly reflect systemic metabolic dysregulation, have demonstrated superior sensitivity in capturing the early risk of HE compared to conventional liver function markers or overt clinical symptoms. This study successfully constructed the first clinical model for predicting the risk of HE in LC patients using TRP metabolites combined with routine biochemical indicators. The results of model validation (discrimination, calibration, and clinical effectiveness evaluation) indicate that compared to traditional single indicators such as blood ammonia, this combined model exhibits higher specificity and sensitivity. The nomogram integrates blood ammonia (BA) and serum serotonin (SER) to provide an intuitive tool for individualized risk estimation. In clinical practice, clinicians can estimate an individual patient’s probability of hepatic encephalopathy by summing the points assigned to their BA and SER values. This risk estimate is intended for clinical risk stratification rather than direct therapeutic decision-making. Patients with higher predicted risk may benefit from closer clinical monitoring or earlier neurocognitive assessment, whereas those at lower risk may be managed with routine surveillance. This provides clinical doctors with an efficient and convenient risk assessment tool, opening up new avenues for early warning of HE. By accurately identifying high-risk patients, this model can provide key decision support for clinical early intervention (such as intestinal decontamination, nutritional support, metabolic regulation, etc.), and is expected to play an important role in the future clinical diagnosis and treatment of liver disease, promoting the improvement of HE prevention and treatment level.

However, several limitations of this study should be acknowledged. First, this was a single-center cohort study, which may limit the generalizability of the findings to other populations and clinical settings. Second, although subgroup analyses were performed, the sample sizes in certain subgroups were relatively limited, which may have reduced statistical power and constrained more detailed stratified analyses. Third, this study did not include an independent external validation cohort. Although bootstrap internal validation was applied to mitigate overfitting and optimism bias, further external validation in large, multicenter cohorts is warranted to confirm the robustness and generalizability of the proposed model. Fourth, owing to the cross-sectional design of this study, longitudinal assessments were not available; consequently, causal relationships and the temporal predictive value of the model could not be fully evaluated. Finally, the clinical application of metabolomics faces challenges, including high detection costs and technical complexity, which currently hinder its widespread implementation.

Conclusions

This study revealed significant alterations in serum tryptophan metabolites across different stages of liver cirrhosis and hepatic encephalopathy. Tryptophan, serotonin, 3-hydroxykynurenine, and quinolinic acid were identified as key metabolic markers associated with disease progression. A clinical prediction model incorporating blood ammonia, serotonin, and prothrombin activity demonstrated good discrimination and clinical utility, offering a valuable tool for early identification and risk stratification of HE in cirrhotic patients.

Author contributions

XC and LW contributed equally to this work; they drafted the manuscript and participated in study design and data analysis. LN and YH were responsible for data collection. NF supervised the study and critically revised the manuscript. All authors read and approved the final version of the manuscript.

Funding

This work was jointly supported by the National Natural Science Foundation of China (82570697), Special Funding for the Construction of Innovative Provinces in Hunan (2021SK4031), Natural Science Foundation of Hunan Province (2024JJ5362, 2024JJ9401, 2025JJ81075, 2025JJ81085, 2023JJ60370), Major research projects come from National Clinical Key Specialty (Z2023081, Z2023158), Postgraduate Scientific Research Innovation Project of University of South China (233YXC046), and Scientific Research Project of University of South China (Documents of the university of south China [2019]02).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This study has been approved by the Ethics Committee of Nanhua Hospital Affiliated to Nanhua University (Approval No.: 2024-Ky-233).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.