ABSTRACT
-
Background/Aims
Acute liver failure (ALF) has high mortality predominantly due to compromised immune system and increased vulnerability to bacterial and fungal infections.
-
Methods
Plasma lipidome and fungal peptide-based community (mycobiome) analysis were performed in discovery cohort (ALF=40, healthy=5) and validated in a validation cohort of 230 patients with ALF using high-resolution-mass-spectrometry, artificial neural network (ANN) and machine learning (ML).
-
Results
Untargeted lipidomics identified 2,013 lipids across 8 lipid group. 5 lipid-species—phosphatidylcholine (PC)[15:0/17:0], PC[20:1/14:1], PC[26:4/10:0], PC[32:0] and TG[4:0/10:0/23:6]—significantly differentiated ALF-NS (FC>10, P<0.05, FDR<0.01). Mycobiome alpha/beta diversity was significantly higher and showed 4 phyla and >20 species significantly dysregulated in ALF-NS linked with lipid metabolism, fatty acid elongation in ER, and others (P<0.05). Lipid and mycobiome diversity values in ALF-NS were strongly correlated (r2>0.7, P<0.05). Multi-modular correlation network showed striking associations between lipid, fungal peptide modules, and clinical parameters specific to ALF-NS (P<0.05). Cryptococcus amylolentus CBS6039 and Penicillium oxalicum 1142 directly correlated with phosphatidylcholine, triglycerides, and severity in ALF-NS (r2>0.85, P<0.05). POD-fungus and POD-lipids showed direct association with infection, necrosis, and hepatic encephalopathy (Beta>1.2, P<0.05). POD-lipid (AUC=0.969 and HR=1.99 [1.02–2.04]) superseded POD-fungus and severity indices for early-mortality prediction. Finally, significant increase in PC (15:0/17:0) level showed highest normalized importance, and ANNs and ML predicted early mortality with >95% accuracy, sensitivity, and specificity. Interestingly, fungal surveillance protein Clec7a was significantly downregulated (>2-fold), leading to a notable increase in fungal infection-mediated choline/phosphatidylcholine and associated enzymes (FC>1.5; Kennedy cycle). This contributed to phosphatidic acid-mediated hyper-inflammation in ALF-NS.
-
Conclusions
In ALF, the plasma lipidome and mycobiome are dysregulated. Increased circulating phosphatidylcholine could stratify ALF predisposed to early mortality or require emergency liver transplantation.
-
Keywords: Liver failure; Fungal infections; Omics; Lipidome; DILI; Phosphatidylcholine
Study Highlights
• Our study reveals severe disruptions in circulating lipids and the mycobiome in ALF, strongly correlating with disease severity and outcomes, highlighting their role in pathogenesis.
• Downregulation of fungal Surveillance CLR Protein “Clec7a” leads to fungal infections and contributes to an increase in plasma phosphatidylcholine, choline, and phosphatidic acid mediated hyper-inflammation in ALF patients.
• Lipid and fungal peptide integration reveals heightened Kennedy pathway-driven inflammation and mitochondrial failure, contributing to poor outcomes in ALF.
• Lipid panel showed >90% diagnostic efficiency and a 1.99 hazard ratio, making it the best early mortality predictor in ALF using machine learning.
Graphical Abstract
INTRODUCTION
Acute liver failure (ALF) is associated with infections, multi-organ failure (MOF), and high mortality, reaching up to 80% in untreated patients [
1]. Emerging evidence indicates that patients with ALF with fungal infections are highly predisposed to early mortality, largely due to limited knowledge about fungal infections and their contribution to liver failure [
2]. Therefore, studying the fungome is critical for understanding ALF pathogenesis and segregating patients who are at risk of early mortality and require emergency treatment [
1,
2].
Alterations in microbial composition (bacteria and fungi) within the bloodstream have been proposed as potential contributors to disease progression [
3]. In patients with ALF, invasive fungal infections (IFIs) exacerbate inflammation and oxidative stress, contributing to systemic deterioration [
3].
Candida, Cryptococcus, and
Aspergillus species are among the most common opportunistic fungal pathogens encountered in advanced liver disease patients with high model for end-stage liver disease (MELD) scores [
4]. Some fungi, such as
Cryptococcus neoformans and
Histoplasma capsulatum, generate lipid immunomodulators [
5]. Additionally,
Candida albicans, C. neoformans, and
Paracoccidioides brasiliensis are linked to hyperinflammation through their production of prostaglandin E
2 [
4,
6]. IFIs can lead to multi-organ dysfunction and increase the early mortality risk by 26% (candidiasis) and 50% (aspergillosis) [
3]. Addressing IFIs in patients with liver disease is crucial for improving outcomes and minimizing early mortality.
In liver failure, the increased accumulation of pro-inflammatory cytokines, leukotrienes, acylcarnitines, bile acids, and lipid mediators signals lipotoxicity-mediated mitochondrial dysfunction [
7]. Total cholesterol (TC), triglycerides (TGs), and high-density lipoprotein (HDL) decrease as ALF severity increases [
8]. Additionally, elevated circulating levels of phosphatidylcholine (PC) and phosphatidylethanolamine (PE) affect energy metabolism and ALF progression [
9,
10], and increased lyso-PC modulates immune checkpoints in ALF [
7]. These findings highlight the contribution of lipid-mediated changes to ALF pathophysiology.
However, studies directly linking the ALF lipidome and fungal infections remain scarce [
4,
11]. A detailed investigation of the circulating lipid landscape in patients will improve our understanding of ALF pathogenesis, particularly fungal-induced lipid-mediated changes, disease progression, and patient stratification for predicting poor outcomes.
In this study, plasma samples from patients with ALF and healthy controls were analyzed to characterize key lipid species and fungal peptides with diagnostic value for detecting ALF non-survivors (ALF-NS) at baseline. Using a weighted lipid/fungal co-expression network analysis (WL/FCNA) and multi-modular correlation clustering (MMCC), we identified an ALF-NS-associated panel of lipid species and fungal peptides with striking associations with clinical parameters. This study identifies critical fungal and lipid species linked to hyperinflammation in patients with ALF. Furthermore, we demonstrate the utility of lipidomics, fungal peptides, and machine learning (ML) in stratifying patients with ALF predisposed to early mortality.
MATERIALS AND METHODS
Sample collection
This longitudinal study recruited 300 participants at the Institute of Liver and Biliary Sciences, New Delhi, India, between December 2018 and December 2021. Each ALF diagnosis was based on the presence of jaundice with hepatic encephalopathy within four weeks, along with an increased international normalized ratio (INR) >1.5 [
12]. Patients with splenomegaly, clinical ascites, known hepatitis B or C infection, or regular alcohol intake were excluded. Patients with ALF were clinically followed until death or up to 90 days post-discharge. Blood samples were collected at the time of patient enrollment (baseline). Death due to liver failure was defined as progressive hepatic failure or related complications, including cerebral edema, renal failure, infection, sepsis, or MOF during an ICU stay or within 30 days of diagnosis. Those patients were classified as ALF-NS, and those who survived for at least 90 days were categorized as ALF survivors (ALF-S).
Discovery cohort
Plasma samples from 40 patients with ALF (n=40; survivors=8, non-survivors=32) and 5 healthy controls (n=5) were collected prospectively. These samples were analyzed using untargeted lipidomics and mycobiome (fungome) analyses to identify lipid species that distinguish ALF patients from healthy controls. Differentially expressed plasma lipid species and fungal peptides were identified using a multi-parametric segregation procedure, incorporating fold change (FC),
P-values, AUROC values, and a random forest (RF) analysis (mean decrease in accuracy, MDA). The top five lipid species and fungal peptides were selected for further analysis (
Fig. 1A).
Validation cohort
To validate the top five lipid species, an independent cohort of 230 patients with ALF (n=230; survivors=84, nonsurvivors=146) was obtained from the National Liver Disease Biobank, Delhi. Additionally, healthy controls (n=25) and severe alcohol-associated hepatitis (SAH) patients (n=200) were included as disease controls to determine the specificity of the identified lipid biomarkers (
Fig. 1A).
Machine learning
The robustness of the identified lipid biomarkers was further evaluated using ML. An artificial neural network (ANN) analysis was conducted to determine whether the identified lipid parameters outperformed clinical variables known to be associated with early mortality in ALF. For this analysis, all 270 ALF samples were randomized to eliminate bias. The dataset was then divided into a training cohort (70%) and a testing cohort (30%). The analyst was blinded to sample identities. Sensitivity, specificity, accuracy,
P-values, and normalized importance scores for predicting early mortality were calculated using the top five lipid species, as detailed in the
Supplementary Methods.
All samples were stored at –80°C until analysis. This study was conducted in accordance with the Helsinki Declaration and was approved by the Institutional Review Board of our institution (IRB approval no. IEC/2019/70/NA06). Written informed consent was obtained from all participants.
Enrichment of weakly abundant plasma proteins
To enrich low-abundance plasma proteins, albumin was removed using a dye-binding affinity chromatographic method on an AKTA Pure purification system [
13] (details in the
Supplementary Methods).
Untargeted lipidomic and fungal peptide-based community analyses of plasma samples were performed using ultra-high-performance liquid chromatography-high-resolution mass spectrometry (UHPLC-HRMS). The detailed protocol is provided in the
Supplementary Methods,
Supplementary Figure 1, and our recent publication [
14].
Results are presented as mean±standard deviation unless stated otherwise. Statistical analyses were performed using GraphPad Prism (version 6; GraphPad Software Inc., La Jolla, CA, USA) and SPSS (version 20; IBM Co., Armonk, NY, USA). A
P-value <0.05, adjusted using the Benjamini-Hochberg correction, was considered statistically significant. For group comparisons, the unpaired two-tailed Student’s t-test and Mann–Whitney U-test were used for two-group comparisons, and one-way ANOVA and Kruskal–Wallis tests were used for multiple-group comparisons. Correlations were assessed using a Spearman correlation analysis (
r²>0.5,
P<0.05, FDR<0.01 were considered significant). Annotated lipid and fungal peptides were first subjected to global analysis using Weighted Lipid Co-expression Network Analysis (WLCNA) and Weighted Fungal Coexpression Network Analysis (WFCNA) [
15]. ALF-specific lipid and fungal modules were correlated with clinical parameters using R software. Additional statistical assessments included probability of detection (POD) for non-survival, multivariate AUROC, Cox regression analysis, and Kaplan–Meier survival analysis. Shannon diversity indices for lipid and fungal species were calculated using R software (R Foundation for Statistical Computing, Vienna, Austria), followed by a linear regression analysis with respect to clinical complications. The lipid biomarker candidates were further validated using ANN and ML models, as detailed in the
Supplementary Methods and
Supplementary Figure 2.
RESULTS
Demographic profile
In this pilot study, 270 patients with ALF were divided into the discovery cohort (n=40; ALF-NS=32, ALF-S=8) and the validation cohort (n=230; ALF-NS=146, ALF-S=84) and were compared with healthy controls (discovery cohort, n=5; validation cohort, n=25). The etiologies of ALF in the discovery and validation cohorts were drug-induced liver failure in 42.8% and 38.58% and viral in 57.2% and 61.44%, respectively. The demographic profiles of the two cohorts were comparable. Levels of cholesterol, serum alanine aminotransferase (ALT), INR, total bilirubin, and serum aspartate aminotransferase (AST) were elevated in patients with ALF, particularly in ALF-NS. Serum albumin levels and platelet counts were significantly decreased (
P<0.05) compared with the healthy controls. Total leukocyte count, percentage of neutrophils, and MELD score were significantly higher in ALF-NS than in ALF-S or healthy controls (
P<0.05). The overall 90-day mortality rate in patients with ALF was approximately 65%, and the 30-day mortality rate was ~50% (
Supplementary Table 1).
Impaired liver function significantly affects lipid metabolism and circulating lipid levels [
16]. Therefore, an untargeted lipidomic analysis was conducted to characterize the plasma lipid landscape (lipid species and lipid-like molecules) in patients with ALF. Analysis using both positive and negative electrospray ionization identified 2,013 lipids (1,781 lipid species, 232 lipid-like molecules) associated with eight lipid groups and 35 subgroups in the study cohort (
Supplementary Table 2). ALF-NS uniquely expressed zymosterol and monogalactosyldiacylglycerol and exhibited higher glycerolipid levels than the ALF-S group (>22% vs. 16.11%) and healthy controls (12.82%;
Fig. 1B). We identified 369 differentially expressed lipid species (DELs) in patients with ALF compared with healthy controls (242 upregulated, 127 downregulated;
P<0.05). Additionally, 347 DELs (244 upregulated, 103 downregulated) were differentially expressed in non-survivors compared with survivors (FC>1.5;
P<0.05, FDR<0.01;
Fig. 1C,
Supplementary Table 3). A partial least squares discriminant analysis (PLS-DA) and supervised clustering analysis clearly segregated ALF-NS from the ALF-S and healthy controls (
Fig. 1D,
Supplementary Figs. 3,
4). ALF-NS showed a significant increase and direct correlation in phosphatidylserine (P-serine), phosphatidylinositol (P-inositol), and glycosylated sphingolipids, which are linked to inflammation, apoptosis, and disease severity (
Fig. 1E). Furthermore, ALF-NS displayed dysregulated expression of PUFA-derived lipid mediators (
Supplementary Fig. 5). Among the DELs, the highest MDA was observed for PC(32:0), PC(26:4/10:0), PC(20:1/14:1), PC(15:0/17:0), and TG(4:0/10:0/23:6), which were the most important lipid species differentiating ALF-NS from ALF-S. The POD based on the top five lipid species (AUC=0.99) was >80% (
Fig. 1F,
Supplementary Fig. 6;
P<0.05). These findings suggest that patients with ALF possess distinct lipid signatures associated with inflammation and apoptosis, that could serve as biomarkers for early mortality. However, these observations warrant validation in larger cohorts.
Patients with ALF have compromised immune systems, and are highly susceptible to microbial (bacterial) and mycobial (fungal) infections, that can exacerbate inflammation and systemic stress [
2]. A fungal peptide analysis of baseline plasma samples identified four fungal phyla and more than 20 dysregulated genera in ALF-NS, with a significant increase in
Ascomycota and
Basidiomycota compared with the ALF-S and healthy controls (
P<0.05;
Fig. 2A,
Supplementary Fig. 7,
Supplementary Table 4). Fungal taxonomic richness (alpha diversity) was highest in ALF-NS (
P<0.05), and the principal coordinate analysis (beta diversity) clearly distinguished ALF-NS from the ALF-S and healthy controls (
Fig. 2B,
Supplementary Figs. 8,
9). A supervised clustering analysis further segregated ALF-NS from the ALF-S and healthy controls (
Supplementary Fig. 10). Notably, an increase in Ascomycota- and Basidiomycota-associated peptides in ALF-NS directly correlated with clinical parameters related to liver functionality and the lipid profile, highlighting a strong association between fungal peptides and the clinical phenotype of ALF-NS (
Fig. 2C). Functionally, ALF-NS-specific fungal peptides were linked to lipid metabolism, fatty acid elongation in the ER, defense mechanisms, and transport (
Fig. 2D). A linear discriminant analysis (LDA) identified a significant increase in
Cladophialophora psammophila CBS 110553, Penicillium oxalicum 114-2, and others in ALF-NS (
Fig. 2E,
Supplementary Fig. 11). The top six fungal peptides demonstrated the highest MDA (
Supplementary Figs. 12,
13). The diagnostic efficacy based on these peptides (AUC=1) predicted mortality with >80% accuracy (
P<0.001) for ALF-NS (
Supplementary Fig. 14). These findings underscore a significant increase in fungal diversity in ALF-NS that is, functionally associated with energy and lipid metabolism, indicating infection and lipid dysfunction. The identified fungal peptides could potentially segregate ALF-NS from ALF-S, but further validation is needed.
To evaluate the effects of different ALF etiologies, we compared lipidomic and fungal peptide signatures between viral and drug-induced liver injury survivors and non-survivors. No significant differences were observed in plasma signatures (lipids and mycobiome) across etiologies, suggesting that biomolecular signatures are robust and specific for stratifying ALF-NS (
Supplementary Figs. 15-
19).
Fungal pathogens dysregulate host lipid metabolism to increase their own survival [
17]. Understanding the interplay between fungal infection and lipid dysregulation could provide valuable insights for the management of fungal infections in patients with ALF. As evident from our study, fungal alpha diversity (Shannon Diversity Index) correlates directly with the alpha diversity of circulating lipids (R
2>0.8,
P<0.05) in ALF-NS (
Fig. 3A) suggesting that changes in the fungal alpha diversity bring concordant changes in circulating lipids. Furthermore, fungal alpha diversity was directly associated with infection and necrosis. In contrast, the alpha diversity of lipids demonstrated notable associations with hepatic encephalopathy and MOF in ALF patients (
Fig. 3B).
Using W[L/F]CNA in Perseus software [
15], we identified ALF-NS-specific lipid and fungal peptides that clustered into distinct modules, indicating that they potentially share biological functions. The identified ALF-NS modules were cross-correlated to form a multi-modular correlation network (MMCN) integrating lipidomic and fungal data (
Fig. 3C–
3F).
WLCNA
A total of 2,013 lipids (1,781 lipid species, 232 lipid-like molecules) were clustered into 19 modules (soft threshold>9, scale-free topology fit index>0.85;
P<0.05). Of them, 9 modules (pink, brown, purple, and others) were specific to ALF-NS (
Fig. 3C).
WFCNA
A total of 118 fungal peptides were clustered in two modules (soft threshold>9, scale-free topology fit index>0.85;
P<0.05). A module–trait relationship analysis identified the turquoise module, specific to ALF-NS, in which lowest common ancestor [LCA] species such as
Yarrowia lipolytica and
Penicillium brasilianum were functionally associated with lipid transport, defense mechanisms, and other processes. In contrast, the gray module, specific to ALF-S, was associated with
Aspergillus calidoustus, Alternaria, and other LCA species, that were functionally linked to energy metabolism dysfunction (
Fig. 3D).
MMCN
The mean module intensities of lipid modules and fungal modules were cross-correlated to form an MMCN using Pearson correlation coefficients. The MMC analysis revealed striking associations between the ALF-NS–specific lipid modules (pink, brown, purple, and other) and the fungal peptide module (turquoise;
P<0.05, r
2>0.5;
Fig. 3E). Furthermore, glycerolipids and glycerophospholipids within the pink, purple, magenta, and tan modules showed a direct correlation with the functionality of the turquoise module, which was associated with energy metabolism, nucleotide transport and metabolism, and other processes (
Fig. 3F,
Supplementary Fig. 20). The surge in pathogenic fungal peptides, along with their associated functions and lipid groups, appears to be directly linked to poor prognosis in patients with ALF.
Overall, these results highlight a significant connection between fungal peptide function and lipid species diversity in ALF.
Correlation between plasma lipidome and fungal peptides highlights prominent increase in mitochondria impairment and inflammation
The non-survivor–specific lipid modules (pink, brown, purple, and others) were grouped into lipid classes containing PC, ceramides, polyketides, prenol lipids, lysophosphatidylglycerol, TG, monoglycerides (MG), diacylglycerol (DG), lyso-PE, lysophosphatidylserine, sphingomyelins, fatty acyls (FA), and sterol lipids. These lipid modules correlated directly with the fungal turquoise module, which is associated with energy and lipid metabolism (R²>0.5,
P<0.05;
Fig. 4A). The non-survivor–specific fungal peptides correlated directly with circulating DELs, suggesting a strong association between the mycobiome and circulating lipid species in ALF-NS (
Fig. 4B,
Supplementary Figs. 21,
22). The ALF-NS–specific fungal turquoise module was associated with four fungal phyla and more than 30 LCA, exhibiting a positive correlation with distinct lipid classes. Notably, four lipid classes—DG, MG, PC, and PE—were commonly linked to glycerolipid and glycerophospholipid metabolism (
Figs. 4C,
4D). The lipid and fungal modules specific to non-survivors showed direct correlations with severity indices (MELD, KCH) and liver function parameters (AST, ALT, and others;
Fig. 4E). Furthermore, ALF-NS–specific lipids were significantly enriched in glycerophospholipid metabolism and TNF-alpha signaling pathways, among others (
Fig. 4F). These findings indicate that fungal and lipid modules in ALF-NS are interlinked and contribute to mitochondrial dysfunction and inflammation.
Based on the FC, RF (MDA analysis),
P-values, and AUROC values, the top six fungal peptides (LCA) and top five lipid species were used to estimate the POD early mortality in patients with ALF (FC>1.5,
P<0.05;
Fig. 5A). Clinically, both POD-lipid and POD-fungus (fungal peptides) showed a significant positive correlation with liver function markers (AST, ALT, bilirubin), severity indices (MELD, KCH), and the lipid profile parameter very-low-density lipoprotein (R²>0.85,
P<0.05;
Fig. 5B).
POD-fungus and POD-lipid were directly associated with infection, necrosis, and hepatic encephalopathy, in patients with ALF (
Fig. 5C). The diagnostic efficiency of POD-lipid (AUC=0.969;
Fig. 5C) was superior to that of POD-fungus and MELD, and KCH scores. Additionally, univariate and multivariate Cox regression analyses confirmed POD-lipid as the best predictor of early mortality in ALF (HR 1.99, 95% CI 1.02–2.04;
Figs. 5D, 5E). Kaplan-Meier survival analysis showed significantly higher early mortality (<30 days) in patients with ALF with a POD-lipid cutoff >25% (log-rank
P<0.05,
Fig. 5F). Collectively, these findings suggest that the five selected lipid species are strong predictors of early mortality, but that finding should be validated in a larger number of patients with ALF.
To validate the top five lipid species associated with non-survival in ALF, we used two validation approaches, HRMS-LC/MS and ML. These tests compared a validation cohort of 230 patients with ALF (146 non-survivors, 84 survivors) with a disease control group (
Fig. 6A). A quantitative HRMS-LC/MS analysis confirmed significantly increased expression of the top five lipid species in the validation cohort, consistent with the discovery cohort (
Fig. 6B, 6C). For ML validation, five supervised algorithms were used: LDA, k-nearest-neighbors (kNN), support vector machines (SVMs), classification and regression trees (CART), and RF. Subsequently, the accuracy, sensitivity, specificity, and
P-value were computed to assess the putative predictors for early mortality in ALF. In all, 30 trained and tested ML models were generated using these algorithms to assess the five lipids individually and together. Fourfold (outer) nested repeated (five times) tenfold (inner) cross-validation (with randomized stratified splitting) was used to train and test the ML models, which had optimized hyperparameters. Accuracy and kappa statistics were computed for all five lipid species in the plasma validation cohort (
Fig. 6D). Among all lipid species, PC (15:0/17:0) demonstrated the highest predictive capability (>95%) for early mortality, and RF emerged as the most effective algorithm for stratifying patients with ALF (
Fig. 6E). Additionally, ANN analysis using a supervised multilayer perceptron algorithm identified PC (20:1/14:1) and PC (15:0/17:0) as the most robust predictors, with a normalized importance score exceeding 90% (
Fig. 6F,
Supplementary Figs. 23,
24). These findings validate the utility of PC (20:1/14:1) and PC (15:0/17:0) as reliable indicators of early mortality in patients with ALF.
Our study identified significant downregulation of the Clec7a protein, which increases susceptibility to fungal infection.
Downregulation of Clec7a protein, a key C-type lectin receptor major CLRs involved in fungal recognition, impairs antifungal immune responses [
18]. This suppression weakens the host’s ability to detect and combat fungal pathogens, leading to an increased risk of fungal infections (
Fig. 7A) [
5].
We identified a panel of lipid species with a diagnostic efficiency of 96% (AUC>0.95) compared with other severity indices. Increased accumulation of TGs induced lipotoxicity-mediated inflammation and ROS production [
19]. Furthermore, increased accumulation of PC is linked to inflammation via the Kennedy pathway [
20] and plays a role in fungal pathogenicity within host organisms [
20]. Additionally, its conversion to phosphatidic acid activates inflammatory pathways [
11].
In patients with ALF, the Kennedy pathway is upregulated by both fungal infections and immune activation, leading to increased choline and PC levels in both host and fungal cells (e.g.,
Candida sp., Aspergillus sp.). Increased PC and its linked enzymes drive phosphatidic acid accumulation, which exacerbates hyperinflammation (
Fig. 7B,
Supplementary Figs. 25–
27) via pathways such as MAPK and NF-kappa B (
Fig. 7C). These findings support the role of LPA in modulating the innate immune response and contributing to ALF-associated hyperinflammation (
Fig. 7C).
Our results demonstrate that fungal infection-mediated choline and PC accumulation (via the Kennedy cycle) leads to increased phosphatidic acid levels, driving hyperinflammation in patients with ALF (
Fig. 7C). Patients with ALF displaying such dysregulation at baseline are predisposed to early mortality.
DISCUSSION
ALF is characterized by severe acute liver injury, leading to impaired liver functions, which increases vulnerability to bacterial and fungal infections [
1]. Recent literature suggests that patients with ALF typically encounter fungal infections at a late stage of the disease. However, patients who develop fungal infections have a higher probability of early mortality than those who do not. Therefore, it is important to study the mycobiome and lipidome landscapes of ALF patients to identify specific fungal or lipid signatures that could enable early detection of ALF patients predisposed to early mortality [
21].
In this pilot study, HRMS-based lipidomics and fungal peptide community analyses were performed on baseline plasma samples from patients with ALF. Our primary aim was to identify and validate lipid and fungal peptide signatures with diagnostic value for predicting early mortality in patients with ALF. We sought to understand the association between fungal composition and the complex metabolic network of lipids in baseline plasma samples to identify mycobiome-associated lipid species and elucidate their contribution to ALF pathophysiology. Then, we validated the identified molecular signatures using HRMS, ANNs, and ML in the validation cohort.
Plasma lipidome and mycobiome profiles were found to be significantly dysregulated in patients with ALF. Increased plasma levels of PC emerged as an early predictor of mortality, aiding in the identification of patients requiring urgent liver transplantation.
In patients with ALF, low lipid levels, particularly HDL, correlate with disease severity and poor outcomes [
16]. This dysregulation can be attributed to the abrupt loss of liver function, which disrupts normal lipid metabolism and exacerbates the severity of ALF [
8,
16]. Disrupted lipid homeostasis caused by liver diseases alters metabolic processes such as cellular signaling and apoptosis, leading to inflammation [
21]. Excessive accumulation of saturated fatty acids and their intermediate products contributes to lipotoxicity, linking chronic ER stress to decreased autophagic flux and progression of liver disease [
22].
Previous studies have linked elevated PC levels to insulin resistance, necrosis, and inflammation, and reduced levels of lyso-PCs 16:0, 18:0, and 18:1 have been associated with poor prognosis in ALF [
7,
9]. In hepatocellular carcinoma, increased lipogenesis and PC synthesis coincide with decreased β-oxidation and mitochondrial failure [
23]. Additionally, aberrant PC/PE ratios have been associated with altered energy metabolism and advanced liver diseases [
9,
24]. In this study, the elevated PC levels found in ALF-NS patients could indicate an energy-deprived state and disrupted β-oxidation process.
Patients with ALF have an increased risk of IFI, leading to systemic inflammation and an increased risk of early mortality [
2,
5]. This observation was validated by the decreased expression of the Clec7a receptor protein, a critical component of host defense against opportunistic fungi such as
Candida, in ALF-NS [
18]. Although comprehensive studies on mycobiome alterations in patients with ALF are lacking, it is well established that patients critically ill with ALF frequently experience gut dysbiosis that increases their susceptibility to fungal infections [
2,
25].
Significant alterations in mycobiome diversity promote inflammation in patients with ALF. In this study, we observed a notable increase in mycobiome diversity, particularly within the
Ascomycota and Basidiomycota phyla, in ALF-NS [
25]. Specifically, ALF-NS exhibited a significant increase in fungal species such as
Cladophialophora, Penicillium, and
Candida that are associated with immune dysfunction and intestinal integrity loss and known to contribute to disease severity and increased mortality in acute liver diseases [
4,
25]. Dysregulation of gut metabolic pathways was also found to be linked to disease severity in patients with ALF, particularly through disruptions in lipid metabolism and fatty acid elongation in the ER [
9,
22]. Additionally,
Cryptococcus and
Aspergillus infections have been associated with high 90-day mortality rates [
5]. Together, our results suggest that gut mycobiome alterations can contribute to lipid homeostasis disturbances and energy deprivation in patients with ALF.
The Shannon diversity indices of fungal peptides and lipid molecules in plasma correlated directly with each other. Interestingly, fungal diversity was directly associated with infection and necrosis, whereas lipid diversity was linked to hepatic encephalopathy, aligning with the pathophysiology and severity of ALF [
6,
17,
21].
An integrated analysis of lipid and fungal peptide profiles using the WLCNA and WFCNA results revealed a striking MMCC network between lipid and fungal peptides in ALF-NS. This network highlighted critical lipid modules (L_Pink, L_Brown, etc.) and a fungal module (turquoise) enriched with PC, PE, MG, and DG phospholipids [
26].
The levels of glycolipids, which are integral to fungal plasma membranes, significantly regulate fungal virulence and disease severity [
26]. Furthermore, fungal metabolites such as bile acids contribute to chronic liver diseases [
4,
27]. Altered levels of lipids such as PC, PE, DG, and MG and their ratios influence key cellular processes, leading to inflammation and oxidative stress [
26]. Studies have shown that
Chlamydia pneumoniae infections affect TC and TG metabolism in mice [
28]. Additionally,
Candida, Cryptococcus, and
Aspergillus species modulate lipid metabolism through phospholipase enzyme secretion [
29]. Invasive fungi such as
Candida and
Cryptococcus use the Kennedy pathway for phospholipid biosynthesis, especially for PE and PC, which are essential for maintaining membrane integrity and virulence; our findings underscore its crucial role in fungal biology and virulence [
20]. These observations collectively suggest a complex interplay between fungal infections and lipid metabolism, highlighting potential links between infections and metabolic disorders.
We also observed direct associations between ALF-NS–specific fungal peptide and lipid modules and liver function parameters (AST, ALT), severity indices (MELD, KCH), and lipid profiles.
Among the predictive models for early mortality in ALF, POD-lipid emerged as the best predictor (AUC=0.96), with an HR of 1.99. A cutoff at the 25th percentile correlated significantly with mortality in patients with ALF. Interestingly, POD-fungus was associated with infection and necrosis, whereas POD-lipid was linked to progression of hepatic encephalopathy. These findings highlight the potential role of circulating fungal peptides and lipids as crucial factors influencing poor outcomes in patients with ALF [
5].
Finally, we used HRMS and different statistical measures to validate the identified lipid panel in a separate cohort of 230 patients with ALF. We observed a significant increase in lipid expression, similar to the discovery cohort. ML demonstrated accuracy, sensitivity, and specificity >95%, with the RF model emerging as the best predictive model (accuracy >95% using 2,000 trees and 6 randomly sampled variables per node). The out-of-bag error rate of 5% indicates strong generalization capability. Our findings were cross-validated using ANN, which identified PC (15:0/17:0) and PC (20:1/14:1) as the most influential predictors of early mortality.
A major limitation of this study is its monocentric design, which focused solely on the liver. Because ALF affects multiple organs, further research is needed to explore its systemic effects. In conclusion, we present novel data from plasma lipidome and fungal peptide–based community analyses in patients with ALF, providing a comprehensive understanding of baseline plasma lipidome and mycobiome profiles. Our analyses identified key lipid species and fungal peptides that significantly distinguish ALF-NS from ALF-S. In validation testing, increased plasma PC levels effectively segregated patients with ALF who were predisposed to early mortality.
Our results also show that downregulation of the fungal surveillance CLR protein Clec7a leads to fungal infections, contributing to increased plasma PC, choline, and phosphatidic acid levels via the Kennedy cycle in hepatocytes. Such dysfunction should be regulated to ameliorate hyperinflammation in patients with ALF [
18]. The involvement of phosphatidic acid in excessive inflammation emphasizes its potential as a therapeutic target. Furthermore, segregation of patients with ALF who are at higher risk of early mortality based on the identified fungal and lipidomic signatures will improve the management and choice of therapeutic strategies for patients with ALF.
Future studies should focus on longitudinal monitoring of lipid and fungal markers across ALF cohorts to identify evolving disease patterns, evaluate phosphatidic acid and inflammation as therapeutic targets, and investigate Clec7a pathways to enhance fungal immunity without triggering excessive inflammation. By integrating lipidomics, mycobiome analysis, and targeted therapies, future research can advance personalized treatment strategies, improve survival rates, and enhance our understanding of ALF pathophysiology.
FOOTNOTES
-
Authors’ contribution
JSM and NS conceptualized the work. NS, SP, GT, MY, NS, BM, VB, SB, RS, YM, SY, AK, AG, RM, SS and JSM were involved in sample enrolment, processing, experimental work and data analysis. The manuscript was drafted by NS, SP, and JSM. SS and SKS proofread the manuscript. Professor Sarin proofread the manuscript and provided critical inputs. This manuscript has been seen approved by all authors.
-
Acknowledgements
The work was supported from the project by the Indian Council of Medical Research (5/4/8-3/CD/JS/2021-NCD-II and project ID: 2020-4958).
-
Conflicts of Interest
The authors have no conflicts to disclose.
SUPPLEMENTAL MATERIAL
Supplementary material is available at Clinical and Molecular Hepatology website (
http://www.e-cmh.org).
Supplementary Figure 1.
Overview of the study design for lipidome (with blue arrows), and fungome (with green arrows) in the discovery and validation cohorts.
Supplementary Figure 2.
Flow chart of statistical analysis performed for Integrated Analysis of Lipidome and Fungome (fungal peptide analysis) in baseline plasma and also for elucidating key lipidome indicators of poor outcomes in acute liver failure patients.
Supplementary Figure 3.
Partial Least Squares - Discriminant Analysis (PLS-DA) plot showing clear separation of lipidome in acute liver failure and healthy plasma samples with ALF:HC and NS:S revealing component variance along with the VIP scores of important features identified by PLS-DA and loading plot between the selected components.
Supplementary Figure 4.
Heatmap showing hierarchical clustering of 50 different lipid species from ALF vs. healthy and NS vs. S (distance measure using Euclidean and clustering algorithm using ward D).
Supplementary Figure 6.
Fifteen different lipid features identified by Random Forest and ranked by the mean decrease in classification accuracy from ALF vs H and NS vs S.
Supplementary Figure 7.
Fungal peptide-based community abundance in the study group (healthy vs. ALF and NS vs. S) at the phylum and species level (P<0.05).
Supplementary Figure 8.
Alpha diversity indices (Shannon/Simpson index) and principal coordinate analysis (PCoA; beta diversity) in baseline plasma fungal peptide (Fungome) of ALF vs. healthy and NS vs. S.
Supplementary Figure 12.
Random Forest analysis for non-survival, survival, and healthy with 500 classification trees for fungal peptides: OOB error of 0 was due to small sample size of n=40 in ALF and n=5 in healthy control.
Supplementary Figure 13.
Fifteen different fungal features identified by Random Forest and ranked by the mean decrease in classification accuracy from ALF vs H and NS vs S.
Supplementary Figure 24.
Neural network generation model summary and parameter estimates for IBM SPSS software using top 5 lipid species, fungal peptides, and clinical parameters.
Figure 1.(A) Design of the study. (B) Relative lipidome abundance associated with 8 lipid groups in the study groups (ALF-S, ALF-NS, Healthy control) (P<0.05). (C) Volcano plot showing differentially expressed lipids in baseline plasma samples from ALF vs. Healthy and NS vs. S groups (FC>1.5, P<0.05, FDR<0.01). Upregulated expression is shown in red, and downregulated expression is shown in green (P<0.05). (D) PLS-DA and heatmap showing clear segregation of the Healthy (green), ALF survivor (red), and ALF non-survivor (dark blue) groups on the basis of lipid species signatures. (E) Debiased sparse partial correlation of lipid groups (red, positive correlation; blue, negative correlation; P<0.05). (F) AUC (1) value for predicting ALF-NS (POD>80%) based on the panel of the five best lipids. ALF, acute liver failure; PLS-DA, partial least squares discriminant analysis; NS, non-survivors; FC, fold change; FDR, false discovery rate; AUC, area under the curve; POD, probability of detection; S, survivors.
Figure 2.(A) Relative fungal peptide–based taxonomic classification abundance in the study groups at the phylum and genus levels (P<0.05, FDR<0.01). (B) Alpha-diversity indices (Shannon/Simpson indices) and principal coordinate analysis (beta-diversity) in the fungal peptide–based community found in the study groups (P<0.05) at the feature level. (C) Correlations between clinical parameters and different fungal peptide phyla, along with their expression status (red bar=upregulated, blue=downregulated). (D) Cluster of Orthologous Groups function analysis of fungal peptides in the study groups. (E) LDA of fungal peptides in the study groups (P<0.05). FDR, false discovery rate; LDA, linear discriminant analysis.
Figure 3.(A) Fungal alpha diversity (Shannon diversity index) correlates with the alpha diversity of circulating lipids (R2>0.8, P<0.05). (B) Linear regression of Shannon-Fungus and Shannon-Lipid with the clinical complications of ALF patients. (C) Weighted lipid correlation network analysis heatmap showing module–trait relationships, represented as mean values for each group. The color scale on the right indicates correlations from −1 (green) to 1 (red). (D) Weighted fungal peptide-based-taxonomic-classification correlation network analysis heatmap showing module–trait relationships represented as mean values for each group. The color scale on the right indicates correlations from −1 (green) to 1 (red) and the fungal LCA (F_turquoise and F_grey) correlation with the metabolic pathway (R2>0.5, P<0.050. (E) MMCC plot of lipid and fungal modules. The red rectangle shows a direct correlation between the lipid and fungal modules. (F) Correlation between the lipid-group module specific to ALF-NS and the metabolic pathway (R2>0.5, P<0.05). ALF, acute liver failure; LCA, lowest common ancestors; MMCC, multi-modular-correlation-cluster; NS, non-survivors.
Figure 4.(A) Correlation between lipid species specific to ALF-NS and the metabolic pathway (R2>0.5, P<0.05). (B) Fungal peptides positively correlated with lipids (blue bar=positive correlation, R2>0.5, P<0.05). (C) Venn diagram shows lipid species and 4 fungal phyla common in ALF-NS. (D) Alluvial plot of fungal phyla with 4 common lipid species specific to ALF-NS. (E) Clinical correlations between fungal modules and lipid modules and the POD of the top fungal peptides and lipids (POD>80%, R2>0.5, P<0.05). (F) KEGG pathway analysis of upregulated lipid species specific to ALF-NS. ALF, acute liver failure; NS, non-survivors; POD, probability of detection; KEGG, Kyoto Encyclopedia of Genes and Genomes.
Figure 5.(A) Panel of the top 5 indicators of lipid species and top 6 indicators of fungal peptides selected based on RF value, FC>1.5, and P-value<0.05 for calculating POD non-survivors. (B) Plot showing clinical correlations (liver functionality and lipid profile) with POD (r2>0.85, P<0.05). (C) Linear regression comparison of POD-lipid and POD-fungus against clinical complications in ALF patients. (D) Multivariate AUROC analysis against severity indices (MELD, KCH) and POD (lipid and fungus) (P<0.01). (E) Univariate and multivariate analyses of clinical parameters, top lipid species and fungal peptides, and POD (lipid and fungus) were conducted. POD-lipid had the best results, with a hazard ratio of 1.99. (F) Assessment of 30-day survival based on POD-lipid (25% cutoff) in ALF patients. RF, random forest; FC, fold change; POD, probability of detection; MELD, model for end-stage liver disease.
Figure 6.(A) Schematic representation of the validation study conducted using two methods, HRMS-LC/MS and machine learning, on validation cohort 1 (plasma) and validation cohort 2 (disease control, SAH. (B) Quantitative assessment of the top 5 lipid species in validation cohort 1 indicates significant upregulation in ALF-NS (P<0.05). (C) Quantitative assessment of the top 5 lipid species in validationcohort 2, comparing the baseline plasma of SAH patients (n=200) with that of ALF patients (P<0.05). (D) Accuracy and kappa values of the different ML models tested for indicator identification in the study groups. (E) The accuracy, specificity, sensitivity, and p-values of five different lipid species, individually and together, in the validation cohort, along with a confusion matrix for the different ML models. (F) Normalized importance of the top 5 lipid species and clinical parameters according to an artificial neural network model. ALF, acute liver failure; ML, machine learning; NS, non-survivors; SAH, severe alcohol-associated hepatitis.
Figure 7.(A) The CLR protein Clec7a is downregulated in the ALF-NS group compared with the ALF-S group (P<0.05). (B) Bar graphs show that the levels of phosphatidic acid and phosphatidylcholine are significantly increased in the ALF-NS group (P<0.05). (C) Graphical abstract shows that fungal infection contributes to hyperinflammation by increasing the level of plasma phosphatidylcholine. ALF, acute liver failure; NS, non-survivors; S, survivors.
Abbreviations
artificial neural networking
aspartate aminotransferase
area under the receiver operating characteristic curve
classification and regression trees
differentially expressed lipid
gamma-glutamyl transferase
linear discriminant analysis
mitogen-activated protein kinase
model for end-stage liver disease
normalised importance (%)
partial least squares discriminant analysis
serum alkaline phosphatase
REFERENCES
- 1. Stravitz RT, Lee WM. Acute liver failure. Lancet 2019;394:869-881.
- 2. Belicard F, Pinceaux K, Le Pabic E, Coirier V, Delamaire F, Painvin B, et al. Bacterial and fungal infections: a frequent and deadly complication among critically ill acute liver failure patients. Infect Dis (Lond) 2023;55:480-489.
- 3. Barros N, Rosenblatt RE, Phipps MM, Mansour MK. Invasive fungal infections in liver diseases. Hepatol Commun 2023;7:e0216.
- 4. Lahmer T, Pecanha-Pietrobom PM, Schmid RM, Colombo AL. Invasive fungal infections in acute and chronic liver impairment: A systematic review. Mycoses 2022;65:140-151.
- 5. Ikuta KS, Mestrovic T, Naghavi M. Global incidence and mortality of severe fungal disease. Lancet Infect Dis 2024;24:e268.
- 6. Noverr MC, Phare SM, Toews GB, Coffey MJ, Huffnagle GB. Pathogenic yeasts Cryptococcus neoformans and Candida albicans produce immunomodulatory prostaglandins. Infect Immun 2001;69:2957-2963.
- 7. Trovato FM, Zia R, Artru F, Mujib S, Jerome E, Cavazza A, et al. Lysophosphatidylcholines modulate immunoregulatory checkpoints in peripheral monocytes and are associated with mortality in people with acute liver failure. J Hepatol 2023;78:558-573.
- 8. Manka P, Olliges V, Bechmann LP, Schlattjan M, Jochum C, Treckmann JW, et al. Low levels of blood lipids are associated with etiology and lethal outcome in acute liver failure. PLoS One 2014;9:e102351.
- 9. van der Veen JN, Kennelly JP, Wan S, Vance JE, Vance DE, Jacobs RL. The critical role of phosphatidylcholine and phosphatidylethanolamine metabolism in health and disease. Biochim Biophys Acta Biomembr 2017;1859:1558-1572.
- 10. Loomba R, Quehenberger O, Armando A, Dennis EA. Polyunsaturated fatty acid metabolites as novel lipidomic biomarkers for noninvasive diagnosis of nonalcoholic steatohepatitis. J Lipid Res 2015;56:185-192.
- 11. Paul B, Lewinska M, Andersen JB. Lipid alterations in chronic liver disease and liver cancer. JHEP Rep 2022;4:100479.
- 12. Williams R. Classification, etiology, and considerations of outcome in acute liver failure. Semin Liver Dis 1996;16:343-348.
- 13. Tripathi G, Sharma N, Bindal V, Yadav M, Mathew B, Sharma S, et al. Protocol for global proteome, virome, and metaproteome profiling of respiratory specimen (VTM) in COVID-19 patient by LC-MS/MS-based analysis. STAR Protoc 2022;3:101045.
- 14. Sharma N, Pandey S, Yadav M, Mathew B, Bindal V, Sharma N, et al. Biomolecular map of albumin identifies signatures of severity and early mortality in acute liver failure. J Hepatol 2023;79:677-691.
- 15. Rudolph JD, Cox J. A network module for the Perseus software for computational proteomics facilitates proteome interaction graph analysis. J Proteome Res 2019;18:2052-2064.
- 16. Taylor R, Zhang C, George D, Kotecha S, Abdelghaffar M, Forster T, et al. Low circulatory levels of total cholesterol, HDLC and LDL-C are associated with death of patients with sepsis and critical illness: systematic review, meta-analysis, and perspective of observational studies. EBioMedicine 2024;100:104981.
- 17. Guzman G, Niekamp P, Tafesse FG. The squeaky yeast gets greased: The roles of host lipids in the clearance of pathogenic fungi. J Fungi (Basel) 2020;6:19.
- 18. Tang J, Lin G, Langdon WY, Tao L, Zhang J. Regulation of C-type lectin receptor-mediated antifungal immunity. Front Immunol 2018;9:123.
- 19. Heeren J, Scheja L. Metabolic-associated fatty liver disease and lipoprotein metabolism. Mol Metab 2021;50:101238.
- 20. Gibellini F, Smith TK. The Kennedy pathway--De novo synthesis of phosphatidylethanolamine and phosphatidylcholine. IUBMB Life 2010;62:414-428.
- 21. Agmon E, Stockwell BR. Lipid homeostasis and regulated cell death. Curr Opin Chem Biol 2017;39:83-89.
- 22. Han J, Kaufman RJ. The role of ER stress in lipid metabolism and lipotoxicity. J Lipid Res 2016;57:1329-1338.
- 23. Sangineto M, Villani R, Cavallone F, Romano A, Loizzi D, Serviddio G. Lipid metabolism in development and progression of hepatocellular carcinoma. Cancers (Basel) 2020;12:1419.
- 24. Calzada E, Onguka O, Claypool SM. Phosphatidylethanolamine Metabolism in Health and Disease. Int Rev Cell Mol Biol 2016;321:29-88.
- 25. Jiang L, Stärkel P, Fan JG, Fouts DE, Bacher P, Schnabl B. The gut mycobiome: a novel player in chronic liver diseases. J Gastroenterol 2021;56:1-11.
- 26. Rella A, Farnoud AM, Del Poeta M. Plasma membrane lipids and their role in fungal virulence. Prog Lipid Res 2016;61:63-72.
- 27. You N, Zhuo L, Zhou J, Song Y, Shi J. The role of intestinal fungi and its metabolites in chronic liver diseases. Gut Liver 2020;14:291-296.
- 28. Marangoni A, Fiorino E, Gilardi F, Aldini R, Scotti E, Nardini P, et al. Chlamydia pneumoniae acute liver infection affects hepatic cholesterol and triglyceride metabolism in mice. Atherosclerosis 2015;241:471-479.
- 29. Djordjevic JT. Role of phospholipases in fungal fitness, pathogenicity, and drug development - lessons from cryptococcus neoformans. Front Microbiol 2010;1:125.
, Shiv Kumar Sarin4
