This study was conducted in Hainan Province, China, from October 2022 to July 2023. It involved 167 participants: 64 healthy controls (HCs), 48 with ALD, and 55 with MAFLD. Participants were adults aged 18 ~ 65, diagnosed with either ALD or MAFLD according to international consensus guidelines (Eslam et al., 2020; Dunn and Shah, 2016). Diagnosis was further refined using ultrasound assessments by a blinded physician, categorizing HS as mild (diffuse echogenic enhancement) or moderate to severe (bright echogenicities with ultrasound beam attenuation). Subgroups within the MAFLD cohort included 20 with moderate to severe HS (msMAFLD) and 35 with mild HS (miMAFLD), along with 37 non-obese (nobMAFLD) and 18 obese individuals (obMAFLD). For ALD, subgroups of 28 with AFL and 20 with ALC were analyzed. This research received ethical approval from the Institutional Review Board of Hainan Medical University (Approval No. HYLL-2023-453), conforming to the Declaration of Helsinki guidelines. All participants provided written informed consent.

Participants with MAFLD met criteria including varying degrees of HS on ultrasound and at least one of the following: BMI ≥ 23 kg/m², type 2 diabetes (fasting glucose ≥7.0 mmol/L), or metabolic dysfunction. ALD criteria required a history of significant alcohol intake (>40 g/day) with imaging evidence of HS or cirrhosis. Exclusions were applied for viral hepatitis, drug-induced liver disease, severe cardiovascular or renal diseases, recent antibiotics or other gut microbiota-affecting drugs, abnormal diets, and significant recent dietary changes.

Detailed criteria for inclusion and exclusion of HCs are detailed in the Supplementary material.

Detailed demographic and clinical data were collected, including dietary habits assessed via a standardized food frequency questionnaire (Sun et al., 2021). Blood samples were obtained after overnight fasting for biochemical analyses conducted at our central laboratory. Variables such as serum K⁺, Na⁺, Cl⁻, Ca⁺, alanine aminotransferase (ALT), aspartate aminotransferase (AST), glutamyl transpeptidase (GGT), alkaline phosphatase (ALP), total bilirubin (TBIL), direct bilirubin (DBIL), indirect bilirubin (IBIL), creatinine (Cr), uric acid (URIC), blood urea nitrogen (BUN), high-density lipoprotein cholesterol (HDL-C), triglycerides (TG), cholesterol (CHOL), low-density lipoprotein cholesterol (LDL-C), fasting plasma glucose (FPG), along with blood cell counts, including neutrophils (Neu), lymphocytes (Lym), monocytes (Mono), red blood cell distribution width (RDW), platelets (PLT), and other standard blood tests, were analyzed. Detailed information is summarized in Supplementary Dataset 1.

Fecal samples were immediately processed into three 100 mg aliquots, snap-frozen in liquid nitrogen, and stored at −80°C until analysis.

Fecal DNA was extracted using the Fecal Genome DNA Extraction Kit (DP712-02, TIANGEN, China) (Karlsson et al., 2014). DNA quality was quantified with Qubit (Invitrogen, United States), and PCR amplified using ITS1FI2/ITS2 primers. Following pooling, the samples were sequenced on an Illumina NovaSeq 6000 (PE250) by Shanghai Biotree Biotech Co., Ltd. The sequencing primers were excised from demultiplexed raw sequences using cutadapt (v1.9). Paired-end reads were merged via FLASH (v1.2.8). Reads failing to meet quality standards (quality scores <20), those <100 bp in length, and reads with over 5% “N” records were trimmed utilizing the sliding-window algorithm in fqtrim (v0.94). Subsequently, quality filtering was executed to obtain high-quality clean tags using fqtrim. Chimeric sequences were filtered using Vsearch software (v2.3.4), and DADA2 was utilized for denoising and generating amplicon sequence variants (ASVs).

Categorical variables were analyzed employing Pearson’s chi-squared test, and outcomes are reported as numbers and percentages, unless specified differently. Not continuous outcome data are detailed as medians with upper and lower quartiles in brackets, unless otherwise indicated. Fungal diversity was assessed using the Shannon and inverse Simpson indices, and β-diversity was visualized through principal coordinate analysis (McMurdie and Holmes, 2013; Segata et al., 2011). Data were analyzed using the Kruskal-Wallis and Wilcoxon rank-sum tests, with significance adjusted for multiple comparisons using Holm’s method. Significant fungal taxa differences were identified using Linear Discriminant Analysis Effect Size (LEfSe) (McMurdie and Holmes, 2013; Segata et al., 2011). Statistical tests were two-tailed, with a significance level of p ≤ 0.05 indicating statistical significance.

As summarized in Table 1, median BMI was significantly higher in both liver disease cohorts compared to HCs, with MAFLD patients showing a notably higher BMI than ALD patients. Key biochemical parameters such as DBIL, AST, GGT, FPG, and URIC were elevated in liver disease groups compared to HCs and ALD patients exhibited significantly higher levels of these markers than MAFLD patients. Additionally, levels of ALT, TG, and LDL-C were increased, whereas HDL-C was lower across both disease groups vs. HCs, achieving statistical significance. Exceptions included TBIL, TP, ALB, and RBC, which showed a trend toward significance when comparing HCs to the ALD group, and CHOL, which exhibited a similar trend when comparing HCs to the MAFLD group. Furthermore, IBIL levels were notably higher in the ALD group than in the MAFLD group.

Analysis of alpha diversity showed no significant differences across groups (Shannon Index, p = 0.95; Inverse Simpson Index, p = 0.95), as detailed in Supplementary Figures S1A,B. However, significant distinctions in community composition were observed using MANOVA based on the Jaccard index (p ≦ 0.001), illustrated in Figure 1A. Bray–Curtis dissimilarity further supported these findings. Notably, the fungal microbiome showed a significant contrast between the ALD and MAFLD groups according to the Jaccard index (p ≦ 0.001) (Figure 1B) and Bray–Curtis dissimilarity analysis further emphasized this disparity. Linear Discriminant Analysis (LDA) identified specific fungal taxa differentiating ALD from MAFLD. Notably, genera such as Trichosporon and Davidiella, and species like Trichosporon unclassified and Davidiella unclassified, were prevalent in ALD, while Rhizopus and Candida albicans (C. albicans) were more common in MAFLD (Figures 1C,D). Likewise, the relative abundance of the genus Periconia and species C. albicans was low in the ALD group compared with the MAFLD group (Supplementary Figures S2A,B). The relative abundances of the Mucor and Malassezia were significantly increased in the ALD cohort in relation to MAFLD (Supplementary Figures S2C,D).

In ALD patients, significant reductions were noted in enzymes like aryl-alcohol dehydrogenase [NADP(+)], and carbonyl reductase (CBRs, NADPH) compared to HCs, with further decreases in MAFLD patients (Figure 2A). Metabolic pathways including the Leloir pathway and pantothenate and coenzyme A biosynthesis were downregulated in ALD compared to HCs, with more pronounced reductions in MAFLD (Figure 2B). Conversely, pathways related to adenosine nucleotide and pyrimidine biosynthesis were upregulated in both disease groups (Figure 2B).

We determined the mean decrease accuracy per random forest analysis to identify fungal genera and species with the highest feature importance for detecting ALD vs. MAFLD. The top genera distinguishing between ALD and MAFLD included Davidiella, Malassezia, Asterotremella, Pleosporales_unclassified, and Periconia (Figure 3A). The top species were Malassezia_restricta, Davidiella_unclassified and Asterotremella_pseudolonga (Figure 3B). In the second step, we determined genus, species, and combination discriminant values for ALD and MAFLD, respectively (Figure 3C). The highest area under the curve (AUC) value for a single fungal predictor was 0.81 for Pleosporales_unclassified, 0.79 for Davidiella and Asterotremella, 0.76 for Malassezia, Davidiella_unclassified, and Asterotremella_pseudolonga, 0.7 for Malassezia_restricta. We determined that the fungal signature consisting of species Asterotremella_pseudolonga, Malassezia_restricta and genus Malassezia had the highest discriminatory power with an AUC of 0.94 (Figure 3C and Supplementary Table S1). The sensitivity, specificity, accuracy, positive, and negative predictive value (PPV and NPV) for this fungal signature were 0.87, 1, 0.94, 1, and 0.89, respectively (Supplementary Table S1).

Periconia and Periconia_byssoides were predominantly found in non-obese MAFLD compared to obese MAFLD or HCs (Figures 4A,B). Similarly, Rhizopus_microsporus_var_chinensis, Rhizopus_microsporus_var_rhizopodiformis, and Pleosporales_unclassified showed elevated levels in MAFLD patients with moderate to severe HS relative to those with mild HS or HCs (Figures 4C,D).

Distinct fungal profiles were associated with different stages of ALD. Genera like Pleosporales_unclassified and species including Candida_albicans, Auricularia_polytricha, Pleosporales_unclassified, and Scopulariopsis_brevicaulis were significantly elevated in patients with ALC compared to AFL or HCs (Figures 5A,B).