Disentangling environmental and disease-specific signatures in the gut microbiome of psoriasis: discovery of Fimenecus sp. as a novel biomarker and characterization of the gut virome

Abstract

Background

The contribution of the gut microbiome to the pathogenesis of psoriasis remains a subject of debate, with inconsistent findings across studies likely confounded by environmental factors. This study aimed to statistically disentangle the effects of a shared household environment from disease-specific microbial signatures in psoriasis. Our objective was to identify novel, multi-kingdom biomarkers, encompassing bacteria and viruses, that hold significant diagnostic and therapeutic potential.

Methods

We conducted a nested case-control study, performing shotgun metagenomic sequencing on stool samples from 143 participants. The cohort comprised 98 psoriasis patients, 28 healthy cohabiting relatives, and 17 unrelated healthy controls. A comprehensive multi-kingdom analysis of bacteria, viruses, and their associated metabolic pathways was implemented. To ensure the robustness of our findings, a two-stage discovery-validation strategy was employed to identify distinct microbial features associated with psoriasis.

Results

Our analysis revealed that the shared household environment was the predominant factor shaping the overall gut microbiome structure. Despite this strong confounding effect, we successfully identified a novel bacterial species, Fimenecus sp000432435, as a robust biomarker for psoriasis, achieving an area under the curve (AUC) of 0.84. Genomic functional prediction indicated that this species encodes pathways with the potential for B-vitamin and secondary bile acid biosynthesis. Furthermore, characterization of the gut virome identified five disease-associated bacteriophages. Among these, vBin_422 exhibited a significant negative correlation with the abundance of Fimenecus sp000432435, suggesting a potential ecological interaction. Notably, the biotin biosynthesis pathway was negatively correlated with disease severity, whereas specific viral taxa showed a positive correlation with systemic inflammatory markers within the patient cohort.

Conclusions

Controlling for environmental confounders reveals that psoriasis is associated with sparse but distinctmicrobial signatures rather than broad dysbiosis. Fimenecus sp000432435 is a promising candidate for non-invasive diagnostics, while the characterized virome opens new therapeutic avenues targeting bacteriophage-bacteria interactions in psoriasis management.

Trial registration

ChiCTR-IOR-17011075. Registered 6 April 2017, http://www.chictr.org.cn/showproj.aspx?proj=17334

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08013-4.

Background

Psoriasis is a chronic, immune-mediated inflammatory skin disease affecting 2–3% of the global population [1]. Despite the success of biologic agents targeting the IL-23/Th17 axis [2], approximately 18–23% of patients exhibits primary non-response to IL-17 inhibitors [3], with long-term studies demonstrating that only 66% of patients remain on treatment at two year [4], highlighting a critical need for a more nuanced understanding of the disease’s etiology. While familial aggregation has traditionally attributed to shared genetics, large-scale cohort studies suggesting that shared environmental exposures may be an equally, if not more, significant driver of disease risk [5, 6].

This clinical complexity has directed research towards the gut microbiome as a key modulator of systemic inflammation via the “gut-skin axis” [7], a paradigm where intestinal dysbiosis compromise gut barrier integrity, leading to the systemic translocation of microbial components that can trigger or exacerbate cutaneous inflammation [8–10]. However, the search for a consistent psoriasis-specific gut microbial signature has yielded a landscape of conflicting results. As documented in recent systematic reviews [11, 12], discrepancies in findings regarding community structure and diversity suggest that uncontrolled confounding variables may be obscuring the true relationship between the gut microbiome and psoriasis.

The shared household environment is a well-documented factor influencing gut microbial composition [13], yet its impact has not been systematically quantified in psoriasis research. This represents a critical methodological challenge: how to effectively disentangle these environmental effects from genuine disease signatures. Consequently, our study posits that to uncover reliable biomarkers, it is first essential to adopt a study design capable of statistically controlling for this cohabitation effect [14].

Here, we conduct a shotgun metagenomic analysis of a unique cohort comprising psoriasis patients, their cohabiting healthy family members, and unrelated healthy controls. This nested case-control design provides a powerful framework to statistically disentangle microbial signatures of the shared environment from those specifically linked to the psoriatic disease state. We hypothesized that the shared environment is a more powerful driver of overall microbiome structure than the disease state itself, and that after accounting for this, specific and stable microbial taxa associated with psoriasis would emerge. By disentangling these effects, this work aims to identify reliable microbial biomarkers and provide a refined framework for future mechanistic studies.

Methods

Study participants and sample collection

The 98 patients with psoriasis and 45 healthy controls in total were included in the dermatology clinic in Guangdong Provincial Hospital of Chinese Medicine. Participants were excluded if they met any of the following criteria: (1) pregnancy or lactation; (2) history of serious hematologic, respiratory, or circulatory diseases; (3) severe infectious diseases, tumors, or other malignancies; (4) primary or secondary immunodeficiency; (5) antibiotic treatment within the previous 3 months; (6) systemic hormonal treatment within the previous 4 weeks; or (7) treatment with biological agents within the previous 6 months. Patients completed questionnaires, underwent a dermatological examination, and were evaluated disease severity using psoriasis area and severity index (PASI), body surface area (BSA) and visual analogue scale (VAS) scores. To control for potential household effects on the gut microbiome, each psoriasis patient contributed one cohabiting household member (1:1 paired design, n = 28 pairs). Cohabitants were selected based on: (1) minimum cohabitation duration of one year; (2) availability and willingness to participate; and (3) absence of psoriasis or chronic inflammatory conditions. Cohabitants were primarily spouses/domestic partners, with the remainder being adult siblings or parents, prioritizing relationships with greatest environmental sharing. This 28-pair dataset served as the discovery dataset for identified microbiota markers and established an accurate and diagnostic specific model. Patients must meet the criteria for the diagnosis of Psoriasis referred to in the Clinical Guidelines of Psoriasis 2008 reported by the Chinese Medical Association.

Metagenomic sequencing and data processing

Stool samples were collected using OMNIgene GUT OM-200 kits (DNA Genotek Inc., Canada) and stored at −80 °C. Microbial DNA was isolated using the NucleoSpin Soil kit (MACHEREY-NAGEL, Germany). Sequencing libraries were prepared and sequenced on an Illumina HiSeq X Ten platform to generate 150-bp paired-end reads.

Raw sequencing reads were quality-filtered using Trimmomatic (v0.33) [15], and host DNA was removed by mapping against the human reference genome (hg19) using Bowtie2 (v2.3.4.3). Bacterial taxonomic and functional profiles were generated using MetaPhlAn4 [16] and HUMAnN 3.0 [17], respectively. For virome analysis, viral contigs were assembled, identified, and quality-controlled using the Metagenomic Viral Pipeline (MVP) [18], with functional annotation against PHROGS and PFAM database.

Statistical analysis

Statistical analyses were performed in R. Alpha diversity was calculated using the `MicrobiomeStat` package [19] (https://github.com/cafferychen777/MicrobiomeStat) and compared between group with sex and BMI. Beta diversity was visualised using principal coordinates analysis (PCoA) on Bray-Curtis dissimilarity and tested for significance with permutational multivariate analysis of variance (PERMANOVA). The bacterial co-occurrence network was constructed based on Pearson correlation coefficients of species-level abundances using the ggClusterNet package [20].

A two-stage discovery-validation strategy was used for biomarker identification. Differentially abundant taxa were screened in the discovery cohort using LEfSe [21] and rigorously validated in the second cohort with linear models (MaAsLin2 [22], LinDA [23]) adjusting for sex and BMI. To assess the diagnostic potential of the primary bacterial biomarker, a random forest classifier was trained on the balanced discovery cohort (28 Pso vs. 28 RC) and independently evaluated on the validation cohort (70 Pso vs. 17 HC). Species abundances were CLR-transformed with a pseudocount of 1. Feature selection employed a two-step approach: collinearity removal using `findLinearCombos()` followed by the Boruta algorithm (v8.0.0) [24]. The Boruta algorithm identifies important features by comparing their importance against shadow features—randomly permuted null variables serving as a statistical baseline; only features consistently outperforming shadow features are classified as “Confirmed”. A Random Forest classifier was trained using 5-fold cross-validation with 5 repeats, with hyperparameter tuning via grid search for `mtry` (1–10). Model performance was evaluated using multiple metrics, including the area under the receiver operating characteristic curve (ROC-AUC), precision, recall, F1-score, balanced accuracy, and Cohen’s Kappa. Given the class imbalance in the validation set, Precision-Recall (PR) analysis and Synthetic Minority Over-sampling Technique (SMOTE) sensitivity analysis were additionally performed to confirm the robustness of the results. DeLong’s test was used to compare ROC curves. The complete details of the data processing pipeline, including normalization procedures, feature engineering strategies, hyperparameter tuning, and class imbalance handling, are provided in Additional file 3: Supplementary Methods.

Partial Spearman correlations, adjusting for sex and BMI via a linear model residual-based approach, were used to assess associations between microbial features and host clinical phenotypes. The functional potential of key bacterial species was annotated using EggNOG-mapper [25]. A p-value < 0.05 was considered statistically significant.

Peripheral blood mononuclear cells (PBMC) isolation and flow cytometry

PBMC were isolated from a subset of 26 psoriasis patients using Ficoll-Paque density gradient centrifugation. This immunological profiling was designed to explore within-patient correlations between microbiome features and immune phenotypes, rather than to compare immune profiles between patients and healthy individuals. Following in vitro stimulation with PMA/ionomycin in the presence of Brefeldin-A, cells were stained for surface and intracellular markers, including CD4, CD25, IFN-γ, TNF-α, IL-17 and FoxP3. Data were acquired on a BD FACSAria III flow cytometer and analyzed using FlowJo software.

Results

Cohort characteristics reveal significant baseline differences between patient and control groups

The study enrolled 143 participants: 98 psoriasis patients (Pso) and 45 healthy controls, the latter stratified into 28 cohabiting relatives (RC) and 17 unrelated controls (HC) (Additional file 2: Supplementary Table S1). Patients presented with moderate disease severity (mean PASI 6.4 ± 4.4; mean BSA 8.7 ± 10.9) (Table 1). Significant baseline differences were observed between the psoriasis and combined control groups, including a higher proportion of males in patient group (74.5% vs. 30.2%, Chi-squared test p < 0.001), a higher BMI (24.0 ± 4.0 vs. 21.9 ± 3.2, student t test p = 0.002) and differences in smoking status (27.4% vs. 2.8%, Chi-squared test p = 0.004). However, smoking and alcohol consumption data exhibited substantial missing values (20% in healthy vs. 3% in patients), which would have reduced the effective sample size and exacerbated group imbalance if included in multivariable adjustment. This sex and BMI imbalance was also observed within the 28 cohabiting patient-relative pairs (Additional file 2: Supplementary Table S2). Consequently, all subsequent statistical analyses were adjusted for sex and BMI as potential confounders, both of which had complete data across all participants.

Table 1.

Statistics of the cohorts in this study

Index	Psoriasis subjects	Healthy subjects	p
Sample size	98	45	-
Age (years)	39.1 ± 12.5	43.2 ± 13.9	0.10
Sex (%male)	74.5%	30.2%	<0.001*
BMI	24.0 ± 4.0	21.9 ± 3.2	0.002*
PASI	6.4 ± 4.4	-	-
Median PASI	5.4	-	-
BSA	8.7 ± 10.9	-	-
Median BSA	6.15	-	-
Smoking (Y/N)	26/69	1/35	0.004*
Drinking (Y/N)	21/74	3/33	0.12

* p < 0.05

Overview of the gut metagenomic and viromic landscape

Shotgun metagenomic sequencing of the 143 stool samples produced a high-quality dataset, yielding a mean of 61.3 ± 7.1 million host-depleted reads per sample. Taxonomic profiling identified 1382 distinct bacterial species, with the community structure typical of the human gut, dominated by the phyla Bacteroidetes (mean relative abundance ~72%) and Firmicutes (mean relative abundance ~21%) (Additional file 1: Supplementary Figure S1). For the virome, our assembly-based workflow identified 11,843 unique viral operational taxonomic units (vOTUs), the vast majority (98.3%) of which were classified as double-stranded DNA (dsDNA) bacteriophages of the order Caudoviricetes. Functional annotation of the 350,035 predicted viral genes highlighted a substantial proportion of “viral dark matter”, as most genes could not be assigned a specific function (Additional file 2: Supplementary Table S3). Genes annotated via PHROGS database were dominated by core phage lifecycle processes such as DNA, RNA and nucleotide metabolism (6.1% of all predicted genes), head and packaging (4.5%), and tail (4.1%) (Additional file 1: Supplementary Figure S2a). Complementary annotation against the PFAM database provided protein family information for 67,874 genes, with the most prevalent families similarly related to core enzymatic functions (e.g., P-loop NTPase) and viral structures (e.g., Phage-related proteins) (Additional file 1: Supplementary Figure S2b). Furthermore, analysis against the ADS database identified 1105 genes associated with Anti-Defense Systems, overwhelmingly dominated by anti-CRISPR-Cas functions (Additional file 1: Supplementary Figure S2c). The absence of RNA-dependent RNA polymerase (RdRP) genes confirmed the DNA-dominant nature of the virome captured.

Household environment primarily drives gut microbiome community structure

Analysis of beta diversity revealed that the shared household environment was a more significant factor than disease status in shaping community composition. A PERMANOVA on the bacteriome found no significant overall community structure between the three study groups (Pso, RC, and HC) (p = 0.098), with the grouping factor explaining only 1.9% of the total variance (R² = 0.019; Fig. 1a). Critically, patients’ bacteriomes were significantly more similar to their cohabiting relatives (RC) than to unrelated healthy controls (HC) (Fig. 1b–c). This environmental modulation was also evident in alpha diversity; after confounder adjustment, patients had significantly lower Shannon diversity than unrelated controls (adjusted p = 0.027), but this difference disappeared when comparing patients to their cohabiting relatives (Fig. 1d). Network topological properties, which differed between patients and unrelated controls, showed convergence when the analysis was restricted to cohabiting pairs, indicating these structural characteristics were also strongly influenced by shared environmental factors (Additional file 1: Supplementary Figure S3).

Fig. 1.

Comparative analysis of gut bacteriome diversity and community composition among Pso, RC, and HC groups. (a) Principal coordinates analysis (PCoA) plot of bacterial species composition based on bray-curtis dissimilarity. Each point represents an individual sample. The accompanying table displays the results of the PERMANOVA analysis. (b) Boxplots comparing the broad group-wise bray-curtis dissimilarity (beta diversity) of bacterial communities between the three groups (representing general pairwise distances between any two individuals across each group pair). (c) Dendrogram and stacked bar plots illustrating the microbial community composition across the Pso, RC, and HC groups. The dendrogram showed the hierarchical clustering of the groups based on their microbial profiles. The stacked bar plots represented the relative abundance of major bacterial phyla. Each bar reflected the proportional composition of these phyla within each group. (d) Alpha diversity of the gut bacteriome across the three cohorts, measured by the Shannon and pielou indices. Statistical significance was assessed using a general linear model adjusted for confounders. Pso: psoriasis group (n = 98), RC: relatives control group (n = 28), HC: healthy control group (n = 17)

This powerful cohabitation effect was mirrored in the gut virome. Although no significant overall group difference was observed (PERMANOVA, R² = 0.016, p = 0.18; Fig. 2a), pairwise analysis confirmed that viromes from individuals within the same household were significantly more homogenous (Fig. 2b). However, in contrast to the bacteriome, viral alpha diversity remained stable across all groups after confounder adjustment (Fig. 2c). These findings collectively establish that environment influence masks community level disease signals, necessitating our paired-sample design, to identify true microbial features of psoriasis.

Fig. 2.

Comparative analysis of virome diversity and community composition among Pso, RC, and HC groups. (a) Principal coordinates analysis (PCoA) plot of viral community composition (vOTU level) based on bray-curtis dissimilarity. Each point represents an individual sample. The accompanying table displays the results of the PERMANOVA analysis. (b) Boxplots comparing bray-curtis dissimilarity (beta diversity) across three comparison categories. To characterize the cohabitation effect mentioned above, distances were categorized as: inter-control (distances between any two unrelated healthy individuals from either RC or HC groups), inter-patient (distances between any two unrelated psoriasis patients), and intra-pair (specifically representing the distances within each of the 28 matched patient-relative cohabiting pairs). (c) Alpha diversity of the gut virome across the three groups, measured by the Shannon and pielou indices. Statistical significance was assessed using a general linear model adjusted for confounders. Pso: psoriasis group (n = 98); RC: relatives control group (n = 28); HC: healthy control group (n = 17)

A novel Fimenecus species is associated with psoriasis

To overcome the strong environmental effects, we employed a stringent multi-stage. An initial screening on the discovery set (n = 28 patient-relative pairs) using LEfSe and linear models yielded a shortlist of eight candidate species (Additional file 1: Supplementary Figure S4a-b). To further prioritize this list, an alternative rigorous differential abundance analysis (LinDA) confirmed that a single uncharacterized species, s__GGB51647_SGB4348, was the most significantly enriched feature in psoriasis patients (Fig. 3a). This biomarker was taxonomically classified as Fimenecus sp000432435.

Fig. 3.

Identification and validation of Fimenecus sp000432435 as a biomarker for psoriasis (a) Differential abundance analysis between the 28-paired Pso and RC groups at species level using LinDA (linear regression framework for differential abundance analysis), with adjusted for sex and BMI. The plot displays the coefficient (indicating direction and magnitude of abundance differences) against the mean abundance. Each point represents a species, with its size corresponding to mean abundance, and color indicating prevalence (darker colors represents higher prevalence). Fimenecus sp000432435 is the most significantly enriched species in the Pso group. Significance was determined using a FDR threshold of <0.1. (b) Feature importance analysis using a random forest classifier combined with the Boruta algorithm to identify key bacterial species distinguishing psoriasis patients from controls. The analysis was performed on the training set of 28-paired Pso and RC groups. The x-axis represents the importance score of each feature. Species colored in red were significantly enriched in the Pso group, while those in blue were enriched in the RC group, based on prior differential abundance analysis. (c) Receiver operating characteristic (ROC) curve analysis evaluating the classification performance of different models on the independent validation set (70 Pso and 17 HC individuals). Model 1 (green line) includes the nine species identified by Boruta. Model 2 (cyan line) includes a refined set of three species after addressing collinearity. Model 3 (red line), built using only Fimenecus sp000432435, achieved a superior AUC of 0.84 (95% CI: 0.74–0.94), significantly outperforming the other models. (d) Prevalence of Fimenecus sp000432435 across the psoriasis (Pso, n = 98), cohabiting relative control (RC, n = 28), and unrelated healthy control (HC, n = 17) groups. The waffle chart visually represents the detection rate, where each square corresponds to one individual. Colored squares indicate samples in which the species was detected, while grey squares indicate non-detection. The accompanying table presents the detailed statistical results from the pairwise Fisher’s exact test for prevalence. For each comparison, the table lists the sample sizes (N), raw p values (P), and FDR adjusted p values

To evaluate the diagnostic potential of this finding, we constructed and compared several random forest classifiers (see Additional file 3: Supplementary Methods for full pipeline details). A model built using a panel of nine species selected by the Boruta algorithm showed only moderate performance on the validation set (AUC = 0.66, Fig. 3b–c, Additional file 1: Supplementary Figure S5a). A refined three-species model showed improved but not significantly different performance (AUC = 0.76, p = 0.36; Fig. 3c; Additional file 1: Supplementary Figure S5b). Remarkably, a classifier built solely using Fimenecus sp000432435 achieved an AUC of 0.84 (95% CI: 0.74–0.94; Fig. 3c), significantly outperforming both the nine-species (p = 0.01) and three-species models (p = 0.04) in a Delong’s test. This single-species model also demonstrated robust performance across multiple evaluation metrics, achieving a precision of 0.93, and an F1-score of 0.87 (Additional file 3). Furthermore, Precision-Recall analysis confirmed the model’s reliability under the imbalanced validation set conditions (PR-AUC = 0.83), and SMOTE sensitivity analysis verified that these results were not artifacts of class distribution (PR-AUC = 0.96 after rebalancing; Additional file 3). This superior performance of the single-species model underscores its power as a standalone biomarker. This strong association was further confirmed by its prevalence pattern: the detection rate of Fimenecus sp000432435 was significantly higher in psoriasis patients compared to the combined control group (relatives and healthy controls). Quantitatively, the presence of this species was associated with a substantial increase in psoriasis risk, yielding an Odds Ratio (OR) of 10.51 (95% CI: 4.26–27.57, Fisher test p < 0.001). While the biomarker displayed moderate specificity (60.0%), its strong enrichment in the disease group underscores its potential as a risk indicator (Fig. 3d).

The genomic functional profile of the psoriasis-associated Fimenecus sp000432435 highlights predicted pathways relevant to host-microbe interactions

Genomic analysis of Fimenecus sp000432435 revealed its genomic coding capacity was significantly concentrated in 15 KEGG pathways (adjusted p < 0.05, Fig. 4a–b; Additional file 2: Supplementary Table S4). These included foundational metabolic processes (e.g., purine and pyrimidine metabolism) and notably, potentially involved in producing host-interactive molecules, including secondary bile acid biosynthesis and the biosynthesis of B vitamins (riboflavin B2 and biotin B7). Interestingly, a community-level analysis of the entire metagenome revealed that the biotin biosynthesis pathway (MetaCyc ID: PWY-5005) was the only functional pathway significantly enriched in the psoriasis discovery set, though it did not meet the significance threshold in the validation cohort (Additional file 1: Supplementary Figure S6).

Fig. 4.

Functional annotation and enrichment analysis of microbial species Fimenecus sp000432435. (a) Overview of the functional annotation of the representative genome sequence for Fimenecus sp000432435. The chart summarizes the annotation results from EggNOG-mapper, showing the number of predicted protein-coding genes and the proportion that could be assigned to COG (clusters of orthologous groups) and KEGG (Kyoto encyclopedia of genes and genomes) functional categories. Of these, 114 genes had KEGG annotation, involving 33 KEGG module functions. (b) KEGG pathway over-representation analysis of the annotated genes. The bubble plot displays the top 10 pathways that were significantly enriched (adjusted p < 0.05) as determined by a hypergeometric test using R package `microbiomeprofiler`. The y-axis lists the enriched pathways, while the x-axis shows the GeneRatio (the ratio of genes from the input list that are annotated to the pathway). The size of each bubble corresponds to the number of genes from the input list found in the pathway (Count), and the color intensity reflects the statistical significance, with darker colors indicating lower adjusted p values. The analysis highlights enrichment in both foundational metabolic pathways and pathways for producing host-interactive molecules like B vitamins and secondary bile acids

Five viral taxa, including a functionally distinct phage, are linked to disease status

An initial screening on the discovery set using LinDA identified 71 candidate viral taxa nominally associated with psoriasis (p < 0.05; Fig. 5a). To stringently validate these candidates, an independent LinDA analysis was performed on the validation set, we identified five dsDNA bacteriophages that were consistently associated with disease status across both cohorts (four psoriasis-enriched and one healthy-enriched; Fig. 5b). One psoriasis-enriched viral bin, vBin_422, was functionally distinct. The vBin_422 was composed of three distinct vOTUs: PsV.C10_k141_47179_13834_26417, Control.A15_k141_143498, and PsV.A21_k141_37751. Its genome contained a unique collection of eight rare genes organized into three distinct functional groups: replication/regulation (helicase, glycosyltransferase, and an Lpa-like transcriptional activator), host interaction (Ref-like RecA filament endonuclease, Doc-like toxin, and NinI-like serine-threonine phosphatase), and morphogenesis (head formation proteins PmgG-like and PmgS-like) (Fig. 5c). The convergence of these rare functional genes, particularly those implicated in host interaction, suggests that vBin_422 may possess a distinct functional profile enriched in putative host-interaction elements, distinguishing it from control-associated viruses.

Fig. 5.

Discovery, validation, and functional characterization of psoriasis-associated viral taxa. (a) Results of the initial discovery screen for disease-associated viruses. The bar chart displays the effect size (coefficient) from the LinDA analysis performed on the 28 psoriasis (Pso) and relative control (RC) pairs. Each bar represents one of the 71 viral features that were nominally associated with psoriasis (p < 0.05). Positive coefficients indicate enrichment in psoriasis, while negative coefficients indicate enrichment in controls. The five high confidence features that were subsequently confirmed in the validation set are explicitly labeled. (b) CLR-transformed abundance of the five validated viral taxa in both the discovery and validation cohorts. The violin plots overlaid with boxplots show the abundance distributions for psoriasis (Pso) and control (RC or HC) groups. The consistent differential abundance across two independent sets of samples underscores the robustness of these viral biomarkers. p-values from the LinDA models are displayed for each comparison. (c) Gene map of the psoriasis-enriched viral bin, vBin_422. The map is organized by the three constituent vOtus that compose this bin, as detailed in the main text. Each arrow represents a predicted gene. Genes predicted to be rare are highlighted with a bold outline. Genes are colored according to their predicted functional category: replication and regulation (green), host interaction (orange), Morphogenesis (blue), auxiliary functions (pink), and hypothetical protein of unknown function (grey)

Exploratory analysis reveals specific associations between microbial signatures and host immunity

To explore potential associations between microbial features and host immunity, we performed partial Spearman correlation analysis within the patient cohort (n = 26 patients with PBMC data), controlling for confounders. Notably, as healthy control blood samples were not available, this analysis examines within-group associations rather than disease-specific differences, and revealed complex ecological and clinical associations (Fig. 6a–b). Fimenecus sp000432435 abundance was strongly positively correlated with two psoriasis-enriched viruses (PsV.C35_k141_2633, ρ = 0.60, p < 0.001; Control.A16_k141_7489_9501_52544, ρ = 0.58, p < 0.001) but showed a significant negative correlation with the functionally unique phage vBin_422 (ρ = −0.36, p = 0.008), indicating a moderate inverse association (Fig. 6a).

Fig. 6.

Correlation analyses of key microbial features. (a) Correlations between the primary bacterial biomarker, Fimenecus sp000432435, and the five validated viral taxa. The lollipop plot displays the partial Spearman correlation coefficients (ρ) for each virus-bacterium pair, calculated while controlling for host sex and BMI. Each point represents a virus, colored to indicate the direction and statistical significance of the correlation (p < 0.05). The analysis reveals both significant positive (e.g., orange-red) and negative (e.g., cyan) relationships, as well as non-significant ones (grey). (b) Correlations between key microbial features and host phenotypes. The heatmap displays partial Spearman correlation coefficients (ρ) from an analysis associating microbial features (rows) with host clinical and immunological markers (columns), adjusted for host sex and BMI. The microbial features include the key bacterium, the five validated virus features, and the biotin biosynthesis pathway (pathway ID: PWY-5005). Red cells indicate positive correlations and blue cells indicate negative correlations, with color intensity corresponding to the magnitude of the correlation coefficient. Statistically significant associations are marked with asterisks (* for p < 0.05; ** for p < 0.01)

Clinically, several viral taxa correlated with markers of disease severity and inflammation. The abundance of vBin_422 correlated positively with both the VAS and CD8+ T cell counts. Another virus, Control.A16_k141_7489_9501_52544, was positively correlated with CD8+ T cells and the key psoriatic cytokines IFN-γ and TNF-α. In contrast, the community-level abundance of the biotin biosynthesis pathway (PWY-5005) exhibited a significant negative correlation with the VAS score and CD8+ T cell counts. The bacterial biomarker, Fimenecus sp000432435, did not correlate with any severity scores but showed a specific positive association with peripheral CD4+ T cell counts (Fig. 6b).

Discussion

Our study addresses a critical challenge in psoriasis microbiome research: the confounding effect of the shared household environment. We first confirmed that cohabitation is a key driver of gut microbiome composition, exerting a more profound influence on community structure than disease status. This finding aligns with previous landmark studies [13, 26] and likely explains the inconsistent results reported in the psoriasis literature. By employing a nested case-control design to statistically control this environmental noise, we were able to isolate microbial signatures robustly associated with the disease itself.

We identified a single bacterial species, Fimenecus sp000432435, as a distinct and predictive biomarker for psoriasis. A random forest model using this species alone achieved an AUC = 0.84 that significantly surpassed multi-species models. This suggests the gut microbial signature of psoriasis may be characterized by a focal, high-impact change rather than a diffuse, community-wide dysbiosis. This biomarker contrasts with previous reports highlighting taxa such as Prevotella copri [27], a discrepancy we attribute to cohort-specific environmental confounders. Genomic analysis of Fimenecus sp000432435 revealed enrichment in pathways related to B-vitamin biosynthesis (biotin and riboflavin) and secondary bile acid metabolism. These pathways are of potential interest given their documented roles in immune regulation. Biotin deficiency has been shown to enhance Th17 differentiation in human CD4+ T cells through upregulation of RORγt and suppression of Foxp3 [28], while riboflavin-derived metabolites serve as antigens for mucosal-associated invariant T (MAIT) cells [29], which have been identified as a source of IL-17A in psoriatic skin [30]. Whether Fimenecus sp000432435 influences psoriasis pathobiology through these metabolic capacities, however, cannot be determined from our metagenomic data and requires further functional validation. Notably, while the biotin biosynthesis pathway was enriched in psoriasis patients compared to their cohabiting relatives in our discovery cohort, its abundance within patients was negatively correlated with disease severity (VAS score). This patter – where greater pathway abundance associates with milder symptom – is directionally consistent with the reported protective effects of biotin against Th17-mediated inflammation, though this pathway did not reach significance in the validation cohort and should be interpreted with caution.

The ability to pinpoint such a novel, uncharacterized taxon stems directly from methodological advancements. We utilized MetaPhlAn 4, whose expanded genomic database and refined algorithms permit a far deeper profiling of the microbial community, bringing previously unclassified ‘microbial dark matter’ into focus [16]. While Fimenecus sp000432435 is a newly appreciated member of the gut microbiome, its potential link to host-environment interactions is beginning to emerge. Intriguingly, a recent large-scale “sociobiome” study identified this exact species as being associated with individual and neighborhood socioeconomic status [31]. This external finding powerfully reinforces the central thesis of our work: that potent environmental factors, extending even to the level of social determinants of health, can profoundly shape the gut microbiome. It raises a compelling question for future research: could Fimenecus sp000432435 act as an immunological mediator of psoriasis, or is it primarily a sensitive indicator of a specific environmental exposure or lifestyle pattern, which in turn is the true driver of disease risk? The discovery of this species, therefore, not only provides a promising biomarker but also opens a new window into the complex interplay between environment, microbiome, and psoriatic disease.

This work also provides the first characterization of the gut virome in psoriasis, revealing a disease-associated signature composed of five dsDNA bacteriophages. This finding opens a new dimension for understanding the gut-skin axis, suggesting that bacteriophages, as dynamic regulators of bacterial communities [32], may play a pivotal role in the pathogenesis of psoriasis. Most notably, we discovered a strong negative correlation between the abundance of Fimenecus sp000432435 and a functionally distinct, psoriasis-enriched phage, vBin_422. This finding implies a potential ecological link between the viral and bacterial signatures. While the negative correlation could be consistent with phage-mediated lysis or competitive exclusion, we acknowledge that cross-sectional data cannot definitively establish predator-prey dynamics, and longitudinal studies are required to elucidate the precise ecological nature of this relationship.

Finally, our exploratory analysis of microbe-immune interactions provided preliminary insights into the distinct roles of bacterial and viral signatures. We observed that the psoriasis-enriched Fimenecus sp000432435 was positively correlated with peripheral CD4+ T cell counts. Notably, this correlation did not extend to specific Th17 cells, regulatory T cell subsets, or systemic inflammatory cytokines. This dissociation suggests that Fimenecus might be associated with a generalized expansion of the CD4+ T cell pool rather than specifically skewing the compartment toward a Th17 phenotype. In parallel, the viral signatures displayed heterogeneous immunological profiles. The primary viral biomarker, vBin_422, correlated positively with CD8+ T cell counts, whereas another disease-associated phage, Control.A16_k141_7489_9501_52544, exhibited broader correlations with both CD8+ T cells as well as key inflammatory cytokines including IFN-γ, TNF-α. These observations align with the concept that specific gut phages may interface with adaptive immunity, and in some cases, associate with systemic inflammation, a phenomenon potentially linked to phage DNA sensing mechanisms [33]. We underscore, however, that these findings represent statistical associations derived from a cross-sectional dataset. Future validation using gnotobiotic models is essential to determine whether these microbial features actively influence immune regulation or merely reflect the inflammatory environment of the host.

The principal strength of our study lies in its robust design, complemented by a rigorous analytical framework to parse disease-specific signals from strong environmental noise and correlated these with host immunological markers, a strategy validated in previous psoriasis research [34]. However, its limitations must be acknowledged. As discussed, the cross-sectional design precludes causal inference, and thus experimental validation in future prospective studies will be necessary to confirm these computational predictions. Furthermore, our immunological analysis focused on within-patient correlations between microbial features and immune parameters; future studies incorporating healthy controls would be needed to determine whether these associations are psoriasis-specific. Additionally, our virome characterization, derived from bulk shotgun metagenomics, is primarily sensitive to dsDNA viruses. This is a common and practical approach, but a comprehensive survey of the entire viral landscape, including ssDNA and RNA viruses, would necessitate specialized sample-processing protocols [14]. Finally, while the cohabiting control design effectively minimizes shared household-level confounders, we did not systematically collect detailed metadata on all individual behavioral factors, such as precise dietary habits or physical activity. Consequently, the possibility of residual confounding from unmeasured lifestyle variables cannot be entirely excluded.

Conclusion

In conclusion, by meticulously controlling environmental confounders, our study reveals a distinct and sparse microbial signature in psoriasis. We pinpoint a single bacterial species with significant potential as a clinical biomarker and provide a foundational characterization of the associated gut virome. These findings shift the focus from broad dysbiosis to the influence of specific microbial entities and their ecological interactions, paving the way for new diagnostic strategies and targeted therapeutic interventions for psoriasis.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

AUC

Area under the ROC Curve

BSA

Body Surface Area

BMI

Body Mass Index

CI

Confidence Interval

dsDNA

Double-stranded DNA

FDR

False Discovery Rate

HC

Heathy Control

LEfSe

Linear Discriminant Analysis Effect Size

MaAsLin

Multivariate Association with Linear Models

PASI

Psoriasis Area and Severity Index

Pso

Psoriasis

PBMC

Peripheral blood mononuclear cell

RC

Relatives control

ROC

Receiver operating characteristic

VAS

Visual Analogue Scale

vOTUs

viral Operational Taxonomic Units

Author contributions

ChuL and XF designed the study and supervised the entire investigation. ChuL, HD, DY, CuiL and SY contributed to patient recruitment and sample collection. QQ, JD, JY, CW, XS and LH performed the bioinformatic analysis. QQ, JD, YD, YL and YC contributed to the primary interpretation of analytical outcomes. JD and QQ wrote the first draft of the manuscript. JD and QQ wrote sections of the manuscript. All authors contributed to manuscript revision, read, and approved the final manuscript.

Funding

This project was supported by the grants from National Natural Science Foundation of China (No.U23A6012, U20A20397), Incubation Program for the Science and Technology Development of Chinese Medicine Guangdong Laboratory (HQL2024PZ019), Science and Technology Planning Project of Guangdong Province (2020B1111100005), Science and Technology Planning Project of Guangzhou (No.2024A03J0055), Innovation Team and Talents Cultivation Program of National Administration of Traditional Chinese Medicine (No. ZYYCXTD-C-202204), Guangdong Hospital of Chinese Medicine Youth Grant (Intersection of Chinese Medicine and Mathematics, YN2022QN31), Guangzhou Basic and Applied Basic Research Scheme (2024A03J0732), State Key Laboratory of Dampness Syndrome of Chinese Medicine Special Fund (SZ2024KF06), the Fund of State Key Laboratory of Dampness Syndrome of Chinese Medicine (SZ2021ZZ28, SZ2024KF06, SZ2024KF14), Special Funding for Scientific Research in Traditional Chinese Medicine at Guangdong Provincial Hospital of Traditional Chinese Medicine (YN2024GZRPY084), State Key Laboratory of Traditional Chinese Medicine Syndrome (QZ2025ZZ13), and Construction Project of Guangdong Provincial Distinguished Traditional Chinese Medicine Practitioner Lu Chuanjian’s Heritage Studio (Document No. Yue Zhong Yi Ban Han [2023] 108). Clinical Cooperation Project of Traditional Chinese and Western Medicine for Major and Difficult Diseases (ZDYN-2024-A-020).

Data availability

The metagenome sequencing data have been submitted to CNGB Nucleotide Sequence Archive (CNSA) under accession number CNP0000322 (http://db.cngb.org/cnsa/review/show/CNP0000322_20190704_0cc726b4). Computer code used for the microbiome data analysis is available at github link: https://github.com/qqwxp1987/psoriasis_microbiome.

Declarations

Ethics approval and consent to participate

This study was followed by the Declaration of Helsinki Principles and approved by the ethical committee of the Guangdong Provincial Hospital of Traditional Chinese Medicine (GPHCM B2016-138-02). Written informed consent was obtained from the volunteers after the nature and possible consequences of the studies were explained. The protocol was registered under Chinese Clinical Trial Registry (ChiCTR-IOR-17011075. Registered 6 April 2017, http://www.chictr.org.cn/showproj.aspx?proj=17334).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests