ORIGINAL RESEARCH article

Front. Microbiol., 29 June 2026

Sec. Microorganisms in Vertebrate Digestive Systems

Volume 17 - 2026 | https://doi.org/10.3389/fmicb.2026.1839847

Gut microbiota signatures associated with lithium treatment and clinical response in patients with bipolar disorder

  • 1. Department of Biomedical Sciences, Microbiology and Virology Unit, University of Cagliari, Monserrato, Italy

  • 2. Department of Biochemistry, Institute of Chemistry, University of São Paulo (USP), São Paulo, Brazil

  • 3. Department of Biomedical Sciences, Section of Neuroscience and Clinical Pharmacology, University of Cagliari, Cagliari, Italy

  • 4. Section of Pathology, Department of Biomedical Sciences, University of Cagliari, Cagliari, Italy

  • 5. Unit of Clinical Pharmacology, University Hospital Agency of Cagliari, Cagliari, Italy

  • 6. Section of Psychiatry, Department of Medical Sciences and Public Health, University of Cagliari, Cagliari, Italy

  • 7. Unit of Clinical Psychiatry, University Hospital Agency of Cagliari, Cagliari, Italy

  • 8. Department of Pharmacology, Dalhousie University, Halifax, NS, Canada

Abstract

Lithium, the gold-standard treatment for bipolar disorder, exhibits highly heterogeneous clinical responses and no validated biological predictors of responsiveness are currently available. Emerging evidence suggests that the gut microbiota influences mood disorders and psychotropic drug response, raising the hypothesis that specific microbial signatures may modulate lithium responsiveness. In this study, we characterized taxonomic and functional gut microbiota profiles in 77 patients with bipolar disorder, of whom 40 were receiving lithium (20 responders, 20 non-responders) and 37 were treated with valproate as the main mood stabilizer (valproate, n = 31; lamotrigine, n = 6), with the aim of identifying potential microbial markers of clinical response. Microbiota composition was assessed through 16S rRNA sequencing targeting the V3-V4 region. Differential abundance was evaluated using Analysis of Composition of Microbiomes with Bias Correction (ANCOM-BC2), and the functional potential was inferred using the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2 (PICRUSt2). Our finding showed that lithium treatment was associated with a selective gut microbiota reorganization, including reduced Actinobacteria (Actinomycetota) phylum, notably Coriobacteriia class, and enrichment in Firmicutes (Bacillota) taxa, including Selenomonadales, Megamonas and Clostridia taxa, alongside reductions in primary fermentative and biosynthetic pathways. This shift, characterized by a reduction of primary fermenters and enrichment of secondary fermenters and SCFA-producing taxa, suggests a more efficient fermentative ecosystem in lithium-treated patients. Responders showed enrichment in methanogenic taxa (Methanobrevibacter) and Clostridiales vadinBB60 group when compared with patients treated with other mood stabilizers; however, these differences were not observed in the direct comparison between lithium responders and non-responders. While causal relationships cannot be inferred, these findings indicate treatment-associated microbial patterns and support further investigation into microbiota-directed adjunctive therapies.

1 Introduction

Lithium has represented for decades the reference pharmacological intervention for bipolar disorder (BD) and still remains the mainstay treatment, being able to significantly reduce the risk of relapse and suicide in affected individuals (). However, clinical response to long-term Li treatment is highly heterogeneous, its mechanism of action is only partly understood, and recent meta-analyses have shown that, despite numerous molecular and pharmacogenomic studies, no biomarkers have yet been universally validated to predict Li responsiveness (; ).

In parallel, in recent years, research on the gut microbiota (GM) has highlighted its central role in modulating the gut-brain axis, influencing psychopathological phenotypes and the response to psychotropic treatments (). Review and experimental studies have reported that alterations of the GM are associated with mood disorders, including major depressive disorder and BD (), and that psychiatric medications, including Li, may influence the microbial profile (; ). Together, these findings open the possibility that the GM may harbor predictive biomarkers of pharmacological response or act as an active modulator of therapeutic efficacy.

Lithium exerts multifactorial pharmacological effects through the modulation of multiple targets, including GSK-3β (Glycogen Synthase Kinase-3 beta) and enzymes of the inositol pathway (), through which indirectly affects a number of pathways and mechanism that lead to neurotrophic, neuroprotective and anti-inflammatory properties (; ; ; ; ; ; ; ). Given these pleiotropic hallmarks it is reasonable to hypothesize that Li may also influence the brain-gut axis. Supporting this hypothesis, preclinical studies indicate that Li carbonate can mitigate intestinal mucosal inflammation by modulating the GM and activating regulatory T cells (Treg) through GPR43-dependent mechanisms, leading to an increase in Short-Chain Fatty Acids (SCFAs)-producing bacteria ().

An in vivo experimental study conducted in a porcine model, evaluating different dietary sources of Li (Li carbonate and Li-enriched mushrooms) showed that therapeutic Li doses led to a significant increase in intestinal microbial diversity and richness, particularly in the colon, and to an enrichment of taxa considered beneficial for host health, such as Lactobacillus, Ruminococcaceae, Enterorhabdus, Muribaculaceae, and Coprococcus in pigs. Additional taxa, such as Prevotellaceae and Bacteroidales, were enriched in pigs receiving Li-enriched mushrooms, while Li carbonate increased Clostridia, Ruminococcus, Burkholderia and Bacteroidales at recommended dosages ().

similarly showed that Li, valproate and aripiprazole increased microbial diversity and richness in animal models without impairing intestinal barrier function, with enrichment in Clostridium, Peptoclostridium, Intestinibacter, and Christensenellaceae ().

A recent systematic review on pharmacological treatments in BD reported that Li and antipsychotics were associated with significant variations in intestinal bacterial genera. While SCFA-producing species such as Eubacterium biforme, E. rectale and Clostridium Cluster IV, involved in maintaining the integrity of the intestinal barrier and in modulating the inflammatory response, were favored, opportunistic taxa such as Clostridium perfringens, Klebsiella pneumoniae and Campylobacter hominis increased in other contexts, suggesting possible therapy-associated dysbiosis (). These mixed findings highlight the complex and potentially bidirectional effects of Li on the GM and reinforce the need for targeted studies in patients with BD.

Current evidence therefore supports the hypothesis that psychotropic treatments, including Li, may promote the growth of beneficial bacteria; however, findings remain conflicting and preliminary. These discrepancies likely reflect heterogeneity in study populations, stratification by dietary habits, treatment duration, metabolic comorbidities, concomitant medications and other covariates.

Furthermore, given the gastrointestinal and systemic side effects of Li () and its interactions with diet, medications and comorbidities, its influence on the GM may not be uniformly beneficial. Such side effects could either be independent of its microbiota-modifying properties or, conversely, could potentially serve as indicators for therapy monitoring (). It is plausible that Li promotes a favorable microbial balance in some patients, whereas in others, particularly in non-responders or patients with gastrointestinal side effects, the impact may be minimal or even unfavorable.

In this context, investigating whether the microbiota is differentially modulated in Li-responders and non-responders is essential, as such information may help identify microbial signatures predictive of therapeutic success.

Valproic acid (VPA), frequently used as a clinical comparator in clinical trials for mood stabilization, also exhibits diverse molecular mechanisms, including anticonvulsant and epigenetic actions related to histone deacetylases inhibition, and can affect metabolism and the microbiome (). Consequently, VPA represents an appropriate comparison group to evaluate how two mood stabilizers with distinct mechanisms and safety profiles influence the GM.

From a clinical and translational perspective, comparing the GM profile of patients with BD treated with Li to that of patients treated with valproic acid as the main mood stabilizers, while stratifying responders and non-responders to Li, has three potential relevant outcomes: I) identifying microbial biomarkers associated with Li response to support personalized medicine; II) clarifying microbiota–metabolite–immune mechanisms influencing Li efficacy and tolerability; III) enabling future microbiota-targeted interventions, such as dietary strategies, probiotics, prebiotics and microbial therapeutics, to complement pharmacological treatment.

The existing literature on interactions between psychiatric pharmacology and the microbiome suggests a promising yet still emerging field, with a clear need for well-characterized cohort studies that rigorously control for confounders, such as drug exposure, treatment duration, disease history, metabolic variables, diet and antibiotic use.

To explore these interactions, we compared the GM profile of patients with BD treated with Li (n = 40) with a control group treated with other mood stabilizers (valproate, n = 31; lamotrigine, n = 6). We further stratified Li-treated patients into responders (n = 20) and non-responders (n = 20) to identify candidate microbial signatures associated with therapeutic responsiveness. This design allowed us to investigate whether specific microbial signatures differ in relation to both Li treatment and clinical response.

Note that the lack of an untreated control group is largely due to the difficulty in recruiting truly drug-naïve patients, as pharmacological treatment is often initiated before a definitive diagnostic assessment is completed. It would have allowed a more precise discrimination between microbiota alterations potentially related to bipolar disorder and those associated with pharmacological exposure.

2 Materials and methods

2.1 Study design and characteristics of the sample

Patients recruited by trained clinical psychopharmacologists and psychiatrists at the Unit of Clinical Pharmacology, University Hospital Agency of Cagliari, and at the Unit of Clinical Psychiatry of the University of Cagliari. The recruitment pipeline, diagnostic tools, and inclusion and exclusion criteria were agreed between the two centers and were as such harmonized. Patients had a diagnosis of BD type I or BD type II according to the criteria of the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (APA) (). The exclusion criteria included the presence of other psychiatric or severe organic disorders, ongoing chemo/radiotherapy, severe psychopathology, gastrointestinal diseases, treatment with corticosteroids, proton pump inhibitors, antimicrobials, antibiotics, prebiotics and/or probiotic in the 3 months preceding sample collection, and assumption of substance abuse.

The study included 77 patients, of whom 40 were receiving Li at the time of recruitment (20 responders and 20 non- responders), while 37 were treated with other mood stabilizers (valproate, n = 31; lamotrigine, n = 6).

At recruitment, patients had been euthymic for at least 1 year and were stable on their current treatment. After enrollment, patients underwent multidimensional assessment, including clinical, anthropometric and lifestyle evaluation. At the same time, GM composition and predicted functional activity were assessed from fecal samples collected during the euthymic phase.

Patients receiving maintenance treatment with Li or other mood stabilizers were further stratified according to long-term treatment response. Response to treatment was evaluated using the Retrospective Criteria of Long-Term Treatment Response in Research Subjects with Bipolar Disorder (Alda scale) (), which provides a validated quantitative measure of clinical improvement during mood stabilizer treatment, with total scores ranging from 0 to 10 after adjustment for potential confounding factors. Briefly, this scale quantifies the degree of improvement in the course of treatment (A criterion or A score) expressed as a composite measure of change in frequency and severity of mood symptoms. The A score is weighed against 5 factors (B criteria) which allow one to determine if the observed improvement is a result of the treatment rather than a spontaneous improvement or an effect of additional medication. Specifically, the B criteria consider: the number of episodes before/off the treatment (B1), the frequency of episodes before/off the treatment (B2), the duration of the treatment (B3), the compliance during period(s) of stability (B4) and the use of additional medication during the period of stability (B5). The total score (TS) is obtained by subtracting the B score from the A score.

Based on the results of the inter-rater agreement (), a total score ≥7 was considered indicative of a favorable response (; ). More specifically, the study by , which evaluated the inter-rater agreement [Kappa (κ)] of twenty-nine ConLiGen sites who took part in a two-stage case-vignette rating procedure, showed that the highest K was obtained for the TS cut off point of 7. This confirmed what was previously shown by , further supporting the use of this cut-off for binary definition of full response to lithium and to other mood stabilizers. This same value has been used and validated in many other studies across the years (; ). Considering the small number of responders in the group of non-Li treated patients (responders to valproic acid, n = 3; responders to lamotrigine, n = 1), responders and non-responders to Li were compared with overall sample of patients treated with other mood stabilizers for both clinical variables and gut microbial profile.

2.2 Clinical, anthropometric, and lifestyle evaluation

All patients underwent clinical assessment, including age, sex, age at disease onset, illness duration in years, bipolar sub-diagnosis, treatment response status, use of concomitant mood stabilizers and number of depressive and manic episodes. Anthropometric evaluation, including measurement of Body Mass Index (BMI), was performed following current standards (), as previously described ().

Sex was recorded as a biological variable (male/female) based on sex assigned at birth as reported in the medical records. In this study, the term sex refers to biological attributes (chromosomal, hormonal and anatomical characteristics), while gender identity was not assessed.

Age was categorized in two different ways. First, participants were divided into three groups: early adulthood (25–45 years), midlife (46–60 years) and late life ( >60 years). Second, age was dichotomized into two groups: adults (25–60 years) and late life ( >60 years).

Among lifestyle factors, physical activity, smoking status, adherence to the Mediterranean diet (assessed through its validated score) were evaluated. Comorbidities were additionally recorded. Nutritional assessment was based on administration of the standardized and validated Mediterranean Diet Score, (MDS, range 0–55) questionnaire ().

2.3 Gut microbiota analysis

2.3.1 Sample collection

A stool sample from each patient was self-collected during the euthymic phase and delivered to the laboratory within 3 h. Fresh samples were aliquoted and stored at -80°C until further processing.

2.3.2 Genomic DNA extraction, bacterial DNA quantification, and 16S libraries preparation and sequencing

Genomic DNA extraction from stool samples and bacterial DNA quantification were performed as previously described (). The library preparation protocol has been detailed elsewhere (). Barcoded 16S amplicon libraries targeting the V3-V4 hypervariable region of the bacterial 16S rRNA gene were generated using specific primers and the Nextera XT index kit (Illumina, Inc., San Diego, CA, United States). Library size and quality were verified using the D1000 reagents kit (Agilent Technologies, Santa Clara, CA, United States) on the Tapestation 4200 system (Agilent Technologies, Santa Clara, CA, United States). Quantification, normalization, pooling of 16S libraries and the PhiX control preparation were performed as previously described (; ). The combined 16S library and PhiX control were denatured and sequenced on the MiSeq Illumina platform using a v3 paired-end run (2 × 300 cycles) and the MiSeq v3 Reagent Kit (Illumina).

2.4 Bioinformatic and statistical analysis

Participant data were summarized as mean ± standard deviation (SD) for continuous variables and n (%) for categorical variables. Between-group comparisons were performed using Welch’s t-test or ANOVA for normally distributed continuous variables, Mann–Whitney U or Kruskal-Wallis tests for non-normally distributed variables, and Chi-square or Fisher’s exact tests for categorical variables. Post-hoc pairwise comparisons with Benjamini-Hochberg correction were applied when global differences were significant. Effect sizes were calculated for significant comparisons. Statistical significance was set at p < 0.05.

Sequences were processed using QIIME2 (). Quality control was performed using FastQC () and MultiQC (). Primers were removed using cutadapt (q2-cutadapt QIIME 2 plugin) (). Using DADA2 (q2-dada2 QIIME 2 plugin), reads were trimmed and filtered by quality (). High-quality reads were denoised and merged to produce amplicon sequence variants (ASVs). ASVs were aligned using MAFFT (), and a phylogenetic tree was constructed using FastTree2 () within the q2-phylogeny plugin of QIIME2. The p-sampling-depth parameter was set to 18,000, below the minimum read count across samples, ensuring normalization and comparability of data. Taxonomic assignment of ASVs was performed using a Naive-Bayes classifier through the q2-feature-classifier plugin of QIIME2, trained on the specific primer pairs used in this study and on the SILVA (v132) database (). We are aware that some of the taxon names that we use have changed [for example, Actinobacteria is now named Actinomycetota; Euryarchaeota is now named Methanobacteriota ()]. However, we chose to report the old names because they are those given in reference to the database that we used (), thus ensuring consistency and reproducibility.

Alpha and beta diversity metrics were calculated using the q2-diversity plugin. Statistical differences in alpha diversity were assessed using the linear regression model for normally distributed metrics (Observed ASVs, Chao1, and Faith’s Phylogenetic Diversity), a beta regression model for metrics ranging between 0 and 1 (Evenness and Simpson), and rank-based linear model for non-normal distribution (Shannon), adjusting for relevant clinical covariates. The beta diversity was visualized through Principal Coordinates Analysis (PCoA). Dotted ellipses indicated 95% confidence intervals for a multivariate t-distribution, computed with the stat ellipse function from ggplot2 (R package) (). Significance differences in beta diversity distances were tested using PERMANOVA test, with the adonis2 function from the vegan package in R, with p ≤ 0.05 considered statistically significant.

Differential abundance analysis was performed using ANCOM-BC2 (Analysis of Composition of Microbiomes with Bias Correction v.2.9.1) in R (). The analysis was controlled for potential confounders, including age, years of illness, years of treatment, use of additional mood stabilizers, antipsychotic treatment at recruitment, number of manic episodes, total number of episodes, physical activity, MedDiet score and hypertension.

Predicted functional metagenomic pathways were inferred using the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2 (PICRUSt2 v.2.5.2) () and classified according to the Metabolic Pathway (MetaCyc) database (). Statistical comparisons of pathway abundances were performed using the ANOVA-Like Differential Expression (ALDEx2) (; ), using the Wilcoxon rank-sum test to compare two groups (patients treated with Li and patients treated with other mood stabilizer) and the generalized linear model (glm) function to compare three groups (Li-responders, Li non-responders, and patients treated with other mood stabilizer). Benjamini-Hochberg correction was applied for multiple testing. All microbiome analyses (statistics and plotting) were performed in R, and all plots were generated using the ggplot2 package ().

3 Results

3.1 Anthropometric, metabolic, lifestyle, and health status evaluation

The study included 77 patients: 40 (14 men and 26 women) were treated with Li and 37 (15 men and 22 women) were treated with valproic acid as main mood stabilizer (see Table 1). The two groups did not differ for sex, BMI, smoking status, adherence to the Mediterranean diet, age at illness onset, subtype diagnosis, treatment response, use of other mood stabilizers, number of depressive episodes, and presence of diabetes.

TABLE 1

VariableLi-treated
(n = 40)		Other mood stabilizers
(n = 37)		Statistical test usedTest
(t, W, or χ2)		P-value
Sex (n; %)Chi-squareχ2 = 0.107510.743
Man14; 35.0015; 40.54
Woman26; 65.0022; 59.46
Age (M ± SD)57.45 ± 12.5350.30 ± 10.06Welch’s t-testt = 2.77120.007**
Age (n; %)Chi-squareχ2 = 10.6150.005**
Early adulthood9; 22.5010; 27.03
Midlife11; 27.5021; 56.76
Late life20; 50.006; 16.22
Age (n; %)Chi-squareχ2 = 8.35660.004**
Adult20; 50.0031; 83.78
Late life20; 50.006; 16.22
Age at onset of the disease (M ± SD)25.67 ± 8.6328.62 ± 9.19Mann–Whitney UW = 601.50.159
Illness in years (M ± SD)31.77 ± 12.7021.67 ± 11.53Welch’s t-testt = 3.65780.0005**
Sub-diagnosis (n; %)Chi-square0.867
BDI24; 60.0025; 67.57
BDII13; 32.5011; 29.73
Treatment responders (n; %)
Yes20; 50.003; 8.11
No20; 50.0028; 75.68
Use of additional mood stabilizers (n; %)Chi-squareχ2 = 1.41760.234
Yes13; 32.506; 16.22
No27; 67.5028; 75.68
Number of depressive episodes (M ± SD)7.76 ± 7.1110.59 ± 10.00Mann–Whitney UW = 5090.234
Number of manic episodes (M ± SD)7.34 ± 7.163.68 ± 4.85Mann–Whitney UW = 8910.006**
Anthropometric data
BMI (M ± SD)25.93 ± 4.3427.68 ± 6.06Mann–Whitney UW = 5690.316
BMI (n; %)Chi-squareχ2 = 1.38150.501
Normal weight20; 50.0012; 32.43
Overweight12; 30.0013; 35.13
Obesity8; 20.008; 21.62
Lifestyle factors
Physical activity (n; %)Chi-squareχ2 = 3.92310.048*
No20; 50.0026; 70.27
Yes19; 47.508; 21.62
Smoking status Yes (n; %)Chi-squareχ2 = 1.32710.249
No16; 40.0020; 54.05
Yes23; 57.5015; 40.54
MedDietAdherence (n; %)Fisher’s exact-0.770
High5; 12.505; 13.51
Moderate30; 75.0025; 67.57
Low5; 12.507; 18.92
MedDietScore (M ± SD)7.65 ± 1.827.32 ± 2.34Mann–Whitney UW = 8100.473
Comorbidities
Diabetes (n; %)Fisher’s exact-1.000
No38; 95.0036; 97.30
Yes2; 5.001; 2.70
Hypertension (n; %)Fisher’s exact-0.015*
No31; 77.5036; 97.30
Yes9; 22.501; 2.70

Comparison of demographic, clinical, anthropometric, lifestyle, and comorbidity characteristics among patients treated with Li and those treated with other mood stabilizers.

Data are expressed as mean (M) ± Standard Deviation (SD) for continuous variables and as number (percentage) for categorical variables. Between-group comparisons were performed using the Welch’s t-test for normally distributed continuous variables, the Mann–Whitney U test for non-normally distributed variables, and the Chi-square or Fisher’s exact test for categorical variables. Significant results (p < 0.05) are highlighted in bold. *: statistically significant with adjusted p-value < 0.05; **: statistically significant with adjusted p-value < 0.01. M, Mean; SD, Standard Deviation; Li, Lithium; BMI, Body Mass Index; BD, Bipolar Disorder; BDI, Bipolar Disorder I; BDII, Bipolar Disorder II; MedDiet, Mediterranean Diet. t, Welch’s t-test statistic; W, Mann–Whitney (Wilcoxon rank-sum) test statistic; χ2 = chi-square test.

Significant differences were found with respect to age, both when considered as a continuous variable and when stratifying the study population into age groups. Most participants were adults, with a higher representation of late life among patients treated with Li (50%) and midlife among those treated with other mood stabilizers (56.76%). Using the dichotomized age classification, half of the Li-treated participants were adults (25–60 years) and half belonged to the late-life group ( >60 years), whereas 83.8% of participants treated with other mood stabilizers were adults and 16.2% were classified as late-life group. Additional significant differences between the two cohorts regarding illness duration (in years), number of manic episodes, level of physical activity and presence of hypertension were observed.

When stratifying the study population of patients treated with Li into groups according to their response to therapy (see Table 2), among Li-responders six patients (30%) were men and 13 (65%) were women, while among non-responders, seven (35%) were men and 13 (65%) were women. Significant age differences were observed between groups, mainly driven by the comparison between Li-responders and patients treated with other mood stabilizers, with a mean age of 60.25 years in Li-responders, compared to 54.65 years in Li non-responders and 50.30 years in the group treated with other mood stabilizers.

TABLE 2

VariableLi R
(n = 20)		Li NR
(n = 20)		Other mood stabilizers
(n = 37)		Statistical test; correction methodF or χ2 statisticEffect size methodGlobal p-valuePost-hoc pairwise comparisonsEffect size
Sex (n; %)Pearson’s Chi-squared; BH0.47286Cramer’s V with MC0.789-
Man7; 35.007; 35.0015; 40.54
Woman13; 65.0013; 65.0022; 59.46
Age (M ± SD)60.25 ± 11.8554.65 ± 12.8750.30 ± 10.06ANOVA; Tukey HSD5.075Cohen’s d0.009Li R vs. Li NR: 0.267;
Li R vs. Other: 0.006**;
Li NR vs. Other: 0.353
Li R vs. Li NR: 0.453;
Li R vs. Other: 0.929;
Li NR vs. Other: 0.392
Age (n; %)Fisher’s exact; BHCramer’s V with Monte Carlo0.013Li R vs. Li NR: 0.356;
Li R vs. Other: 0.016*;
Li NR vs. Other: 0.124		Li R vs. Li NR: 0.217;
Li R vs. Other: 0.451;
Li NR vs. Other: 0.296
Early adulthood3; 15.006; 30.0010; 27.03
Midlife5; 25.006; 30.0021; 56.76
Late life12; 60.008; 40.006; 16.22
Age (n; %)Pearson’s Chi-squared; BH11.598Cramer’s V with Monte Carlo0.003Li R vs. Li NR: 0.206;
Li R vs. Other: 0.002**;
Li NR vs. Other: 0.070		Li R vs. Li NR: 0.200;
Li R vs. Other: 0.450;
Li NR vs. Other: 0.264
Adult8; 40.0012; 60.0031; 83.78
Late life12; 60.008; 40.006; 16.22
Age at onset of the disease (M ± SD)24.25 ± 8.2227.10 ± 9.0028.62 ± 9.19Kruskal-Wallis; Wilcoxon BH3.3705Rank Biserial0.185-
Illness in years (M ± SD)36.00 ± 12.8627.55 ± 11.3121.67 ± 11.53ANOVA; Tukey HSD9.553Cohen’s d0.0002Li R vs. Li NR: 0.068;
Li R vs. Other: 0.0001**;
Li NR vs. Other: 0.018		Li R vs. Li NR: 0.698;
Li R vs. Other: 1.193;
Li NR vs. Other: 0.513
Years of treatment19.75 ± 10.1918.53 ± 10.4611.41 ± 6.59Kruskal-Wallis; Wilcoxon BH9.5107Rank Biserial0.009Li R vs. Li NR: 0.746;
Li R vs. Other: 0.010*;
Li NR vs. Other: 0.037*		Li R vs. Li NR: 0.060
95% CI: [-0.29, 0.41];
Li R vs. Other: 0.530
95% CI: [0.23, 0.74];
Li NR vs. Other: 0.410
95% CI: [0.08, 0.66]
Sub-diagnosis (n; %)Pearson’s chi-squared; BH0.64024Cramer’s V with Monte Carlo0.726-
BDI12; 60.0012; 60.0025; 67.57
BDII8; 40.005; 25.0011; 29.73
Use of additional mood stabilizers (n; %)Fisher’s exact; BH12.736Cramer’s V with Monte Carlo0.004Li R vs. Li NR: 0.010*;
Li R vs. Other: 0.695;
Li NR vs. Other: 0.010*		Li R vs. Li NR: 0.480;
Li R vs. Other: 0.104;
Li NR vs. Other: 0.388
Yes2; 10.0011; 55.006; 16.22
No18; 90.009; 45.0028; 75.68
Antipsychotic treatment at recruitmentPearson’s Chi-squared; BH22.731Cramer’s V with Monte Carlo8,00E-06Li R vs. Li NR: 4.98E-05**; Li R vs. Other: 0.001**; Li NR vs. Other: 0.174Li R vs. Li NR: 0.704; Li R vs. Other: 0.489; Li NR vs. Other: 0.225
Yes4; 20.0018; 90.0024; 67.86
No16; 80.002; 10.0010; 27.03
Antidepressants treatment at recruitmentFisher’s Exact; BHCramer’s V with Monte Carlo0.683––
Yes3; 15.005; 25.009; 24.32
No17; 85.0015; 75.0025; 67.57
Number of depressive episodes (M ± SD)9.38 ± 8.136.24 ± 5.8310.59 ± 10.00Kruskal-Wallis; Wilcoxon BH2.7234Rank Biserial0.256––
Number of manic episodes (M ± SD)8.71 ± 9.326.06 ± 4.153.68 ± 4.85Kruskal-Wallis; Wilcoxon BH7.9569Rank Biserial0.019Li R vs. Li NR: 0.921;
Li R vs. Other: 0.164;
Li NR vs. Other: 0.014*
Li R vs. Li NR: -0.020, -0.02
95% CI: [-0.39, 0.35];
Li R vs. Other: 0.270
95% CI: [-0.05, 0.55];
Li NR vs. Other: 0.470
95% CI: [0.18, 0.69]
Total number of episodes (M ± SD)39.15 ± 16.8637.85 ± 14.5115.81 ± 12.76Kruskal-Wallis; Wilcoxon BH35.416Rank Biserial2,00E-08Li R vs. Li NR: 0.989;
Li R vs. Other: 2.00E-6**; Li NR vs. Other: 2.00E-6**		Li R vs. Li NR: -0.005, -0.005
95% CI: [-0.35, 0.34]; Li R vs. Other: 0.790
95% CI: [0.65, 0.89]; Li NR vs. Other: 0.780
95% CI: [0.63, 0.88]
Anthropometric data
BMI (M ± SD)26.07 ± 4.0625.79 ± 4.7127.68 ± 6.06Kruskal-Wallis; Wilcoxon BH1.0241Rank Biserial0.599––
BMI (n; %)Fisher’s exact; BH–Cramer’s V with Monte Carlo0.922––
Normal weight10; 50.009; 45.0012; 32.43
Overweight6; 30.005; 25.0013; 35.13
Obesity4; 20.004; 20.008; 21.62
Lifestyle factors
Physical activity (n; %)Pearson’s Chi-squared; BH14.854Cramer’s V with Monte Carlo0.001Li R vs. Li NR: 0.003**;
Li R vs. Other: 0.001**;
Li NR vs. Other: 0.903		Li R vs. Li NR: 0.487;
Li R vs. Li Other: 0.488;
Li NR vs. Other: 0.017
No5; 25.0015; 75.0026; 70.27
Yes14; 70.005; 25.008; 21.62
Smoking status Yes (n; %)Pearson’s Chi-squared; BH5.1271Cramer’s V with Monte Carlo0.077––
No5; 25.0011; 55.0020; 54.05
Yes14; 70.009; 45.0015; 40.54
MedDietAdherence (n; %)Fisher’s exact; BH–Cramer’s V with Monte Carlo0.928––
High3; 15.002; 10.005; 13.51
Moderate15; 75.0015; 75.0025; 67.57
Low2; 10.003; 15.007; 18.92
MedDietScore (M ± SD)8.35 ± 1.606.95 ± 1.797.32 ± 2.34Kruskal-Wallis; Wilcoxon BH6.4217Rank Biserial0.040Li R vs. Li NR: 0.025*; Li R vs. Other: 0.102;
Li NR vs. Other: 0.525		Li R vs. Li NR: 0.480
95% CI: [0.16, 0.71]; Li R vs. Other: 0.290
95% CI: [-0.02, 0.55];
Li NR vs. Other: -0.100
95% CI: [-0.40, 0.21]
Comorbidities
Diabetes (n; %)Fisher’s exact–Cramer’s V with Monte Carlo1.000––
No19; 95.0019; 95.0036; 97.30
Yes1; 5.001; 5.001; 2.70
Hypertension (n; %)Fisher’s exact; BH–Cramer’s V with Monte Carlo0.008Li R vs. Li NR: 0.451; Li R vs. Other: 0.017*; Li NR vs. Other: 0.178Li R vs. Li NR: 0.180; Li R vs. Li Other: 0.397; Li NR vs. Other: 0.230
No14; 70.0017; 85.0036; 97.30
Yes6; 30.003; 15.001; 2.70

Comparison of demographic, clinical, anthropometric, lifestyle, and comorbidity characteristics among Li-responders (Li R), Li non-responders (Li NR), and patients treated with other mood stabilizers.

Clinical, demographic, anthropometric, lifestyle, and comorbidity characteristics of participants by treatment group (Li-Responders, Li Non-Responders and Other mood stabilizers). Continuous variables are presented as mean ± SD and categorical variables as n (%). Global p-values indicate the overall differences among groups. Post-hoc pairwise comparisons are shown only for variables with statistically significant global p-values. Effect sizes are reported for pairwise comparisons. Significant results (p < 0.05) are highlighted in bold, as *p < 0.05, **p < 0.01. Categorical variables were analyzed with Pearson’s Chi-squared or Fisher’s Exact tests (BH correction for multiple comparisons). Continuous variables were tested for normality (Shapiro-Wilk); normally distributed variables were analyzed using ANOVA with Tukey HSD post-hoc tests, non-normal variables with Kruskal-Wallis and Wilcoxon post-hoc tests (BH correction). Effect sizes are reported for significant comparisons: Cohen’s d for ANOVA, rank biserial for Wilcoxon, and Cramer’s V for categorical variables. M, Mean; SD, Standard Deviation.

When stratifying by age category, significant differences between Li-responders and patients treated with other mood stabilizers were also detected. The majority of Li-responders were classified as late life group (60%), while Li non-responders were more evenly distributed across early adulthood (30%), midlife (30%), and late life (40%). In contrast, patients treated with other mood stabilizers were predominantly in midlife (56.8%), with a smaller proportion classified as late life group (16.2%). These findings suggest that older age was more frequent among Li-responders, whereas younger and middle-aged adults were more represented in the group treated with alternative stabilizers.

Similarly, significant between-group differences were observed for illness duration, treatment duration, use of additional mood stabilizers, use of antipsychotics at recruitment, total number of mood episodes, total number of manic episodes, engagement in physical activity, Mediterranean diet score and prevalence of hypertension. A significantly higher proportion of Li non-responders (55%) used additional mood stabilizers, compared with Li-responders (10%) and patients treated with valproic acid as the main mood stabilizer (16.2%). Likewise, 90% of Li non-responders were taking antipsychotic medications, a proportion significantly higher than in Li-responders (20%), whereas Li-responders showed significantly lower use of antipsychotics compared with patients treated with other mood stabilizers (67.86%).

Conversely, a lower proportion of Li-responders (25%) reported engaging in physical activity compared with both non-responders (75%) and patients treated with other stabilizers (70.27%). However, adherence to the Mediterranean diet was significantly higher among Li-responders than among non-responders. Responders and non-responders also exhibited a significantly longer duration of treatment (in years) compared with patients treated with other mood stabilizers, as well as a higher number of total mood episodes. Furthermore, a greater proportion of Li non-responders were concurrently taking additional mood stabilizers compared with responders and with patients treated with valproic acid as the main mood stabilizer.

3.2 Gut microbiota analysis

Samples were normalized to 18,000 reads (the lowest number of reads per sample). After processing the reads through DADA2, approximately 75% passed the QC filter and ∼60% on average remained in the last step (passed QC, non-chimeric reads) (see Supplementary Figures 3, 4).

3.2.1 Alpha and beta diversity analysis

Alpha diversity analysis did not reveal significant differences in species richness or overall diversity between patients treated with Li and those receiving other mood stabilizers. No statistically significant differences were observed for Observed ASVs (p = 0.7213), Shannon (p = 0.2162), Faith’s Phylogenetic Diversity (p = 0.538), Chao1 (p = 0.6540) and Simpson indices (p = 0.178). However, Pielou’s evenness was significantly lower in the Li-treated group compared to patients receiving other mood stabilizers (p = 0.0462), suggesting a less uniform distribution of microbial taxa in these individuals (see Figure 1A). Means, SD, F statistic, R-squared, and p-values of each analysis can be found in the Supplementary Table 1.

FIGURE 1

To visualize the beta diversity among the same groups, we obtained the PCoA using the following distances: Aitchison, Jaccard, Bray-Curtis dissimilarity, unweighted UniFrac, and weighted UniFrac metric (Figure 1C).

The first two axes of the PCoA explained approximately 20% of the variance for the unweighted UniFrac distance, while the variance explained for Bray-Curtis and weighted UniFrac distances was approximately 20 and 42%, respectively.

To assess whether the microbial community composition significantly differed between treatment groups, a PERMANOVA test was performed (see Supplementary Table 3). No significant differences were detected for Aitchison (p = 0.157), Jaccard (p = 0.113), Bray-Curtis (p = 0.062), unweighted UniFrac (p = 0.113) or weighted UniFrac (p = 0.074) distances.

We investigated the alpha and beta diversity of the GM across three clinical groups: patients who responded to Li treatment (Li-responders), patients who did not respond to Li (Li non-responders), and patients treated with other mood stabilizers (Other mood stabilizers). No significant differences were observed among groups for Observed ASVs (p = 0.6400), Pielou’s evenness (p = 0.2809), Shannon (p = 0.8994), Faith’s Phylogenetic Diversity (p = 0.5072), Chao1 (p = 0.5982) and Simpson indices (p = 0.6659) (see Figure 1B and Supplementary Table 2).

PCoA plots based on multiple distance metrics (Aitchison, Jaccard, Bray–Curtis, Unweighted UniFrac, and Weighted UniFrac) revealed an overall overlap in the microbial composition between groups. No significant differences were observed for Aitchison (p = 0.601), Jaccard (p = 0.194), Bray-Curtis (p = 0.561), unweighted UniFrac (p = 0.416) or weighted UniFrac (p = 0.311) distances (see Figure 1D and Supplementary Table 4).

3.2.2 Compositional analysis of intestinal microbiota

Taxonomic profiling of GM is shown in Figure 2. At the phylum level, both Li-treated patients and those on other mood stabilizers exhibited a predominance of Firmicutes (Bacillota) and Bacteroidota, with substantial interindividual variability. Interestingly, Actinobacteria (Actinomycetota) appeared to be more abundant in patients treated with other mood stabilizers compared to those treated with Li.

FIGURE 2

Further stratification of Li-treated patients into responders and non-responders revealed additional differences in microbiota composition. At the class and order levels, taxa such as Clostridia and Bacteroidia were dominant across groups.

At the family and genus levels, clearer divergences were observed. Families like Ruminococcaceae and Lachnospiraceae were abundant in all groups, though their proportions differed across subgroups.

Differential abundance analysis at different taxonomic levels revealed several taxa with significant changes in relative abundance between patients treated with Li and those on other mood stabilizers (Figures 3A–D and Supplementary Table 5). Using the Benjamini-Hochberg method to correct for multiple testing, significant differences were observed with a p-value threshold of < 0.05.

FIGURE 3

In patients treated with Li, we observed a significant reduction in the relative abundance of the Actinobacteria (Actinomycetota) phylum and several taxa within it, including Coriobacteriia, Coriobacteriales, and the Senegalimassilia genus. Within the Firmicutes (Bacillota) phylum, we observed an increase in the Selenomonadales order and the Megamonas. genus, as well as the Tyzzerella, Ruminiclostridium 9, Flavonifractor, Unclassified Lachnospiraceae 958, all belonging to the Clostridia class. The analysis was controlled for potential confounders, including age, hypertension, physical activity, illness in years and number of manic episodes.

The same analysis based on the response to Li treatment revealed significant alterations in the GM especially in Li-responders when compared with patients treated with other mood stabilizers. In particular, we observed an increase in Li-responders in the Euryarchaeota (Methanobacteriota) phylum and in the correlated taxa Methanobacteria, Methanobacteriales, Methanobacteriaceae, and Methanobrevibacter, and in the Clostridiales vadinBB60 group_uncultured genus belonging to the Firmicutes (Bacillota) phylum. No statistically significant differences were observed in either Li responders or Li non-responders compared with patients treated with other mood stabilizers (Figures 4A–E and Supplementary Table 5).

FIGURE 4

3.2.3 Functional metagenome prediction analysis

The functional analysis revealed 42 MetaCyc pathways statistically decreased in patients treated with Li compared to patients treated with other mood stabilizers (q < 0.05). For Li response, no statistically significant differences were found between responders and non-responders, as none of the associations exceeded the conventional significance threshold (q < 0.05) and all adjusted p-values were > 0.9. However, comparisons between Li-responders and others, as well as between Li non-responders and others, were suggestive at q < 0.10 (see Supplementary Figure 5 and Supplementary Table 6).

The predictive functional analysis of the GM revealed a distinctive metabolic profile associated with Li treatment. In line with the observed taxonomic alterations, Li-treated patients showed a significant reduction in numerous pathways involved in bacterial structural biosynthesis, carbohydrate metabolism, primary fermentation, and the biosynthesis of nucleotides and essential cofactors compared with those treated with other mood stabilizers. Among the most markedly reduced pathways were peptidoglycan maturation (meso-diaminopimelate containing), peptidoglycan biosynthesis IV (Enterococcus faecium), and peptidoglycan biosynthesis V (β-lactam resistance). In parallel, a reduction was observed in multiple carbohydrate degradation pathways, including galactose degradation I (Leloir pathway), sucrose degradation IV (sucrose phosphorylase), sucrose degradation III (sucrose invertase), lactose and galactose degradation I, starch degradation V, and the superpathway of glucose and xylose degradation, as well as in fermentative pathways such as mixed acid fermentation, heterolactic fermentation, acetylene degradation, and the Bifidobacterium shunt (see Figure 5 and Supplementary Table 7).

FIGURE 5

4 Discussion

In the present study, we analyzed the GM of patients with bipolar disorder treated with Li, comparing it with that of patients receiving other mood stabilizers, with the aim of identifying microbial signatures associated with treatment exposure and clinical response. Alpha- and beta-diversity analyses did not reveal significant differences between groups, except for a reduction in Pielou’s evenness in Li-treated patients compared with those treated with other mood stabilizers. This suggests not a global depletion of the microbial ecosystem, rather a selective reorganization in which specific taxa become favored over others.

One of the most consistent findings was the significant reduction in the Actinobacteria (Actinomycetota) phylum, particularly the Coriobacteriia class, the Coriobacteriales order and the Senegalimassilia genus in Li-treated patients. This result is particularly interesting in light of the work by Painold et al., who reported an increase in taxa belonging to Actinobacteria and Coriobacteriia in patients with BD compared with controls (). Actinobacteria and taxa within the Coriobacteriia class, including Eggerthellaceae and Coriobacteriaceae families, are involved in lipid metabolism, in the processing of dietary aromatic compounds and correlate with serum cholesterol levels (; ; ).

Members of the Actinobacteria phylum, particularly Bifidobacterium spp., are thought to contribute to primary fermentation processes by participating in the degradation of resistant starch and plant-derived polysaccharides via glycosyl hydrolases (). It has also been shown that members of the Actinobacteria phylum, acting as primary fermenters, contribute to intestinal lactate production, which can subsequently be utilized by specific bacterial groups, known as lactate utilizers, for butyrate synthesis (). Based on our results, the reduction in Actinobacteria (Actinomycetota) phylum in Li-treated patients merely suggest a decrease in microbial lactate production and could potentially indicate a protective effect of Li, consistent with pathophysiological models linking excess lactate, both intestinal and cerebral, to BD (). However, this hypothesis should be investigated in future multi-omics studies to clarify the functional implications of these microbial changes.

Scientific evidence shows that genera within the Coriobacteriia class, such as Collinsella and Eggerthella, are associated with increased intestinal permeability, IL-17–mediated inflammation, unfavorable metabolic phenotypes and, in some cases, greater severity of chronic inflammatory disorders (; ; ; ; ; ). Conversely, Senegalimassilia remains poorly characterized, with studies reporting contradictory associations with metabolic and inflammatory markers (; ; ; ; ). Therefore, changes in Senegalimassilia abundance may reflect broader shifts in host metabolic organization and microbial ecosystem interactions, rather than having a direct causal effect. Overall, the reduction in Coriobacteriia taxa in Li-treated patients suggests that Li may indirectly modulate the gut environment in a direction less favorable to taxa with pro-inflammatory potential, plausibly through lithium’s known anti-inflammatory and immunomodulatory properties ().

These findings align with evidence from who documented an enrichment of lactate-producing bacteria (Lactobacillus, Enterococcus, Streptococcus, Bifidobacterium) in major depressive disorder (MDD), BD and schizophrenia. Although these microorganisms can contribute to beneficial cross-feeding pathways that generates acetate/propionate/butyrate from lactate, potentially adverse effects may emerge when lactate accumulates, a phenomenon linked to multiple disorders including neurotoxicity and mitochondrial dysfunction. Moreover, increased intestinal permeability may facilitate the systemic lactate translocation, contributing to neuropsychiatric manifestations (). Elevated brain lactate has been reported in several mood disorders, including bipolar disorder (; ). Within this framework, the reduction of primary fermenters in Li-treated patients merely suggest a rebalancing of fermentative metabolism.

Lithium treatment was also associated with an increase in several Firmicutes (Bacillota) taxa, including the Selenomonadales order and its Megamonas genus, as well as multiple taxa belonging to the Clostridia class, such as Flavonifractor, Ruminiclostridium 9 (Oscillospiraceae family), Tyzzerella and Unclassified Lachnospiraceae 958 (Lachnospiraceae family). The Lachnospiraceae and Oscillospiraceae families are among the most relevant SCFA producers in the human gut, particularly butyrate and acetate (; ; ), with key roles in epithelial barrier integrity, immune modulation and control of neuroinflammatory processes mediated through the gut-brain axis (; ; ). McGuinness et al. also described consistent reductions in major butyrate producers in psychiatric disorders (). Thus, the enrichment of secondary fermenters in Li-treated patients suggests a shift from primary fermentation toward secondary fermentation sustained by cross-feeding, in line with lithium’s capacity to enhance fermentative efficiency and to modulate regulatory T cells through GPR43-dependent mechanisms ().

Megamonas contributes to the production of acetate, propionate and lactate, providing essential substrates for cross-feeding (). Its abundance has been associated with metabolic conditions, although it may decrease in hepatic steatosis and increase in response to prebiotic interventions (; ). Tyzzerella possesses the ability to produce aromatic amines from amino acids and express β-N-acetyl-hexosaminidase, involved in the synthesis of human milk oligosaccharide precursors. Although associated in some studies with unfavorable metabolic markers (; ; ), it has also been reported to increase in contexts characterized by improved psychological wellbeing or lower inflammatory stress (), suggesting a context-dependent physiological role. Overall, the increase in Megamonas and Tyzzerella may reflect a reorganization of the intestinal metabolic network in Li-treated patients.

The Flavonifractor genus, also increased in the Li-treated group, has been reported as enriched in bipolar disorder independently of treatment (; ), suggesting that its higher abundance may reflect a disorder-related trait shared by patients with BD rather than a Li-induced effect. Future studies including a healthy control group will be essential to better define the role of this taxon in the pathology of the disorder.

The analysis of Li-response subgroups revealed that the most significant differences emerged when comparing Li-responders with non-Li treated patients, whereas no statistically significant differences in the direct comparison between responders and non-responders to Li or between Li non-responders and patients treated with other mood stabilizers were found.

Responders showed an increase in methanogenic taxa, including the Euryarchaeota (Methanobacteriota) phylum and the correlated Methanobacteria, Methanobacteriales, Methanobacteriaceae and Methanobrevibacter taxa, as well as the Clostridiales vadinBB60 group_uncultured genus, belonging to the Firmicutes (Bacillota) phylum.

Methanogens consume hydrogen, supporting continuous fermentation cycles and increasing SCFA yield (; ; ), while enhancing ecosystem stability by reinforcing cooperative metabolic interactions within the microbiota. Their clinical relevance remains debated, having been linked to both pathological and healthy phenotypes (; ; ); however, their presence has also been associated with intestinal ecosystem stability and functional resilience in long-lived populations (). Methanogenesis also favors acetate production (), a key SCFA involved in insulin sensitivity, energy homeostasis, epithelial barrier integrity and neuroinflammation. In the present study, methanogens did not significantly differentiate responders from non-responders; therefore, no direct association with treatment response can be inferred and their specific contribution remains to be established.

The enrichment of the Clostridiales vadinBB60 group, belonging to the class Clostridia within the Firmicutes (Bacillota) phylum, in responders compared with patients treated with other mood stabilizers is also notable. This potential SCFA-producing group () has been proposed as protective in postpartum depression () and shown to be reduced in patients with depression compared with controls (). It has also emerged as a potential biomarker of therapeutic response in microbiome-based interventions in murine models of Alzheimer’s disease (). Although its functions remain incompletely defined, associations with dopaminergic/serotonergic metabolites () and resilience to colitis (; ) have been described. In this study, the Clostridiales vadinBB60 group did not show significant differences between responders and non-responders; therefore, its potential role remains to be clarified and warrants further investigation.

Functional prediction analysis confirmed that Li was associated with a reduction in primary biosynthetic and fermentative pathways, including peptidoglycan pathways, nucleotide and cofactor biosynthesis, carbohydrate degradation pathways and primary fermentative pathways such as mixed acid fermentation, heterolactic fermentation and Bifidobacterium shunt (; ; ). This pattern is coherent with the observed reduction in primary fermenters such as Actinobacteria (Actinomycetota) and Coriobacteriia () and supports the hypothesis of a shift toward secondary fermentation sustained by SCFA-producing taxa, such as Lachnospiraceae and Oscillospiraceae ().

In line with alterations in gut microbial metabolic potential observed in Li-treated patients, peripheral metabolomic profiling reported by in the same cohort showed treatment-associated changes in systemic energy metabolism. In particular, authors observed higher systemic levels of glucose, pyruvate, lactate, glutamine and α-ketoglutarate, along with decreased glutamate, metabolites previously implicated in bipolar disorder and modulated by Li exposure.

While our microbiome analysis indicated a reduction in primary carbohydrate fermentative capacity and a shift toward taxa associated with cross-feeding and secondary fermentation, these microbial metabolic tendencies reflect changes in intestinal substrate processing rather than direct contributors to systemic metabolite pools. At the same time, the plasma metabolomic profile described by Piras et al. suggests a rearrangement of host energy metabolism characterized by elevated circulating glucose, pyruvate and lactate, consistent with increased glycolytic flux and a relative down-regulation of mitochondrial oxidative pathways in Li-treated patients.

Taken together, the microbial and metabolomic findings may represent complementary but independent signatures of lithium exposure at different biological levels, warranting future multi-omics studies with paired data to determine whether they reflect coordinated host-microbiota metabolic interactions.

5 Conclusion and future perspectives

Overall, our findings indicate that Li treatment is associated with a selective reorganization of the intestinal microbiota in patients with bipolar disorder, characterized by a reduction in primary fermenters and an enrichment of taxa involved in secondary fermentation and SCFA production. Functional predictions further support this ecological shift, showing a decrease in biosynthetic and primary fermentative pathways.

The identification of taxa enriched in lithium responders, including Methanobrevibacter and the Clostridiales vadinBB60 group, was derived from comparisons with patients treated with other mood stabilizers rather than from direct analyses between responders and non-responders. For this reason, no direct association with treatment response can be inferred, and their specific contribution remains to be established.

Given the observational and cross-sectional nature of the study, causal inferences cannot be established. Future research should adopt longitudinal designs including pre- and post-treatment assessments and integrate multi-omic approaches, such as fecal metabolomics, microbial transcriptomics, and targeted quantification of SCFAs and lactate, to clarify the mechanistic links between specific microbial configurations and therapeutic response. Furthermore, interventional studies are warranted to determine whether targeted modulation of the GM, through dietary strategies, prebiotics, probiotics or microbiota-directed therapeutics, may enhance Li tolerability or efficacy.

Exploring this host-microbiota interface may open new avenues for precision medicine approaches in the treatment of bipolar disorder, integrating psychopharmacology with microbiome-informed strategies.

6 Limitations of the study

This study presents several limitations that should be taken into account when interpreting the results. A major limitation is the absence of an untreated clinical control group, which would have enabled a clearer distinction between microbiota features associated with bipolar disorder and those attributable to pharmacological treatments. Without a reference control, it remains uncertain whether some of the observed taxonomic patterns represent disorder-specific traits or treatment-related modifications.

Furthermore, the cross-sectional design prevents establishing causal relationships between microbial composition, predicted metabolic functions and clinical response.

An additional limitation concerns the clinical heterogeneity between treatment groups. Although the covariates were included in the adjusted ANCOM-BC2 models and several microbiota-related confounders were controlled through exclusion criteria, residual confounding cannot be fully excluded.

Moreover, although the cohort was clinically well characterized, the sample size was relatively small, particularly for stratified comparisons between responders and non-responders, reducing statistical power and potentially masking subtler associations.

Finally, functional prediction obtained with PICRUSt2 rely on inferred gene content and do not necessarily reflect real microbial activity. Although useful for generating hypotheses, these predictions require validation with direct functional analyses such as metatranscriptomics, metabolomics, or targeted SCFA quantification.

Statements

Data availability statement

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://www.ebi.ac.uk/ena, PRJEB108658.

Ethics statement

The studies involving humans were approved by the Ethics Committees of the University Health Agency of Cagliari (Prot. NP/2020/4412 on 28 October 2020). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

Author contributions

VP: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review & editing. LG: Writing – review & editing, Formal analysis, Methodology. SS: Writing – review & editing, Formal analysis, Methodology. GS: Methodology, Investigation, Writing – review & editing. ClP: Methodology, Conceptualization, Writing – review & editing. CrP: Methodology, Conceptualization, Writing – review & editing. MS: Data curation, Writing – review & editing. AM: Data curation, Writing – review & editing. DC: Data curation, Writing – review & editing. GS: Resources, Writing – review & editing. RA: Resources, Writing – review & editing. CC: Resources, Writing – review & editing. MC: Resources, Writing – review & editing. PP: Resources, Writing – review & editing. MP: Resources, Writing – review & editing. FS: Resources, Writing – review & editing. MZ: Conceptualization, Resources, Writing – review & editing. BC: Conceptualization, Resources, Writing – review & editing. MM: Writing – review & editing, Conceptualization, Resources. JS: Writing – review & editing, Methodology. LA: Conceptualization, Methodology, Resources, Writing – review & editing. AS: Writing – review & editing, Conceptualization, Methodology, Resources, Supervision. AM: Writing – review & editing, Conceptualization, Funding acquisition, Methodology, Resources, Supervision.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This research was funded by the EU funding within the NextGenerationEU-MUR PNRR Extended Partnership Initiative on Emerging Infectious Disease (Project no. PE00000007, INF-ACT).

Acknowledgments

We acknowledge the CeSAR (Centro Servizi Ricerca d’Ateneo) core facility of the University of Cagliari and Marta Costa for assistance with the generation of 16S sequencing data. The authors also thank Bianca M. Manzin for her support in the English editing of the manuscript. We sincerely thank all the volunteers who took part in this study.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2026.1839847/full#supplementary-material

References

Summary

Keywords

16S rRNA, bipolar disorder, gut microbiota, lithium, microbiota gut–brain axis, next generation sequencing, patient-centered approach

Citation

Palmas V, Gama LCM, Sato Guima SE, Sanna G, Pisanu C, Piras C, Spada M, Meloni A, Congiu D, Severino G, Ardau R, Chillotti C, Contu M, Paribello P, Pinna M, Suprani F, Zompo MD, Carpiniello B, Manchia M, Setubal JC, Atzori L, Squassina A and Manzin A (2026) Gut microbiota signatures associated with lithium treatment and clinical response in patients with bipolar disorder. Front. Microbiol. 17:1839847. doi: 10.3389/fmicb.2026.1839847

Received

26 March 2026

Revised

23 May 2026

Accepted

03 June 2026

Published

29 June 2026

Volume

17 - 2026

Edited by

Gratiela Gradisteanu Pircalabioru, University of Bucharest, Romania

Reviewed by

Tiago Nardi, National Cancer Institute Foundation (IRCCS), Italy

Lawrence Shih Hsin Wu, China Medical University, Taiwan

Jahwarhar Izuan Abdul Rashid, National Defence University of Malaysia, Malaysia

Updates

Copyright

© 2026 Palmas, Gama, Sato Guima, Sanna, Pisanu, Piras, Spada, Meloni, Congiu, Severino, Ardau, Chillotti, Contu, Paribello, Pinna, Suprani, Zompo, Carpiniello, Manchia, Setubal, Atzori, Squassina and Manzin.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Vanessa Palmas,

† These authors have contributed equally to this work

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Share on WeChat

Scan with WeChat to share this article

Article metrics