Ranjan Sinha, Inti Pedroso, Shreyas V. Kumbhare, Bharat Joshi, Santosh K. Saravanan, Dattatray S. Mongod, Simitha Singh-Rambiritch, Tejaswini Uday, Karthik M. Muthukumar, Carmel Irudayanathan, Chandana Reddy-Sinha, Parambir S. Dulai, Daniel E. Almonacid
Now published in Journal of Personalized Medicine doi: 10.3390/jpm12081237
Mental health diseases are a major component of morbidity and health care costs. Obesity and gut disorders are comorbid with mental health, with the gut microbiome hypothesized to play a key mechanistic role in linking them. Pharmacological and behavioral interventions are currently used to treat mental health disorders, but they have limited efficacy. Dietary and weight-loss interventions have been shown to provide different benefits. Still, there exists conflicting evidence regarding their effects which may be due to an individual’s genetic or microbiome factors modulating the improvement of symptoms.
We aimed to identify genetic and gut microbiome factors that explain the improvement in mental health after a dietary and lifestyle intervention for weight loss.
We recruited 369 individuals participating in the Digbi Health personalized digital care program, for which we evaluated the association between 23 genetic scores, the abundance of 178 gut microbiome genera, and 42 gut-brain modules (pathways related to neuroactive metabolites produced by gut microbes) with the presence/absence of anxiety/depression or sleep problems at baseline and improvement on anxiety, depression, and insomnia after losing at least 2% body weight.
The mean BMI and age of the study cohort were 34.6 and 48.7, respectively, and there was an overrepresentation of individuals with functional gastrointestinal disorders (84%). On average, the individuals lost 5.4% of body weight at the time of follow-up (mean of 88 days), and more than 95% reported improvement in at least one outcome. We found significant correlations between genetic scores with anxiety and depression at baseline, gut microbial functions with sleep problems at baseline, and genetic scores and gut microbial taxa and functions with anxiety, depression, and insomnia improvement. Among the gut microbial functions identified, the abundance of butyrate synthesis genes was associated with less than average improvement in depression symptoms, the abundance of kynurenine synthesis genes was associated with less than average improvement in anxiety symptoms, and the abundance of genes able to synthesize and degrade neuroactive hormones like nitric oxide was associated with greater than average improvement in depression and insomnia symptoms. Among the genetic scores identified, anxiety or depression at baseline were associated with genetic scores for alcohol use disorder and major depressive disorder, and greater than average improvement in anxiety and depression symptoms was associated with an obstructive sleep apnea genetic score. Furthermore, a type 1 diabetes genetic score was associated with a greater than average improvement of insomnia symptoms, whereas a type 2 diabetes genetic score was associated with a less than average improvement of insomnia symptoms. We compared the relative ability of demographic, genetic, and microbiome factors to explain baseline and improvement in mental health and found that genetic and microbiome factors provide value above demographic variables alone. Medication and recreational drug use do not confound microbiome associations with mental health.
The digital therapeutics care program significantly decreased body weight and concomitantly decreased self-reported mental health symptom intensity. Our results provide evidence that genetic and gut microbiome factors help explain interindividual differences in mental health improvement after dietary and lifestyle interventions for weight loss. Thus, individual genetic and gut microbiome factors provide a basis for designing and further personalizing dietary interventions to improve mental health.
Poor mental health is a significant determinant of health-related quality of life with important implications at individual and population levels. Pharmacological and behavioral interventions prevent and treat individuals at risk of or suffering from mental health disorders, but their efficacy is limited, and many experiencing improvements will relapse . The COVID19 pandemic resurfaced and worsened the mental health crisis and brought awareness to society of its relationship with obesity and other chronic health conditions . There is a great need to develop cost-effective interventions that provide significant short and long-term therapeutic benefits to individuals suffering from mental health, especially those with multiple comorbid conditions. Digital therapeutics have gained increased attention as a strategy to provide care to large numbers of individuals, and emerging evidence suggests its effectiveness for several chronic diseases [3, 4].
The epidemiological literature provides compelling evidence indicating that genetic and non-genetic factors contribute to the etiology of mental health disorders [5–7]. Additionally, substantial evidence exists linking mental health with digestive and gut disorders. For instance, two meta-analyses found a higher rate of anxiety and depression among IBS patients [8, 9], and a recent genetic study identified genetic factors linking IBS with mental health disorders . The increasing prevalence of mental health has been linked to the increased rate of obesity and lifestyle risk factors. A meta-analysis found a bidirectional relationship between the two, with depressed individuals having a 37% increased risk of being obese. Those who were obese had an 18% increased risk of being depressed . Furthermore, human genetic studies have shown a causal link between higher BMI and depression [12, 13] and higher BMI and anxiety, among other psychological and psychiatric disorders .
Recently, extensive epidemiological studies coupled with gut microbiome sequencing are reinforcing the importance of the gut-brain axis and identifying the gut microbiome factors underlying it. A gut microbiome and behavioral study found associations between the gut microbiome’s taxa and functions with depression status and quality of life . There has been growing evidence, including clinical trials focused on understanding probiotics’ effects, suggesting bidirectional gut-brain-axis communication in mental health .
From an intervention point of view, the current evidence shows that inter-individual variation in the gut microbiome composition is mainly due to non-genetic factors, i.e., diet, exercise, and medication  and, therefore, the gut microbiome primarily contributes to the non-genetic etiology of mental health disorders. However, exceptions to this pattern exist [18, 19]. Thus, modulation of the gut microbiome by pre- and pro-biotics and other dietary or lifestyle interventions is an essential avenue for preventing and treating mental health and other conditions (for instance, see ). There is mixed evidence regarding the effect of weight loss on mental health, with reports providing contradictory evidence [21–25]. It is plausible that interindividual differences, mediated by genetic or non-genetic factors such as the gut microbiome, can explain why some individuals improve their mental health after weight loss and others do not.
Digbi Health has implemented and commercialized a personalized digital care intervention that uses its multi-omics platform to provide dietary and lifestyle recommendations personalized using genetic and microbiome information. The intervention has been shown to provide body weight loss in over 70% of individuals , reduction of fasting blood glucose level by 17.55% and an average reduction of HbA1c level by 6.27% , and a significant reduction in the symptomatology of Functional Gastro-Intestinal Disorders (FGIDs) . This study addresses if gut microbiome taxa or functions and genetic markers explain body weight loss’s effect on mental health. We focused our analysis on 369 overweight and obese participants on Digbi Health personalized digital care dietary and lifestyle intervention who reported baseline and follow-up mental health outcomes on depression, anxiety, and insomnia. Our results provide evidence for inherited genetics and gut microbiome factors predisposing individuals to improving mental health after weight loss. They offer an opportunity to personalize and develop tailored dietary recommendations to tackle obesity and mental health disorders.
E&I Review Services, an independent institutional review board, reviewed and approved IRB Study #18053 on 05/22/2018. Additionally, IRB Study #21141 was determined to be exempt from IRB review by E&I Review Services on 08/06/2021. Research material derived from human participants included self-collected buccal and fecal swabs. Informed consent was obtained electronically from study participants.
This study aimed to identify genetic and gut microbiome factors associated with mental health improvement after a dietary and lifestyle intervention. We included the study subjects that reported mental health conditions, particularly anxiety or depression when initiating the intervention. Study subjects were asked to rate their improvement in mental health after they had achieved at least 2% body weight loss through the digital therapeutic program. This low body weight loss threshold allowed us to have a broad range of body weight loss in this cohort. Additionally, our previously published research indicated that two thirds of high BMI individuals lost at least 2% body weight 120 days after initiating the intervention. As genetic information, we considered genetic scores for traits that are known comorbidities of anxiety, depression, and obesity (see details below). We added to the outcomes of the study sleep problems at baseline and insomnia at follow-up. Both are integral parts of anxiety and depression and are known to be associated with digestive and gut issues, like functional gastrointestinal disorders. As for the gut microbiome, we considered taxonomic annotation summarized at the level of genera and predicted functional pathways previously linked with mental health (see details below).
Participant enrollment, intervention, and phenotype data collection
Study participants were recruited from February 2020 to October 2021 among those who achieved 2% or more body weight loss from the date when enrolled in the Digbi Health personalized digital care program. The study participants were provided with an online questionnaire that included questions regarding anxiety or depression at baseline as well as sleep problems at baseline. Those who indicated a positive answer to having anxiety or depression were asked to rate the intensity of their anxiety and depression on a scale from 0 (minimum) to 5 (maximum). Those that indicated a positive answer about having sleep problems at baseline were asked to rate the intensity of their insomnia on a scale from 0 (minimum) to 5 (maximum). Additionally, as part of the enrollment process for Digbi’s program study, participants provided information on the presence or absence of symptoms associated with FGIDs, prescribed and over-the-counter medications or supplements, alcohol intake, recreational drug usage, and demographic information, including age, gender, and height. Body mass index at baseline and follow-up was calculated using the closest weight measurement to the enrollment date and date on which the survey was answered (within ± 14 days). Individuals were classified as having an FGID if they reported one or more of the following seven conditions: constipation, diarrhea, gassiness, bloating, heartburn/acid reflux, abdominal pain (cramping/belly pain) and IBS [29, 30].
To identify those participants consuming either antidepressants, anti-anxiolytics, antibiotics, or antimycotics, we mapped the subjects’ reported medications names to the chembl database using the database API . We further obtained the ATC codes (https://www.whocc.no/atc_ddd_index/) from which we identified medications prescribed for depression (N06AA, N06AB, and N06AX), anxiety (N05B) and antibiotics and antimycotics (J01B, J01C, J01D, J01E, J01F, J01G, J01M, J01R, J01X, J02, and J04).
Digbi Health’s intervention has been described elsewhere . In a nutshell, personalization of plans is achieved by analyzing participants’ genetics, gut bacteria, lifestyle, and demographics. Based on these data, the program encourages participants to make incremental lifestyle changes focused on reducing sugar consumption and timing meals to optimize insulin sensitivity, reduce systemic inflammation by identifying possibly inflammatory and anti-inflammatory nutrients, and increase fiber diversity to improve gut health. Behavioral changes are implemented with the help of virtual health coaching and the app, ensuring they are habit-forming.
Sample collection and processing: Genome SNP array and gut microbiome sequencing
Subjects self-collected saliva samples using buccal swabs (Mawi Technologies iSwab DNA collection kit, Model no. ISWAB-DNA-1200) and fecal samples using fecal swabs (Mawi Technologies iSWAB Microbiome collection kit, Model no. ISWAB-MBF-1200). Sample collection was completed by following standardized directions provided to all subjects in an instruction manual. Saliva DNA extraction, purification, and genotyping using Affymetrix’s Direct to Consumer Array version 2.0 (“DTC”) on the Affymetrix GeneTitan platform was performed by Akesogen Laboratories in Atlanta, GA. Genotype calling was performed using Axiom Analysis Suite Software version 5.1.1. Genotypes were set to ‘missing’ if the sample did not meet the confidence threshold for making the call or if the probeset did not fall into a recommended category. Specific details about the probeset metrics can be found in the Axiom Analysis Suite User Guide or the Genotyping Data Analysis Guide (ThermoFisher Scientific 2022). Sample processing of baseline (pre-intervention) fecal samples was followed by the 16S rRNA gene amplicon sequencing performed at Akesogen Laboratories in Atlanta, GA. DNA extraction was performed using Qiagen MagAttract Power Microbiome DNA Kit on an automated liquid handling DNA extraction instrument.
Microbiome data analyses
The V3-V4 region of the 16S rRNA gene was amplified and sequenced on the Illumina MiSeq platform using 2 x 300 bp paired-end sequencing . Sequence reads were demultiplexed, and ASVs generated using DADA2 in QIIME2 version 2020.8 . Primers and low-quality bases (Q <30) were trimmed off the reads from forward and reverse reads. Taxonomic assignment was performed using the vsearch consensus method against the 99% non-redundant Silva 138 reference database . We excluded hits to Mitochondria, Chloroplast, Eukaryota, and unassigned taxa at the phylum level . Six samples for participants who reported antibiotic consumption were excluded from the downstream analysis. Any ASV not seen more than once in at least 10% of the samples were filtered out to remove ASVs with a small mean and trivially large coefficient of variation . The abundance matrix was rarefied at even depth (n=36,000 reads per sample with 500 iterations) using the ‘q2-repeat-rarefy’ plugin from QIIME2 . The abundance matrix was then agglomerated at the genus level, resulting in 178 taxa across 344 samples. We further filtered any taxa with read counts less than 10. Furthermore, the raw abundance data were subjected to centered-log ratios (CLR) transformation .
The microbial functional prediction was performed using the q2-picrust2 plugin (version 2021.2) in QIIME2 . The abundance matrix comprising KEGG Orthologs was then used to obtain the abundance of specific pathways related to neuroactive metabolites produced by gut microbes, described as ‘Gut-Brain Modules’ (GBMs) in an earlier study . Predicted gene abundance encoding these metabolites was derived using the Omixer-RPM package (version 0.3.2) .
Genetic data analyses
Probe level genotype calls were unified at the variant level and formatted in VCF format, and multiallelic sites with discordant genotype calls among different probes were set as missing. VCF files were left normalized with bcftools version 1.14 (using htslib version 1.14). Beagle version 5.3 [40, 41] was used for phasing and imputation using the 1KG project as reference . Processing of the 1KG data for imputation included only SNPs and Indels, removing sites with allele counts <2 and left normalization with bcftools, and converting filtered and phased VCF files to bref3 format using the bref3.28Jun21.220.jar script provided by the Beagle software suite. We included 13,478,023 variants on downstream analyses deriving from sites imputed with r2>=0.8 or chip-genotyped. All analyses were orchestrated by a SnakeMake pipeline .
We merged our un-imputed genotype data with the 1KG data, including only sites with MAF > 1% and genotype missingness < 10%. We performed PCA analysis and calculated the first 20 PCs using Plink2  on the combined dataset. We focused on the main five super populations defined by the 1KG project to estimate the sample’s ancestry composition. We trained a random forest classifier as implemented on scikit-learn  with max depth = 8 and number of trees = 100 using 75% and 25% of the dataset for training and test, respectively. We obtained a Matthew’s correlation coefficient of 0.98 showing good performance as has been reported elsewhere for this strategy .
Twenty-three traits were selected based on being digestive system traits comorbid with mental health disorders, anxiety and sleep in particular, or mental health disorders that are known comorbidities of digestive system conditions focused on IBS, IBD, and obesity [47, 48]. We reviewed each trait’s genetic and genome-wide association studies and extracted summary statistics, including chromosome, position, effect allele, effect size, and ancestry of discovery and replication populations. Table S1 provides a detailed description of each trait and associated references as data sources. Genetic scores were calculated by multiplying the beta value or logarithm of the odds ratio by the number of risk alleles each individual had and the mean overall genetic variants included on the panel. All genetic scores were coded to be interpreted such that a larger genetic score is associated with increasing inherited genetic predisposition to the condition.
To identify genetic and microbiome predictors associated with each outcome, we performed univariate analyses utilizing a multiple regression with demographic variables. We used logistic regression for sleep problems at baseline and anxiety or depression at baseline. We used Poisson regression to improve intensity scores, with the outcome being the intensity at T1 and offset being the intensity at T0. In all models, we included as demographic variables FGID: binary, the self-reported status of the previous diagnosis of a functional gastrointestinal disorder; gender: binary, male or female; BMI at T0: continuous variable; Age: continuous variable; weight loss: categorical variable, categorized as those with no change, lost 0 to 5%, 5-10% or more than 10% of their body weight at T1 in relation to T0; and five principal components (continuous variable) calculated using the genetic ancestry analyses described previously. Linear regression was fitted using the statsmodels python package  v0.13.2 using the Binomial and Poisson families with the identity and links for logistic and Poisson regression, respectively, and with the HC3 covariance matrix as suggested earlier . These regression models identified microbiome and genetic factors measured at baseline that are associated with increased prevalence of mental health at baseline and different levels of improvement in mental health at follow-up. On the logistic models (linear models with Binomial family link), a regression coefficient greater than zero is interpreted as an increasing prevalence of self-reported illness with an increasing abundance of microbiome factors or a higher value of the genetic scores. On the Poisson regression models, a regression coefficient greater than zero is interpreted as a higher abundance of microbiome factors or a higher value of genetic scores being associated with less than average improvement in the outcome.
We applied correction for multiple hypothesis tests using the False Discovery Rate (FDR)  as implemented on the function “multiple tests” of the statsmodels python package and selected statistically significant results with an FDR ≤ 0.15.
We performed model comparisons between four different models by building regression models for demographic predictors only (D), demographic and microbiome predictors (D+M), demographic and genetic predictors (D+M), and all three sets of predictors (D+M+G). For microbiome and genetic predictors, we included in the models’ variables identified on univariate analyses with an FDR ≤ 0.15. We performed a singular value decomposition (SVD) analysis with the microbiome variables, bacterial genera, and functional pathways to avoid collinearity in the regression models. We used the singular vectors as predictors on the regression models. We selected the same number of singular vectors as the number of microbiome variables used for the SVD. We report adjusted pseudo r-squared values, which allow comparison between models with a different number of predictors; in particular, we used Pratt’s method , which is recommended for cases where the ratio between the number of samples and predictors is less than 100 . Model comparisons were performed using Cox-Snell pseudo r-squared adjusted using Pratt’s methods to correct for the different number of predictors in the different models. To estimate the mean, median, standard deviation, and percentiles of the pseudo r-squared values, we performed bootstrap with 1001 bootstrap replicates. Bootstrap values were bias-corrected by subtracting the absolute value of the difference between the mean pseudo r-squared of the 1001 bootstrap replicates and the value obtained from the original non-bootstrapped data. This ensures that the mean of the bootstrap distributions is the same as the pseudo r-squared obtained from the original non-bootstrapped data. The bootstrap analysis is not meant for hypothesis testing but to provide uncertainty estimated around the pseudo r-squared reported.
To test the effect of potential confounders, particularly mental health medications and demographics, on the association of gut microbiome with mental health outcomes, we performed a multivariate analysis using PERMANOVA with 999 bootstrap iterations based on Bray-Curtis dissimilarity with the vegan package in R . Age explained the highest variation in the microbiome at baseline, and hence to reduce its confounding effect, PERMANOVA models were run by controlling for age by stratifying it as blocks (strata=age). We additionally performed linear regression models with CLR transformed taxa abundances as outcomes and included potential confounders and their interaction with mental health as predictors.
Our study sample consisted of 369 individuals recruited from the Digbi Health research study cohort that reported anxiety/depression or sleep problems when starting Digbi Health personalized digital care intervention (Figure 1). The study subjects were enrolled on average 88.3 days (median = 64 and std = 67.7 days) by the time lost at least 2% body weight and were sent a questionnaire about their mental health and answered the questionnaire on average 1.7 days (median = 0 and std = 7.7 days) after receiving it. We obtained microbiome and genetic data for 344 and 348 individuals, respectively, 328 of whom submitted both sample types. For the baseline models of anxiety or depression, and sleep problems, all these 328 participants were studied. For the improvement models, we started with the 147, 148, and 163 individuals that reported their change in intensity for anxiety, depression, and insomnia, respectively. Most of the individuals (>95%) reported improvement or maintenance of symptoms, and only 4% (6 out of 148) for depression, 2.7% (4 out of 147) for anxiety, and 1.8% (3 out of 163) for insomnia reported higher scores at T1 compared with T0 (Figure 2 and Table S2). Due to the small sample size of individuals with worsening symptoms, we excluded them from additional analyses. Likewise, we also excluded individuals reporting improvements in 4 and 5 scale points, which were in total 8 for anxiety, 7 for depression and 6 for insomnia (Table S3). Thus, improvement models were based on 135 responses for anxiety, 135 responses for depression, and 154 responses for insomnia. We included in the analyses of the gut microbiome 178 bacterial genera and 42 functional pathways, and 23 genetic scores.