This research was conducted in accordance with the Good Clinical Practice guidelines and the Declaration of Helsinki. The protocol was approved by the Clinical Research Ethics Committee of the Cork Teaching Hospitals (Protocol Number APC102). Informed consent was obtained from all participants prior to enrollment into the study.
Participants and study design
This study was designed as a single-blind, randomized, controlled study. Healthy adult (male and female) participants with poor dietary habits, aged 18–59 years were recruited from the Cork area between February 2018 and November 2020. The CONSORT flow diagram showing subject enrollment and allocation is depicted in Supplementary Fig. 1. Participants were block randomized (block of 4, stratified by gender) into either intervention (DIET) or control (CONT) group using randomly permuted blocks and were instructed to follow their respective diet for 4 weeks (detailed description of diets below). Other than following the diet, participants were instructed to refrain from introducing any probiotic or other nutritional supplements and to keep physical activity consistent throughout the study.
Participants were not eligible to participate if they had any significant acute or chronic illness, were taking any medication (except for contraceptive pills), were peri-menopausal, menopausal or post-menopausal, were pregnant, or lactating, were not fluent in the English language, were vegan, a habitual daily smoker or had taken any pro-, pre- or antibiotic 4 weeks prior to enrollment in the study. Poor dietary habits were determined by conducting a dietary recall with a trained dietitian. At the initial screening visit, participants were screened for any psychiatric disorder using the MINI International Neuropsychiatric Interview (MINI, Version 7.0.2) and demographic data were collected. Participants also completed the National Adult Reading Test (NART) [21] as a brief measure of verbal IQ as well as the State-Trait Anxiety inventory [22] to measure baseline anxiety levels.
Dietary intervention
Psychobiotic diet
The foods in focus of the psychobiotic diet included those known to influence the microbiota, namely, whole grains, prebiotic fruits and vegetables, fermented foods, and legumes while discouraging consumption of “unhealthy” foods such as sweets, fast food or sugary drinks.
The dietary intervention consisted of one individual, 30-min long education session (baseline) as well as a 15-min long (after 2-weeks) refresher session facilitated by a registered dietitian. During the first session, habitual dietary habits were assessed based on a 7-day food diary the participants completed in the week leading up to the baseline visit. Following this assessment, participants were educated on the components of the study diet, which included consumption of fruits and vegetables high in prebiotic fibers (6–8 servings per day, e.g., onions, leeks, cabbage, apples, bananas, oats), grains (5–8 servings per day) and legumes (3–4 servings per week) as well as fermented foods (2–3 servings per day, e.g., sauerkraut, kefir or Kombucha). For fermented foods, one serving equaled 200 ml or one cup. All other serving sizes were according to the standardized portion size guidance of the Health Service Executive Ireland. Advice including strategies for meal planning, sample menus and recipes that include the specified foods of the study diet as well as on nutrition label reading was given to the participants. Additionally, participants were educated on current Irish food pyramid guidelines and were advised to stay within a 2000–2200 kcals daily intake for females and 2400–2800 kcals for males, which corresponds to the recommended calorie intake per day for moderately active adults according to the Irish dietary guidelines. All education materials (Supplementary Material) were provided to the participants as handouts to take home. Facilitators and barriers to adhering to the study diet were discussed and all remaining questions were answered. In the second session, dietary intake was reviewed again in form of a 7-day food record. Participants were reminded of the components of the diet and were instructed to increase consumption of specific foods if guidelines were not met. Again, all concerns and questions of the participants were discussed. Between study education sessions, participants were encouraged to contact the study dietitian with questions or concerns.
Control diet
A dietitian also facilitated the education for the control group and the session was matched in time and dietitian-contact to the diet intervention. A review of dietary habits was completed like the diet group; however, minimal input was provided and focused mainly on the HSE food pyramid. At the 2-week follow up, support was offered to the participants and dietary intake was reviewed using the 7-day food record.
All participants were debriefed after the study and participants in the control arm were provided information about the psychobiotic diet.
Sample collection
Questionnaire data (as described below) and collection of biological samples was completed pre- and post-intervention.
Dietary intake quantification
Dietary intake was quantified using a 7-day food record and food frequency questionnaire (FFQ) that was validated for an Irish adult population [23]. The food diary was completed three times, first in the week leading up to the baseline visit, then in the second week of intervention and finally in the week leading up to the final visit. Instructions on how to complete the food diary (including portion sizes, preparation of food) were provided to participants and the food diary was reviewed by a study dietitian. The FFQ was completed at the pre- and post-intervention visits and asked participants to estimate the intake of specified foods over the previous month. Food groups were derived from the FFQ, by converting each response to a corresponding frequency factor and summing the frequency factors over all food items in a food group to get the average serving per day.
Self-report questionnaires
Assessment of perceived stress
The Cohen’s perceived stress scale (PSS) [24] was completed by the participants pre- and post-intervention.
General health assessment
Gastrointestinal (GI) health was assessed using a GI symptom visual analogue scale. Changes in stool type were captured by the Bristol stool chart [25]. A physical exam (height, weight, blood pressure) was completed at the study visits. Sleep quality was measured using the Pittsburgh Sleep Quality Index (PSQI [26]).
Biological samples
Sample collection
Plastic containers containing an AnaeroGen sachet (Oxoid AGS AnaeroGen Compact, Fischer Scientific, Dublin) were used to collect freshly voided fecal samples. Participants were instructed to collect the fecal sample as close to the study visit as possible and to keep in a refrigerator until delivery to the laboratory, where it was aliquoted and stored at −80° for later analysis.
Whole blood was collected into 4-ml lithium-heparin-containing tubes (Greiner Bio-One, 454029). Plasma samples were collected into 3-ml K3EDTA tubes (Cruinn Diagnostics Limited, Dublin). Plasma samples were centrifuged at 1500 × g for 10 min. The supernatant was aliquoted and stored at −80 °C for later analysis.
Participants were instructed to collect midstream urine samples during the first urination in the morning into a sterile collection tube. Urine samples were stored in the participants refrigerator until delivery to the laboratory. In total, 4 ml of samples were aliquoted into 1 ml of 0.25% sodium azide and stored at −80 °C until further analysis.
Fecal microbiota analysis
DNA extraction
DNA was extracted from 200 mg of fecal sample using the QIAamp Fast DNA Stool Mini Kit (Qiagen, Germany) combined with repeated bead beating steps, according to methods previously described [27, 28]. DNA concentrations were quantified using Qubit High Sensitivity Kit (Invitrogen). Extracted DNA was then stored at −20 °C until prepared for metagenomic sequencing.
Shotgun sequencing library preparation
Whole genome shotgun sequencing library was performed the Nextera XT kit and the library was prepared according to the Nextera XT DNA Library Preparation Guide (Illumina). Briefly, DNA concentrations were diluted to 0.2 ng/μl and fragmented by incubating at 55 °C for 7 min. Paired-end Nextera XT indexes were added and 12 cycles of amplification process were completed. Samples were purified with AMPure XP beads according to the manufacturer’s instructions. In order to calculate molarity, DNA concentration was quantified using the Qubit dsDNA High Sensitivity Assay and amplicon size was measured by Agilent Bioanalyser 2100. Lastly, 1 mM of libraries were pooled before loading on the Illumina NextSeq platform for 150 bp paired-end sequencing.
Blood inflammatory profile
Cytokine levels from unstimulated and stimulated blood samples were quantified using the V-PLEX Proinflammatory Panel 1 (human) Kit (MSD, K15049D). Cytokine quantification was done according to the manufacturer’s guidelines with one modification, where 100 μl sample was added directly onto the plate without dilution. C-reactive protein levels were quantified using the kit following the manufacturers guidelines (MSD, K151STD) using a 1000-fold dilution. Sensitivity of each assay is shown in Supplementary Table 1. Specificity of each assay was <0.6%.
Fecal metabolomics
Sample analysis was carried out as follows:
SCFA
SCFA/metabolite filtrate were acidified using hydrochloride acid, and deuterium labeled internal standards were added. All samples were analyzed in a randomized order. Analysis was performed using a high polarity column (Zebron™ ZB-FFAP, GC Cap. Column 30 m × 0.25 mm × 0.25 μm) installed in a GC (7890B, Agilent) coupled with a quadrupole detector (5977B, Agilent). The system was controlled by ChemStation (Agilent). Raw data were converted to netCDF format using Chemstation (Agilent), before the data were imported and processed in Matlab R2014b (Mathworks, Inc.) using the PARADISe software as previously described [29].
Untargeted metabolomics of semi-polar metabolites
The analysis was carried out using a Thermo Scientific Vanquish LC coupled to Thermo Q Exactive HF MS. An electrospray ionization interface was used as ionization source. Analysis was performed in negative and positive ionization mode. The UPLC was performed using a slightly modified version as previously described [30]. Peak areas were extracted using Compound Discoverer 3.1 (Thermo Fisher Scientific). Identification of compounds were performed at four levels; Level 1: identification by retention times (compared against in-house authentic standards), accurate mass (with an accepted deviation of 3 ppm), and MS/MS spectra, Level 2a: identification by retention times (compared against in-house authentic standards), accurate mass (with an accepted deviation of 3 ppm). Level 2b: identification by accurate mass (with an accepted deviation of 3 ppm), and MS/MS spectra, Level 3: identification by accurate mass alone (with an accepted deviation of 3 ppm).
Lipidomics
The lipid filtrate were diluted 50 times in eluent mix. Compound extraction was performed in the same way as described for semi-polar metabolites. Data were processed using Compound Discoverer 3.0 (Thermo Fisher Scientific). The retention time of compounds within the same lipid class is estimated using the relation between retention time and the chain length and number of double bonds. Identification of compounds were performed at four levels: Level 1: identification by accurate mass, MSMS spectra and estimated retention time; level 2a is based accurate mass and estimated retention; level 2b is based on accurate mass and MSMS spectra from an external library; level 3: accurate mass and elemental composition alone.
Blood and urine targeted metabolomics
Blood and urine metabolomics analysis, targeting key metabolites resulting from human-gut-microbiota co-metabolism of dietary essential amino acids tryptophan, tyrosine, phenylalanine and branched chain amino acids were performed at Fondazione Edmund Mach (Trento, Italy) according to established methods in the laboratory using UHPLC-ESI-MS/MS analysis [31].
Cortisol awakening response
Morning saliva samples (the morning of the day of study visit) were collected for assessment of the cortisol awakening response using the Salivette system (Sarstedt, Germany). A total of four samples was collected, with the first one collected upon wakening, the second one 30 min later, and the third and fourth one in 15 min increments. Participants were instructed to not brush their teeth until after all saliva samples were collected, to not eat or drink anything prior to the first sample, and to avoid eating and drinking 15 min prior to the remaining samples. Saliva samples were centrifuged at 1500 × g for 5 min, aliquoted, and stored at −80 °C for later analysis.
Salivary cortisol was analyzed in duplicates using the ENZO Life Sciences enzyme-linked immunosorbent assay kits (Exeter, UK) per manufacturer’s instructions. The lower limit of detection was 0.919 nmol/l. Inter- and intra-assay coefficients of variability were 13.4 % and 10.5%.
Statistical analysis
All data were analyzed using SPSS 25 (IBM, Armonk, NY, USA). Data analysis was performed using the per protocol analysis. Outliers were identified by visual inspection of box blots showing extreme outliers at ±3 standard deviations and consideration was given to removal for extreme outliers. Normality of data was examined using the Shapiro–Wilk statistic and data, if necessary, was transformed using the following transformations: natural log transformation achieved normality for energy, fat, MUFA, cholesterol, magnesium, iron, vitamin A, thiamine, vitamin B6, biotin, folate, fruit, sweets, and vegetables; square root transformation was used to transform omega 6, sodium, manganese, selenium, vitamin D, vitamin C, condiments, sauces and soups, grains, high fat dairy products, other vegetables, processed meat, and refined carbohydrates. PSQI component scores, IPAQ, GI health measures, nutrition outcome variables not listed above, and inflammatory markers were not normally distributed and transformation did not achieve normality.
Differences in outcome measures pre- and post-intervention were analyzed using analysis of covariance and post hoc paired- (within group) or unpaired- (between group) sample two-tailed t-test. Homogeneity of variance was checked using Levene’s test of equality of error variances. Age, gender and BMI were included as co-variates for the analysis of study outcome measures. If parametric assumptions were violated and data transformation did not normalize data distribution, the non-parametric Kruskal-Wallis and post hoc paired- (within group) or unpaired- (between group) Wilcoxon singed-rank test (two-tailed) were applied. For cortisol measurements, area under the curve were calculated with respect to ground (AUCg) and increase (AUCi) [32]. Categorical data were assessed using chi-squared test. p < 0.05 was considered significant. Data is expressed as mean ± SEM or median (IQR) unless otherwise indicated.
Bioinformatics analysis
We performed quality checks on raw sequences from all fecal samples using FastQC. Shotgun metagenomic sequencing data were then processed through an analysis workflow that utilizes Huttenhower Biobakery pipeline, including Kneaddata, MetaPhlAn3 and HUMAnN3 to obtain species, genes and pathways abundance matrix. Briefly, quality filtering and host genome decontamination (human) was performed using Trimmomatic and Bowtie2 via Kneaddata wrapper program with following parameters: ILLUMINACLIP:/NexteraPE-PE.fa:2:30:10, SLIDINGWINDOW:5:25, MINLEN:60, LEADING:3, TRAILING:3. Taxonomic and functional profiling of the microbial community was performed using MetaPhlan3 and HUMAnN3 using default parameters. Next, gene abundance matrix was further collapsed by Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology term and Gene Ontology term mapping via “humann_regroup_table” function provided within HUMAnN3.
Further microbiome data-handling was done in R (version 4.1.2) with the Rstudio GUI (version 1.4.1717). The iNEXT library was used to compute alpha diversity for the first three hill numbers (Chao1, Shannon entropy and Simpson Index). Differences in alpha diversity were assessed using linear mixed effect models using the lme4 package. Principal component analysis was performed on centered log-ratio transformed (clr) values as a visual companion to the beta diversity analysis. Zeroes were imputed using the “unif” method. Beta diversity was computed in terms of Aitchison distance (Euclidean distance of clr-transformed counts) and differences in beta diversity were assessed using the PERMANOVA implementation from the vegan library using 10,000 permutations. Microbial volatility was measured as the Aitchison distance between pre- and post-intervention [33].
Gut-brain modules (GBMs) and gut-metabolic modules (GMMs) were calculated using the R version of the Gomixer tool [34]. Differential abundance of taxa and functional modules was assessed by fitting linear mixed effects models on the clr-transformed count tables. To correct for multiple testing (FDR) in tests involving microbiome features, Storey’s q value post hoc procedure was performed with a q value of 0.2 as a cut off. (Code availability: Custom R scripts are available online at https://github.com/thomazbastiaanssen/Tjazi). Plotting was handled using ggplot2.
Metabolomics analysis
Sample analysis was carried out by MS-Omics as follows. Peaks were quantified using area under the curve. Biostatistics were run in R with RStudio GUI. Missing values were imputed by taking 95% of the minimum observed abundance per metabolite. Metabolites that were detected in fewer than 10% of samples were dropped from the analysis. Principal component analysis was performed on CLR-transformed values using the ALDEx2 library. The number of permutations was set to 1000. PERMANOVA followed by a pairwise PERMANOVA was used to find structural differences between treatments on a compositional level. To correct for multiple testing in tests involving metabolomics features, Storey’s q value post hoc procedure was performed with a q value of 0.2 as a cut off (Code availability: Custom R scripts are available online at https://github.com/thomazbastiaanssen/Tjazi). Metabolites that were found to be altered by diet were mapped to their Human Metabolome Database identifier and subjected to pathway analysis using the MetaboAnalyst online pipeline, choosing the human KEGG library as a reference [35]. Custom Metabolomics figures were generated using ggplot2. Fecal lipids that were significantly altered were subjected to over representation analysis using LIPEA (https://lipea.biotec.tu-dresden.de/home, accessed October 2021) based on a database source including KEGG.
Dietary adherence score
A dietary adherence score was calculated based on the ModiMedDiet score as described by [36], with higher scores indicating more adherence to the dietary recommendations. Briefly, a score per serving was assigned to fruit intake (3.33 per serving), vegetables (1.67 per serving), grains, nuts and seeds (2 per serving), legumes (3.33 per serving), and fermented foods (5 per serving). A maximum score of 10 could be achieved for all six categories (fruit; vegetables; grains; legumes, buts and seeds; fermented foods; “unhealthy foods” (e.g., sweets, chocolate, biscuits, fried food) based on the education provided to the participants, summing up to a maximum score of 60. No additional points were given for exceeding the recommended intake. For the unhealthy foods category, 10 points were assigned for 0–3 servings per week, 5 points for 3–6 servings per week and 0 points for ≥6 servings per week.
We performed regression models to predict the changes in outcomes measures based on dietary adherence scores or performed correlation analysis if regression model was deemed inappropriate to explore the relationship between dietary adherence and changes in study outcome measures. Model fit for the regression analysis was determined by checking multicollinearity was checked with VIF being close to 1, residuals were normally and randomly distributed, and no autocorrelation was detected using Durbin Watson statistic of around 2, and influential observations were checked using Cook’s Distance.
