Several lines of evidence suggest that gut microbiota play an important role in mental health [1-3]. Patients with depression have altered gut microflora. Conversely, supplementation with certain probiotics in patients with depression produces anti-depressant effects [2-8]. Mice exposed to chronic stress, which causes depressive-like behavior, have altered gut microbiota composition including a reduction of Lactobacillus [9, 10]. In contrast, administration of certain Lactobacillus species (Lactobacillus spp.), such as Lactobacillus plantarum, Lactobacillus casei, Lactobacillus reuteri, or Lactobacillus-derived factors, improve stress-induced depressive-like behaviors [8-16], suggesting that certain Lactobacillus bacteria have a potential to confer neurobehavioral effects by modification of stress-induced changes in the brain. Lactobacillus paracasei is a member of the genus Lactobacillus [17]. Recent reports show that supplementation of live or heat-killed Lactobacillus paracasei produced anti-depressive-like effects in mice chronically treated with corticosterone [18], in rats treated with maternal separation [19], and in mice exposed to chronic unpredictable stress [20]. Thus, different Lactobacillus species produce anti-depressive effects, although the mechanisms by which different Lactobacillus species modify brain function are not clearly understood.

Several mechanisms have been proposed for how probiotics affect brain function: a list of studies show that probiotics release bacterial metabolites, which stimulate residual gut immune cells to release neuroactive cytokines [8, 21-25]. Recent studies, including our own, have reported that bacteria-derived extracellular vesicles (EV) mediate the beneficial effects of probiotics [26-30]. Lactobacillus plantarum-derived EVs have an ability to restore glucocorticoid-induced down-regulation of Bdnf and Nt4/5, via up-regulation of Sirt1 in cultured hippocampal HT22 cells and produce anti-depressive effects in mice with stress-induced depression [10, 11]. However, the mechanisms by which different Lactobacillus strains modify brain function are not clearly understood.

In the present study, we investigated whether Lactobacillus paracasei-derived EV (Lpc-EV) produces genome-wide transcriptional responses in HT22 neuronal cells in vitro, and whether Lpc-EV treatment produces anti-depressive effects in a stress-induced animal model of depression.

Animals

Seven-week-old male C57BL6 mice were purchased from Daehan BioLink (Eumsung, Chungbuk, Republic of Korea). They were housed in pairs in standard clear plastic cages in a temperature (23~24°C)- and humidity (50~60%)-controlled environment under a 12-h light/dark cycle (lights on from 07:00 to 19:00), and were allowed ad libitum access to water and food.

Chronic stress was induced using restraints as described previously [31, 32]. Briefly, mice were individually restrained using a well-ventilated, 50-ml polypropylene conical tube for 2-h daily for 14 days. Control mice housed in pairs were maintained in their home cages without disturbance. Lpc-EV was administered to mice via the intraperitoneal route, each dose being 6 μg of Lpc-EV in 100 μl of injection volume, as described previously [11].

Mice were handled in accordance with the animal care guidelines of Ewha Womans University. The experimental procedures for treatment with restraint and EVs were approved by the Ewha Womans University Animal Care and Use Committee (IACUC 15-012).

Preparation of EVs from Lactobacillus paracasei

Bacterial culture and EV isolation were carried out as described previously [26]. In brief, Lactobacillus paracasei was cultured in MRS broth (MBCell, Seoul, Republic of Korea) for 18 h at 37°C with gentle shaking (150 rpm). The bacterial culture was centrifugated at 10,000×g for 20 min, and the supernatant was collected and passed through a 0.22-μm bottle-top filter (Corning, NY, USA) to remove remaining cells or cell debris. The filtrate was concentrated using a MasterFlex pump system (Cole-Parmer, IL, USA) and a 100-KDa Pellicon 2 Cassette filter membrane (Merck Millipore, MA, USA), and then passed through a 0.22-μm bottle-top filter again. EVs were pelleted from the resulting filtrate by ultracentrifugation at 150,000×g for 3 h at 4°C. Pellets were washed and resuspended in PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, and 1.8 mM KH₂PO₄). The relative amount of Lpc-EV was quantified on the basis of protein concentration, which was measured using a BCA protein assay kit (Thermo Fisher Scientific, MA, USA). Resuspended Lpc-EVs were stored at -80°C until use.

HT22 cell culture and drug treatment

HT22 cells were cultured as described previously [11, 33]. HT22 cells were grown in DMEM (LM-001-05; Wel Gene Inc., Gyeongsan-si, Republic of Korea) supplemented with 10% heat-inactivated fetal bovine serum (FB02-500; Serum Source, NC, USA) and antibiotics (penicillin and streptomycin) (LS-202-02; Wel Gene Inc.) at 37°C in a humidified incubator gassed with 95% air and 5% CO₂. Cells grown to 70~80% confluence were treated with corticosterone (400 ng/ml) or Lpc-EV (10 μg/ml) in DMEM containing 1% fetal bovine serum. After 24 h, cells were washed with phosphate-buffered saline (PBS) and harvested.

Microarray analysis

Microarray analysis was carried out as described previously [32, 34, 35]. Briefly, control HT22 cells (CON), HT22 cells treated with GC (GC), and HT22 cells treated with GC and Lpc-EV (GC+Lpc-EV) groups were prepared. Total RNA from each group was isolated using RNeasy Mini Kit columns (Qiagen, Hilden, Germany). The quality and quantity of the purified RNA were measured using the Nanodrop 1000 V3.7.1 Spectrophotometer; NanoDrop Technologies, Wilmington, DE, USA) and, additionally, Agilent Bioanalyzer 2100 Expert (Agilent Technologies, Palo Alto, CA, USA). Purified total RNA had a quality between 1.8 and 2.0 for the 28s/18s rRNA ratio and between 1.8 and 1.9 for the A260/280 ratio.

Purified total RNA (400 ng) was converted to first- and second-strand cRNA, which was then converted to biotin-labeled cRNA samples. Biotin-labeled cRNAs (7.5 µg) were fragmented in an array fragmentation buffer by heating to 94°C for 35 min. Each of the fragmented, biotin-labeled cRNA samples (6.0 µg) was hybridized to an Agilent SurePrint G3 Mouse GE 8×60K Microarray, containing 29,116 mouse gene transcript counts. After washing, the array signals were amplified with Amersham Fluorolink streptavidin-Cy3 (GE Healthcare Bio-Sciences, Little Chalfont, UK) and scanned using GeneChip^® HT Scanner and AGCC software (Affymetrix GeneChip^® Command Console, Version 3.2.2). Microarray data were obtained from two independent sets of samples and were analyzed.

Microarray signals were converted into log₂ scale values and normalized with Robust Multi-array Average (RMA) method implemented in Affymetrix^® Power Tools (APT). False discovery rate (FDR) was obtained by adjusting p-value using Benjamini-Hochberg algorithm. The log₂ expression values of microarray data were used to determine the differences in the expression levels of genes in control (CON) vs. GC, and GC vs. GC+Lpc-EV. Macrogen Inc. (Seoul, South Korea) carried out the microarray process of hybridization, raw data collection, including signaling reading of the internal quality control probes and extraction of the scanned raw data.

Gene Ontology enrichment analysis

Converted microarray signal values were analyzed using the Rank-Rank Hypergeometric Overlap (RRHO) method as described previously [35-37]. Microarray signal values of CON vs. GC and GC vs. GC+Lpc-EV were ranked by signed log₁₀-transformed p-values. A RRHO geographic map of microarray signal values was constructed by placing the gene expression profile of the CON vs. GC group on the x-axis and that of the GC+Lpc-EV group on the y-axis using the RRHO website (http://systems.crump.ucla.edu/rankrank/). The ranked value was colored red if the mean expression level was higher than that of the comparison group, and blue if it was lower. A total of 29,116 transcript signals were analyzed and plotted on a RRHO geographic map. The genes that were up- or down-regulated by GC or Lpc-EV were selected by a top 20% cutoff. For the selected genes, serial hierarchical clustering was used to construct a dendrogram to visualize GC- and Lpc-EV-dependent gene expression profiles with a heatmap using TIGR Multiexperiment Viewer (MeV), version 4.9 (http://www.tm4.org/mev.html).

Serial k-means clustering and following by Gene Ontology (GO) enrichment analyses were carried out as described previously [32, 35]. Identified genes were grouped into multiple functional clusters using k-means clustering with increasing k values, followed by Gene Ontology (GO) enrichment analysis. K-means clustering with k=7 was used to form clusters of genes with representative GO terms, such as “Response to steroid hormones”, “Histone modification”, and/or the known key stress signature genes Mkp-1, Fkbp5, and Mecp2. The GO term hierarchy for the significant probe list was assigned based on the Mouse Genome-Database (http://www.informatics.jax.org), and GO was assessed via STRING (http://string-db.org).

Quantitative real-time PCR

Real-time PCR was used to verify gene expression patterns for selected genes in vitro and in vivo. Quantitative real-time PCR was carried out as described previously [10, 11]. Briefly, hippocampal tissues were homogenized using pellet pestles (Z359971, Sigma-Aldrich) in TRIzol reagent (15596-018, Invitrogen). HT22 cells grown in 6-well plates were harvested and homogenized in TRIzol reagent, and the total RNA was isolated. After treating with DNase I to avoid genomic DNA contamination, 2 μg of total RNA was used for conversion to cDNA using a reverse transcriptase system (Promega, Madison, WI, USA). Quantitative real-time PCR was carried out with 4 μl of 1/8 diluted cDNA, 10 μl of 2X iQTM SYBR Green Supermix (Bio-Rad Laboratories, Foster City, CA, USA), and 1 μl each of 5 pmol/μl forward and reverse primers in 20 μl of volume using the CFX 96 Real-Time PCR System Detector (Bio-Rad Laboratories). Transcript levels were normalized relative to Gapdh levels. Expression levels were analyzed by calculating the delta delta Ct values, which are the difference between the delta Ct values of the treated group and the control group. The delta delta Ct values were then expressed as 2 raised to the power of delta delta Ct using the CFX Manager Software (Bio-Rad Laboratories).

The following primers were used:

tBdnf (total form), forward 5’-TGGCTGACACTTTTGAGCAC-3’ and reverse 5’-GTTTGCGGCATCCAGGTAAT-3’; Bdnf1, forward 5’-CAGTGACAGGCGTTGAGAAA-3’ and reverse 5’-AACGCCCTCATTCTGAGAGA-3’; Bdnf2, forward 5’-GCAAGTGTTTATCACCAGGATCTAGC-3’ and reverse 5’-CGCCTTCATGCAACCGAAGTATG-3’; Bdnf3, forward 5’-GCTGCCTTGATGTTTACTTTGAC-3’ and reverse 5’-GCAACCGAAGTATGAAATAACATAG-3’; Bdnf4, forward 5’-GCTTTGTGTGGACCCTGAGTT C-3’ and reverse 5’-GCCGCCTTCATGCAACCG-3’; Nt3, forward 5’-TACTACGGCAACAGAGACG-3’ and reverse 5’-GTTGCCCACATAATCCTCC-3’; Nt4/5, forward 5’-AGCGTTGCCTAGGAATACAGC-3’ and reverse 5’-GGTCATGTTGGATGGGAGGTATC-3’; Ngf, forward 5’-AGCATTCCCTTGACACAG-3’ and reverse 5’-GGTCTACAGTGATGTTGC-3’; Nt3, forward 5’-TACTACGGCAACAGAGACG-3’ and reverse 5’-GTTGCCCACATAATCCTCC-3’; TrkB, forward 5’-AAGGACTTTCATCGGGAAGCTG-3’ and reverse 5’-TCGCCCTCCACACAGACAC-3’; Mkp-1, forward 5’-GCGCTCCACTCAAGTCTTCT-3’ and reverse 5’-AGAGGGGTACTACAGGAGCT-3’; Fkbp5, forward 5’-AATCAAACGGAAAGGCGAGG-3’ and reverse 5’- -3’; Hsp90a, forward 5’- -3’ and reverse 5’-CCAATCGGAATGTCGTGGTC-3’; p53, forward 5’-TCACTGCATGGACGATCTGT-3’ and reverse 5’-GACAAAAGATGACAGGGGCC-3’; Mecp2, forward 5’-ACAGCGGCGCTCCATTATC-3’ and reverse 5’-CCCAGTTACCGTGAAGTCAAAA-3’; Cbp, forward 5’-GGTTGCCTATGCTAAGAAAGT-3’ and reverse 5’-GATGCCTTGCTTATGTAAACG-3’; p300, forward 5’-GACTGAGCAGCGATAATG-3’ and reverse 5’-CAAGGTGTCTCTAGTGTATG-3’; Hdac1, forward 5’-CAGTGTGGCTCAGATTCCCT-3’ and reverse 5’-GGGCAGCTCATTAGGGATCT-3’; Hdac2, forward 5’- GGGACAGGCTTGGTTGTTTC-3’ and reverse 5’-GAGCATCAGCAATGGCAAGT-3’; Hdac3, forward 5’-AGAGAGGTCCCGAGGAGAAC-3’ and reverse 5’-ACTCTTGGGGACACAGCATC-3’; Sirt1, forward 5’-GATCCTTCAGTGTCATGGTTC-3’ and reverse 5’-ATGGCAAGTGGCTCATCA-3’; Sirt7, forward 5’-CTGGAGATTCCTGTCTACAACCG-3’ and reverse 5’- AGTGACTTCCTACTGTGGCTGC-3’; Kdm4a, forward 5’-GTCTGGCCTCTTCACTCAGT-3’ and reverse 5’- TACCATTCACGTCTGCTCCA-3’; Kdm4b, forward 5’-CGGGGCTTTTCACCACAGTAC-3’ and reverse 5’- GTACAGGGAGCCACTGATGT-3’; SUV39H1, forward 5’-TGGTTAAGTGGCGTGGGTAT-3’ and reverse 5’- TTGTTCCCAACGCTGAAGTG-3’; Setdb1, forward 5’-GGTGGTTGAAGAGCTGGGTA-3’ and reverse 5’-TCACTTCCCTGGATGCATCA-3’; Gapdh, forward 5’-AGAAGGTGGTGAAGCAGGCATC-3’ and reverse 5’-CGAAGGTGGAAGA GTGGGAGTTG-3’.

Behavioral tests

All behavioral tests were monitored with a video tracking system (SMART; Panlab, Barcelona, Spain) and/or a webcam recording system (HD Webcam C210, Logitech, Newark, CA, USA).

Sociability test

The sociability test was carried out as described previously [32, 33]. Briefly, a U-shaped two-choice field was prepared by partitioning an open field (40×40 cm²) with a wall (20 cm in width and 20 cm in height) at the center point. A subject mouse was allowed to freely explore the two-choice field with an empty circular grid cage (12 cm in diameter and 33 cm in height) on each side for 5 min and was then returned to the home cage. After 10 min, a social target was loaded into a circular grid cage on one side, and the subject mouse was allowed to explore both fields for 10 min. The trajectory of the mouse’s movements and time spent in each field were recorded using a video tracking system. The field with the circular grid cage containing a social target and the field containing an empty grid cage were defined as the target field and non-target field, respectively.

Tail suspension test

The tail suspension test was carried out as described previously [32, 33]. Mice were suspended individually by fixing their tails with adhesive tape to the ceiling of an enclosed shelf 50 cm above a bottom floor. While recording using a webcam recording system, cumulative immobility time was measured for 6 min. Immobility was defined as the time the animal spent suspended with all limbs motionless.

Forced swim test

The forced swim test was performed as described previously [32, 33]. Mice were placed in a Plexiglas cylinder (15 cm in diameter×27 cm height) containing water at a temperature of 24°C. Mice were placed in the cylinder for 6 min, and the cumulative immobility time was measured for the last 5 min. Immobility was defined as the time the animal spent floating with all limbs motionless.

Statistical analysis

Two-sample comparison was carried out using the student’s t-test. Multiple comparisons were performed by one-way ANOVA followed by the Newman-Keuls post hoc test. All data are represented as the mean±SEM, and statistical significance was accepted at the 5% level.

Lpc-EV counteracted glucocorticoid-induced genomic responses in HT22 cells

We investigated whether Lpc-EV has the ability to modify glucocorticoid (GC)-induced changes in the expression of neurotrophic factors. GC (corticosterone) treatment in HT22 neuronal cells decreased the expression of Bdnf, Nt3, Nt4/5, Ngf, and TrkB. In contrast, Lpc-EV treatment counteracted GC-induced decreased expression of Bdnf, Nt3, Ngf, and TrkB (Fig. 1A).

Next, we explored the underlying mechanisms by which Lpc-EV restored GC-induced changes in the expression of neurotrophic factors. To the end, we applied two different approaches. First, we tested whether any known factors regulating the expression of neurotrophic factors are involved in Lpc-EV effects. Several classes of stress-related cellular or epigenetic factors, such as Mkp1, Fkbp5, Hsp90a, p53, MeCP2, Hdac2, and Sirt1 have been examined for their role in regulating the expression of neurotrophic factors [11, 33, 38-40]. GC treatment in HT22 cells decreased or tended to decrease the expression of p53, Mecp2, and Sirt1, while increasing the expression of Mkp1, Fkbp5, and Hsp90a. In contrast, Lpc-EV treatment blocked GC-induced reduced expression of p53 and Mecp2, and GC-induced increased expression of Mkp1 and Fkbp5 but not significantly Hsp90a, Hdac2, and Sirt1 (Fig. 1B, C).

Recently, it has been reported that MeCP2 is a critical player in regulating the expression of Mkp1 and Fkbp5, and neurotrophoic factors [39-41]. Therefore, it was investigated whether MeCP2 is a key regulator of the effects of Lpc-EV on the expression of neurotrophic factors. siRNA-mediated knockdown of Mecp2 blocked Lpc-EV-induced recovery of GC-induced changes of Bdnf, Nt4/5, TrkB, p53, Mkp1, and Fkbp5 but not Nt3 and Ngf (Fig. 1D~G). These results suggest that Lpc-EV has the ability to counteract GC-induced genomic changes of those genes through Mecp2 upregulation.

Second, we used a genome-wide approach to identify genes whose expression was changed by GC and their altered expression was reversed by Lpc-EV. We obtained the microarray signal values of the genes whose expression was changed in HT22 cells treated with GC (GC) compared to control HT22 cells (CON) (GC vs. CON), and the genes whose expression was changed in HT22 cells treated with GC and Lpc-EV (GC+Lpc-EV) compared to HT22 cells treated with GC (GC+Lpc-EV vs. GC). Then, we constructed the geographic map of gene expression profiles using the obtained the signal values (Fig. 2A, B) Using the Rank-Rank Hypergeometric Overlap (RRHO) method [36, 37]. Of the total 29,116 transcripts in the microarray, 8,158 transcripts (28.0%) were upregulated by GC, and their altered expression was reversed by Lpc-EV treatment and these transcripts were located in Quadrant A of the RRHO map. Another 8,592 transcripts (29.5%) were down-regulated by GC and their altered expression was reversed by Lpc-EV treatment, and these transcripts were located in Quadrant D of the RRHO map (Fig. 2A~C). Of those transcripts, the top 20% of transcripts by rank based on expression differences in each quadrant were selected. Then, 1,632 in Quadrant A and 1,728 in Quadrant D were selected further by annotating the 20%-ranked transcripts with Mus musculus genes. A cutoff of 0.4, which is a medium confidence interaction score in the STRING database, was used to finally select 998 transcripts in Quadrant A and 756 transcripts in Quadrant D that were most likely to be actively responding transcripts to GC and Lpc-EV. These transcripts were then placed in further analyses.

Serial K-means clustering and Gene Ontology (GO) enrichment analysis indicated that the selected genes (998 genes in Quadrant A and 756 genes in Quadrant D) could be grouped into multiple functional clusters that enriched genes involved in stress responses and epigenetic modifications. In principle, all the selected genes actively responded to both GC and Lpc-EV. In the present study, we focused on the following gene enrichment pathways. The 998 genes in Quadrant A could be grouped into 7 clusters (Fig. 2D; Supplemental Table S1 and S2), one of which (Cluster 5) included a number of the genes covered by the GO terms “regulation of neuron death”, “regulation of stress-activated MAPK cascade”, “response to steroid hormone”, “regulation of oxidative stress-induced cell deaths” and “Histone H3 acetylation” (Fig. 2D, F, G; Supplemental Table S2). Cluster 5 contained the stress signature genes Mkp1 (Dusp1) and Fkbp5 (Supplemental Table S2), whose expression was increased by GC and their increased expression was reversed by Lpc-EV treatment (Fig. 1).

The 756 genes in Quadrant D could be grouped into 8 clusters (See also Supplemental Table S1 and S3), two of which (Clusters 3 and 4) included a list of the genes covered by the GO terms, “histone modification” and “histone acetylation” (Cluster 3); and “regulation of neuron death”, “stress-activated protein kinase signaling cascade”, “regulation of transcription from RNA polymerase II promoter in response to stress” and “regulation of long-term neuronal synaptic plasticity” (Cluster 4) (Fig. 2E, H; Supplemental Table S3). Cluster 4 contained Ngf and Mecp2 (Supplemental Table S3), whose expression was decreased by GC and their decreased expression was reversed by Lpc-EV in HT22 cells (Fig. 1).

GC- and Lpc-EV-dependent changes in the expression of the selected genes were analyzed using a dendrogram. A dendrogram of the expression patterns of the 135 genes in Cluster 5 (Quadrant A) highlighted the genes for stress-activated Mapk cascades and neuronal cell death and whose expression was upregulated by GC and their altered expression was reversed by Lpc-EV (Fig. 3A). Similar analyses of the 101 genes in Cluster 3 and the 99 genes in Cluster 4 (Quadrant D) were used to build the dendrograms of genes for histone modification factors and stress-activated protein kinase signaling cascade and whose expression exhibited GC-dependent downregulation and Lpc-EV-induced reversion of the altered expression (Fig. 3B, C).

We focused on and investigated the changes in the expression of the genes with the functional modules representing histone modification factors in Clusters 3 and 4 in Quadrant D (Fig. 3B, C; Supplemental Table S1 and S3). Real-time PCR analysis indicated that GC treatment decreased the expression of p300, Sirt1, Sirt7, and Suv39h1 and increased the expression of Hdac2. In contrast, Lpc-EV treatment blocked GC-induced down-regulation of p300, Sirt1, and Suv39h1 but not Sirt7, also inhibiting GC-induced up-regulation of Hdac2 (Fig. 4A, B). siRNA-mediated knockdown of Mecp2 blocked Lpc-EV-induced restoration of GC-induced changes of p300, Hdac2, and Sirt1, whereas its effect on Suv39h1 was insignificant (Fig. 4C, D). These results suggest that Lpc-EV counteracts GC-induced changes in the expression of histone modification factors through Mecp2 upregulation.

Lpc-EV treatment reversed stress-induced changes in the expression of Mkp-1, Fkbp5, and Mecp2 in the hippocampus

Next, we investigated whether Lpc-EV treatment produces genomic responses against stress-induced changes in the brain of mice. Mice exposed to chronic restraint stress (CRST) for 2-h daily over 14 days exhibit persistent depressive-like behaviors [10, 33, 34]. CRST-treated mice had increased expression of Mkp-1 and Fkbp5, and decreased expression of Mecp2 in the hippocampus (Fig. 5A~C). In contrast, post-stress treatment with Lpc-EV in CRST-treated mice reversed stress-induced increased expression of Mkp-1 and Fkbp5, also reversing decreased Mecp2 expression (Fig. 5A~C).

The epigenetic factors, Mecp2, Sirt1, Suv39h1, and Hdac2 act as upstream regulators of Mkp-1, Bdnf, Nt3, Nt4/5, and TrkB in the brains of CRST mice [11, 33, 41]. We tested whether Lpc-EV treatment had an impact on stress-induced changes of epigenetic factors. CRST mice had decreased expression of Cbp, p300, Sirt1, Sirt7, and Suv39h1, and increased expression of Hdac2. In contrast, post-stress treatment with Lpc-EV in CRST-treated mice partially rescued stress-induced changes in the expression of Hdac2, Sirt1 and Suv39h1 (Fig. 5D, E). Lpc-EV also rescued stress-induced changes in the expression of Bdnf, Nt3, Nt4/5, Ngf, and TrkB in the hippocampus (Fig. 5F).

Collectively, these results suggest that Lpc-EV cargo contents have the ability to counteract stress-induced changes in the expression of Mkp-1, Fkbp5, Mecp2, Sirt1 and Suv39h1, also counteracting stress-induced changes in the expression of Bdnf, Nt3, Nt4/5, Ngf, and TrkB in the hippocampus.

Lpc-EV treatment relieved stress-induced depressive-like behavior

Mice exposed to CRST showed reduced time to explore the cage with a target mouse over an empty cage in the social interaction test (SIT) (Fig. 6A~C), thus exhibiting reduced social interaction in the SIT, and showed increased immobility time in the tail suspension test (TST) and the forced swim test (FST) (Fig. 6D, E), which are consistent with previous reports [10, 33, 34]. In contrast, post-stress treatment with Lpc-EV for 14 days in CRST mice reversed stress-induced reduced social interaction (Fig. 6A~C), and stress-induced increased immobility in the TST and FST (Fig. 6D, E). These results suggest that Lpc-EV produces anti-depressive effects at the behavioral level.

Probiotics are believed to be beneficial for psychiatric disorders, yet the underlying mechanisms by which specific probiotic strains produce neuroactive effects are not clearly understood. In the present study, we demonstrate that Lpc-EV produced genome-wide changes against GC-induced transcriptional responses in HT22 cells. Microarray analysis followed by the RRHO method led us to identify 1,754 genes that were up- or down-regulated by GC, and their altered expression was reversed by Lpc-EV treatment in HT22 cells (Fig. 1~4; Supplemental Table S1~S3). Lpc-EV-induced actively responding genes include those which contain functional modules representing “responses to stress or steroid hormones”, “histone modification”, and “regulating MAPK signaling pathways”. Our results raise the following important issues on Lpc-EV effects (Fig. 2; Supplemental Table S3).

First, Lpc-EV induced genome-wide transcriptional changes by directly acting on HT22 neuronal cells without intermediate mediators. Considering that bacterial EVs contain proteins/peptides, carbohydrates, lipids, nucleic acids and bacterial metabolites [42-44], the bioactive components in Lpc-EV should contain some of such components, although which cargo components in Lpc-EV produce those massive transcriptional responses yet remains to be characterized in the future. More discussion will be presented in the following sections.

Second, Lpc-EV cargo contents have the ability to reverse GC-induced changes in the expression of epigenetic factors. Lpc-EV-induced up-regulation against GC-induced changes in neurotrophic factors, Mecp2, and Sirt1 (Fig. 1) are similar to those of Lactobacillus plantarum-derived EVs [10, 11]. Lpc-EV also up-regulated other epigenetic factors, including Hdac1, Hdac2, and Suv39h1 (Fig 4; Supplemental Table S3). Considering that epigenetic factors can serve as the basic regulators of chromosome structures and the transcription of numerous genes, Lpc-EV-induced changes in epigenetic regulation of genes likely contribute genome-wide effects.

Third, Lpc-EV counteracted GC-induced transcriptional responses. Genome-wide microarray analysis followed by the RRHO method led to the identification of 1,754 GC-responsive genes (998 genes in Quadrant A and 756 genes in Quadrant D) and their altered expression was reversed by Lpc-EV treatment (Fig. 1~4). Regarding the massive and diverse genome-wide effects of Lpc-EV, it is possible that bioactive components in Lpc-EV likely drive multiple arrays of signaling events to restore a homeostatic ability in GC-challenged HT22 cells.

The findings that Lpc-EV produced similar effects on the expression of Mkp-1, Fkbp5, Mecp2, Sirt1, and Suv39h1 epigenetic and nuclear factors, and of Bdnf, Nt3, Nt4/5, Ngf, and TrkB in HT22 cells (Fig. 1~4) and in the hippocampus of CRST-treated mice (Fig. 5) raise the possibility that Lpc-EVs contain neuroactive components that can directly modify brain cells via transcriptional responses. Bacterial EVs carry various molecular cargo materials, such as bacterial nucleic acids, enzymes, peptides, short-chain fatty acids, and biometabolites [42, 43]. Because bacterial EVs consist of lipid bilayer-encapsulated vesicles with a diameter of 20~250 nm, they can easily cross the mucosal barrier of the intestine and cell membrane [42]. Indeed, orally delivered Cy7-labeled Staphylococcus aureus EVs [43] or Cy7-labeled Helicobacter pylori EVs [45] in mice were distributed in the brain and many other tissues. Treatment with Lactobacillus casei-derived EVs in the human colorectal cell line Caco-2 decreases TLR9 gene expression and increases the anti-inflammatory cytokines IL-4 and IL-10 [46]. Treatment with Lactobacillus plantarum-derived EVs increases BDNF expression in HT22 cells [11]. These results suggest that bacterial EVs contain cargo materials that may induce signaling events and transcriptional responses by directly acting on eukaryotic host cells including brain cells. Recently, it has been reported that EVs derived from Lactobacillus acidophilus or Escherichia coli exhibit antitumor activity, which is dependent on interferon-gamma (IFN-γ) production. Trypsin-sensitive EV surface proteins appear to be the active component of EVs that induce IFN-γ production [47]. Propionibacterium freudenreichii EVs have anti-inflammatory properties, and the surface proteins in these EVs are responsible for their probiotic effects [48]. However, it is not known at present whether EVs from different Lactobacillus species contain distinct cargo materials that induce similar or distinct transcriptional responses or whether EVs from different Lactobacillus species have similar or common cargo materials that induce the same pathways leading to the observed transcriptional responses.

In the present study, we delivered EVs through the intraperitoneal (IP) route, which is physiologically linked to the circulatory system, and thereby attempted to abate a possible complication involving intestine-related mechanisms. IP-injected Lactobacillus plantarum-derived EVs increased BDNF expression in the hippocampus of mice, similarly as that induced in cultured neuronal cells [10, 11]. IP-injected molecules in rodents rapidly and effectively enter the circulatory system, and, therefore, IP administration can be chosen as a justifiable route for proof-of-concept studies [49]. IP-injected small molecules (MW<5,000) and fluids are predominantly absorbed from the visceral peritoneum by diffusion through the mesenteric capillaries, drain into the portal vein, and enter the systemic circulatory system. Whereas IP-injected large molecules (MW>5,000), proteins, blood and immune cells are taken up mainly through the lymphatic system [49]. Although it needs to be investigated which is a preferred path for IP-injected probiotic-EVs to enter the circulatory system, we speculate they might go through any of the two routes as EVs are nano-size lipid vesicles. IP-injected fluorescent lipophilic dye-labeled EVs are accumulated in many organs in the body, including the brain, liver, and kidney, although accumulation levels vary among different regions [50]. Therefore, the results of the present study, together with those from our recent studies [10, 11, 44, 45], support the notion that bacterial EVs can serve as important mediators for communication from gut microbiota to the brain.

This study was supported by a grant (2022M3E5E8017561) for PL Han, a grant (2022R1C1C2011198) for EH Lee, and a grant (2022R1I1A1A01070983) for JY Park from the Ministry of Science, ICT and Future Planning, Republic of Korea and partly by MD Healthcare Inc.