The gut microbiome modulates the transformation of microglial subtypes

Single-cell nucleus RNA-seq profiling of Hip and PFC

A schematic of nuclei isolation and the snRNA-seq workflow from the Hip and PFC is shown in Fig. 1a. Using the droplet-based single-nucleus method, we captured 72,226, and 67,698 nuclei from the Hip and PFC, respectively, in the 9 mice (3 per group). We obtained an average of 45,455 reads per nuclei in the Hip and 50,245 reads in PFC after stringent quality control (Supplementary Table S1).

After dimensionality reduction and graph-based clustering (UMAP), we identified 36 distinct clusters in the Hip (Supplementary Fig. 1b) and 29 clusters in PFC (Supplementary Fig. 1c). Then we annotated major cell types in the two regions based on the expression of well-established marker genes [49]. Here, excitatory neurons (n = 43,956 and 41,837 in the Hip and PFC, respectively; marked by Grin2a, Syt1, Grin1), interneurons (n = 6036, 7702; Gad1, Gad2), oligodendrocyte (n = 6021, 2984; Plp1, Mog, Mbp), OPC (Oligodendrocyte precursor cells; n = 2089, 1994; Pdgfra, Vcan), microglia (n = 2343, 2353; Csf1r, Ctss, C1qa), and astrocytes (n = 6803, 5862; Aqp4, Gja1) were clearly identified (Supplementary Fig. 1d, e and Fig. 1b, c).

The absence of gut microbiota changed glial cells proportion in the Hip and PFC

Initially, we calculated the proportion of major cell types in two brain regions. In Hip, we found that the microglial proportion was significantly lower in GF compared to SPF (P = 0.013), and microbial colonization failed to rescue this change (GF vs. CGF, P = 0.662; Fig. 1d). Meanwhile, hippocampal astrocyte, oligodendrocyte and OPC proportion were downregulated in the GF group relative to the SPF group (GF vs. SPF, P = 0.005 for astrocyte, 0.0394 for oligodendrocyte and 0.015 for OPC), astrocyte and OPC were reversed by microbial colonization in the CGF group (GF vs. CGF, P = 0.0003, 0.034), excluded oligodendrocyte. Furthermore, in contrast to glial cells, excitatory neuron increased in GF (GF vs. SPF, P = 0.0444), microbial colonization reversed this trend in CGF (GF vs. CGF, P = 0.0342). In PFC, the proportion of microglia in the GF group relative to the SPF group trended upward (P = 0.062), which could be reversed by microbial colonization (Fig. 1e). We did not find any difference in the composition ratio of the remaining cell types between the three groups. These observations demonstrate that the presence or absence of the gut microbiome primarily impacted the relative composition of glia in the Hip and PFC.

The absence of gut microbiota resulted in cell-specific transcriptomic changes

Next, we performed a DEGs analysis between GF and SPF (Supplementary Fig. 2a, b). We identified 4999 and 6122 DEGs across the major six cell types in the Hip and PFC, respectively, based on the total captured gene in each type (Supplementary Table S6 and Supplementary Fig. 2c). In these two brain regions, glial cells had more DEGs than neurons (Supplementary Fig. 2d, e). The top DEGs in the two regions were mainly involved in mitochondrial dysfunction and the RNA translation process even across cell types (Supplementary Fig. 2f, g).

To further uncover the cell-specific transcriptomic changes modulated by gut microbiota, we identified 846 and 1333 cell-specific DEGs across the six major cell types in the Hip and PFC (Fig. 2a, b), respectively. This observation highlights that the single-cell-level resolution is vital to uncover how the gut microbiome modulates transcriptional changes in the brain. The function enrichment pathways of these cell-specific genes were also significantly different. For example, the altered microglial cell-specific DEGs were enriched for neuroinflammatory and complement system signaling pathways in the Hip (Supplementary Fig. 3a), such as alterations of chemokine receptor-Cx3cr1, interferon gamma receptor-Ifngr1, interleukin receptor-Il6ra, and complement family-C1qa, C1qb, and C1qc, respectively. In contrast, in the PFC, microglial cell-specific DEGs were enriched for RhoGDI and IL-8 signaling pathways (Supplementary Fig. 3b), for example, Map2k1, Nfatc3, and Rhot1. Together, our DEGs analysis showed that the absence of gut microbiota resulted in cell-specific transcriptomic changes.

Fig. 2: Gut microbiota mainly affected microglia in the two brain regions.
figure 2

a, b Venn diagrams describe 846 and 1333 cell-specific DEGs across six major cell types in the Hip (a) and PFC (b). c DEGs (GF vs. SPF) downsampling analysis showed greater microglial gene dysregulations in the Hip (10 times repetition; Microglia vs Astrocyte, ***P = 4.9471E–11, Microglia vs Oligodendrocyte, *P = 0.0395, Microglia vs OPC, ***P = 3.4168E–21, Microglia vs Excitatory neuron, ***P = 1.0866E–19, Microglia vs Interneuron, ***P = 6.7238E–24; P values are from two-tailed Student’s t-test). d DEGs (GF vs. SPF) downsampling analysis showed greater microglial gene dysregulations in PFC, Microglia vs Astrocyte, ***P = 2.5005E–14, Microglia vs Oligodendrocyte, ***P = 1.1953E–17, Microglia vs OPC, *P = 4.019E–20, Microglia vs Excitatory neuron, ***P = 1.1021E–13, Microglia vs Interneuron, ***P = 1.562E–13; 10 times repetition; P values are from two-tailed Student’s t-test).

Microglial transcriptomes were preferentially influenced

Here we explored which cell types were preferentially modulated by the gut microbiome. Disregarding different cellular counts, oligodendrocytes, astrocytes, and microglia mainly contributed to DEGs detected in the Hip (Supplementary Fig. 2d), while the majority of DEGs in PFC were derived from excitatory neurons, interneurons, and microglia (Supplementary Fig. 2e). To rule out the inherent confounding effects of unequal captured cell ratios (neuron: glia ratio = 2.89:1 in PFC and 3.2:1 in Hip), the DEGs burden analysis [41] was carried out by comparing the same number of nuclei across all cell types for ten times by downsampling the data. Accordingly, we found that microglia had the largest number of DEGs in both the Hip and PFC (Fig. 2c, d), suggesting that microglia were preferentially impacted among the six major cell types in the two regions. These findings aligned with the disparate microglial ratios also found in the two brain regions.

We determined whether microglial DEGs were brain-specific. Venn diagram analysis showed that there were 370 genes shared in two regions, while 563 genes only changed in the Hip, and 694 in PFC (Supplementary Fig. 4a). Functionally, we found that the GF mice were enriched for mitochondrial dysfunction, oxidative phosphorylation, inflammasome pathway, and NRF2-mediated oxidative stress response, and depleted for synaptogenesis, synaptic long-term potentiation, and synaptic long-term depression signaling in the Hip. In the PFC, some pathways such as NRF2-mediated oxidative stress response were consistently enriched in GF relative to SPF mice. However, synaptic-related pathways such as the synaptogenesis signaling and long-term synaptic potentiation showed opposite changes in the PFC relative to Hip (Supplementary Fig. 4b). Our results suggest that microglial transcriptional changes caused by the gut microbiome vary in a brain region-specific manner.

The gut microbiome mainly modulated microglia-astrocyte communication

We conducted CellPhoneDB database [44] (cellphonedb V1.10 R package) analysis to uncover potential ligand-receptor pairs between cells to understand how gut microbiome absence influenced communication between microglia and other major cell types. Detailed data are shown in Supplementary Table S2. Microglia communicated mostly with astrocytes, followed by oligodendrocytes, in the Hip of the SPF group. Similar cell-to-cell communications were also found in the GF group, but the lack of a gut microbiome in that group resulted in decreased interaction intensities. Microbial colonization slightly increased the microglia-astrocyte communication (Supplementary Fig. 5a). For example, we found diminished communication between microglial Entpd1 to astrocytic Adora1 in the GF group, which was restored in the CGF group. The CD39 (Entpd1) and Adora1 pair can regulate neuronal activity via its participation in adenosine metabolism [50]. In the SPF group, cellular communication between microglia and other cells was weaker in the PFC than in Hip. Interestingly, lack of a gut microbiome led to significantly increased microglia-astrocyte communication, ranking first in the communication between microglia and other cells. Furthermore, microbial colonization failed to modulate the communication between microglia and other cells (Supplementary Fig. 5a). In summary, the gut microbiome mainly influenced microglia-astrocyte communication in the Hip and PFC of GF and CGF mice.

Microbial colonization rescued microglial gene alterations

Here, we identified 933 and 1064 microglial DEGs by comparing the GF and SPF groups. Interestingly, we found that most of DEGs from these two groups (74.91% and 78.76%) could be rescued by microbial colonization. These rescued genes were associated with chemical synaptic transmission, and cell-cell adhesion in the Hip (Supplementary Table S3 and Supplementary Fig. 5b). For example, we observed six rescued cadherin family genes (e.g., Cdh8, Cdh9, Cdh11 and Cdh12), mainly involving in cell adhesion, and seven rescued gamma-aminobutyric acid (GABA) receptor genes (e.g., Gabarap, Gabarapl2, Gabra2, Gabrb1). In addition, genes enriched in fundamental molecular processes like protein binding and transport, were rescued in the PFC by microbial colonization, such as 6 reversed genes (e.g., Eif1, Eif1b, Eif2s3y, Eif4e, Eif4h, and Eif5a) belonging to eukaryotic initiation factor (Supplementary Fig. 5c). Our findings suggest that microbial colonization effectively reversed microglial transcriptomic changes in the Hip and PFC.

Gut microbiota modulated mutual transformation of microglial subpopulations

Having demonstrated that gut microbiome absence mainly affects microglia, we wanted to further clarify if or how microglial subpopulations changed. Therefore, we performed a microglial re-clustering analysis, which yielded 10 and 6 subpopulations in the Hip and PFC, respectively. In the Hip, two microglial subpopulations (Hip_M1, M4), with a composition ratio of 83.98%, were significantly enriched in GF compared to SPF (0.53%), and microbial colonization could effectively reverse these changes in the CGF group (0.44%; P = 5.2523E–8, one-way ANOVA; Fig. 3a, b). For Hip_M0, a contrasting pattern was observed between the three groups (0.27% in GF, 34.99% in SPF, 45.18% in CGF; P = 0.014, one-way ANOVA; Fig. 3a, b). QuSAGE was used to identify functional gene sets of each subpopulation. This analysis showed that the anti-inflammatory and regulatory T cells (Treg) gene sets were most activated in Hip_M1 and Hip_M4 (Supplementary Fig. 6a–c), such as enrichment of Entpd1, Mif and Tgfb1 (Supplementary Fig. 7a–d), which were inhibited in Hip_M0 (Supplementary Fig. 6, d).

Fig. 3: The gut microbiome modulates the mutual transformation between microglial subpopulations.
figure 3

a UMAP plot depicting microglial subpopulations divided by groups in the Hip. Graph clustering shows that GF cells were located primarily in region #1, and SPF and CGF cells were located in region #2. b Bar chart showing that subcluster 0 (Hip_M0) was enriched in SPF and CGF groups and diminished in GF group, subclusters 1 (Hip_M1) and 4 (Hip_M4) were enriched in the GF group and reversed by microbiome colonization. c UMAP plot depicting Hip_M1 and Hip_M4 located mainly at region #1, Hip_M0 at region #2. Combined with Pseudotime analysis (Supplementary Fig. 8a, c, e), the arrow showed a shifting trend of transforming relationships between Hip_M0, Hip_M1 and Hip_M4. d UMAP plot depicts microglial subpopulations divided by groups in the PFC. Graph clustering shows that GF cells contributed more plots in region #1 than SPF and CGF cells. Furthermore, cells in region #2 primarily belonged to SPF and CGF groups. e The bar chart showed that subcluster 0 (PFC_M0) was enriched in SPF and CGF groups and decreased in the GF group; subcluster 2 (PFC_M2) was enriched in the GF group and reversed by colonization. f UMAP plot depicted PFC_M2 mostly located at region #1, PFC_M0 at region #2. Combined with Pseudotime analysis (Supplementary Fig. 8 b, d, f), the arrow showed a shifting trend of transforming the relationship between PFC_M0 and PFC_M2.

In the PFC, the composition ratio of PFC_M2 (37.36%) was enriched in GF relative to SPF (10.23%), and could be effectively rescued in the CGF group (11.53%; P = 0.011 between the three groups, one-way ANOVA; Fig. 3d, e). Meanwhile, the composition ratio of PFC_M0 was significantly reduced in GF (13.10%) compared to SPF (46.85%) and CGF (51.19%), and enriched in SPF and CGF (46.85% in SPF, 51.19% in CGF; P = 0.015, one-way ANOVA; Fig. 3d, e). QuSAGE analysis showed that the anti-inflammatory and Treg gene sets, such as Entpd1, Mif, Vegfa and Tgfb1 (Supplementary Fig. 7e, f), were most activated in PFC_M2 (Supplementary Fig. 6e, g), and inhibited in PFC_M0 (Supplementary Fig. 6e, f). Next, Pseudotime analysis was conducted to further explore these mutual transformational relationships. Hip_M1 and Hip_M4 were more located at the start of developmental trajectories; however, Hip_M0 was more located on the middle and end (Supplementary Fig. 8a, b). In PFC, PFC_M0 and PFC_M2 showed the same trend (Supplementary Fig. 8c, d). It suggested that gut microbiota modulate transformational control between different microglial subpopulations, displaying the shift from Hip_M1&4 to Hip_M0, and PFC_M2 to PFC_M0 (Fig. 3c, f). Together, these findings demonstrated that the gut microbiome modulated the mutual transformation of microglial subpopulations in the two regions.

To ensure these findings, we further carried out an independent snRNA-seq analysis of Hip among GF, SPF and CGF (n = 3/group), named batch 2 (B2). Generally, two batches of single-cell transcriptome data were highly consistent in the number of capturing cells (Supplementary Fig. 9a) and identification of the cell types (Supplementary Fig. 9b). In the analysis of microglia subtypes, we found that the three hippocampal microglial subpopulations observed in batch 2 were also highly similar to B1_Hip_M0, M1 and M4 (Supplementary Fig. 9c; Fisher’s exact test; Supplementary Table S7). Odd ratio of B2_Hip_M2 and M3 versus Hip_M0 were 133.6 (FDR = 3.00E–04) and 91.4 (FDR = 2.34E–05), B2_Hip_M0 versus Hip_M1 and M4 were 99.2 (FDR = 3.14E–89) and 27.7 (FDR = 3.90E–29). Furthermore, proportion of these subpopulations showed mutual transformation among groups (Supplementary Fig. 9d, e), B2_Hip_M2 and M3 increased in SPF and CGF, decreased in GF like B1_Hip_M0, moreover, B2_Hip_M0 showed opposite trend like B1_Hip_M1 and M4.

The microglial genes rescued by microbial colonization are linked with AD and MDD

To explore potential associations between the rescued genes and representative neuropsychiatric disorders, the DisGeNET database [51] was used for Disease Enrichment analysis. In both the Hip and PFC, the microglial genes rescued by microbial colonization were linked to neuropsychiatric diseases such as AD (n = 173 in Hip and 174 in PFC), MDD (n = 98 and 61), and autism (n = 116 and 52; Fig. 4a, b).

Fig. 4: Microglial genes regulated by the gut microbiota are linked with AD and MDD.
figure 4

a, b In Hip (a) and PFC (b), reversed microglial genes were closed to neuropsychiatric diseases, such as AD and MDD. c 19 genes overlapped between plague-induced genes (PIGs), disease-associated microglia (DAM), and reversed genes, including Apoe, Cx3cr1, Trem2, Fcer1g, C1qa, Fcrls, C1qb, Itm2b, C1qc, Man2b1, Cd9, Olfml3, Cst3, Ctsb, Ctsl, Gusb, Ctss, Hexa, and Ctsz. d Ten genes overlapped between microglial DEGs in depression like macaque (unpublished data), MDD risk gene in the DisGeNet database and reversed genes in microglia (converted to HUGO Gene Symbol), including FKBP5, AUTS2, ERBB4, NEGR1, NRG3, RABGAP1L, SLC1A3, ANK3, CTTNBP2, and ITGB5.

Single-cell studies of AD and MDD were selected to further confirm these findings. AD had the highest numbers of rescued microglial genes in the Hip and PFC, and we had a long-term interest in MDD. We found that, although the cell types associated with distinctive diseases were different, a large number of disease risk genes aligned with the microglial genes rescued by microbial colonization (Fig. 4a). In particular, for AD, 19 microglial genes overlapped between PIGs [52] (plague-induced genes), DAM [22], and reversed genes, including Apoe, Fcer1g, C1qa, Frcls, C1qb, Itm2b, C1qc, Man2b1, Cd9, Olfml3, Cst3, Trem2, Ctsl, Ctsb, Ctss, Gusb, Ctsz, Hexa, and Cx3cr1 (Fig. 4d). As for MDD, we used our single-cell analysis of dorsolateral PFC from a non-human primate depression model (unpublished data, not shown), and matched them against MDD risk genes in the DisGeNet database. We found 10 overlapped genes, including FKBP5, AUTS2, ERBB4, NEGR1, NRG3, RABGAP1L, SLC1A3, ANK3, CTTNBP2, and ITGB5 (Fig. 4e). These findings demonstrated that microglial genes reversed by microbial colonization were mainly linked with AD and MDD, suggesting that microbial modulation of these key microglial genes via the gut-brain axis may be a potential therapeutic strategy for AD and MDD.

Cross-species analysis showed that microglia subpopulations regulated by gut microbiota were associated with AD and MDD

Next, we further verified whether the transcriptomic changes of these 5 microglial subpopulations were linked with AD and MDD. We performed cross-species analysis of the association between these 5 microglial subpopulations and these two disorders by using animal and human sc/snRNA-seq data from five publications [22,23,24, 53], and snRNA-seq analysis from our non-human primate depression model (unpublished data, not shown). The marker genes of 5 microglia subpopulations were compared with the microglial marker genes or enriched DEGs associated with these diseases. We found that transcriptomic changes of these microglial subpopulations were highly associated with AD and MDD across human, mouse, and macaque (Fig. 5a). Furthermore, the noted DAM was highly similar to Hip_M1 (Fisher’s exact test, FDR = 3.52E–30, odd ratio = 7.379; Supplementary Table S4) and Hip_M4 (Fisher’s exact test, FDR = 3.32E–08, odd ratio = 3.896). In addition, we found that only PFC_M2 was significant relative to MDD-associated microglia in Macaca (Fisher’s exact test, FDR = 0.001, odd ratio = 2.605). This cross-species analysis provided evidence that microglia subpopulations’ transcriptomic changes modulated by gut microbiota were highly linked with AD and MDD.

Fig. 5: Cross-species analysis described that microglia subpopulations regulated by the gut microbiota were associated with AD and MDD.
figure 5

a Circos graph shows microglial subpopulations highly associated with AD and MDD cross humans, mice and macaca (proportion of segment means –logFDR, Fisher’s exact tests, FDR-BH corrected). b The OFT showed no differences in locomotion activity in the three groups (SPF, n = 17; GF, n = 16; CGF, n = 16; data are mean ± SEM; NA means P = 0.109; P values are from ANOVA test). c GF mice displayed decreased immobility time in FST and restored partially in CGF mice (n = 6 in GF and CGF, n = 7 in SPF; data are mean ± SEM; GF vs. SPF, ***P = 3.00E–06; GF vs. CGF, **P = 0.0011, SPF vs. CGF, ***P = 6.10E–05; P values are from ANOVA post hoc analysis-Tamhane T2 test). d The decrease of spontaneous alternation rate of GF mice in the Y-maze test compared to SPF mice was restored in CGF mice (n = 10/group; data are mean ± SEM; GF vs. SPF, *P = 0.022; GF vs. CGF, *P = 0.012; P values are from ANOVA post hoc analysis-LSD test).

Behavioral tests support the association between gut microbiota, AD, and MDD

We used an animal behavioral test panel related to AD and MDD including OFT, Y-maze and FST to confirm the above findings. There was no difference in locomotion activity between the three groups (P = 0.109; Fig. 5b). However, the percent immobility time was significantly decreased in the FST of GF compared to SPF mice (P = 6.9879E–11), suggesting impacts on behavior despair. Microbial colonization increased the percent immobility time in CGF mice, although it was not completely restored to the same level as in SPF mice (Fig. 5c). In the Y-maze test, the spontaneous alternation rate was significantly decreased in GF relative to SPF (P = 0.023), and this change could be completely reversed by microbial colonization (CGF) (P = 0.013; Fig. 5d). This behavioral test suggested a close association between the gut microbiome and short-term memory changes. Together, our behavioral studies support the single-cell observations that transcriptomic changes of microglial subpopulations were highly linked with AD and MDD.

Read more here: Source link