Bacteria Invade the Brain After Intracortical Microelectrode Implantation
Microbiome composition can be measured directly by sequencing the 16S rRNA gene from bacterial DNA isolated from feces or other tissue28. Therefore, our first step was to identify and compare the composition of microbes present in both fecal matter and the brain tissue of naïve unimplanted mice and microelectrode-implanted mice.
Here, the V3-V4 region of the gene for the 16S rRNA small subunit was sequenced using bacterial DNA extracted from biopsy punches of unimplanted non-treated control brains 2 weeks after housing separation. Similar sequences were assigned to operational taxonomic units (OTUs, representing species-level observations) and their read counts were used to compare microbial composition between unimplanted brains and implanted control brains both 4- and 12-weeks post-implantation (Fig. 1). Variation in microbial communities manifests primarily via differences in prevalence (presence/absence of an OTU in a sample) and/or differences in abundance (proportion of sample reads derived from an OTU). Significant differences in the prevalence and the abundance of microbes were observed in the brain following microelectrode implantation. Observations at the genus level will also be discussed as changes to the genera.
Across all samples within a group, brain tissue from the unimplanted control group contained 25 total genera (four unique), whereas the 4-week control group contained 112 total genera (72 unique, 93 invading), and the 12-week control group contained 36 total genera (zero unique, 21 invading) (Fig. 1A). Nineteen of the 25 genera found in the unimplanted brain were found in the mouse brain tissue 4 weeks after microelectrode implantation, while an additional 93 gut-derived genera were found in the brains of implanted mice 4 weeks after microelectrode implantation. By 12-weeks post-implantation, 72 of the 93 invading genera were no longer found in implanted brain tissue, while 21 of the invading genera could still be detected in the brain. Interestingly, 13 genera found in the unimplanted brains were also found at both time points post-implantation, while only two of the original genera found in the unimplanted brains were able to repopulate the brain after becoming absent for some duration post-implantation (Fig. 1A).
Antibiotic Treatment Facilitates Invasive Microbe Diversity
An additional cohort of mice was treated with antibiotics to deplete fecal (gut) microbiota. Mice with differential expression of microbes could then be used to examine the correlation between the composition of the microbes invading the brain and microelectrode recording performance. Antibiotic-treated mice were provided with an antibiotic cocktail of Ampicillin, Clindamycin, and Streptomycin in their drinking water following established protocols29. Antibiotic-treated mice displayed significant alterations to the gut microbiome as early as one week after the start of treatment, which continued throughout the study (Supplemental Fig. S1A-C). Antibiotic treatment had no discernible effect on microbial composition in the unimplanted brain (Supplemental Fig. S1D).
Firmicutes was the most abundant phylum found in the brain at 4-weeks post-implantation, with Bacteroidota dominating the unimplanted and 12-weeks post-implantation brains (Fig. 1B). Linear discriminant analysis Effect Size (LEfSe) is a method used in biomarker discovery to identify taxa most likely to be associated with differences between experimental groups30. Higher values of the log-linear discriminant analysis score indicate greater enrichment of taxa within a particular group (Supplemental Fig. S2). The unimplanted brain was characterized by an abundance of microbes from the phyla Bacteroidota (genus Muribaculaceae) and Firmicutes (genus Lactobacillus). The 4-week post-implantation brain was characterized by an abundance of microbes from the phylum Firmicutes, from the class Clostridia. The 12-week post-implantation brain was characterized by an abundance of microbes from the phyla Bacteroidota (genus Bacteroides) and Proteobacteria.
The Shannon Diversity Index, a measure of alpha (within a sample) diversity, provides a quantitative assessment of the species richness and evenness of a bacterial community sample; it is robust to sample composition, with a higher value indicating greater sample diversity31. The Shannon Diversity Index varied significantly by implantation status but not by treatment group, with the highest values observed in the 4-week post-implantation group (Fig. 1B). The number of observed OTUs varied significantly within treatment group with the largest number of OTUs observed at 4 weeks, in which 60 ± 22 OTUs were observed in rarefied control brains and 45 ± 24 OTUs in rarefied antibiotic brains (Fig. 1B). At 12 weeks post-implantation, control brains contained 17 ± 3 OTUs and antibiotic contained 19 ± 3 OTUs, compared to 18 ± 3 OTUs in unimplanted control brains and 15 ± 1 OTUs in unimplanted antibiotic brains (Fig. 1B).
OTUs were then classified as “native” (detected in the unimplanted brain) or invading (not detected in the unimplanted brain) to investigate differences in abundance and taxonomy. At 4-weeks post-implantation, invading microbes composed 62.8% ± 27.7% of the relative abundance in the control group (n = 6) and 54.1% ± 37.1% of the relative abundance in the antibiotic group (n = 5) (Fig. 1C). At 12-weeks post-implantation, invading microbes made up 8.9% ± 2.8% of the relative abundance in controls (n = 7) and 12.1% ± 5.4% of the relative abundance in the antibiotics group (n = 6) (Fig. 1C). The populations varied significantly based on implantation status. The most abundant invading microbes were from the phylum Bacteroidota in the unimplanted group, Firmicutes at 4 weeks, and Proteobacteria at 12 weeks. The invading Firmicutes bacteria were largely gone by 12 weeks.
Given the complexity of comparing multiple groups with feature-rich microbial data, we applied the dimension reduction technique of non-metric multidimensional scaling (NMDS) to visualize the relationship between samples across the three implantation statuses (unimplanted, 4-weeks, and 12-weeks post-implantation) and two treatment groups (control vs. antibiotic-treated) in 2-dimensional space (Fig. 1D). The unweighted UniFrac distance measures Beta (between-groups) diversity by considering the phylogenetic information of observed microbes32. We used the UniFrac distance to quantify the degree of difference between samples by calculating the fraction of branch length in the de novo-assembled phylogenetic tree that is unique to either of the two samples being compared. Samples that are more like each other have fewer unique evolutionary relationships and appear closer together in the 2-dimensional ordination space, as will experimental groups. Here, we found three significantly distinct clusters, largely segregated by implantation status, suggesting that the brains of unimplanted, 4-week post-implanted, and 12-week post-implanted animals have distinct microbial populations (Fig. 1D). Two antibiotic-treated and one control brain from the 4-week post-implantation are like that of the 12-week group, suggesting that there is variation in the rate at which the brain-bacterial environment may stabilize. Of note, none of the implanted brains we examined returned to a bacterial composition resembling the unimplanted brain tissue, regardless of treatment or time point.
Correlation Between Gut and Brain Isolated Microbes
Given that daily administration of antibiotics via drinking water altered the composition of microbes in the fecal matter (Supplemental Fig. S1A-C), we next examined the composition of microbes in the brain tissue for microelectrode implanted mice. Analysis of Compositions of Microbiomes with Bias Correction (ANCOM-BC) was used to model microbial communities via a linear regression framework and test for differential abundance by implantation status (unimplanted, 4-weeks and 12-weeks post-implantation) and treatment group (antibiotic, control). Notably, there were 16 genera found at lower abundance in the antibiotic-treated brain at 4-weeks post-implantation, 15 of which were not detected in the unimplanted brain (Fig. 2A). In addition, 13/16 genera were from the class Clostridia of the phylum Firmicutes, found in the tissue adjacent to microelectrodes in this study. At 12-weeks post-implantation, we found there to be no differentially abundant genera between antibiotic-treated and control brains (Fig. 2A).
Fecal and brain bacteria were compared to understand the origin of invading bacteria. There were 198 OTUs observed in the implanted brain, of which only 32 were detected in the unimplanted brain. Of the 166 OTUs invading the brain, 49 were observed in the gut samples, lending credibility to the hypothesis that gut-derived bacteria invade the brain after IME implantation. However, these 49 gut-derived bacteria represent only a portion of the invading microbes, suggesting that the remaining majority come from a different source in the body’s microbiome outside of the gut and brain (Fig. 2B). At 4 weeks post-implantation in the control group, 13.7% ± 4.9% of total reads (27.7% ± 16.0% of invading reads) were gut-derived and 49.0% ± 22.8% of total reads (72.3% ± 16.0% of invading reads) were of unknown origin. In the 4-week antibiotic-treated group, 24.3% ± 26.7% of reads (48.5% ± 25.4% of invading reads) were gut-derived and 29.8% ± 27.1% (51.5% ± 25.4% of invading reads) were of unknown origin. At 12 weeks post-implantation in the control group, 3.7% ± 1.9% of reads (42.1% ± 22.6% of invading reads) were gut-derived and 9.0% ± 2.8% (57.9% ± 22.6% of invading reads) were of unknown origin, while for the antibiotic group, 5.6% ± 3.9% of reads (44.1% ± 19.7% of invading reads) were gut-derived and 12.1% ± 5.4% of reads (55.8% ± 19.7% of invading reads) were of unknown origin.
Antibiotic Treatment Impacts Intracortical Microelectrode Performance
Functional, single-shank, silicon 16-channel intracortical microelectrodes were implanted into the primary motor cortex to obtain awake neural recordings. Animals were separated into two cohorts consisting of untreated control and antibiotic-treated groups. Biweekly recordings and analysis indicate that arrays implanted in antibiotic-treated animals performed significantly better than the control group based on measurement of the proportion of active electrodes, or active electrode yield (AEY), at week 0 (day of implantation), week 1, week 4, and week 5 (Fig. 2A). The largest difference in AEY was observed in week 4 (79% for antibiotic-treated animals vs. 62% for the control group).
Antibiotic-treated and control animals declined significantly in performance over time, consistent with historical data (Fig. 2B)4,34,35. When grouped into known phases for the maturation of the neuroinflammatory response36,37, antibiotic-treated mice performed significantly better in the acute (weeks 0–5) phase of implantation (80% AEY for antibiotics vs. 67% AEY for control), exhibited no difference during the sub-chronic (weeks 6–11) phase (52% for antibiotics vs 54% for control), and displayed a significant decline in performance at the chronic (week 12) time period (42% for antibiotic vs 56% for control, Fig. 2B).
During the sub-chronic implant period, the peak-to-peak voltage (Vpp) (97.9 µV ± 43.7 µV for antibiotic vs. 81.1 µV ± 34.1 µV for control, Fig. 2C), noise levels (12.4 ± 2.7 µV µV for antibiotic vs. 11.3 µV ± 2.7 µV for control, Fig. 2D), and spike rate (12.0 ± 17.8 for antibiotic vs. 6.8 ± 6.9 for control, Fig. 2E), were all significantly higher in the antibiotic group compared to the control. SNR showed no significant change across groups or time points (Fig. 2F).
Antibiotic Treatment Impacts the Neuroinflammatory Response to Intracortical Microelectrodes
Neuroinflammation has long been associated with intracortical microelectrode failure3,16,38. Here, we utilized one of the most advanced methods reported to date to assess the intracortical microelectrode-tissue interface, both spatial proteomics (with and without cell specificity) 39, and spatially-resolved whole mouse transcriptomics40,41. Our goal was to begin to understand the potential relationship between invasive microbes in the brain, neuroinflammation, and microelectrode recording performance.
Spatial and cell-specific neural proteomic evaluation of the implant site (up to 270 µm from the implant) provides a robust view of the health of the brain tissue and the effect of treatment on inflammation. Comparisons between antibiotic and control were made at 4- and 12-weeks post-implantation, along with temporal comparisons within the antibiotic and control groups (4-week antibiotic: n = 4, 4-week control: n = 3, 12-week antibiotic: n = 3, 12-week control: n = 3). All comparisons between antibiotic and control were made with control as the baseline. A negative fold change indicates lower expression in antibiotic-treated mice compared to the control group (“downregulation”), while a positive indicates higher expression in antibiotic-treated mice compared to the control group (“upregulation”). Across all comparisons, 28 of the 39 possible proteins were differentially expressed in at least one comparison. Table 1 shows the full list of 39 proteins examined and the 6 proteins used for quality control and normalization. The complete area of interest (AOI) represents the tissue within 270 µm of the implant site. Within the AOI, the inner AOI is the tissue adjacent to the implant site to 90 µm from the implant site; the middle AOI is the tissue 90 µm to 180 µm from the implant site; and the outer AOI is the tissue 180 µm to 270 µm from the implant site. Neuron (NeuN-positive) and astrocyte (GFAP-positive) cell-specific regions were analyzed for each AOI. All twelve potential combinations of the AOIs used for comparison in this study can be visualized in Fig. 7A and are summarized in Table 2 (See Methods for an in-detail explanation).
The proteomic analysis of antibiotic-treated mice compared to control at 4-weeks post-implantation indicated that in all cases of differential protein expression, the proteins were decreased in expression in tissue from the antibiotic-treated mice compared to the untreated control group (Fig. 3, Table 2). Seven of the twelve AOI comparisons indicated differential protein expression (Fig. 3). Without cell-specific segmentation of the AOI, the complete AOI (0-270 µm), inner, and outer AOI all showed differential protein expression. Specifically, protein expression of our panel showed the downregulation of 18 proteins for the full AOI (ATG12, ATG5, BAG3, CD163, CD31, CD40, CD68, CSF1R, MAP2, NeuN, NfL, OLIG2, P62, PLA2G6, SYP, TMEM119, ULK1, and VIM), six in the inner region (ATG12, CD68, MAP2, SYP, TMEM119, ULK1), none in the middle, and 19 proteins in the outer region (ATG12, CD31, CD40, CD45, CD68, CSF1R, CTSD, GPNMB, ITGAX, MAP2, NeuN, NfL, P62, PLA2G6, SYP, TMEM119, ULK1, VIM, and VPS35) (Fig. 3A-D, Table 2).
For neuron-specific comparisons, there were 15 total proteins downregulated for the full AOI (ATG12, BAG3, CD31, CD68, CSF1R, GPNMB, MAP2, NfL, OLIG2, P62, PLA2G6, SYP, TMEM119, ULK1, and VIM), none in the inner region, 18 downregulated in the middle region (ATG12, BAG3, CD11b, CD31, CD39, CD45, CD68, GPNMB, Ki-67, MAP2, OLIG2, P62, PLA2G6, SYP, TFEB, ULK1, VIM, and VPS35), and eight downregulated in the outer region (BAG3, CD31, CD68, CSF1R, GPNMB, MAP2, SYP, and ULK1) (Fig. 3E-H, Table 2).
In astrocyte-specific comparisons, no proteins were differentially expressed in the full AOI, middle, or outer regions (Fig. 3I-L). However, seven proteins were downregulated in the inner region of the astrocyte-specific AOI (ATG12, CD40, CD68, MAP2, MerTK, TFEB, and ULK1) (Fig. 3I-L, Table 2). Table 2 summarizes comparisons including all twelve AOIs at 4-weeks post-implantation.
The proteomic analysis of brain tissue from antibiotic-treated mice compared to control at 12-weeks post-implantation indicated only one differentially expressed protein, CD163, which is a marker that indicates the transition from pro-inflammatory M1 to M2 anti-inflammatory macrophage phenotype (Supplemental Fig. S3) 42. CD163 was indicated to be upregulated in the antibiotic-treated group, compared to the untreated control group in the astrocyte-specific collection, in the area between 180–270 µm from the microelectrode-tissue interface. This upregulation to the M2 phenotype may promote preservation of viable neural tissue near the implant site 43.
Within treatment groups, examination of temporal changes in protein expression from 4-weeks to 12-weeks post-implantation showed that five of the twelve AOI comparisons of antibiotic-treated mice indicated differential protein expression (Supplemental Fig. S4, Supplemental Table S1). Two of the more noteworthy comparisons were identified within examinations of the inner ring. There were eight differentially expressed proteins in the astrocyte-specific inner ring, and ten differentially expressed proteins in the non-specific inner ring, suggesting the largest differential expression between the temporal comparison of the antibiotic-treated animals to be in the tissue closest to the microelectrode implant site. Temporal comparison of the untreated control mice also demonstrated that five of the twelve comparisons indicated differential protein expression (Supplemental Fig. S5, Supplemental Table S2). However, with the untreated control mice, the two groups with the largest differential expression only indicated three or four differentially expressed proteins each, with the remaining comparisons only showing one differentially expressed protein each. The spatial organization of differential protein expression was evenly distributed between AOIs, one total, two inner, one middle, and one outer AOI each demonstrated differential protein expression.
Spatial transcriptomic evaluation of the implant site was performed to understand how many genes were differentially expressed and involved in which pathways and molecular processes. Very few studies have been performed with spatial transcriptomic analysis of the intracortical microelectrode-tissue interface to date40,41. However, the NanoString GeoMx system used here is uniquely capable of collecting the entire tissue of interest, rather than orientated spherical regions forming a grid within the tissue being analyzed. Here, the whole mouse transcriptome was first filtered using quality control steps in the NanoString GeoMx software (see Methods for details on filtering) leaving a total of 8259 genes. Of the 8259 genes included in our analysis, 490 were differentially expressed at 4-weeks post-implantation, and 1375 genes were differentially expressed at 12-weeks post-implantation (Fig. 4A-B). Out of all differentially expressed genes, only 52 are shared between the 4- and 12-week time points, indicating consistent temporal changes.
To further our understanding of the response to altering the composition of the brain-invasive gut microbiome and the implications on neuroinflammation and brain health, we completed a pathway analysis using the iPathways software. The differential gene expression detected in our study implicated dozens of biological pathways and functions related to neural health. Here, for brevity and focus, we only discuss pathways in which a high proportion of genes associated with the pathway were differentially expressed at either 4-weeks or 12-weeks post-implantation, or pathways in which many differentially expressed genes switched between up- and down-regulated between the 4- and 12-week timepoints.
At 4-weeks post-implantation, there were 20 differentially upregulated genes of 122 genes associated with ribosomal protein function in antibiotic-treated animals compared to control (Fig. 4C, Supplemental Fig. 6A). At 12-weeks post-implantation there were 19 differentially expressed genes in the same pathway. However, at 12-weeks post-implantation only six genes were upregulated and 13 were downregulated in the antibiotic-treated animals compared to the control (Fig. 4D, Supplemental Fig. 6B).
Neurodegeneration is an important pathway to consider as it relates to long-term neural health. Healthy, firing neurons can only be detected within ~ 150 µm from the intracortical microelectrode site44. Consequently, evidence of neurodegenerative pathways near the implant site detected by spatial transcriptomic analysis is of prominent interest. At 4-weeks post-implantation, there were 17 differentially expressed genes associated with the neurodegenerative pathway (Fig. 4C), with 49 differentially expressed genes at 12-weeks post-implantation. At 4-weeks post-implantation, four genes are differentially expressed in ubiquitin-proteasome system (UPS) disruption (three upregulated, one downregulated, Supplemental Fig. 7A), and five genes are upregulated in the mitochondrial dysfunction pathway (Supplemental Fig. 7B) (Fig. 4C). At 12-weeks post-implantation, 10 genes associated with UPS disruption were differentially expressed (eight upregulated, one downregulated, Supplemental Fig. 8A), 23 genes associated with mitochondrial dysfunction and mitophagy were differentially expressed (15 downregulated, eight upregulated, Supplemental Fig. 8B), and six associated with tau protein accumulation were differentially expressed (five upregulated, one downregulated, Supplemental Fig. 8C) (Fig. 4D). In addition to pathway analysis, gene ontology (GO) offers added insight into the effect of microbiome alteration via antibiotic treatment on cellular and biological functions in the brain. A discussion of additional pathway analysis can be found in the Supplemental Information.