Single-nuclei RNA sequencing and identification of major brain cell types
To sequence a relatively higher proportion of the cell types comprising the BBB, fluorescence activated nuclear sorting was combined with snRNAseq [45], in post-mortem midbrain sections of 14 schizophrenia and 15 control cases (Figure 1 A.I-A.II). Unbiased cluster analysis of nuclear transcriptomic profiles from 178 009 single nuclei revealed 19 clusters (S. Figure 1 A). These were annotated into 11 different main brain cell types based on the expression of cell type-specific marker genes (S. Figure 1 B). The majority of the sequenced nuclei derived from astrocytes (36.3% ± 14.6, S. Table 10). A cluster containing both pericytes and SMCs nuclei was identified based on the co-expression of PDGFRB and ACTA2 (S. Figure 1 B). Fibroblasts and endothelial nuclei highly expressed LAMA2 and PECAM1, respectively. Ependymal nuclei were identified because of their high expression of Doublecortin Domain-Containing Protein 1 (DCDC1) (S. Figure 1 B) in addition to their enrichment in genes expressed by mouse ependymal cells (S. Figure 1 C) [49].
Transcriptional changes associated with schizophrenia are limited and specific to the ependymal cells and pericytes
To evaluate potential contributions of the cells of the BBB to schizophrenia brain pathology, we first extracted pericytes-SMCs, fibroblasts, astrocytes, ependymal and endothelial nuclei (57 522 nuclei in total) from the complete dataset and determined variable genes within this selection in order to identify different cellular populations (Figure 1 B.I and B.II). Unsupervised clustering of the BBB nuclei resulted in 11 clusters that were annotated into 7 main BBB cell types (Figure 1 B.III). The relative proportions of the different BBB cell types are indicated in S. Table 11. Pericytes (2.7% ± 1.7) and SMCs (1.1% ± 0.8) nuclei segregated into two different clusters; pericytes highly expressed DLC1 and PDGFRB whereas SMCs highly expressed MYH11, ACTA2 and TAGLN (Figure 1 B.III, S. Table 3). Ependymal nuclei (3% ± 2), probably derived from the cerebral aqueduct, highly expressed the cilia-related gene DNAH9. Fibroblasts (6.9 ± 5.9) abundantly expressed ABCA9 and endothelial nuclei (4.4 ± 3.6) were identified based on the expression of endothelial marker genes, such as FLT1 [51]. We identified a small cluster of mesenchymal nuclei (MSCs; 0.79 ± 2.48) expressing SLIT2 and PHLDB2 (Figure 1 B.III, S. Table 3). Astrocytes comprised 81.1 % (± 13.5) of the BBB nuclei and segregated into five different clusters (Figure 1 B.II-BIII). No prominent donor effects were observed, as all samples contributed to nearly all clusters (Figure 1 C).
To test for differences in the relative proportions of the major BBB cell types between schizophrenia and controls, we used a generalized linear model, and no significant differences were observed (Figure 1 D). Next, we compared the transcriptomic profiles of the BBB cells in schizophrenia with respect to controls. Differences in the transcriptome between schizophrenia and controls were evaluated in each of the cell types, independently. With an absolute log2FC > 0.3 and adjusted p value < 0.05, we identified 14 differentially expressed genes (DEGs; S. Figure 2 A-D). The largest difference in gene expression (highest log2FC, S. Table 6) was observed for NRXN1, in the MSCs cluster, with a reduced expression in schizophrenia as compared to control MSCs (S. Figure 2 C). The ependymal cluster depicted the largest number of DEGs between schizophrenia and controls, including reduced PDE4D expression, and increased FOXP2 and EML6 expression in schizophrenia (S. Figure 2 A). Pericytes were the second cluster with the highest number of DEGs between schizophrenia and controls (S. Figure 2 B), with increased LRBA and reduced DOCK9 expression in schizophrenia pericytes. The expression of NRXN1, PDE4D, LRBA and FOXP2 in the samples did not significantly correlate with any of the case-related variables (S. Table 7), suggesting that changes in the expression of these genes are mainly due to the effect of diagnosis.
Taken together, our results suggest that the relative proportions of the different main classes of BBB cell types is unaltered in the midbrain of the patients with schizophrenia and that the few schizophrenia-associated changes in gene expression are in the ependymal, pericytes and MSCs nuclei.
Sub-clustering analysis of the major BBB cell types
To investigate if sub-populations of BBB cell types were altered in relation to schizophrenia, we ran sub-clustering analysis across all the BBB cell types containing more than 1 000 nuclei and compared their proportions between schizophrenia and controls. Sub-clustering analysis of pericytes, fibroblasts and ependymal nuclei did not reveal a schizophrenia-associated subpopulation (data not shown).
Sub-clustering analysis revealed four different endothelial sub-populations
Midbrain endothelial sub-populations (Figure 2 A; 2 244 endothelial nuclei) depicted enriched expression of genes highly expressed by different human lung endothelial sub-populations [52] (Figure 2 B), indicating the presence of nuclei corresponding to arteries, capillaries, and two sub-types of veins in our dataset. Capillary nuclei comprised the largest midbrain endothelial population (56.7% ± 12.1; S. Table 11) and abundantly expressed CLDN5 (Figure 2 C, S. Table 4). Arterial nuclei represented 20% ± 8.2 of the midbrain endothelial cells (S. Table 11) and highly expressed previously reported arterial markers such as VEGFC, EFNB2 and FBLN5 [53, 54], and novel markers such as, IGFBP3, DKK2 and ROR1 (Figure 2 C, S. Table 4). Venular nuclei segregated into two sub-populations, both highly expressing the venular marker gene TJP1. Remarkably, Endo_veins1 (16.2% ± 7.3, S. Table 11) abundantly expressed the glutamate transporter SLC1A, implying a possible role in the regulation of brain glutamate levels [55]. Endo_veins2 represented a smaller proportion of endothelial nuclei (7% ± 6.3, S. Table 11) and abundantly expressed SERPINE, VCAM1 and ICAM1 (Figure 2 C, S. Table 4), suggesting their involvement in the regulation of brain vascular capture and permeability to circulating immune cells [28, 56]. Accordingly, “Cell migration”, “Response to stimulus” and “Signal transduction” were the top enriched biological processes in Endo_veins2 (Figure 2 D, S. Table 8).
To test for the presence of schizophrenia-related endothelial sub-populations, we compared the proportions of the endothelial sub-clusters between schizophrenia and controls, but no differences were observed (Figure 2 E). Recently, some schizophrenia cohorts have been stratified based on the expression of inflammatory markers, such as IL-18, IL-8, IL-6, IL-1β, SERPINA3 and TNFα, in blood and CNS [50, 57-59], and a greater proportion of individuals with schizophrenia (~40%) are classified as having a “high inflammation” status compared to controls (~10%) [3, 60]. As may be expected, in schizophrenia patients with a pro-inflammatory signature, possible alterations in the brain vasculature and BBB permeability are consistently reported, reflected by altered expression of structural and functional brain endothelial cells marker genes [57, 58] (discussed in [8]). In our dataset, 4 out of the 14 schizophrenia cases were previously identified as high-inflammation cases, based on the expression of inflammatory cytokines in their cortical grey matter [50]. Considering that brain endothelial cells may be affected by chronic inflammation, which could be reflected by changes in the abundance of particular endothelial sub-types associated with high inflammation in schizophrenia, we compared the proportion of the endothelial sub-populations in schizophrenia-high inflammation with respect to controls, but no differences were observed (data not shown).
In summary, these data provide a characterization of different endothelial sub-populations in the human midbrain, and none of them were differentially represented in the schizophrenia samples.
Sub-clustering analysis revealed six astrocyte sub-populations
Variation in the sequence and expression activity of genes highly expressed by astrocytes has been associated with schizophrenia [61, 62]. We performed sub-clustering analysis of the astrocyte nuclei to describe the different sub-populations of astrocytes in the human midbrain and to determine whether there is a contribution of a particular sub-population of astrocytes to schizophrenia pathophysiology.
With low clustering resolution, the astrocyte nuclei segregated into six sub-clusters (Figure 3 A). Based on the expression of previously reported maker genes of astrocyte sub-types, such as GFAP for fibrous and SLC1A2 for protoplasmic astrocytes [17, 63] (Figure 3 B, S. Table 5), and in gene ontology enrichment analysis (Figure 3 C, S. Table 9), we annotated the six sub-clusters into two protoplasmic (52.11 %, S. Table 11), two fibrous (42.86 %, S. Table 11) and two astrocyte sub-populations associated with immune functions (4.93 %, S. Table 11). In Ast_protoplasmic2 compared to Ast_protoplasmic1, the genes SHISA9, GRIA4, NTN1 and GREB1L were more abundantly expressed (Figure 3 B). Ast_fibrous2 and Ast_fibrous1 nuclei were transcriptionally similar and shared several marker genes, like GFAP, ADAMTSL3, SLC38A, CPAMD8, among others (Figure 3 B, S. Table 5); however, these genes were more abundantly expressed in Ast_fibrous2 (Figure 3 B). The nuclei of the two immune-related astrocyte sub-clusters depicted increased expression of the complement component 3 (C3) (Figure 3 B, S. Table 5). Ast_immune1 represented a small percentage of the astrocyte nuclei (2.5% ± 2.2, S. Table 11) and highly expressed genes coding for interferon-inducible proteins, such as GBP2 [64], IFIT3 and STAT1 (Figure 3 B, S. Table 5). We evaluated the expression of two lists of genes related to an astrocytic reactive phenotype, across the different astrocyte sub-types (S. Figure 3). The first set corresponded to genes highly expressed by a sub-type of astrocytes identified in multiple sclerosis active lesions [65] (S. Figure 3 A.I-A.II) and the second set contained commonly up-regulated genes in human astrocytes subjected to different stressful stimuli [66] (S. Figure 3 B), resembling the transcriptome of a “general” astrocyte reactive phenotype. Ast_immune1 nuclei were enriched in both gene sets, as compared to the other midbrain astrocyte clusters, indicating that Ast_immune1 may correspond to reactive astrocytes that are present in low abundance in the human midbrain, normally and in schizophrenia. The second identified immune related astrocyte sub-population, Ast_immune2, also represented a small proportion of the astrocytes (2.5% ± 1.4, S. Table 11) and highly expressed the genes coding for the Dedicator of Cytokinesis Proteins 2 and 8, DOCK2 and DOCK8 (Figure 3 B, S. Table 5). These proteins are guanine nucleotide exchange factors that activate Rho-family small GTPases on the plasma membrane of leukocytes and dendritic cells, modelling their migration [67]. Congruently, “Cell activation”, “Small GTPase mediated signal transduction” and “Immune system process” were the top three enriched biological processes in Ast_immune2 (Figure 3 C, S. Table 9).
To evaluate whether a sub-population of midbrain astrocytes may be involved in schizophrenia, we compared the proportion of these sub-clusters between schizophrenia and controls. We did not observe changes in the proportion of any of the described astrocyte sub-population associated with schizophrenia (Figure 3 D), suggesting that the number of different astrocyte sub-types in the human midbrain is not altered in schizophrenia.
Recent evidence indicated that astrocytes contribute to the cortical pro-inflammatory signals in a subset of the schizophrenia patients with neuro-inflammation [68]. Also, high GFAP expression was detected in midbrain post-mortem tissue of high-inflammation schizophrenia patients [7]. To test whether a particular sub-population of midbrain astrocytes may be contributing to schizophrenia pathophysiology in the cases with cortical inflammation, we compared the proportion of all the different sub-populations of astrocytes in the schizophrenia-high inflammation with respect to controls (S. Figure 4 A). We did not observe an increased proportion of the astrocyte sub-clusters associated with immune function in the high inflammation schizophrenia cases, which suggests that a differential amount of these astrocytes is not contributing pro-inflammatory signals in schizophrenia. However, the proportion of a sub-type of protoplasmic astrocytes (Ast_protoplasmic2) was increased in the schizophrenia-high inflammation group as compared to controls (S. Figure 4 A). To identify potential functional differences in the schizophrenia-high inflammation-related protoplasmic astrocytes (Ast_protoplasmic2) with respect to Ast_protoplasmic1 (“normal” protoplasmic astrocytes), we compared these two clusters in terms of their gene expression profiles (S. Figure 4 B). The genes more abundantly expressed in the schizophrenia-high inflammation related astrocytes were enriched in “Regulation of neurotransmitter receptor activity”, “Postsynaptic density membrane” and “AMPA glutamate receptor activity” gene ontology terms, among other terms that seem more related to glutamate neurotransmitter regulation than directly to inflammation (S. Figure 4 C). Together, these observations suggest that inflammation might be interfering with the activities of the astrocytes in the midbrain grey matter of the patients with schizophrenia, contributing to schizophrenia pathophysiology.