A single cell atlas of prion disease in the murine cortex and hippocampus

To profile the response of brain cell sub-populations to prion disease, we performed single cell RNA sequencing of cortical and hippocampal cells isolated from 8 RML infected mice when they reached clinical endpoint criteria at time points ranging from 150–172 dpi (Fig. 1A). For comparison, we also sequenced cortical cells from 4 mock mice collected at 147, 168, 186 and 189 dpi and hippocampal cells from 5 mock mice at 110, 147, 168, 186 and 189 dpi. We provide information on the mouse number, brain region, treatment and number of days post infection for each sequencing library included in our analysis in Additional file 1: Table S1. We were unable to exactly match the ages of Mock and RML mice used because we processed cortical and hippocampal brain tissues from one mouse per day. This was done to ensure consistency in the preparation of single-cell suspensions from all mice used in the study. Reagent clogs in the microfluidics of the chromium controller during separation of single cells reduced the usable dataset in the case of the hippocampal samples to 7 RML and 2 mock mice. These two mock mice used were collected at 110 and 147 dpi. Following pre-processing and quality control, the resulting 21 single-cell RNAseq datasets were integrated to produce an “atlas” of brain cells during prion infection, consisting of 147,536 cells that were classified into 39 transcriptionally distinct clusters via Seurat’s graph based clustering approach (Fig. 1B).

Fig. 1

A single cell atlas of brain cells from the cortex and hippocampus from mice infected with RML scrapie or mock infection. A Schematic representation of workflow for single-cell sequencing of cortical and hippocampal cells during prion disease. B UMAP projection of all 147,536 cells sequenced from the cortex and hippocampus of RML and mock infected mice. Cells were clustered using graph based clustering and cell types were assigned to each cluster using SCType (Additional file 4) followed by manual inspection of marker genes (Additional file 2) identified for each cluster. The total number of cells isolated from RML and mock infected mice that were assigned to each cluster is provided in Additional file 3. C Normalized expression level of canonical marker genes for major brain cell types were plotted on the UMAP projection to verify identities of each cell type. These include P2ry12 (microglia), Gfap (astrocytes), Rbfox3 (mature neurons), Cd163 (perivascular macrophages), Pdgfra (oligodendrocyte progenitor cells), Spag17 (ependymal cells), Dcx (immature neurons), Mki67 (neural progenitor cells) and Cldn5 (vascular cells). micro—microglia; endo—vascular endothelial cells; astro—astrocytes; peri—pericytes; g.neu—mature glutamergic neurons; im.neu—immature neurons; opc—oligodendrocyte progenitor cells; smc—vascular smooth muscle cells; pvm—perivascular macrophages; epen—ependymal cells; lymph—lymphocytes; vlmc—vascular leptomeningeal cells

We used automated classification of brain cell types with SCType (based on custom reference markers; Additional file 4) combined with manual inspection of marker genes (Additional file 2) to assign a cell type identity for each cluster. We noted that 3 of the clusters (8, 15 and 35) had unusually high expression of genes associated with technical-artefacts (e.g. high mitochondrial gene expression, Malat1 etc.) [15, 32]. Consequently, we could not identify clear brain cell type specific markers, so we removed these clusters from the final atlas. In total, we identified populations of astrocytes, microglia, perivascular macrophages, oligodendrocyte progenitor cells, glutamatergic neurons, immature neurons, endothelial cells, pericytes, vascular smooth muscle cells, vascular leptomeningeal cells, lymphocytes, and ependymal cells. We verified the identities of these cell types by examining the expression of the canonical marker genes P2ry12 (microglia), Gfap (astrocytes), Rbfox3 (mature neurons) Cd163 (perivascular macrophages), Pdgfra (oligodendrocyte progenitors), Spag17 (ependymal cells), Dcx (immature neurons), Mki67 (neural progenitors) and Cldn5 (endothelial cells) (Fig. 1C). Mature oligodendrocytes were absent from the dataset as expected, given the use of myelin removal beads to minimize clogs in the microfluidics of the chromium controller during cell separation. We then used these clusters representative of cellular sub-types to characterize differences related to brain pathobiology during prion infection. Given its relevance to disease, we examined the expression of Prnp across the single cell atlas, and found it to be most highly expressed by astrocytes, neurons, and surprisingly, ependymal cells (Additional file 1: Fig. S1).

Transcriptional changes and altered cell sub-type composition during prion disease

To characterize transcriptional responses to prion disease, we performed differential expression analysis between all cells isolated from prion-infected versus mock-infected mice within each cluster independently (Fig. 2). According to criteria of FDR p values  < 0.05, average |log2 fold change|> 0.5 and %cell-expression differences > − 0.1 or < 0.1 for increased/decreased genes respectively, we identified differentially expressed transcripts within most of the cell-type-specific clusters (Fig. 2A, Additional file 8). Further examination via hierarchical clustering of log2 fold change values within each cluster revealed the majority of transcriptional changes in response to prion disease showed little overlap between the different clusters, and we concluded that most were specific to cell-types or sub-types. (Fig. 2B).

Fig. 2

Differentially expressed transcripts within each individual cluster of cells isolated from RML and mock infected mice. A Number of transcripts that met differential expression criteria within each cluster when comparing cells isolated from RML and mock infected mice. Differentially expressed transcripts were defined by: FDR p values  < 0.05, |Log2 fold change|> 0.5, > 25% cell expression and 10% increased/decreased cell expression for increased/decreased transcripts respectively. B Hierarchical clustering of Log2 fold changes for all differentially expressed transcripts within each cluster. The full list of differentially expressed transcripts is provided as Additional file 8. astro—astrocytes; endo—endothelial cells; g.neu—mature glutamergic neurons; im.neu—immature neurons; micro—microglia; opc—oligodendrocyte progenitor cells; pvm—perivascular macrophages; epen—ependymal cells; lymph—lymphocytes; pos. —positive; reg. —regulation of

In single cell RNAseq, cells are assigned to specific clusters entirely based on their transcriptomes. Therefore, transcriptional responses to prion disease might also be reflected by differences in the relative frequency of cell clusters between prion- and mock-infected mice. In other words, differences in the relative frequency of sub-clusters of a given cell-type might indicate transitions from one transcriptional state to another during disease. Therefore, we examined the relative proportion of cells assigned to each cluster within the cortex and hippocampus from either RML infected- or mock-infected mice (Fig. 3, Additional file 5). Altogether, there were some striking differences in the relative composition of various cell-types associated with RML disease. Differences in relative frequency can also be influenced by changes to absolute cell count that occur during disease. In the context of prion disease, this is expected for reactive microglia and astrocytes that are well known to proliferate and for vulnerable neurons that decline in number due to cell death [74]. Technical factors can also influence the observed relative frequency, such as difficulties in cell dissociation, cell death during preparation of single-cell suspensions, and liberation of individual cells from debris or cell-to-cell contacts. Therefore, it is challenging to interpret whether the relative frequencies we present for each cell cluster reflect transitions of transcriptional status, or absolute challenges in cell count. Nonetheless, we present these relative frequencies because in many cases, this metric provided clues into the response of brain cell-types to prion disease. Additionally, the low sample size of n = 2 and n = 7 respectively for the Mock-hippocampus and RML-hippocampus groups was a limitation for distinguishing statistically significant disease-associated differences in relative frequency of hippocampal cell transcriptomes. Despite this, many changes in relative frequency were common between the hippocampus and cortex and were more reliable.

Fig. 3

Differences in relative frequency of cell populations isolated from the cortex and hippocampus of RML infected mice compared to mock infected mice. The relative frequency of each cell cluster is plotted for the cortex and hippocampus of RML and mock infected mice. p values s were calculated using the non-parametric Mann–Whitney U tests. * p values  < 0.1, ** p values  < 0.05, *** p values  < 0.01. All p values are provided in Additional file 5. astro—astrocytes; endo—endothelial cells; g.neu—mature glutamergic neurons; im.neu—immature neurons; micro—microglia; opc—oligodendrocyte progenitor cells; pvm—perivascular macrophages; epen—ependymal cells; lymph—lymphocytes; cx—cortex; hp—hippocampus

Vascular dysfunction implicated through abnormal transcription during prion disease

The blood brain barrier is comprised of various vascular cells including endothelial cells, pericytes, smooth muscle cells and vascular leptomeningeal cells [17]—all of which were present in our single cell atlas (Fig. 1). Disruptions to the blood brain barrier are a common feature of aging and neurodegeneration [42]. Consistent with this, prion-altered vascular transcripts were enriched in ontologies related to cell migration, blood vessel morphogenesis and vascular transport (Fig. 2B, Additional file 8). In addition to these transcriptional changes within vascular cell clusters, we also noticed that most vascular cell clusters decreased in relative frequency in association with disease (Fig. 3). We did not observe evidence of cell-death related transcription by vascular cells during prion disease, and so we cannot conclude whether the decrease of vascular cells were related to blood brain barrier breakdown. It is possible that we observed decreased vascular cell frequencies because of microglial proliferation or other technical factors. Many of the prion altered vascular were overexpressed by clusters endo.2, endo.11 and peri.14. Specific examples of notable disease-altered transcripts that represent enriched gene ontologies are listed as follows: Cluster endo.2 overexpressed transcripts related to cell migration (Sema5a, Flt1, Lef1, Pecam1, Gcnt2, Rhoc, Dock1, Ptk2, Rab11a, Igf1r) and angiogenesis (Sema5a, Ramp2, Flt1, Rock1, Rhoj, Tek, Ism1). Cluster endo.11 overexpressed transcripts related to blood brain barrier transport (Slco1c1, Slc16a1, Slc2a1, Mfsd2a). Cluster peri.14 overexpressed transcripts related to actin filament/supramolecular fiber organization (Carmil1, Myo1b, Ubb, Col8a1, Myh11, Mfge8, Aldoa, Svil, Eps8). These signatures of abnormal transcription seem to hint of vascular dysfunction or remodeling that might occur during prion disease, with blood brain barrier transport possibly being altered. Inflammation is well known to cause disruptions to brain vascular cells [42], so we were not surprised to observe evidence of vascular dysfunction in our single cell atlas. However, our analysis cannot determine whether blood brain barrier permeability is altered at the phenotypic level during prion disease, and so more detailed studies are required to investigate this hypothesis.

Oligodendrocyte progenitor cells are transcriptionally modulated during prion disease

Transcriptional alterations to oligodendrocytes are rarely the focus of investigations into prion pathogenesis because mature oligodendrocytes are considered relatively resistant to prion replication [66]. However, mature Olig2+ oligodendrocytes were recently shown to decrease at the advanced stages of disease in a murine model of Creutzfeldt-Jakob disease [3], implying a role in pathogenesis. In our dataset, populations of oligodendrocyte progenitors were increased in relative frequency during RML disease, particularly in the hippocampus (Fig. 3). This is consistent with a previous bulk RNAseq analysis, where we inferred increased oligodendrocyte progenitors through increased Pdgfra abundance during RML disease [72]. However, it is also possible that we observed the increase in oligodendrocyte progenitor frequency due to technical reasons such as resistance to cell death during isolation relative to other cell types in the condition of prion disease. Disease-altered oligodendrocyte progenitor transcripts were enriched in ontologies related to neuron differentiation, cAMP signaling and response to retinoic acid and included downregulation of canonical oligodendrocyte progenitor cell markers Pdgfra and Vcan (Fig. 2B, Additional file 8). These transcripts were disease-altered in cluster opc.19, but not opc.37. A few disease-altered transcripts were also shared between oligodendrocyte progenitor cells with neurons and astrocytes. Notable upregulated disease-altered oligodendrocyte progenitor transcripts were related to cell adhesion (Kirrel3, Ptprt, Tenm1, Unc5d, Lrrc4c, Dscaml1, Cdh8, Cdh18), synapse assembly (Gabrb3, Kirrel3, Dnm3, Gabrb2, Farp1, Ppfibp1, Lrrc4c, Ppfia2) and neurotransmission (Gria2, Neto1, Dlgap2, Gria3, Grin3a). Notable downregulated oligodendrocyte progenitor transcripts were related to gap junctions (Ptprd, Cntnap2, Agt) and nervous system development (Cntnap2, Vcan, Cntn4, Adgrl3). These findings suggest that transcriptional dysfunction oligodendrocyte progenitor cells is an underappreciated component of prion pathogenesis—an avenue that is worthy of further investigation.

Transcriptional changes of microglia and perivascular macrophages during prion disease

Microglia made up the largest population of cells (99,756/147,536 = 67%) assigned in our library and are well described as particularly responsive to prion replication, taking on reactive phenotypes that can both exacerbate pathology through excess inflammatory signaling and can protect against disease through clearance of toxic PrPSc [53, 58, 63]. As expected, altered microglial transcripts were involved in cytokine signaling, phagocytosis and microglial activation (Fig. 2B, Additional file 8). Surprisingly few markers of reactive microglia were increased within individual microglia clusters, although there were a few such as Lyz2, Apoe, Tyrobp and Irf8. Some microglia-specific markers were decreased across some of the individual microglial clusters, such as P2ry12, Tmem119, Csf1r, and Cx3cr1. Homeostatic markers like P2ry12 and Tmem119 are generally reported to decrease in reactive microglia [41, 54]. Similar to microglia, prion altered transcripts within perivascular macrophages were related to inflammatory signaling through cytokines, chemokines and antigen receptors. There was little overlap between prion-altered transcripts of microglia and perivascular macrophages, implying a distinct response to prion disease. Unsurprisingly, some of the most drastic changes in cellular populations during RML disease were seen in microglia (Fig. 3). We noticed a few of the microglial clusters either decreased, or did not change in relative frequency, whereas many of the microglial clusters increased in abundance and we suspected that these corresponded to reactive, or “disease-associated” microglia that are well known to increase during disease [74]. Unlike microglia, we did not observe an expansion of perivascular macrophages in disease, and perivascular macrophages (pvm.22 and pvm.29) appeared to decrease in the hippocampus and cortex. This could possibly reflect clustering of activated perivascular macrophages with the disease associated microglia, or cellular migration to other brain regions.

Distinct transcriptomes reveal a spectrum of microglial activation states during prion disease

Nearly 100,000 microglial transcriptomes were sequenced, making our single cell dataset particularly well suited to characterize the diversity of microglial activation states during prion disease. Furthermore, compared to single nucleus sequencing, our live single-cell sequencing approach can improve detection of transcriptional changes within activated microglia [82]. Therefore, to define transcriptional states of individual microglia subtypes, we subset the dataset to include only the 99,756 microglial transcriptomes (Fig. 4A). We retained the original microglial clusters from the full atlas, and did not perform further sub-clustering. When comparing microglia isolated from infected with mock- infected mice, we could see that some microglial clusters were nearly unique to disease (micro.9, micro.12, micro.17, micro.23, and micro.36) and were strongly associated with disease (Fig. 4A).

Fig. 4

Microglia take on a spectrum of activation states in association with RML disease. A UMAP projections of all 99,756 microglial cells isolated from RML and mock infected mice. B Hierarchical clustering of all marker genes identified within each microglial sub-cluster. Marker genes were grouped using K-means clustering, and each gene cluster was functionally annotated with enriched GO terms using Enrichr. micro—microglia; cx—cortex; hp—hippocampus

We next functionally characterized the microglial subtypes by identifying marker genes highly expressed by each cluster, up to a maximum of 25 per cluster, resulting in 218 transcripts supplied for hierarchical clustering (Fig. 4B). K-means clustering of genes was used to classify the marker transcripts into 8 gene modules that were functionally annotated with representative enriched gene ontologies. Altogether, we found that reactive microglia take on a spectrum of transcriptional states characterized by expression of genes important for various aspects of glial functionality such as phagocytosis, or cytokine signaling. Based on expression of these gene modules (Fig. 4B), and the expression of specific microglial marker transcripts (Additional file 1: Fig. S2), we further classified microglia into 5 subtypes: (1) homeostatic, and the following reactive subtypes: (2) proliferating, (3) phagocytic, (4) type I interferon (IFN) responding, and (5) antigen presenting (MHC). These functional subtypes are similar to what has previously been reported in relation to Alzheimer’s disease [16]. We also classified some of the microglial clusters as representing intermediate transcriptional states between these subtypes. Within the subset microglial dataset, the homeostatic microglial clusters decreased in relative frequency, whereas reactive microglial clusters (proliferating, phagocytic, IFN and MHC subtypes) increased in relative frequency in association with disease (Fig. 5A, Additional file 9). This could indicate conversion of homeostatic microglia into reactive forms. Therefore, we performed a monocle trajectory analysis of the microglial cells to measure transcriptional status as a function of gene “pseudotime” by supplying cells from cluster micro.4 to serve as the “root” for the trajectory (Fig. 5B and C). We noted a circular transcriptional trajectory between homeostatic microglia, intermediately activated microglial states, and proliferating microglia with branches into phagocytic, antigen-presenting and interferon-responding microglial populations. Altogether, our interpretation was that microglia form a continuous spectrum of transcriptional states, where multiple possible transcriptional trajectories can allow homeostatic microglia to reach distinct disease-associated reactive states, thus mirroring the complexity of functionally distinct phenotypes observed in the brain.

Fig. 5

Monocle trajectory analysis of microglia categorized into 5 transcriptional subtypes. A The relative frequency of each microglia sub-cluster is plotted for the cortex and hippocampus of RML and Mock infected mice. p values s were calculated using non-parametric Mann–Whitney U tests. * p values  < 0.1, ** p values  < 0.05, *** p values  < 0.01. All p values are provided in Additional file 9. B UMAP projection plot of all 99,756 microglial transcriptomes with color mapping to transcriptional pseudotime calculated with Monocle. Cells from cluster micro.4 were supplied to serve as the root for calculating transcriptional pseudotime and trajectories. C UMAP projection plot with color mapping to microglial subtypes

Transcriptional signatures of five microglial subtypes associated with prion disease

Homeostatic microglia (micro.3 and micro.4) were marked by high expression of the canonical microglial markers P2ry12, Cx3cr1, Tmem119 in addition to Nav2, a marker of microglia under healthy conditions [75] (Additional file 1: Fig. S2). These microglia had high abundance of gene modules 6, 7 and 8 that were enriched in ontologies related to calcium transport (Cacnb2, Bcl2, Ank2), regulation of glial apoptosis (Prkca) and regulation of microglial migration (P2ry12, Cx3cr1) (Fig. 4A). Homeostatic microglia elicited a disease-associated decrease in relative frequency (Fig. 5A), although this does not necessarily indicate that the absolute cell count of homeostatic microglia decreases in the prion infected brain. Given the possible transcriptional trajectories that could allow homeostatic microglia to reach different reactive states (Fig. 5B and C), this decrease in relative frequency likely corresponds to conversion of homeostatic microglia into intermediate and eventually reactive transcriptional states during disease.

Gene modules 4 and 5 were involved in cytokine signaling, regulation of inflammation and regulation of cell proliferation and were highly expressed by proliferating microglia (micro.0, micro.9 and micro.12) (Fig. 4A) that were demarked by high expression of Jun, Fos and Il1a (Additional file 1: Fig. S2). Together, Jun and Fos encode proteins that form the AP1 transcription factor that induces inflammatory gene expression in microglia [84]. Interestingly, cluster micro.12 was uniquely marked by very high expression of the cytokine Il12b (Additional file 1: Fig. S2). Furthermore, Cd14 was highly expressed by micro.9 and micro.12 and is a co-receptor for LPS that modulates inflammatory signaling, important for microglial responses to tissue damage-associated signals [35]. We also noted an expansion of proliferating microglial clusters micro.9 and micro.12 in association with disease (Fig. 5A). Altogether, these results suggest that proliferating Jun+Fos+ microglia might contribute towards inflammatory cytokine signaling during prion infection. Specific examples of cytokine signaling related transcripts expressed by these microglia include Cd86, Egr1, Pdgfb, Fos, Ptgs2, Cxcl2, F3, Nfkb1, Socs3, Bcl6, Il1b, Ccl4, Il12b, Tnfsf9, and Junb.

Phagocytic microglia (micro.13) were marked by high expression of Aif1, Ftl1 and Fau (Additional file 1: Fig. S2) and highly expressed gene module 2 that was enriched in synapse pruning and microglial activation, containing many microglial activation markers (Fig. 4B). Specific examples of classical microglial activation markers expressed by these microglia include C1qa, C1qb, C1qc, Tyrobp, Trem2, Aif1, B2m, Prdx5, Fcer1g, Cstb, Ctsz, Cd63, and Cd68. Given that this corresponds to a classic signature of reactive microglia, we were not surprised to see that the relative frequency of micro.13 increased in association with disease (Fig. 5A). Interestingly, Aif1+Ftl1+ microglia corresponded morphologically dystrophic iron-accumulating microglia in an Alzheimer’s mouse model [40], possibly providing clues as to the role of phagocytic microglia in prion disease.

Antigen presenting microglia (micro.17 and micro.36) highly expressed phagocytosis related genes (gene module 2, also highly expressed by the phagocytic microglia cluster micro.13) and genes important for antigen presentation (Fig. 4B). These microglia were marked by high expression of Cd74, H2-Aa, Cd52 and Ccl6 (Additional file 1: Fig. S2). The antigen presentation genes that were highly expressed by these microglia were Cd74, H2-Aa, H2-Eb1, H2-Ab1, H2-K1, and H2-D1. Clusters micro.17 and micro.36 showed some of the most dramatic increases in relative frequency in association with disease (Fig. 5A) and were nearly absent the Mock mice. In fact, micro.36 was not detected at all in the hippocampal cells isolated from Mock mice and was only detected in one cortical cell suspension of Mock mice. Therefore, we postulate that these antigen-presenting microglia subtypes represent highly activated reactive microglia that are strongly associated with prion disease. Cd74 is thought of as a marker of M1 microglial activation, and is expressed by highly activated microglia in the diseased-brain [37, 79], supporting this notion.

Trim30a, Oasl2 and Cxcl10 were highly expressed by type I interferon responsive microglia (micro.23, Additional file 1: Fig. S2) that highly expressed gene module 1 (Fig. 4). Examples of type I interferon responsive transcripts expressed by these microglia include Ifitm3, Bst2, Rsad2, Isg15, Ifit1, Gbp2, Ifit3, Ifit2, and Cxcl10. Like the other reactive microglia subtypes, the type I interferon responsive microglia also show a strong disease-associated increase in relative frequency (Fig. 5A). Type I interferon signaling is often thought of as detrimental in the context of brain pathology, but a recent study has suggested that this pathway might protect neurons during prion infection [33].

As expected, microglia that were considered to represent intermediate transcriptional states (micro.1, micro.5, micro.6, micro.7, and micro.25) had varying expression of the different microglial gene modules (Fig. 4B). Interestingly, micro.6 was uniquely marked by very high expression of Serpine1 (Additional file 1: Fig. S2)—an inhibitor of tissue plasminogen activator (tPA) that promotes microglial migration and inhibits phagocytosis in vitro [36]. Intermediate microglial clusters micro.6 and micro.7 were both positively associated with disease through increases in relative frequency (Fig. 5A).

Dysregulation of neuroprotective astrocytes during prion disease

Astrocytes are one of the main cells types responsible for brain homeostasis through neurotransmitter uptake/recycling, potassium buffering, metabolism, and protection against oxidative stress among other neuroprotective functions [6]. In the context of prion disease however, astrocytes are one of the first cells to take on active phenotypes during disease that may have various beneficial or detrimental roles, concomitant with the earliest detectable deposits of PrPSc [81]. We observed a striking global decrease in relative frequencies of astrocyte populations associated with RML disease (Fig. 3) and to our surprise; we did not observe a clearly resolved cell cluster corresponding to reactive astrocytes. Disease-altered astrocyte transcripts were enriched in ontologies related to synapse organization, blood brain barrier transport and sulfur biosynthesis, hinting at modulation of astrocyte homeostasis functions (Fig. 2B, Additional file 8). Most of the disease-altered astrocyte transcripts were upregulated in cluster astro.10, and notable transcripts were the reactive astrocyte marker Gfap, and transcripts related to cell junction assembly (Kirrel3, Gpm6a, Farp1, Cdh2, Cdh20, Ctnnd2, Nrcam, Cdh19), sulfur metabolism (Bcan, Gstm1, Angpt1, Cspg5, Prelp, Chsy3, Gstm5) and cell projection organization (Ntrk2, Fut9, Magi2, Il1rapl1, Atp1b2, Prkd1). We also noticed a notable group of transcripts downregulated in cluster astro.20 that were related to axonogenesis (Robo2, Auts2, Nrxn3, Slit2). This signature of differential transcription indicates dysfunction of the homeostatic astrocytes that were captured by our live single cell approach.

To better characterize the population of astrocytes isolated, we performed a sub-cluster analysis by combining all 7,813 astrocyte transcriptomes (from clusters astro.10 and astro.20) and re-clustering into 11 new astrocyte sub-clusters (Fig. 6A). We examined the relative frequency of these astrocyte sub-clusters among all astrocytes and classified them based on whether they were depleted (“disease-depleted”), unchanged, or increased (“disease-associated”) during disease (Fig. 6C, Additional file 6). The majority of the astrocyte sub-clusters decreased in the prion infected brains, but two (astrocyte sub-clusters 6 and 8) increased, and we suspected that these might correspond to a small population of reactive astrocytes. We examined the expression of astrocyte marker genes across the astrocyte sub-clusters (Fig. 6B and Additional file 1: Fig. S3) and noted that disease-associated astrocyte sub-cluster 8 had high expression of Gfap, Aqp4, Vim, and low expression of Nrxn3, consistent with reactive astrocytes [