We previously developed knock-in mouse models of familial CJD and FFI linked to E200K and D178N mutations, in the endogenous mouse Prnp gene (15)(16). These models developed late onset, progressive diseases that replicate several key pathological features of the respective human diseases, and importantly, differ from each other in pathological changes and affected brain regions. FFI mice experience neuronal loss and reactive astrocytosis in the thalamus and atrophied cerebellum (15). In contrast, CJD mice develop PrPres and spongiosis, hallmarks of the human disease, most prominently in the hippocampus, and PrPres in the molecular layer of the cerebellum (31) (Fig. 1A). PrP in CJD mice had a slightly altered glycoform pattern, suggesting a slightly altered path through the secretory system, but in FFI mice, mono- and unglycosylated PrP were nearly absent and the total amount of all forms was only 25% of normal levels (15), suggesting it is subjected to intensive quality controls and that the mammalian brain responds to these mutant proteins differently. Automated mouse behavioral analysis indicated sleep was fragmented and core body temperature measurements suggested FFI mice had impaired sleep regulation at this age (32), but electroencephalography (EEG) measurements were not attempted then due to biosafety constraints. Consideration of the neuropathological changes and in vivo clinical abnormalities measured by automated behavioral analysis and in vivo magnetic resonance imaging led to the general picture that disease emerged at approximately 16 months of age for both models (31)(15).
Neural activity is mildly affected in old FFI and CJD mice
To rigorously characterize the general neural health and sleep features in these models, we used the same EEG methods (Fig. 1B) we applied previously to the RML (Rocky Mountain Labs) model of acquired PrD (21). Since this was a telemetric recording system, mice could roam freely in their cage, thereby avoiding artifacts from tethering. In that study, theta frequency waves increased as disease progressed, like observations in several human PrDs (33). Notably, sleep was not affected in RML mice, even in late stages (21). Turning to the FFI mice, since we previously observed that behavioral activity was only mildly affected at 16 months of age, which is likely a result of only mildly diminished neural health at that timepoint, to increase the likelihood of detecting EEG abnormalities we studied mice at approximately 21 months of age (mean = 20.8, SD = 2.3). Surprisingly, considering that in FFI mice temperature was dysregulated and that sleep bouts were disrupted according to an automated video-based system (15), sleep was not strongly affected in the current EEG study. During daytime, non-rapid eye movement (NREM) sleep tended to be reduced, and wake tended to be increased, but neither was significant (Fig. 2). To test if sleep control was vulnerable to external manipulation, we measured the response to six hours of sleep deprivation, which showed no significant differences between FFI and control mice (Additional File 1A + B). Since sleep abnormalities are sometimes absent in humans with FFI (5)(34), their absence from this mouse model is not a complete surprise. CJD mice were studied in parallel, and they also showed no abnormality in baseline sleep (Fig. 2C + D) or in response to sleep deprivation (Additional File 1C + D). Nonetheless, theta frequency waves were increased in FFI mice during NREM and REM sleep (Fig. 2F + G), and in CJD mice during wake and REM sleep (Fig. 2H + J), mimicking this potential biomarker of human PrD. Therefore, despite these models showing neuropathological and behavioral changes at this stage, there are only mild changes to theta frequency and the disease is still too mild to cause sleep disruption.
In our recent RML study mentioned above (21), we found that before EEG, behavioral or neuropathological changes emerged, RiboTag profiling identified specific cell types with altered translatomes. To study a similar disease stage as done for that study (56% of disease onset), these RiboTag experiments included mice at nine months of age.
Capture of cell type-specific mRNA with RiboTag
With RiboTag, cell type-specific translatomes are obtained from homogenates of brain tissue by immunoprecipitating HA-tagged ribosomes and the attached translating mRNA. Driver lines expressing Cre directed by the genes encoding Gad2 (35), vGluT2 (17), PV (19) and SST (35) were used to achieve cell type-specific expression of the RiboTag transgene (Fig. 1C). This enabled us to target wider populations of glutamatergic and GABAergic neurons, as well as PV+ and SST+ GABAergic subtypes. Using a selection of cell type marker genes, we recently confirmed by both immunofluorescence and RNA-seq of RiboTag IPs, that these Cre lines lead to specific and selective activation of RiboTag expression (21). Since the cerebellum was affected in both FFI and CJD models, and the remaining part of the brain (hereafter cerebrum) had distinct brain regions that were targeted in each model, the cerebellum and cerebrum from each brain were frozen separately. RiboTag IPs were prepared for all cell types for cerebrum samples, but only for Gad2 and vGluT2 for cerebellum, since in the cerebellum PV-Cre induces RiboTag expression in the same cells as Gad2-Cre, whereas SST-Cre induces RiboTag expression in very few cells. Consequently, we profiled six cell types, encompassing two brain regions, in two genetic PrDs (Fig. 1D, Additional File 2). To verify the isolation of cell type specific translatomes in RiboTag samples, we additionally analyzed total mRNA from the homogenates in parallel (Fig. 1D). The study group was age-matched (mean = 9.3 months, SD = 0.7), double heterozygous for RiboTag and Cre, and homozygous for either FFI, CJD, or WT Prnp alleles (Additional File 3).
As expected, PCA showed differences between total mRNA samples based on the region (cerebellar vs cerebral) but not cell types (Fig. 3A). In contrast, IP samples showed clear differences based on regions and cell types (Fig. 3B). This was apparent through comparisons of expression of cell type marker genes between IP and total mRNA, which revealed the expected enrichment of general GABAergic and glutamatergic neuronal marker genes in samples in which the respective cell types were targeted (Fig. 3C + D). Targeting of specific subclasses of GABAergic neurons was confirmed by upregulation of PV- or SST-specific marker genes in the respective samples, whereas Htr3a (serotonin receptor 3A) and Vip (vasoactive intestinal peptide), GABAergic markers absent from SST and PV neurons, showed the predicted enrichment in Gad2+ and depletion in PV+ and SST+ IPs (Fig. 3C). In the cerebellum, Gad2+ IPs were enriched for marker genes of several cerebellar GABAergic cell types such as Purkinje, basket, Golgi, and stellate cells, while vGluT2+ IPs showed enrichment for granule cell markers (Fig. 3D) (36)(37). As expected, astrocyte and microglia marker genes (38) were depleted in all IP samples. These results indicate that cell type-specific translating mRNA was successfully isolated from the intended neuronal subpopulations.
Prnp expression varies with cell type and sequence
One potential explanation for selective vulnerability is that vulnerable cell types express high levels of toxic protein. To test this possibility, we examined the expression levels of Prnp in the targeted cell types based on TPM values (Fig. 3E). Unexpectedly, Prnp was expressed almost two-fold higher in vGluT2+ neurons than in GABAergic cell types. These differences were detected in all three genotypes. Higher Prnp expression in vGluT2 neurons may partially explain the selective vulnerability in these models since the regions most affected, thalamus and hippocampus, are predominately glutamatergic. This analysis also showed that FFI mice had slightly lower Prnp expression. This tendency was most pronounced in glutamatergic neurons and only significant in cerebral vGluT2+ neurons (Kruskal-Wallis, p = 0.026, chi2 = 7.312). This observation is consistent with the reduced PrP levels previously reported in FFI mouse brains, suggesting that the D178N mutant engages either a different or more intensive quality control mechanism than the E200K mutant. Since the protein levels are reduced much more than the mRNA levels, the protein misfolding may be happening during and after mRNA translation and both get triaged for degradation.
SST+ neurons show pronounced translatome changes in pre-symptomatic stages of CJD and FFI
A general characterization of translatome profiles for disease-targeted cell types in both disease models was done by differential gene expression analysis with the DESeq2 R package (26) (Additional File 4). Since the mice were at a pre-symptomatic disease stage, we expected mild changes to gene expression and therefore defined differentially expressed genes (DEGs) to have a false discovery rate (FDR) ≤ 0.05 without a log fold change (LFC) cutoff (Fig. 4A + B, Additional File 5A + B). Surprisingly, SST+ neurons responded with the highest number of DEGs in both disease models (CJD: 153, FFI: 684), whereas PV+ neurons showed very few DEGs (CJD: 2, FFI: 3). A comparison of shared DEGs between cell types of the same disease revealed that most DEGs were unique to a given cell type, including GABAergic subtypes (Fig. 4C + D). In contrast, SST+ neurons demonstrated a high overlap in DEGs between CJD and FFI, with 55 down- and 58 upregulated genes shared (Fig. 4E). There were few shared genes in other cell types, likely due to the overall low number of DEGs (Additional File 5C). Since little is known about the vulnerability of SST+ neurons to PrDs, many of our analyses focused on these important cells.
In both mutants, SST+ neurons displayed increased expression of many ribosomal protein mRNAs: of 79 ribosomal proteins, 26 were upregulated in CJD (mean log2FC = 0.42, SD = 0.09) and 57 in FFI (mean LFC = 0.44, SD = 0.09) (Additional File 5D). Besides suggesting an increased need to synthesize proteins, the high functional connectivity of these genes is strongly indicative of a coordinated response. To measure the coordination amongst other DEGs we looked for enriched GO classes by applying ORA. Upregulated DEGs In CJD SST+ neurons were associated with translation (ribosomal protein genes), actin cytoskeleton, actin-filament organization, and axonogenesis (FDR ≤ 0.01, Additional Files 6 + 7). In FFI SST+ neurons upregulated DEGs were mostly related to translation (Snu13, Eef1a1, Eef12 and several ribosomal proteins (Additional Files 7 + 8A)). Upregulated DEGs were overrepresented in “myelination”, and cytoskeleton and cell adhesion-related terms, including “actin-binding”, “focal adhesion” and “cell-substrate junction”. Notably, downregulated DEGs in FFI SST+ neurons were also enriched among cytoskeleton-associated terms (“processes related to neurite morphogenesis and organization” term, “microtubule binding” and “motor activity”) and cell adhesion. Additionally, downregulated genes were overrepresented among terms related to synaptic plasticity and ion-channels or receptor components (Additional file 8B). We also found enrichment of GTPase activity, including genes involved in Ras and Rho signaling, such as activators of Rho-family GTPases (Arhgap32,35,44), Rho guanine nucleotide exchange factors (GEFs) (Als2, Agap2, Trio, Dock4), and downstream effectors (Cdc42bpa, Rock2). Rho GTPases are known regulators of actin cytoskeleton dynamics (reviewed here: (39)), including dendritic spine formation and density (40), further indicating a high connectivity between DEGs. Collectively, these results suggest a concerted effort to reorganize the cytoskeleton of SST+ neurons. In summary, CJD and FFI showed a surprisingly high overlap in DEGs and, to a lesser extent, in enriched GO terms, suggesting that these neurons activate similar responses in both diseases.
Gene set enrichment analysis reveals similar functional enrichments in CJD and FFI
A limitation of ORA is that coordinated but statistically insignificant expression changes of several genes within a pathway may have important biological implications but would be excluded. Therefore, we applied a complementary approach, gene set enrichment analysis (GSEA) (41), to assess enrichment of GO terms for biological processes (BP) and KEGG pathways in each cell type, using the piano R package (27) which provides combined enrichment scores summarizing results of several statistical methods. Additionally, separate p-values for different directionalities of change were provided for each gene set. Gene sets with significant up- or down-regulation (FDR ≤ 0.05) were ranked by their consensus score, which was calculated based on adjusted p-values for all six statistical methods applied (Additional File 9). Only gene sets significant in at least three of the six statistical methods are presented in Fig. 5.
Top ranked gene sets for CJD and FFI SST+ neurons showed upregulation of translation-related gene sets and ND-related pathways, including “Alzheimer’s disease”, “Parkinson’s disease”, and “oxidative phosphorylation”. Shared downregulated terms included “axon extension”, “neuron differentiation”, “positive regulation of neuron projection development” and “synapse organization”. FFI SST+ neurons also showed downregulation of pathways and terms related to synaptic function, phosphatidylinositol phosphorylation, and downregulation of “small GTPase mediated signaling transduction” (Fig. 5 column 3). Therefore, the results of these analyses reflect DESeq and ORA results for SST+ neurons despite methodological differences.
Interestingly, GSEA results also showed similar enrichment patterns for PV+ neurons in both disease models (Fig. 5, column 2), which were missed by DESeq and ORA due to the low number of DEGs (Fig. 4A + B). As with SST+ neurons, we found upregulation of translation-related pathways and GO terms, but not changes of ND-related pathways. Both disease models showed upregulation of immune response-related pathways, and downregulation of phosphatidylinositol-signaling, “positive regulation of autophagy”, and “protein processing in the endoplasmic reticulum” (ER). Downregulated GO terms exclusive to FFI PV+ neurons suggested a disruption in synaptic function (“synapse organization”, “Axon guidance”, “positive regulation of neuron projection development”). Interestingly, “neuron migration”, “neuron differentiation”, and “regulation of cell shape” were upregulated in CJD but downregulated in FFI PV+ neurons (Fig. 5, column 2). Compared to the highly similar enrichment patterns observed in PV+ and SST+ neurons, Gad2+ neurons of the cerebrum were less similar between the disease models. In FFI we observed upregulation of ribosome pathway, GTPase signaling (Ras and Rap1), neuron migration and inflammation-related GO terms, whereas in CJD terms related to metabolic processes, mitochondrial translation, proteasome, and DNA repair were downregulated (Fig. 5, column 1). In contrast, Gad2+ neurons of the cerebellum exhibited widespread changes with similar patterns in each model (Fig. 5, column 5). Both disease models showed upregulation of terms related to translation, splicing, RNA and protein transport, and ND related pathways. GO terms and pathways related to phosphatidylinositol and GTPase signaling, inflammation and cellular morphology (“regulation of cell shape”, “Cell adhesion molecules”), neuron migration and differentiation were downregulated in both diseases.
In contrast to the broadly similar changes in GABAergic neurons, vGluT2+ neurons of the cerebrum showed more disease-specific responses (Fig. 5, column 4). For example, in CJD, ND pathways, oxidative phosphorylation, chromatin organization and axon extension are enriched among downregulated genes. In contrast, FFI showed downregulation of ER protein processing and protein export neuron differentiation, and neuropeptide signaling. Notably, positive regulation of GTPase activity, phosphatidylinositol phosphorylation and chromatin organization had different directionalities between the diseases. Shared responses included downregulation of mRNA transport and upregulation of small GTPase activity and aminoacyl-tRNA biosynthesis. Cerebellar vGluT2+ neurons displayed the most pronounced differences between models, with several gene sets exhibiting opposite regulation between CJD and FFI, such as those related to translation, DNA repair, and mRNA transport, which were down in CJD but up in FFI. We further found downregulation of ND pathways, splicing, protein folding, and starvation response unique to CJD. In contrast, FFI vGluT2+ neurons showed upregulation of apoptosis and regulation of mitotic cell cycle, and downregulation of ER protein processing and synaptic function (Fig. 5, column 6). Overall, GSEA revealed high similarities in enriched ontologies and pathways between the diseases for SST+, PV+ and cerebellar Gad2+ neurons, but in vGluT2+ neurons, especially in the cerebellum, it demonstrated disease-specific responses.
Identification of functional modules in an SST+ co-expression network
Since SST neurons are understudied in PrD research, we wondered if they might reveal new insights into therapeutic targets. Thus, we used a network-based approach to further elucidate patterns in gene expression changes in SST+ neurons. Using our SST+ neuron-specific translatome data we constructed an undirected weighted gene co-expression network using pairwise gene correlations (FDR ≤ 0.01, Spearman ρ > 0.82) (Additional File 10). Community analysis using Leiden algorithm (28) generated six major modules (ranging in size from 249 to 2,733 genes) consisting of genes with highly correlated expression patterns across all conditions (Fig. 6A), which were validated by comparison to a random network. As co-expression analysis builds on the assumption that correlation patterns between genes reflect functional connection, we used ORA to determine significantly enriched (FDR ≤ 0.01) ontology terms and pathways among module genes (Additional File 11).
Module 1 consisted predominantly of genes downregulated in both diseases (Fig. 6A), including 241 genes also differentially expressed, and predominantly downregulated, in FFI. Module genes were significantly overrepresented (FDR ≤ 0.01) among terms related to synaptic transmission, protein modifications and transport, response to starvation, neuron projection development and axon guidance. Module genes annotated to these terms also included several genes which we identified as differentially expressed either in both diseases (indicated in bold italics in Fig. 6A) or specific for FFI (italics). Genes annotated to synapse organization, chromatin remodeling, and regulation of dephosphorylation-related terms included FFI-specific DEGs. Interestingly, ORA of module 1 genes also revealed autophagy-regulation (“negative regulation of macroautophagy” and “TORC1 signaling”) and chromatin modifications (“positive regulation of histone ubiquitination”) among the top enriched ontologies (Additional File 12A).
Module 2 genes were enriched for translation, ribosomal biogenesis, and mitochondrial organization (Additional File 12B, Fig. 6A). This is consistent with ORA results from upregulated DEGs identified in CJD and FFI SST+ neurons (Additional Files 6 + 8), as module 2 contains ribosomal protein genes, a large percentage of which were upregulated in both diseases. Additional enriched GO terms related to ER stress, regulation of apoptotic process-related terms, and unfolded protein response (UPR), which included several FFI-specific DEGs such as activating transcription factors 4 and 5, Atf4 and Atf5 (Fig. 6A). These results are consistent with those from GSEA (Fig. 5) and functional analysis of cell type specific DEGs (Additional Files 6 + 8). This indicates genes in modules 1 and 2 might be of particular interest to familial PrD-associated pathological processes as these show highly correlated expression patterns with a high percentage of DEGs and are functionally closely related to identified dysregulated terms.
Genes in Module 3 were mostly downregulated in both diseases and functionally associated with chemical synaptic transmission, nervous system development, and protein modifications (Fig. 6A) but also including translation initiation, regulation of macroautophagy and stress granule assembly among top enriched GO terms (Additional File 12C). Module 4 was highly connected with Module 3 and contained predominantly upregulated genes associated with ER organization, protein targeting and ubiquitination (Fig. 6A, Additional File 12D). Module 4 also contained several mitochondrial genes, in particular encoding ATP synthase subunits, associated with KEGG pathways oxidative phosphorylation, thermogenesis and Alzheimer’s and Parkinson’s disease pathways (Fig. 6A). Module 5 genes showed significant overrepresentation of terms related to mRNA splicing and RNA processing (Fig. 6A, Additional File 12E). No significant enrichment was detected for genes in module 6.
Hub genes point towards two potential therapeutic targets
To find potentially important regulators, we next identified hub genes that display the largest number of co-expressed genes. We defined hubs as the top 1% of genes with the highest degree centrality, i.e., most direct neighbors, in each module of our co-expression network (Table 1). Notably, three hub genes in Module 1 were also differentially expressed in FFI: GATOR1 subunit Depdc5 (DEP Domain containing Complex 5; degree: 560), histone-deacetylase Mta3 (Metastasis Associated 1 Family Member 3; degree: 551) a subunit of the nucleosome remodeling and deacetylase (NuRD) complex, and Gtf3c1 (General Transcription Factor IIIC Subunit 1; degree: 585) a mediator of RNA polymerase III transcription. Since downregulation of these highly connected hub genes suggests they have a central role in the pathological process that may have far-reaching effects on interaction partners, we next aimed to further validate the interaction of hub genes with their co-regulated neighbors. For this we constructed a protein-protein interaction (PPI) network for each hub gene and its first-degree neighbors, to determine whether known interactions between products of co-regulated genes exist. Predicted PPIs were obtained from STRINGdb, considering only interactions with a high combined confidence score ≥ 0.7, and excluding interactions based on text mining and databases.
There were no predicted interactions of Gtf3c1 with its co-regulated direct neighbors, indicating that this method did not provide further insight for this gene. However, the PPI network for Mta3 included 220 of 551 co-regulated genes from our topological network (Additional File 13) while the PPI network for Depdc5 included 230 of 560 co-regulated genes (Fig. 6B, Additional File 14). Both networks additionally showed strong overlap with 145 shared genes and included 30 genes significantly downregulated in FFI (Fig. 6B, blue border) or 2 in both diseases (green border). Pathway and GO enrichment analysis using the STRING Enrichment application (FDR ≤ 0.05) revealed association of Depdc5 PPI-network genes with autophagy, chromatin organization, vesicle-mediated transport, and neurite morphology (axonogenesis, synapse organization), ribonucleoprotein complex biogenesis, and tRNA metabolic process. Depdc5 and its direct neighbors in the PPI network were associated with TORC1 signaling. Given the far-reaching effects of mTOR signaling on metabolic regulation and autophagy, its involvement in ageing and proposed involvement in neurodegeneration, we propose this may be a central regulator behind translatome changes we observed in SST+ neurons in familial PrD. Taken together, this analysis indicates that for both diseases SST+ neurons show the largest response with TORC1 signaling posing a potential underlying regulatory mechanism.