Eirini Kanata, Franc Llorens, Dimitra Dafou, Athanasios Dimitriadis, Katrin Thüne, Konstantinos Xanthopoulos, Nikolaos Bekas, Juan Carlos Espinosa, Matthias Schmitz, Alba Marín-Moreno, Vincenzo Capece, Orr Shormoni, Olivier Andréoletti, Stefan Bonn, Juan María Torres, Isidre Ferrer, Inga Zerr, and Theodoros Sklaviadis
Edited by Bruce S. McEwen, The Rockefeller University, New York, NY, and approved August 13, 2019 (received for review April 13, 2018)
Prion diseases are fatal neurodegenerative disorders characterized by rapidly progressive dementia. Sporadic Creutzfeldt–Jakob disease (sCJD) is the most prevalent. We report that, specific gene-expression alterations utilizing a reliable in vivo mouse model (tg340-PRNP129MM) with sCJD MM1 subtype, correlate with human disease manifestations in the brain cortex related to disease progression. RNA-editing functions mediated by the APOBEC and ADAR deaminases possibly affecting protein expression necessary for normal brain function, are altered in disease stages. Our data provide powerful evidence, derived from a humanized sCJD mouse model and human autopsy material, discerning the critical role of gene expression and RNA-editing signatures, introducing disease-associated targets that can be extrapolated in other neurodegenerative disorders with common clinical and molecular features.
Prion diseases are fatal neurodegenerative disorders caused by misfolding of the normal prion protein into an infectious cellular pathogen. Clinically characterized by rapidly progressive dementia and accounting for 85% of human prion disease cases, sporadic Creutzfeldt–Jakob disease (sCJD) is the prevalent human prion disease. Although sCJD neuropathological hallmarks are well-known, associated molecular alterations are elusive due to rapid progression and absence of preclinical stages. To investigate transcriptome alterations during disease progression, we utilized tg340-PRNP129MM mice infected with postmortem material from sCJD patients of the most susceptible genotype (MM1 subtype), a sCJD model that faithfully recapitulates the molecular and pathological alterations of the human disease. Here we report that transcriptomic analyses from brain cortex in the context of disease progression, reveal epitranscriptomic alterations (specifically altered RNA edited pathway profiles, eg., ER stress, lysosome) that are characteristic and possibly protective mainly for preclinical and clinical disease stages. Our results implicate regulatory epitranscriptomic mechanisms in prion disease neuropathogenesis, whereby RNA-editing targets in a humanized sCJD mouse model were confirmed in pathological human autopsy material.
Sporadic Creutzfeldt–Jakob disease (sCJD) is a fatal human prion disease characterized by the conversion of the cellular prion protein (PrPC) into its pathological conformer (PrPSc), forming protein aggregates in the brain. Pathological hallmarks include gliosis, massive neuronal loss, and spongiform degeneration (1). sCJD is a heterogeneous disease, with different clinico-pathological outcomes (subtypes). sCJD sybtypes are associated with: 1) SNPs at the prion protein-coding gene (PRNP), especially with a SNP at codon 129 resulting in either Met or Val at the corresponding position of the protein; and 2) with PrPSc protease-resistant patterns, classified as type 1 and 2, with the latter displaying a lower molecular mass, proteinase K- (PK) resistant, immunoreactive band of unglycosylated PrPSc at 19 kDa. The most prevalent sCJD subtype (40 to 60% of all sCJD cases), designated MM1, corresponds to PrPC methionine homozygotes displaying type 1 PrPSc. Disease manifestations include cognitive impairment, gait or limb ataxia, aphasia, mental, and visual signs, with a mean disease duration of 4 mo (2).
Prion transmission, infectivity, and mechanisms triggered by the conversion of PrPC to PrPSc are still unknown. Advances in high-throughput transcriptomics and proteomics have identified molecular signatures associated with prion infection. Studies on in vivo prion models are crucial for disease mechanism delineation, since human disease progresses rapidly and only postmortem tissue is available. Molecular data representative of prion disease models are lacking, especially due to noncomprehensive characterization of highly related scrapie and bovine spongiform encephalopathy or other species prion diseases, by inoculation with adapted prion strains (3⇓⇓–6). Such diseases only partially recapitulate the molecular and neuropathological sCJD hallmarks, suggesting an urgent need for novel molecular markers describing early and later disease stages.
We utilized a mouse model of the sCJD subtype MM1, which faithfully recapitulates human disease pathology in a regional- and subtype-dependent manner (7, 8), to investigate the regional- and temporal-dependent disease transcriptome. Transcriptomic profiles revealed regulation of both previously reported prion-associated and novel sCJD-specific related genes. We then confirmed these epitranscriptomic (RNA-editing events) findings in sCJD-affected human brain autopsy material. Our results indicate transcriptome variation, especially at preclinical disease stages of prion-induced pathology, that could clarify contributing mechanisms (i.e., lysosomal, endoplasmic reticulum stress processes) and identify potential therapeutic strategies.
Differential Global Gene-Expression Analysis in the sCJD Mouse Model (tg340PRNP129MM).
We established differential transcriptomic and epitranscriptomic profiles representative of the cortical region of sCJD tg340-PRNP129MM (tg340) mice at preclinical (120 d postinoculation [dpi]) and clinical (180 dpi) disease stages. Mean survival time was 199 ± 7.9 dpi, while 160 dpi was considered as an early clinical disease stage. Heat map and principle component analysis revealed clear clustering for control versus sCJD during infection (SI Appendix, Figs. S1 and S2). Differentially expressed (DE) were considered genes displaying a false-discovery rate (FDR)-adjusted P ≤ 0.05 and a log2 fold-change (FC) cutoff value ≥0.5 (absolute value). Since any FC cutoff would be entirely arbitrary, we selected a target follow-up strategy combining the ranking of all genes based on their (absolute log) FC and the known/believed genes to be involved in the pathogenesis process. At 120 dpi 1,356 genes were identified as DE and at 180 dpi 655; 58 genes displayed deregulation toward the same direction at both time points (38 up-regulated and 20 down-regulated) (SI Appendix, Fig. S3). Preclinical-stage profiles involved neuronal and synaptic pathways, as well as signaling cascades associated with oxidative or endoplasmic reticulum (ER) stress. Clinical disease-stage profiles were relevant to cell survival, proliferation, differentiation, lysosome function, and immune system (SI Appendix, Fig. S4 and Table S1). DE profiles of prion-infected mice (RML, Me7, 22A, 79A, 139A, 22L scrapie strains) (9⇓⇓⇓⇓⇓–15) or sCJD patients’ postmortem brain tissue (16, 17) from gene-expression datasets, revealed common patterns of deregulation between our model and at least 1 of the analyzed murine prion models at preclinical and clinical disease stages (Fig. 1, SI Appendix, Tables S2 and S3, and Dataset S1). We detected similar profile alterations between our study and human sCJD cases. Interestingly, some genes—including Dbi, Rgs4, Abca1, and Gfap—were deregulated similarly in our study, in human sCJD cases and at least 2 scrapie-murine models (Dataset S1).
Correlation of sCJD-mice gene-expression alterations at preclinical and clinical disease stages with corresponding alterations determined in scrapie-murine models. Venn diagrams displaying the overlap of gene-expression alterations at preclinical (A) and clinical (B) stages among different models of prion diseases. Significant overlaps are observed between sCJD-mice and the tested scrapie models. For a complete list of overlapping genes, please refer to SI Appendix, Tables S2 and S3 and Dataset S1.
Gene-Expression Validation in the sCJD Mouse Model and in sCJD Cases.
Twenty-two DE genes, novel and previously identified, were selected for qPCR validation in mouse cortical tissues (SI Appendix, Table S4). Comparison of the differential gene expression in sCJD and control tg340 mice with our RNA-sequencing (RNA-seq) data analysis revealed a strong correlation (R2 and Pearson correlation coefficient [r] were R2 = 0.4567 and r = 0.6758, P R2 = 0.5817, r = 0.7627, P Fig. 2). Next, 14 genes—including PRDX6, A2M, CD44, VIM (Vimentin), HMOX1, CD9, LAMP5, PLIN4, NGEF, SPP1, HSPB1 (hsp27), DSP, SOCS3, and APOE—were cross-validated by qPCR in human postmortem tissue with matching control samples (Fig. 3). We then validated differential protein expression of 9 proteins, including desmoplakin, c-Jun, CD44, aldehyde dehydrogenase 1 family member A1 (Aldh1a1), vimentin, insulin-like growth factor-1 (IGF1), peroxiredoxin 6 (Prxd6), heme oxygenase 1 (HMOX-1), and cystatin C in human postmortem disease with matching control samples (Fig. 4), and of desmoplakin, cathepsin D, CD44, heme oxygenase 1 (Hmox-1), clusterin, the proto-oncogene c-Jun, Aldh1a1, vimentin, and Prxd6, in sCJD and control tg340 mice (Fig. 5). As negative control, we analyzed the levels of early growth response 1 (EGR-1), which was found exclusively regulated at preclinical stages in the sCJD mouse model, and accordingly its expression levels were not altered in sCJD postportem tissue (Fig. 4).
Correlation of RNA-seq data with RT-qPCR experimental validations in sCJD (tg340) mice. Correlation of gene-expression values between RNA-seq and RT-qPCR. R2 and Pearson correlation coefficients (r) for each time point were as follows: (A) 120 dpi: R2 = 0.4567, r = 0.6758; (B) 180 dpi: R2 = 0.5817, r = 0.7627. Statistically significant correlation between in silico and experimental analyses was detected for both time points (***P
Experimental cross-validation of RNA-seq data in postmortem sCJD MM1 subtype cases by RT-qPCR. RNAs from control (n = 12) and postmortem sCJD patients of the MM1 subtype (frontal cortex region n = 8) were retrotranscribed and tested using Taqman probes to determine human gene expression; human genes were selected based on the corresponding analysis in the sCJD (tg340) mouse model (clinical disease, 180 dpi). GAPDH was used for normalization. Following a D’Agostino and Pearson test to verify the normality of the distribution, mean FC values were compared in disease and control cases using the Mann–Whitney U test. *P P P
Experimental cross-validation of RNA-seq data in postmortem sCJD MM1 subtype cases by Western blot. Western blot analysis of Desmoplakin, c-Jun, CD44, Aldh1a1, vimentin, IGF1, EGR-1, PRDX6, HMOX-1, IBA-1, and cystatin C in sCJD patient postmortem brain tissue. GAPDH was used as loading control. Densitometries derived from the quantification of 6 cases per group are shown. Following a D’Agostino and Pearson test to verify the normality of the distribution, mean FC values were compared in disease and control cases using the Mann–Whitney U test. *P P P
Experimental validation of RNA-seq data in sCJD (tg340) mice at protein level. Western blot analysis of clusterin, desmoplakin, c-Jun, cathepsin D, CD44, Aldh1a1, vimentin, Hmox-1, Egr-1, Prdx6, and Iba-1at preclinical (120 dpi) at clinical (180 dpi) disease stages in the tg340 mice inoculated with brain homogenates from control and postmortem sCJD cases of the MM1 disease subtype. Following a D’Agostino and Pearson test to verify the normality of the distribution, mean FC values were compared between diseased and control animals for each time point using the Mann–Whitney U test. GAPDH was used as a loading control protein. Densitometries derived from the quantification of 3 animals per group are shown. *P P P
Editing Analysis in the sCJD Mouse Model (tg340PRNP129MM).
RNA editing and correlation of editing events in the brain with neurodegenerative (ND) and neurological diseases (18⇓⇓–21), suggest differences in global ADAR and APOBEC editome profiles between control and sCJD animals. We determined global RNA-editing profiles (editomes) in control and sCJD mice at different disease stages using an in-house pipeline (SI Appendix, Fig. S5; editome profile lists are provided in Dataset S2; diagrammatic representation of RNA editomes per animal group and time-point is shown in SI Appendix, Fig. S6). DNA–RNA differences and the number of events for the 12 types of differences between RNA reads and genomic DNA sequences showed that putative editing is highly concordant between biological replicates. Strikingly, over 90% of these sites correspond to A-to-G and C-to-T differences, consistent with A-to-I and C-to-U editing. This observation supports the existing knowledge that A-to-I editing is the primary type of RNA editing. Other types of differences are much less abundant. ADAR-mediated editing is predominant in the mouse cortex compared to APOBEC-mediated editing (SI Appendix, Fig. S6, ring A). Similar genomic distributions of editing events were observed for all tested groups and time-points. Most editing events were detected in intergenic regions, 3′UTRs, exons, or introns, while fewer were in 5′UTRs and other noncoding regions (SI Appendix, Fig. S6, rings B and C). Profiling of control and sCJD animals’ editomes for each time point and clinical stage revealed high similarities between the 2 animal groups and only a few transcripts were uniquely edited either in control or in sCJD. In contrast to global editomes, differentially edited transcript analysis showed APOBEC-mediated editing is overrepresented (∼49 to 60%) (Dataset S3) in almost all phenotypic groups and time points (Fig. 6, ring A). Dissimilar genomic distributions highlight an increase of 3′UTR editing events in sCJD animals compared to controls at clinical disease (Fig. 6, rings B and C). The observed increase in 3′UTR editing is mainly attributed to APOBEC, even though A-to-I editing is also increased at this stage. Differentially edited transcripts at clinical disease displayed alterations in other noncoding regions, while few edited exons (∼5%) were detected among the differentially edited transcripts at this time point. At preclinical disease stage, a clear preference of ADARs to edit intronic and 3′UTR regions was observed, while APOBEC editing was mostly detected in exonic and intergenic regions (Fig. 6, rings B and C). Only a few edited transcripts were unique either in control (30% in preclinical and 27% at clinical stage) or in sCJD animals (28% at preclinical and 11% at clinical disease).
Differential editing profiles of control and sCJD-tg340 mice at preclinical and clinical disease stages. Differential editing was identified as indicated in the text. Ring A: Bars indicate the absolute number of RNA editing sites displaying differential editing for each phenotype (cntr, sCJD), RNA-editing mediating enzyme (ADAR, APOBEC), and disease time point (preclinical: 120 dpi, clinical: 180 dpi). Ring B: Genomic distribution (color code legend on top of the graph) of differential editing events for each phenotype, RNA-editing mediating enzyme and disease time point. The bars indicate the absolute number of positions identified. Ring C: Percent genomic distribution (color code legend on top of the graph) of differential editing events for each phenotype, RNA-editing mediating enzyme and disease time-point. The bars visualize the distribution of RNA-editing positions across functional gene regions calculated as a percentage of the total RNA-editing positions identified in each phenotype. For a detailed list of differentially editing events between the studied animal groups and time points, refer to Dataset S3.
Gene ontology (GO) and pathway analyses of the differential editomes at preclinical and clinical disease stages revealed dendrite-associated transcripts involved in cell adhesion and junction signaling at preclinical disease. At clinical disease, immune related transcripts were significantly represented, but no pathway was significantly enriched.
RNA Editing Validation in the sCJD Mouse Model and in sCJD Cases.
Uniquely edited transcripts in each group, or edited at the same position at both groups yet at differential editing rates, were validated as a proof-of-concept to our in silico RNA editing profiling. These included Fkrp, Plekhm2, and Sidt2 (preclinical), as well as Sec61a1 and Rragd (clinical) in mice (Table 1 and SI Appendix, Figs. S7 and S8). Next, we cross-validated differentially edited mouse targets in human postmortem tissue, including 2 APOBEC (PAQR8, B2M) and 2 ADAR targets (CTSS, RRAGD) (Table 1). Sequencing analysis of PCR products or adequate number of clones harboring the transcript regions under study and comparison with the corresponding genomic reference sequence, confirmed the presence of differential editing between control and sCJD animals for most of the tested targets (Table 1 and SI Appendix, Figs. S7 and S8). We confirmed APOBEC differential editing at the preclinical disease stage in Fkrp and Sidt2 transcripts; Fkrp displays editing of C at position Chr7:16809815 (relative to the mm10 reference genome) only in the disease group (SI Appendix, Fig. S7). Furthermore, C editing at position Chr9:45941373 in the Sidt2 transcript is significantly reduced in sCJD animals compared to controls. Sidt2 also undergoes ADAR-mediated editing in A at positions Chr9:45939224 and Chr9:45939169, and similarly to APOBEC editing, displays reduced editing rates in the diseased animals (SI Appendix, Fig. S8). At the clinical disease stage, we identified differential A-to-I editing in the murine Rragd 3′UTR (Chr4:33020417), showing increased editing levels in sCJD-tg340 animals (SI Appendix, Fig. S8).
Summary of potential disease-associated RNA editing events that have been experimentally verified in the tg340-sCJD mouse model and in human sCJD cases in this study
In human samples, we identified APOBEC-catalyzed RNA–DNA sequence differences (RDDs) in the human PAQR8 3′UTR (Chr6:52406406, coordinate given relative to hg38), presenting higher editing in control samples (Table 1 and SI Appendix, Fig. S9). Additionally, we identified a C-to-U editing event in the sCJD B2M transcript (Chr15: 44713183), at a frequency of 10% (3 of 30 tested clones). The same B2M position appears as “pre-edited” at the genomic DNA (gDNA) level of the tested control sample (Table 1 and SI Appendix, Fig. S9). Furthermore, we identified increased A-to-I editing in the 3′UTRs of human CTSS (significantly different sites at positions Chr1:150731266, Chr1:150731240, Chr1:150731200, Chr1:150731196, Chr1:150731185, Chr1:150731144, Chr1:150731143, Chr1:150731137, Chr1:150731136) (see also Table 1 and SI Appendix, Figs. S10–S12) and RRAGD (significantly different sites at positions Chr6:89366492, Chr6:89366490, Chr6:89366416, Chr6:89366335, Chr6:89366271) transcripts in sCJD cases postmortem brain tissue compared to control samples (Table 1 and SI Appendix, Fig. S12).
Utilizing a humanized sCJD mouse model, recapitulating faithfully human disease, we studied preclinical and clinical disease stages. We used RNA-seq analysis to identify transcriptome and RNA editome alterations during disease progression. DE genes at both preclinical and clinical disease stages included previously identified and similarly deregulated genes in different scrapie-infected mice. Comparison of our sCJD mice data with human sCJD case transcriptomic data identified 19 commonly regulated genes, including Abca1, Capg, Car10, Cst3, Dbi, Dhcr24, Diras2, Gabrd, Gfap, Itpkb, Itpr1, Lrig1, Mt1, Npc2, Pcp4, Rab31, Rgs4, Rin2, and Tnfrsf1a, indicating a common expression profile between the utilized sCJD mouse model and human sCJD. Furthermore, commonly deregulated genes were identified between sCJD mice, human sCJD cases, and scrapie murine models, suggesting significant overlap regarding gene-expression perturbations triggered by different strains of animal (scrapie) and human (sCJD) prion diseases. Remarkably, Abca1, Dbi, Gfap, and Rgs4 presented common deregulation patterns among sCJD mice, human sCJD cases (16, 17), RML mice (9), and Me7/79A/22A mice (14). Collectively, we suggest that different prion strains induce both common and differential responses to their hosts, with the highest variability being associated with preclinical disease stages. Corroborating the notion that common and variable gene-network profiles are induced by different strains at preclinical disease, temporal differences correlating with the pace of infection have been identified regarding the inflammatory responses triggered by strain-specific brain PrPSc accumulation in the thalamus of affected animals (22). Early PrPSc accumulation is believed to induce pathogenetic mechanisms, as well as cellular responses aiming at induction of countervailing cellular responses, which fail to restore cellular homeostasis, resulting in disease progression. Majer et al. (10) reported the induction of cell survival and neurite remodeling pathways in hippocampal neurons of RML-infected mice at preclinical disease and suggested asymptomatic disease as a critical time point for intervention. In contrast, clinical disease-related molecular mechanisms, when extensive degeneration has been established, are more likely to reflect infection. We provide proof-of-principle experimental validations of prion disease-specific transcriptomic profiles at the mRNA and protein level using murine and human sCJD autopsy material. Twenty of 34 DE-validated targets were previously associated with prion diseases, and 14 novel ones are being disease-associated herein (SI Appendix, Table S5). Of notice are: PRDX6, found to attenuate prion-related neuropathology in Prdx6 knockout mice infected with the Me7 scrapie strain (23); TREM2, involved in microglia activation upon prion infection (24); cathepsin D (Ctsd), reported to prevent a-synuclein aggregation and toxicity in in vivo models of Parkinson’s disease (25); and cystatin C (Cts3), found to inhibit Aβ oligomerization in both in vivo Alzheimer’s disease models and in in vitro studies (26), supporting similar mechanisms on PrPSc aggregates in prion diseases. Additionally, we introduced the validated contribution of the epigenetic mechanism of RNA editing in disease-associated molecular targets. Triggered by the highly unknown etiology of sCJD and by reports on the effects of RNA-editing perturbations in neoplastic disease (27⇓⇓⇓–31) and more importantly in neurological and ND disorders (19, 32⇓–34), we established differential editome profiles in the cortex of control and sCJD animals, at preclinical and clinical disease stages, based on an in-house bioinformatics pipeline with stringent criteria to exclude false-positive single-nucleotide variants. To limit false-positive in silico RNA-editing identification and increase the validity of our analysis, we have introduced multiple algorithms and filtering steps in addition to well-accepted in the field quality control analyses, including the assessment of whole RDD mismatch spectrum representation and estimation of editing event enrichment in repetitive regions. Furthermore, we searched for potential pseudogenes among the targets detected to undergo RNA editing, which ruled out the possibility of detecting false-positive RNA-editing events due to duplications. We and others describe ADAR-mediated editing as prevalent in the mouse cortex (>50% of total editing events) (32, 35) and identify a significant representation of C-to-U RDDs occurring as a result of APOBECs’ catalytic action. This places our study among those reporting significant APOBEC-mediated RNA editing in murine brain, including only 1 recent study by Cole et al. (36), which has investigated RNA editing in mouse brain microglia, while other previous APOBEC-induced RNA-editing studies focused on other mouse tissues, in particular small intestine, liver (37, 38), and macrophages (39). The validity of our approach is highlighted by the overlap between our results and previously reported editing sites in similar tissues. Several editing events identified here have been reported in mice, including conserved editing events in the mammalian brain and experimentally validated alterations in murine tissues (SI Appendix, Table S6).
Five sites, residing in the 3′UTR of B2m (chr2: 122152682, chr2: 122152902), Cisd2 (chr3: 135406870), Rragd (chr3: 33020417), and downstream of Smim14 (chr5: 65448046) displayed similar editing-rate alterations between our sCJD and an epilepsy mouse model (32). Differential editing characterized by reverse-frequency alterations between these 2 models was observed at the 3′UTR of the Nt5dc3 (chr10: 86837257), Ankrd28 (chr14: 31700662), and Gpm6b (chrX: 166387925) transcripts, as well as in Gria2 exon 11 (chr3: 80706908, synonymous change); the latter is also differentially edited in Alzheimer’s disease patients (19).
Our data identified reduced differential editing with disease progression. Gene expression and protein level analysis of the main RNA-editing enzymes in mouse (ADAR1, ADAR2, APOBEC1, and APOBEC3) did not reveal any significant changes during disease progression between control and sCJD animals, with the exception of decreased ADAR2 mRNA expression in affected animals at clinical disease (P = 0.0231). Similarly, reduced ADAR2 protein expression was observed without reaching statistical significance (P = 0.5183). Our data are in agreement with Srivastava et al. (32), who also reported reduced ADAR2 expression in epileptic mice. This finding correlates with the observed RNA-editing reduction during disease progression reported in our study, even though additional factors possibly affecting ADAR2 access to targeted RNAs may contribute to the observed difference.
The finding that global RNA editing is reduced during disease progression enhances our strategy to focus on preclinical RNA-editing perturbations for delineating driver disease events and identifying novel potentially therapeutic targets. We identified the cAMP signaling pathway and neuronal transcripts as enriched in differentially edited transcripts at the preclinical disease stage, additionally supported by previous studies reporting synaptic dysfunction as an early event in prion pathogenesis (40, 41). Further studies could focus on differentially edited transcripts, such as Gria2, Gria4, Pde4d (involved in synaptic plasticity), and Sidt2 [implicated in lysosome and or Golgi related processes, all of which are impaired in prion diseases (42⇓⇓⇓⇓⇓–48)]. We have identified individual RNA-editing events in clinical disease stage, supporting the notion that they could represent key molecular disease triggers, even though significantly enriched pathways were not observed. Differentially edited transcripts, including the microglia enriched and immune related B2m, as well as ER-related transcripts (49), such as Sez6l2 (50) and Sec61a1 (51), could represent candidate targets for future functional RNA-editing validation studies.
Regarding RNA-editing analyses in human autopsy samples, we cross-validated differentially edited targets (CTSS, RRAGD) identified from the analysis of the utilized mouse model; furthermore, we used a small cohort of RNA samples for expression analysis at the gene level (n = 5 control, n = 3 sCJD) and valuable brain homogenates for expression analysis at the protein level (n = 4 control, n = 4 sCJD), in order to determine the expression levels of the main RNA-mediating enzymes in humans, namely ADAR1, ADAR2, APOBEC1, and APOBEC3G. We did not detect significant differences in the expression levels of these key editing enzymes between sCJD and control samples, with the exception of ADAR2, which could not be detected at the protein level in the control samples. This result seems to be in contrast to the corresponding analysis of the utilized murine model, possibly suggesting differences regarding the main editing enzyme expression levels between our model and human disease. Even though we cannot exclude the possibility of the occurrence of such differences, we should interpret these data by taking into account the rarity of available samples and the variability in their processing and storage.
The human cross-validated RNA-editing targets have been reported by others as possible contributors to ND disease pathogenesis. Specifically, cathepsins and particularly CTSS have been implicated in prion disease pathogenesis (8), as well as in other NDs (52), including multiple sclerosis (53) and Alzheimer’s disease (54). Additionally, RRAGD is involved in amino acid signaling induced mammalian target of rapamycin (mTRORC1) pathway activation (55); mTORC1 controls organism growth through regulation of cellular processes, including autophagy and lysosome biogenesis, and is deregulated in NDs. RRAGD participates in endosome/lysosome biogenesis and thus in cellular clearance, suggesting that differential editing in this target is associated with disease pathogenetic mechanisms referring to cellular clearance processes (56). Finally, the PAQR8 transcript, validated as differentially edited in human sCJD autopsy brain tissue, encodes a member of the adipoQ progesterone receptor family and has been associated with epilepsy (57), suggesting that its aberrant forms could be involved in other brain disorders, including prion diseases. Functional importance of the newly identified and validated targets in human tissue is highlighted by the predicted changes in microRNA binding sites with probable effects on the altered gene expression of the mRNA targets (CTSS, RRAGD, PAQR8) (SI Appendix, Fig. S13).
Our study utilized a faithful sCJD murine model to identify disease progression transcriptomic profiles and corresponding epitrascriptomic (RNA editing) signatures in human autopsy material. Taking into account that RNA editing displays species-specific differences and keeping in mind potential limitations of the murine model utilized in this study regarding human disease aspects representation, we provide experimental evidences that provide proof-of-principle validations of novel disease-associated targets in human.
Materials and Methods
Anticathepsin D, anti–c-Jun, and anti–Egr-1 were from Santa Cruz; anti–Iba-1 was from Wako; anti-CD44 was from Cell Signaling; antivimentin was from Dako; anti-GAPDH, antidesmoplakin, anti– HMOX-1, anti-Prdx6, anti- ALDH1A1, and anti- IGF1 were from Abcam; antiheat-shock protein 27 (hsp27) was from Novus; anticlusterin was from Chemicon, anticystatin C was from Upstate; and anti–α-actin was from Sigma.
Brain tissue was obtained from the Institute of Neuropathology Brain Bank (HUB-ICO-IDIBELL Biobank) following pertinent guidelines of local ethics committees. All procedures involving human participants were in accordance to the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Neuropathological examination and characterization was carried out in every case on paraffin-embedded samples. Detailed neuropathology, inflammatory profiling, and demographics of the cohort is described elsewhere (58, 59). Control samples did not present any infectious, metabolic, or neoplastic disease. No correlation between postmortem delay or sample storage time and levels of analyzed proteins and mRNA were observed. Age- and gender-matched cases were used. For qPCR analysis, 12 control samples (n = 12, 6 males/6 females) corresponding to individuals of mean age 65 ± 9 y and 12 postmortem sCJD samples (n = 12, 7 males/5 females) from patients of the MM1 disease subtype with mean age 66 ± 8 y were used. For Western blot analysis, 6 controls (n = 6, 3 males/3 females, mean age 63 ± 11 y) and 6 postmortem sCJD MM1 (n = 6, 4 males/2 females, mean age 65 ± 9 y) samples were analyzed.
sCJD Mice Model–tg340-PRNP129MM
The tg340 mouse line expressing about 4-fold level of human PrP MM129 (Met at codon 129) on a mouse PrP-null background was generated as described previously (60). Control or postmortem brain tissues (10% [wt/vol] homogenates) from the MM1 subtype of a sCJD patient were inoculated in 6- to 10-wk-old mice in the right parietal lobe, using a 25-gauge disposable hypodermic needle. Mice were observed daily and their neurological status was assessed weekly. Animals were killed at presymptomatic (120 dpi) and symptomatic (180 dpi) stages. Survival time was calculated and expressed as the mean of the survival day postinoculation of all mice scoring positive for PrPSc. Infection rate was determined as the proportion of mice scoring positive for PrPSc from all inoculated mice. All animal experiments were performed in compliance with the French national guidelines, in accordance with the European Community Council Directive 86/609/EEC. Experimental protocol was approved by the Institut National de la Recherche Agronomique Toulouse/École Nationale Vétérinaire de Toulouse ethics committee. For RNA-seq and qPCR analysis, 4 animals per group and time point were used; for immunoblotting, samples from 3 to 4 animals per group and time point were analyzed.
RNA was purified using the miRVANA RNA isolation kit, following the manufacturer’s protocol. RNAs were treated with DNase Set (Qiagen) for 15 min to eliminate genomic DNA contamination. RNA integrity was assessed by its corresponding RNA Integrity Number (RIN value), determined with the Agilent 2100 Bioanalyzer (Agilent).
RNA-Seq and Differential Gene-Expression Analysis.
Libraries were prepared using the TruSeq RNA Sample Preparation v2 kit (Illumina). Library quality was assessed with an Agilent 2100 Bioanalyzer and a Qubit dsDNA HS Assay Kit. RNA sequencing was performed as described in ref. 61. All original RNA-seq data were deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus, GEO accession no. GSE90977. Briefly, RNA-seq data were subjected to an in-house quality control workflow. Read quality was assessed using FastQC (v0.10.1), to identify sequencing cycles with low average quality, adapter contamination, or repetitive sequences from PCR amplification. Alignment quality was analyzed using SAMtools flagstat (v0.1.18), with default parameters. Sequences were aligned to the genome using gapped alignment, as RNA transcripts are subject to splicing and reads might therefore span 2 distant exons. Reads were aligned to the whole Mus musculus mm10 genome using STAR aligner (62) (2.3.0e_r291) with default options, generating mapping files (BAM format). Read counts for all genes and all exons (Ensembl annotation v72) were obtained using FeaturesCount (http://bioinf.wehi.edu.au/featureCounts/). For data visualization, BAM files were converted into WIG and BigWig files using the MEDIPS “MEDIPS.exportWIG” function with a window of 50 bp and RPM normalization. For the differential expression analysis, read counts generated with FeaturesCount were compared between groups using DESeq2 (63). Genes with a ≥ 0.5 log2 FC cutoff and FDR-adjusted P ≤ to 0.05 were considered as differentially expressed.
Differential Gene-Expression Validations.
Real-time qPCR (RT-qPCR).
RNA samples were reverse transcribed using the High Capacity cDNA Archive kit (Applied Biosystems). qPCR assays were performed in duplicate on cDNA samples in a LightCycler 480 System (Roche). Reactions were set up using 20 × TaqMan Gene Expression Assays (SI Appendix, Table S7) and 2xTaqMan Universal PCR Master Mix (Applied Biosystems). The FC was determined using the equation 2−ΔΔCT. Mean FC values were analyzed with appropriate statistical tests, indicated in the corresponding figures, using GraphPad Prism 6.01.
Human and mice tissues were lysed in Lysis Buffer containing 100 mM Tris pH 7, 100 mM NaCl, 10 mM EDTA, 0.5% Nonidet P-40 and 0.5% sodium deoxycolate plus protease and phosphatase inhibitors. After centrifugation at 14.000 × g for 20 min at 4 °C, supernatants were quantified for protein concentration (Bradford, Bio-Rad), mixed with SDS/PAGE sample buffer, boiled, and subjected to 8 to 15% SDS/PAGE. Gels were transferred onto PVDF membranes and processed for specific immunodetection using an electrochemiluminescence reagent. For Western blot comparative analysis 10 human cases and 3 mice samples per condition were analyzed.
For qPCR and Western blot experiments, normality distribution was analyzed following the D’Agostino and Pearson test. Mann–Whitney U tests were used to compare 2 groups of samples. GraphPad Prism 6.01 was used for statistical calculations. Differences between groups were considered statistically significant at *P P P
Bioinformatics Pipelines for Identification of RNA Editing Events.
Quality control and processing of RNA-seq raw data.
High-depth RNA-seq data (26 to 58 million reads per sample, average 33 million reads per sample, GEO accession no. GSE90977) were subjected to quality control analyses using the FastQC software. Adapter sequences, as well as the first 6 bases of the 5′ end and the first 3 bases of the 3′ end of each read were trimmed utilizing the Trim Galore software (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). Adapter sequence contamination was removed and false-positive RNA editing calls originating from random hexamer primer bias were minimized (64, 65), trimming excluded reads displaying lengths lower than 20 bases. Processed reads were aligned against the mouse reference genome (mm10), using TopHat2 software (66) (https://ccb.jhu.edu/software/tophat/index.shtml). Higher mapping rates were allowed by providing a set of known gene annotations for the reference mm10 genome. Multimapping reads were excluded and up to 3 mismatches per read were allowed. A high proportion of acquired reads (98.7 to 99%, average 98.8%) were mapped to the murine reference sequence. Pertinent data for each of the analyzed samples is provided in SI Appendix, Table S8. Calling of single-nucleotide variants for the identification of possible RNA-editing events was performed utilizing 2 approaches based on the REDItools (67) and VarScan (68) suites, respectively. For the REDItools suite, an initial Blat Correction analysis (using the script REDItoolBlatCorrection.py) was performed to identify ambiguous aligning reads and the REDItoolDenovo.py script was subsequently used to call possible RNA editing events. For the VarScan suite, the function mpileup2snp was used to process pile-up files generated by the SAMtools software (http://samtools.sourceforge.net/pileup.shtml). True RNA-editing events and exclusion of false positives was achieved by: 1) Mapping and base quality thresholds of 20 and 25, respectively; 2) a minimum coverage of 10 reads, of which at least 3 contain the variation; 3) a P value lower than 0.05 for FDR for REDItoolDenovo.py (Benjamini) and Fisher’s exact test for VarScan; and 4) minimum editing frequency of 0.01. Annotation was performed using the ANNOVAR software (http://annovar.openbioinformatics.org/en/latest/) (69) providing dbSNP v142. Strand information was extracted from ENSEMBL.gtf files (mouse gene set) and annotated using custom R scripts and the refGenome R package. SNP exclusion, strand selection, and further statistical analysis for extracting and processing possible C-to-U and A-to-I RNA-editing events was performed using custom R scripts (procedure summarized in SI Appendix, Fig. S5). A github repository containing the in-house scripts used can be accessed through the link https://github.com/athanadd/CJD-mice.
Global RNA-editing profiles establishment and identification of differentially edited transcripts between control and sCJD animal groups.
ADAR and APOBEC mediated RNA editing events predicted by Varscan and RedITools within each phenotype group (control, sCJD) and time point (120 dpi, 180 dpi) were used for establishing the corresponding ADAR and APOBEC mediated editomes. To minimize false-positive RNA-editing events, an arbitrary threshold value of 2 samples including a predicted RNA editing event within every sample group was set. Acquired profiles were visualized by means of indexes depicting the distribution of editing events within each phenotype group per time point. Dual phenotype comparisons were performed on RNA-editing frequencies for each time point to identify statistically significant differences in RNA-editing frequency patterns between control and sCJD animals utilizing a 2-tailed unpaired Student’s t test. P values of P ≤ 0.05 when dual phenotype comparisons were performed.
Reverse-transcription reactions were performed using 500 ng of total RNA and both the oligodT and random hexamer primers included in the PrimeScript kit (TAKARA), following the manufacturer’s instructions. Amplification of an 800- to 900-bp transcript region including the positions of interest was performed using the prepared cDNAs and the high-fildelity Q5 DNA polymerase (New England Biolabs), according to the manufacturer’s instructions. Primer sequences for each target are provided in SI Appendix, Table S9. In cases where editing events were predicted to be present in 1 of the tested groups (Fkrp, Plekhm2, and Sec61a1), PCR products were directly sequenced following PCR clean-up (PCR clean-up and gel-extraction kit, Macherey-Nagel). In cases where editing events were predicted to be present in both groups (Sidt2, CTSS, B2M, and PAQR8), a further step of TA cloning was applied before sequencing analysis. For cloning in the pDrive vector (Qiagen), PCR products were subjected to an additional A-tailing step, followed by clean-up and ligation in the pDrive vector using a 20× excess molar ratio of insert relative to the vector. Plasmids corresponding to 30 clones of transformed TOP10 Escherichia coli cells in each case were prepared using the Nucleospin plasmid kit (Macherey-Nagel) and sequenced at both directions. All sequencing reactions were performed by Cemia (https://cemia.eu/) using the v3.1 BigDye Terminator chemistry (ABI) and were analyzed on a 3730 genetic analyzer (ABI). Raw sequencing data were processed with the Bioedit software v7.09.0. RDDs corresponding to RNA-editing events were identified through alignment of cDNA sequencing data with the corresponding reference sequence. Murine and human genomic DNA regions corresponding to the candidate targets were also amplified, sequenced and used as reference sequences for RDDs identification.
Special thanks to Mr. Christos Nikolaou for his long-term technical and administrative contribution to the Neurodegenerative Disease Group at Aristotle University; also to Dr. Alexandros Beis for the creative conversations and common experiences. We also thank the Scientific Computing Center of the Aristotle University of Thessaloniki and the national HellasGrid Certification Authority, through which the RNA-editing identification pipeline was run, for their support; Dr. Alexandra Charalampidou for providing technical support in computational analysis; Ms. Aikaterini Kerezoglou for providing administrative support; and Ms. Georgia Christoforidou (School of Biology, Aristotle University of Thessaloniki) for providing technical support in experimental validations of RNA-editing events. This study was supported by the Research Funding program ARISTEIA II (Grant RNA edit 3739), cofinanced by the European Union (European Social Fund-ESF) and Greek national funds through the Operational Program “Education and Lifelong Learning” of the National Strategic Reference Framework (to T.S.); the Alliance BioSecure Foundation project “RNA editing in CJD” (T.S. and I.Z.); Federal Ministry of Education and Research grants within the German Network for Degenerative Dementia (to I.Z.); the Red Nacional de Priones AGL2015-71764-REDT-MINECO (to F.L., I.Z., J.M.T., and I.F.); the Spanish Ministry of Health-Instituto Carlos III/Fondo Social Europeo CP16/00041 (to F.L.); by the IKYDA Greek-German collaboration Project 57260006 (to F.L., I.Z., D.D., and T.S.); and in part by Grant AGL2016-78054-R (AEI/FEDER, UE). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Author contributions: E.K., F.L., I.Z., and T.S. designed research; E.K., F.L., D.D., K.T., N.B., J.C.E., M.S., A.M.-M., O.A., and J.M.T. performed research; O.A., J.M.T., and I.F. contributed new reagents/analytic tools; E.K., D.D., A.D., K.X., V.C., O.S., and S.B. analyzed data; and E.K., D.D., K.X., and T.S. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Data deposition: All original RNA-seq data were deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession no. GSE90977).
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1803521116/-/DCSupplemental.
- Copyright © 2019 the Author(s). Published by PNAS.