Our analysis of RNA sequence has validated that methylation occurs consistently at functionally important sites across the mitochondrial transcriptome, and that this methylation is highly correlated across the 11 multiallelic sites. Here we show significant hypermethylation at specific sites in the mitochondrial transcriptome (i.e. mt-tRNA p9 sites) in the cerebellar tissue of individuals with Alzheimer’s disease and progressive supranuclear palsy. Similarities in methylation between these two pathologies may suggest a comparable genetic etiology underlying the cerebellar mitochondrial dysfunction that is a hallmark of both diseases (16,17). Interestingly, no significant difference was observed between NC and PA, which may be due to the ambiguity of differentiating healthy from pathological aging. Individuals diagnosed with pathological aging often appear to be cognitively intact despite the presence of neuritic plaques. Similar presentation has also been observed in healthy aged individuals, making it difficult to distinguish between the two diagnoses (18). Moreover, a recent study by Idaghdour and Hodgkinson (2017) revealed significant hypermethylation at mitochondrial p9 sites in cancer. This substantiates the notion that mt-tRNA hypermethylation may be more closely associated with pathology rather than normal aging processes. Whether this hypermethylation is a cause or consequence of pathology is still unclear.
Nevertheless, it is important to acknowledge the technical limitations associated with traditional RNA sequencing workflows. Here we sought to investigate post-transcriptional methylation of mitochondrial tRNAs using total RNA sequence data. Importantly, these RNA molecules are highly folded and small in nature, often making them difficult to capture. Traditional total RNA sequencing protocols, such as that used here often 1) do not enrich for, and 2) usually detail removal of, small RNAs. This means that mature mitochondrial tRNA reads may have been under-represented in this dataset. Therefore, any conclusions made here may not be based on the entire mature mt-tRNA pool. Future studies investigating p9 methylation should aim to use small RNA enrichment protocols, as well as perform cDNA size selection prior to RNA sequencing.
Nuclear genes are associated with p9 methylation
The gene-based genome-wide association analysis revealed that variation in several nuclear encoded genes is significantly associated with methylation at different p9 sites throughout the mitochondrial transcriptome. Two SNPs (rs12343928; rs4877837) mapping to solute carrier family 28 member 3 (SLC28A3) located on chromosome 9 were found to be significantly associated with methylation at p9 site 585. The strength of this result was further substantiated by the clustering of neighbouring SNPs in linkage disequilibrium within the gene region (Figure 1). The protein encoded by this gene is involved in regulating neurotransmission, vascular tone and metabolism of nucleoside drugs, which are processes that are physiologically relevant to our tissue and diseases of interest. Nucleoside drugs are often used to treat age-related diseases such as type 2 diabetes and hypertension, as well as mitochondrial dysfunction in Parkinson’s disease (PD) (19). This is interesting, as PSP pathology also features mitochondrial dysfunction and bears phenotypic similarity to PD, while T2D and hypertension often serve as comorbidities for AD.
A single SNP (rs2034879) that mapped to several genes (SENP8, MYO9A, GRAMD2) on chromosome 15 (likely due to the restrictive window size of ±10kb that was used) reached the suggestive threshold for association with p9 methylation at position 585. SENP8 is known to catalyze pathways associated with neddylation, a post-transcriptional modification analogous to ubiquitination; these pathways are also responsible for the endocytic degradation of APP. Dysfunction of these pathways has been observed in AD leading to characteristic accumulation of APP and Ab (20). MYO9A is an unconventional myosin responsible for the regulation of Rho GTPase activity within neurons, which in turn regulates neuronal morphology and function; for these reasons, the Rho family has been a recent therapeutic target for neurodegenerative diseases such as AD (21). In turn, various Rho GTPases have also been implicated in maintenance of mitochondrial homeostasis and apoptotic signaling (22). GRAMD2 is involved in the organization of endoplasmic reticulum-plasma membrane contact sites (EPCS), which are key modulators of calcium homeostasis (23). Calcium serves as an important regulator of mitochondrial bioenergetics (e.g. activation of different Kreb’s cycle enzymes (24)), dynamics and apoptotic signaling. Subsequent contact sites between the ER and mitochondria, termed mitochondrial associated membranes (MAMs), facilitate uptake of calcium from the ER to the mitochondria (25), although there is not a clear role for GRAMD2 in this process.
Methylation at six of the 11 p9 sites (5520, 7526, 8303, 9999, 10413, 12146) was significantly associated with a single SNP (rs9872864) mapping to two genes TRAIP and IP6K1 located on chromosome 3. TRAIP encodes TRAF interacting protein, an E3 ubiquitin ligase that plays a key role in cell survival and apoptosis. TRAIP serves important roles in cell cycle checkpoints by regulating spindle assembly, appropriate chromosome distribution during cell division, and DNA damage responses (26). Cell cycle checkpoints have been proposed to serve neuroprotective roles; checkpoint dysregulation and cell cycle re-entry of post-mitotic neurons have been observed in a number of tauopathies (27). IP6K1 encodes an inositol phosphokinase responsible for synthesis of 5-diphosphoinositol phentakisphosphate (5-IP7) from hexakisphosphate (IP6). IP6K1 serves as an important upstream regulator for various metabolic processes (e.g. glucose homeostasis, lipolysis), apoptosis and global transcription (28). It has been investigated as a therapeutic target for obesity and type 2 diabetes (29) and variants within the IP6K1 gene region have been previously identified, using tag SNP analysis, as being associated with AD (30). IP6K1 also impacts mitochondrial function by regulating ATP concentration via alteration of the ratio of glycolytic to oxidative phosphorylation (31). Knockout (IP6K1-/-) yeast models have shown decreased mitochondrial respiration, yet increases in ATP, a paradox that is thought to be due to enhanced glycolysis and depletion of metabolic processes requiring ATP (31). Further, its role in lipolysis is pathologically relevant as findings from Wan et al. (32) demonstrate the capacity of Ab accumulation to promote lipolysis, increasing lipid toxicity and lipid peroxidation, leading to subsequent downstream mitochondrial dysfunction (33), a phenotype common to many neurodegenerative diseases.
While several of the genes associated here are loosely connected to mitochondrial function, we did not identify any loci that may directly affect mt-tRNA processing. It is interesting to note that 5 out of the 6 genes identified in the gene-based GWAS had CADD scores >10 (based on the top leading SNP or maximum CADD score of variants within the gene region). In brief, Combined Annotation Dependent Depletion (CADD) scoring is a predictive machine learning tool that allows for estimation of the deleteriousness and/or pathogenicity of causal genetic variants (34). In general, raw CADD scores exceeding 10 are considered to be in the top 10% of deleterious variants in the human genome (34). Our findings here suggest that the genes identified as having significant associations with p9 methylation, may play a role in pathogenesis; though their exact role remains ambiguous at this time.
Hodkinson et al. (15), reported p9 methylation associations with several genes—one of which is MRPP3, a key player in tRNA processing in the mitochondria. We did not replicate the MRPP3 association which is not entirely surprising since their study was conducted in normal adults (40-69 years) without any particular pathological conditions versus our study of aged subjects with neurodegenerative disease. It may follow that the variability in MRPP3 associated with p9 methylation levels described in Hodgkinson et al., is presumably normal, whereas p9 methylation in our cohort is presumably a result of pathological process(es). The resulting genotype associations may be pointing to novel gene variants that are upstream effectors of mitochondrial function. Conversely, it is important to acknowledge the caveat that gene-based association testing was performed together on all individuals of this cohort; the small sample size of each diagnostic category (NC, AD, PSP, PA) left us underpowered to detect significant associations when analyzing by each group separately. Given that we identified p9 hypermethylation in both AD and PSP, it is possible that genes identified here as significantly associated with p9 methylation may be related to individual pathologies and not necessarily post-transcriptional RNA processes.
Mitochondrially-related gene expression is correlated with p9 methylation
We identified 5300 genes significantly correlated with p9 methylation. Using principle component analysis, we identified clustering of gene expression profiles by disease state with AD and PSP, and NC and PA grouping together respectively. Of the top gene expression hits, we identified over-represented GO terms related to a number of cellular components and processes using PANTHER. Interestingly, there was over-representation of mitochondrial components.
In order to identify tissue-specific changes, we performed further enrichment analysis in IPA using a cerebellum tissue reference and all of the transcripts (5300 unique genes) significantly correlated with p9 methylation. Again, several of the enriched pathways identified were linked to mitochondrial function and homeostasis, as well as DNA damage/repair. We also identified a number of pathways related to RNA processing. In addition, some of the GO terms (e.g. DNA repair and cellular response to DNA damage stimulus) and canonical pathways (e.g. Cell cycle control of chromosomal replication, DNA damage induced 14-3-3s signaling) appeared to correlate with functions of one of the top genes (i.e.TRAIP) identified in the gene-based GWAS. Though some of the GO terms and canonical pathways are rather ambiguous, it does imply that p9 methylation is associated with altered transcription of gene-sets important for mitochondrial function.