tRNA-Derived Fragments in Alzheimer’s Disease: Implications for New Disease Biomarkers and Neuropathological Mechanisms

Human hippocampus specimens

Tissues were requested through the National Institutes of Health (NIH) NeuroBioBank (https://neurobiobank.nih.gov/). Thirty-four postmortem hu-man hippocampus samples were obtained from the Human Brain & Spinal Fluid Resource Center (CA, US), the University of Maryland Brain and Tissue Bank (MD, US), the Mount Sinai NeuroBioBank (NY, US), and the Harvard Brain Tissue Resource Center (MA, US). These samples included 14 con-trols, 15 samples from individuals with a neuropathological diagnosis of AD at Braak stage 5∼6, and 5 tissues from individuals at Braak stage 3. The characteristics of the patients were listed in Table 1.

Bioinformatic analysis

A global sncRNA expression profile in the brain of AD patients, deposited in the Gene Expression Omnibus (GEO) database with an accession #: GSE48552, was reanalyzed recently. In brief, the raw data was downloaded and individual sequences with read numbers ≥10 were classified by comparing them to the miRNA database (miRBase; http://www.mirbase.org), the rRNA database (RDP; http://rdp.cme.msu.edu/), the tRNA database (GtRNAdb; http://gtrnadb.ucsc.edu/), and the Exon-Intron Database (EID; http://www.utoledo.edu/med/depts/bioinfo/database.html). In high-throughput sequencing, the cloning frequency of a sncRNA provides a digital measure of its relative expression level. Therefore, we calculated the relative sequencing frequency of each sncRNA by dividing its raw read numbers by the total read numbers of each experimental group, similarly as we previously described in [13].

qRT-PCR

The total RNA was extracted from frozen hippocampus samples using TRIzol reagents (Thermo Fisher Scientific, MA, US). To measure the genes of interest, iScript cDNA Synthesis Kit (Bio-Rad, CA, US) was used to generate cDNA, followed by qPCR, using iTaq Universal SYBR Green Supermix (Bio-Rad) as we previously described [13]. Ribosomal Protein L13 (RPL13), one of the most stable housekeepers in AD autopsy brain tissue was employed for normalization [24]. The primers used to examine ANG, Drosha, Dicer, NSun2, and cleavage and polyadenylation factor I subunit 1 (CLP1) expression are shown in Table 2.

A schematic representation of specific quantifi-cation of tRF5s by qRT-PCR was shown in Supplementary Figure 1. To quantify tRFs, we first made the 3’-hydroxyl of tRF5s by treating total RNA with T4 polynucleotide kinase (T4PNK, NEB, MA, US) according to the manufacturer’s instructions. The treated RNAs were subsequently exposed to a ligation reaction with a 3’-RNA linker using T4 RNA ligase (Thermo Fisher Scientific, MA, US), and the product was used as a template for reverse transcription (RT) with primer against the linker. The RT products were subjected to SYBR Green qPCR (iTaq™ Universal SYBR Green Supermix, Bio-Rad) using a forward primer specific to a tRF5 of interest and a reverse primer specific to a 3’ linker, and 18S was used for normalization. The sequences of the primers and 3’-RNA linker are listed in Table 3.

Western blot

The hippocampus proteins were prepared using RIPA buffer (Thermo Fisher Scientific, MA, US), followed by protein quantification using BCA Protein Assay Kit (Thermo Fisher Scientific, MA, US). The western blot was done similarly, as we previously described, using antibodies against ANG (Santa Cruz sc-74528, TX, US), Drosha (Santa Cruz sc-393591, TX, US), Dicer (Cell Signaling Technology #3363, MA, US), or GAPDH (Santa Cruz-47724, TX, US) antibodies [25].

Northern blot (NB)

Northern hybridization for tRFs was performed as described [26]. Briefly, 5μg RNA was separated in 15%denaturing polyacrylamide gel with 7 mol/l urea and then transferred to a positively charged nylon membrane (Amersham Biosciences, NJ, US). The membrane was hybridized with a ³²P-labeled DNA probe reversely complementary to the tRF of interest in ULTRAhyb-Oligo solution (Life Technologies, NY, US), followed by washing according to the manufacturer’s instructions.

Statistical analysis

The experimental results were analyzed using Graphpad Prism 5 software. Group comparison was done by non-parametric statistics methods since the sample size throughout was small and distributional assumptions were not able to be met. Specifically, an unpaired two-tailed Mann-Whitney U test was used for the comparison of two independent groups, while the Kruskal-Wallis test was used for Fig. 3G, where four groups of patients with various disease stages were compared. Single, two, and three asterisks represent a p-value of <0.05, <0.01, and <0.001, respectively. Means±standard errors (SE) are shown. For correlation analyses, we performed Spearman’s rank correlation test. Spearman’s rank correlation coefficient (R_s) was used to determine correlations. A p-value of less than 0.05 was considered significant.

OPEN IN VIEWER

Fig. 3

The expression changes of tRF5-ProAGG, tRF5-CysGCA, and tRF5-LeuCAG in AD. A-C) qRT-PCR was performed to detect tRF5-ProAGG (A), tRF5-CysGCA (B), and tRF5-LeuCAG (C) in the hippocampus of control and AD patients, described in Fig. 2. D-F) The expression of tRF5-ProAGG (D), tRF5-CysGCA (E), and tRF5-LeuCAG (F) was compared in the subgroups of EOAD and LOAD with respective paired controls as illustrated in the figures. G) Braak stage-dependent expression of tRF5-ProAGG. The expression of tRF5-ProAGG was plotted according to the Braak stages. H) A graphic demonstration of Spearman’s rank correlation between tRF5-ProAGG expression and Braak stage. I) Patient information. ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001. Data are shown as means±SE.

Both tRFs and ANG were upregulated in AD patients

We reanalyzed online data from the GEO DataSets with an accession #: GSE48552. The deposited raw high throughput sequencing data of small RNAs were obtained from the samples of six LOAD patients (72.2±6.0 years old) and six age-matched controls (72.3±10.4 years old). Among samples, we identified 244 tRFs with base mean reads >100. In AD patients, there were considerable changes in tRFs and miRNAs. As shown in Fig. 1A, tRFs and miRNAs accounted for 5.95±2.16%and 32.40±5.92%of the total mapped reads in the control group. The percentages of tRFs and miRNAs had a 2.4- and 1.5-fold increase in the AD group, respectively. However, piRNAs expression was comparable in the control and AD groups, suggesting that the most affected sncRNAs in AD are tRFs. In Table 4, the sequences and base mean of the top 10 expressed tRFs are listed.

OPEN IN VIEWER

Fig. 1

Changes of ribonucleases and sncRNAs in AD patients. A) Reanalysis of sncRNAs deposited in GSE48552. The relative sequencing frequency of tRFs, miRNAs, and piRNAs was calculated by dividing their raw read numbers by the total read numbers of each experimental group reads. B, C) The mRNA expression of ANG and Dicer. qRT-PCR was performed to detect the mRNA expression of ANG (B) and Dicer (C) in the hippocampus. RPL13 was used as an internal control. D) The protein expression of ANG and Dicer. Western blot was performed to detect ANG and Dicer protein expression levels in the hippocampus. GAPDH was used as an equal loading control. GAPDH-normalized band intensity of ANG and Dicer was determined by ImageJ. E) The corresponding patient information for samples used in B-D. All statistical comparisons were performed using an unpaired two-tailed Mann-Whitney U test. ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001 relative to the paired control (CN) group as illustrated. Data are shown as means±SE.

In the sequencing study of GSE48552 [27], miRNAs were the main focus and the RNA sequencing samples were not treated with T4PNK, an enzyme having 3^′-terminal phosphatase activity that removes both a P and cP from the 3^′-end of RNAs to form a 3^′-OH end [28]. Since not all tRFs contain the 3^′-OH end, tRFs without the 3^′-OH end are unable to be ligated to sequencing barcode [21, 29]. Therefore, T4PNK-untreated sequencing likely left these tRFs unsequenced. For this reason, we did not compare the reads for individual tRFs between control and AD groups. Nevertheless, the study suggested aberrant tRF expression in AD (Fig. 1A).

tRFs are usually classified into three groups: tRF5 which is derived from the 5’-end of mature tRNA, tRF3 whose sequence is aligned to the 3’-end of mature tRNA, and tRF-1 which is the 3’-trailer sequence of pre-tRNAs [26]. Notably, the top ten expressed tRFs in the control group belong to all tRF5 of 30∼40 nt in length (Table 4).

To investigate the role of tRFs in AD development, we recently obtained the hippocampus, one of the main areas in the brain affected by AD [30, 31], from the NIH NeuroBioBank. We first investigated whether ribonucleases with a known function in controlling the tRF biogenesis are affected in AD. ANG usually cleaves tRNAs around the anticodon loops in response to stress or viral infections, resulting in 30–40 nt long tRFs [13, 19]. Our qRT-PCR results showed that ANG mRNA was significantly increased in the AD group, compared with the control group (Fig. 1B). Since Dicer is another ribonuclease responsible for generating 20-nt long tRFs in cancer [32, 33], we also quantified the mRNA expression of Dicer. As shown in Fig. 1C, Dicer mRNA expression was comparable between control and AD groups. In this patch of samples (from the Harvard Brain Tissue Resource Center), we were fortunate to get both EOAD and LOAD samples. However, both LOAD and its age-matched group had less than five samples, not enough for power analysis. Therefore, the significance of ANG in the EOAD and LOAD group was not investigated separately. Even though the sample size of some subgroups was not ideal, some hippocampus tissues were relatively big so that protein samples could be prepared for western blot. As shown in Fig. 1D, we run some EOAD samples and their controls in one gel and the LOAD samples and their controls in another one. Both western blots suggested higher expression of ANG in both EOAD and LOAD, compared with their respective control groups. The overall normalized band intensity of ANG also demonstrated ANG to be significant in AD, compared to the healthy controls, while the protein expression of Dicer was comparable between the AD and control groups (Fig. 1D).

Affected tRF5s in the hippocampus in AD

As shown in Table 4, tRFs from the 5’-end of tRNA-GlyGCC and tRNA-GluCTC, namely tRF5-GlyGCC and tRF5-GluCTC, are the two most abu-ndant basal tRFs. These two tRF5 are also abundant in the primate’s cerebellum, hippocampus, and liver [34]. Therefore, we started investigating the expression of tRF5-GlyGCC and tRF5-GluCTC in the hip-pocampus of AD patients. Our qRT-PCR results demonstrated that both tRF5-GlyGCC and tRF5-GluCTC were significantly increased in the AD group by 2.0 and 1.8 folds, respectively (Fig. 2A, B). We also chose a tRF called tRF5-GlyCCC-2, which is a moderately expressed isoform (Table 4), to investigate the impact of AD on its expression. As shown in Fig. 2C, tRF5-GlyCCC-2 was increased in AD by 1.57 folds, suggesting its involvement in AD, as well.

OPEN IN VIEWER

Fig. 2

Changes in the expression of tRF5-GlyGCC, tRF5-GluCTC, and tRF5-GlyCCC-2 in AD. A-C) qRT-PCR was performed to detect tRF5-GlyGCC (A), tRF5-GluCTC (B), and tRF5-GlyCCC-2 (C) in the hippocampus from control and AD patients. All the expression was normalized by the internal control 18 s RNA. D-F). The hippocampus expression of tRF5-GlyGCC (D), tRF5-GluCTC (E), and tRF5-GlyCCC-2 (F) were also analyzed in subgroups of EOAD and LOAD. G) Patient information for samples used in A-F. Unpaired two-tailed Mann-Whitney U tests were performed for statistical comparisons. ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001, relative to the paired control group as illustrated. Data are shown as means±SE.

The experiments starting from Fig. 2 used samples from other NIH NeuroBioBank resource centers in CA, MD, and NY. Although the overall tissues were tiny and not enough to carry out western blot if needed, each subgroup (the EOAD and its age-matched control; LOAD and its age-matched control) had six or more samples. Therefore, we tried to investigate the significance of interested tRFs in the EOAD and LOAD separately. We found that all th-ree tRF5s shown in Fig. 2A-C had a significant increase in EOAD patients, compared with their age-matched controls (young controls). tRF5-GlyGCC, tRF5-GluCTC, and tRF5-GlyCCC-2, respectively, showed a significant increase of 2.6, 2.1, and 2.3 folds in the EOAD group, compared with its age-matched control group (left panels of Fig. 2D-F). The LOAD group did not show a significant difference in the expression of these tRFs, compared with its age-matched healthy control group (right panels of Fig. 2D-F). Among these three tRFs, the expression of tRF5-GluCTC showed an increasing trend toward significance in the LOAD (1.6-fold, p = 0.07, right panel of Fig. 2E), compared with the age-matched control group. More samples are probably needed in the future to define the significance of these tRFs.

Other than the three tRFs mentioned above, other tRF5s were also chosen for the study, as they were reported to be functional in other biological settings. For example, tRF5-ProAGG has been reported to interact with ribosomes and inhibit global translation [35]. tRF5-CysGCA can impede translation initiation, induce the assembly of stress granules, and have neuroprotective effects [36]. tRF5-LeuCAG promotes cell proliferation and cell cycle in non-small cell lung cancer [37]. Hence, we assessed these three tRF5 in the hippocampus of AD pat-ients. We found that tRF5-ProAGG had a significant 2.8-fold increase. However, tRF5-CysGCA and tRF5-LeuCAG were comparable in the control and AD groups (Fig. 3A-C). We also did subgroup analyses of tRFs shown in Fig. 3A-C. We found that tRF5-ProAGG had a significant 2.6- and 2.9-fold increase in the EOAD and LOAD groups, respectively, compared with their paired control groups (Fig. 3D). tRF5-CysGCA indeed showed a significant increase in EOAD (left panel of Fig. 3E), even though the significance was not observed in the overall AD patient samples (Fig. 3B). In contrast, no changes in the expression of tRF5-LeuCAG were observed in either the EOAD or LOAD group (Fig. 3F).

Given the significance of tRF5-ProAGG in AD, we also explored whether it also shows stage-dependent expression. In brief, 27 samples were provided with Braak stage information. 6 samples with Braak stage 0 were used as controls. Our result indicated tRF5-ProAGG showed a significant increase at Braak stages 3 and 6, but no difference at stage 1, demonstrating stage-dependent expression of tRF5-ProAGG (Fig. 3G). We next tested for correlation of tRF5-ProAGG with the Braak stage. As shown in Fig. 3H, the tRF5-ProAGG expression level positively correlated with the Braak stage (R_s = 0.598, p = 0.002).

One more interesting thing about tRF5-ProAGG is its molecular size. Although the melt curve of qRT- PCR for tRF5-ProAGG showed a single peak, the melting temperature of amplified products was 1°C higher than other tested tRFs. We, therefore, seque-nced all tRFs, after qRT-PCR products being inserted into the pGEM®-T Vector (Promega, WI, US), by Sanger sequencing. Unlike other qRT-PCR products, who demonstrated a single product, tRF5-ProAGG clones showed a 32-nt product with the sequence: 5’-GGCTCGTTGGTCTAGGGGTATGATTCTCGCTT-3’ (long form) and an 18-nt product with the sequence: 5’-GGCTCGTTGGTCTAGGGG-3’ (sho-rt form). Furthermore, NB confirmed two isoforms of tRF5-ProAGG (Fig. 4). Both isoforms were increased in the EOAD and LOAD groups.

OPEN IN VIEWER

Fig. 4

Two isoforms of tRF5-ProAGG. Northern blot was carried out to confirm the presence of long and short isoforms of tRF5-ProAGG in hippocampus tissues from EOAD and LOAD patients. Age-matched controls were included. 5s rRNA was used as an equal loading control.

NSun2 were decreased in EOAD

As shown in Fig. 1, ANG was increased in AD. It has been recently suggested that certain nt modifications of tRNAs determine their cleavage by ribonuclease. For example, NSun2-mediated 5-methylcytosine (m⁵C) methylation in tRNAs is essential for their stability [22, 23]. tRNAs lacking m⁵C methylation because of the decreased expression of NSun2 show increased affinity to ANG, and are prone to be cleaved [23]. tRNA modification controlled by CLP1, another multifunctional kinase, also contributes to tRNAs splicing [38]. The loss of CLP1 activity results in the accumulation of tRF5-Tyr, which sensitizes neurons to oxidative stress-induced cell death [38]. The CLP1-/- mice show a progre-ssive loss of spinal motor neurons [39]. In addition, patients with homozygous missense mutations in CLP1 (R140H) suffer from brain atrophy and severe motor-sensory defects [38, 40]. Therefore, we assessed NSun2 and CLP1 mRNAs expression in the hippocampus of EOAD and LOAD patients. Our results showed NSun2 mRNAs were downregulated in EOAD, compared with age-matched control (Fig. 5A), suggesting a possible association between the downregulation of NSun2 with ANG-mediated tRNA cleavage. Regarding CLP1, we did not detect any change in both EOAD and LOAD groups (Fig. 5B).

OPEN IN VIEWER

Fig. 5

The expression of NSun2 and CLP1 expression in EOAD and LOAD patients. Hippocampus RNAs from EOAD and LOAD patients were used for qRT-PCR to quantify NSun2 (A) and CLP1 (B). Their respective age-matched controls were also included. The expression is present after the normalization by RPL13. The analyses and patient information were similar to what is described in Fig. 2. ^*p < 0.05, relative to the paired control group, as illustrated. Data are shown as means±SE.