Re-Analyses of Samples From Amyotrophic Lateral Sclerosis Patients and Controls Identify Many Novel Small RNAs With Diagnostic And Prognostic Potential

Abstract

Amyotrophic lateral sclerosis (ALS) is a highly heterogeneous disease for which accurate diagnostic and prognostic biomarkers are needed. Toward this goal, we reanalyzed two published collections of datasets generated from the plasma and serum of ALS patients and controls. We profiled these datasets for isoforms of microRNAs (miRNAs) known as isomiRs, transfer RNA-derived fragments (tRFs), and ribosomal RNA-derived fragments (rRFs), placing all remaining reads into a group labeled "not-itrs." We found that plasma and serum are rich in isomiRs (canonical, non-canonical, and non-templated), tRFs, rRFs, and members of an emerging class of small RNAs known as Y RNA-derived fragments (yRFs). In both analyzed collections, we found many isomiRs, tRFs, rRFs, and yRFs that are differentially abundant between patients and controls. We also performed a survival analysis that considered Riluzole treatment status, demographics (age at onset, age at enrollment, sex), and disease characteristics (ALSFRS, rD50, onset type) and found many of the differentially abundant small RNAs to be associated with survival time, with some of these associations being independent of Riluzole treatment. Unexpectedly, many not-itrs that did not map to the human genome mapped exactly to sequences from the SILVA database of ribosomal DNAs (rDNAs). Not-itrs from the plasma datasets mapped primarily to rDNAs from the order of Burkholderiales, and several of them were associated with patient survival. Not-itrs from the serum datasets also showed support for rDNA from Burkholderiales but a stronger support for rDNAs from the fungi group of the Nucletmycea taxon. The findings suggest that many previously unexplored small non-coding RNAs, including human isomiRs, tRFs, rRFs, and yRFs, could potentially serve as novel diagnostic and prognostic biomarkers for ALS.

Keywords: ALS; Amyotrophic lateral sclerosis; Burkholderiales; Diagnostics; Nucletmycea; Prognostics; Y RNAs; isomiRs; miRNAs; rRFs; rRNAs; tRFs; tRNAs; yRFs.

Fig. 1

Overview of the re-analysis. We studied four classes of sncRNAs (isomiRs, tRFs, rRFs, and not-itrs) and sought to identify candidate diagnosis and prognosis biomarkers. After preprocessing the raw reads, we examined their differential abundance, enrichment, correlation, and association with survival time. We carried out the survival analysis only on the plasma collection because it included more datasets

Fig. 2

Summaries. A: Sequenced reads vs. reads that survive quality trimming and adapter removal. B: Relative abundance of the four sncRNA classes across samples: percentages do not reach 100 because only sncRNAs that are unambiguously isomiRs or tRFs are used. C-D: Volcano plots for plasma (C) and serum (D) showing which sncRNAs differ statistically significantly between patients and controls

Fig. 3

Candidate diagnostic sncRNAs in plasma and serum. A-D: Examples of several sncRNAs (two isomiRs, one tRF, and one rRF) whose abundance increases in ALS versus controls in the plasma datasets. E–F: An isomiR (miR-423-3p|0|0|) and a 3´-tRF (tRF-18-HR6HFRD2) whose abundance decreases in ALS, in both serum and plasma. The shown p-values are FDR-adjusted (as reported by DESeq2). All panels: encircled green dots mark mean values. Shown abundance levels are after DESeq2 normalization

Fig. 4

Examples of prognostic and diagnostic sncRNAs. The top row highlights a tRF (A), an rRF (B), a canonical isomiR (C), and a non-canonical isomiR (D). The bottom row highlights two non-canonical isomiRs (E–F) that have higher abundance at time T4 compared to T1 in most of the longitudinal samples (17 of 22, or 77%). The shown p-values are FDR-adjusted. Panels E–F: encircled green dots mark mean values. Shown abundance levels are after DESeq2 normalization

Fig. 5

SncRNAs from RNY4. Examples of sncRNAs that arise from RNY4 and are significantly differentially abundant between ALS and controls in both plasma and serum. Only sncRNAs with a mean normalized expression of >= 50 are shown and aligned to the full-length RNY4. Shown abundance represents mean values following DESeq2 normalization

Fig. 6

Unusual sncRNAs with diagnostic/prognostic associations. A: A sncRNA from the 3´ region of RNY4 that decreases in ALS, in plasma and serum (highlighted in Fig. 5). B: A very short sncRNA from an intron of the HOXA10 gene. C: An sncRNA that cannot be mapped to the human genome with a single change (replacement, insertion, or deletion) but maps exactly to a 5S rRNA sequence that is present in multiple bacteria. D: An sncRNA that cannot be mapped to the human genome with a single change (replacement, insertion, or deletion) but maps exactly to a tRNA-Met sequence that is present in multiple bacteria. All shown p-values are FDR-adjusted. Panels A-B: encircled green dots mark mean values. Note that in panel B the outliers are not shown because they fall outside the shown range of the Y-axis. Shown abundance levels are after DESeq2 normalization