Short RNA regulators: the past, the present, the future, and implications for precision medicine and health disparities

Abstract

We herein provide a brief review of the trajectory that the field of short RNA research followed in the last 25 years. We place emphasis on the unexpected discoveries and the ramifications of these discoveries for the field, as well as offer some thoughts about what the next 25 years may bring. Arguably, the uncovered dependence of different types of short RNAs on individual attributes such as a person’s sex, population origin, race, and on tissue type, tissue state, and disease was most unexpected. This dependence has important ramifications in that it will provide a boost to our understanding of the molecular mechanisms of health disparities as well as pave the way for novel approaches to designing improved and personalized diagnostics and therapeutics.

Introduction

December 2018 marked the 25th anniversary of the publication of the first two papers that showed how a short non-protein coding RNA (ncRNA) molecule can post-transcriptionally regulate the abundance of a protein that is needed for the development of the worm Caenorhabditis elegans [1,2]. Seven years after those early publications, a second such molecule was discovered in the worm [3] to be followed soon thereafter by the discovery of similar molecules in other invertebrates, vertebrates and plants [4–8], as well as viruses [9,10]. These molecules became known as microRNA (miRNA) and have since become the best-characterized type of short ncRNA.

As it turned out, miRNAs were the first of several more types of short RNA that exist in the cells of higher organisms. These types follow distinct biogenesis paths and are known to serve a multitude of diverse functions in the cells. While the exact functional roles remain unknown for the vast majority of these molecules, there is overwhelming evidence in support of their importance. Here, we discuss several of those types and provide brief summaries of what is currently known about them. We also review findings showing that many of these molecules capture important biology that is currently uncharacterized.

MiRNA

MiRNA are derived through cleavage of endogenous hairpin transcripts, approximately 70–100 nucleotides (nt) in length into smaller (~ 22 nt) mature molecules [11]. These tiny yet powerful molecules are key components of nearly all-regulatory processes of the cell. They act as guide molecules for the RNA-Induced Silencing Complex (RISC) to target complementary regions of protein-coding messenger RNA (mRNA) and other long ncRNA to post-transcriptionally regulate transcript abundance [12]. As they play important regulatory roles, it is not surprising that they have been associated with a multitude of events in both homeostasis [13–17] and disease [13,18–24].

Over time, four tenets emerged that shaped miRNA research for many years: (i) the human genome encodes a small number of miRNA loci that are evolutionarily conserved; (ii) the mature miRNA interact primarily with mRNA that they target specifically within their 3′ untranslated region (3′UTR); (iii) the interactions between an miRNA and its targets involve Watson-Crick base pairing in the miRNA’s ‘seed’ region; and (iv) each miRNA precursor produces a single functional molecule, the mature miRNA, either from its 5′ arm or its 3′ arm. The development of new high-throughput screening approaches and the increasing availability of data have forced the community in recent years to re-examine the validity of these tenets.

MiRNA: an ever-expanding miRNA-ome and numerous targets per miRNA

The earliest work in the field posited that the number of the genomic loci encoding miRNA precursors in model organisms was approximately 1% of the organism’s protein-coding genes [8,25]. This belief persisted for many years and was reflected by the entries of the miRBase repository [26]. Moreover, each miRNA was predicted to regulate only a few dozen of mRNA through targets in their 3′-UTR [12,27].

Our earlier work represents a multi-faceted departure in this regard. In 2006, we estimated that there must be more than 25 000 human miRNA precursors [28^●●]. We also presented experimental evidence that a miRNA can have more than one thousand targets, and argued through computational analyses that miRNA could equally well target the protein coding sequence (CDS) of mRNA as well as their 5′ untranslated regions (5′-UTR) [28^●●].

Subsequent experimental and computational work shed light on the questions of how many targets an miRNA can have and where these targets are located. Others and we contributed multiple examples to the literature that demonstrated pervasive miRNA targeting of the CDS [29–32,33^●●] and 5′-UTR of mRNA [34–37], as well as that miRNA can target ncRNA [38^●●,39,40]. These findings were corroborated through independent studies that analyzed Argonaute-footprint datasets [41–43]. Moreover, molecular dynamics simulations of the Argonaute-miRNA-target complex showed that the complex remains stable in the presence of non-Watson-Crick interactions and/or bulges in the seed region [44] in agreement with both early reports [28^●●,45,46] and more recent reports in the field [41,44,47]. These additional interactions greatly increase the number of targets that an miRNA can regulate.

Over the course of the last decade, next-generation sequencing (NGS) approaches became mainstream and helped provide an unprecedented look into an organism’s miRNA-ome. NGS helped expand our understanding of this type of ncRNA as well as search for novel miRNA. NGS have since revealed the presence of many thousands of previously unreported human genome loci that harbor miRNA precursors [48,49,50^●●]. Interestingly, and contrary to the tenets, many of these loci are primate-specific or human-specific and not evolutionally conserved.

MiRNA: isomiRs are powerful isoforms with distinct roles

Decidedly, the most unexpected observation in the miRNA field was the realization that a miRNA precursor arm does not produce a single mature miRNA product but rather a ‘cloud of isoforms’. These isoforms are known as isomiRs (Figure 1). Any two isomiRs from the same miRNA arm differ slightly in their lengths, their 5′ termini, 3′ termini, or some combination [51–53].

Figure 1.

Some isomiRs from the 5p arm of miR-183. Shown are a few of the isomiRs that this arm produces. Highlighted is the isomiR that one typically encounters in public databases.

Analyses by others and us showed that nearly 50% of the time the isomiR commonly listed in public databases such as miRBase [26] is not the most abundant isomiR in a given tissue [53,54^●,55]. The immediate implication for the studies that preceded the discovery of isomiRs is that half of the time, the specific molecule that was studied or profiled was not the most abundant for the tissue or disease type under consideration. Moreover, the number of the most abundant isomiRs that can arise from a specific miRNA arm can change radically from one tissue/disease context to the next: for example, see Figure 1A of Ref. [54^●]. These two points are particularly notable considering that we showed that distinct isomiRs from the same miRNA precursor arm have non-overlapping sets of gene targets [56^●●]. Taken together, these observations suggest the presence of many thousands of additional molecules that greatly increase the regulatory power of miRNAs and whose functional roles are currently uncharacterized.

It is worth noting here that the discovery of isomiRs raises an important computational problem. The problem is that of determining whether an isomiR sequence appears elsewhere on the genome, outside the span of the miRNA precursor. Thus, when mapping reads on the genome, exhaustive searches are now required to ensure the miRNA-ness of an isomiR. This is indeed a matter with considerable consequences: at least 8% of the isomiRs from the currently known human miRNA loci are of ambiguous genomic origin (Rigoutsos and Londin, unpublished data).

PiRNA

Piwi-interacting RNA (piRNA) are ncRNA, 24–35 nt in length, which bind to Piwi proteins and repress transposons and other targets to maintain genome integrity in animal germlines [57–63]. The piRNA pathway is highly conserved across phyla, and disruption of the pathway causes elevation of transposons and defects in germline development, eventually resulting in sterility. Our recent study identified Papi as a significant piRNA biogenesis factor [64^●], which was later found to function in PNLDC1-catalyzed 3′-end maturation of piRNAs [65^●]. Our work also revealed a previously uncharacterized link between cell-cell contact and piRNA biogenesis [66^●], as well as clarified the biogenesis pathway of tRNA-derived piRNAs, in which 5′-tRNA halves serve as direct piRNA precursors [67^●●]. The role of the piRNA pathway in somatic tissues has been demonstrated in lower eukaryotes [16], and human Piwi expression has been shown in various somatic cancers [68]. However, whether the piRNA pathway has a role in human somatic tissues/diseases is controversial as the majority of annotated human ‘piRNA’ sequences, deposited in piRNA databases, have been found to correspond to other ncRNA sequences [69].

tRNA

Since their discovery in the 1950’s, transfer RNA (tRNA) have been best known as abundant adapter components of the translational machinery, converting mRNA codon information into amino acid sequences. Moreover, it had been thought that each tRNA genomic locus produced a single product, the mature tRNA. As was the case with miRNA, NGS revealed a different picture. Others [70–76] and we [55,77^●●,78^●●,79,80^●●,81^●,82^●●] showed that each tRNA precursor produces ‘clouds of tRNA-derived fragments’, which are known as ‘tRF’ and co-exist with the mature tRNA [83,84].

From a structural standpoint, these fragments fall into six subtypes (Figure 2): (i) 5′-tRNA halves (5′-tRH), which are ~34 nt long and produced through cleavage at the anticodon by Angiogenin (ANG) [85^●●,86,87^●●,88]; (ii) 3′-tRNA halves (3′-tRH), which are the tail-half of the mature tRNA that remains after ANG cleavage at the anticodon; (iii) 5′-tRF, which are typically ~20 nt long and produced by cleavage at the D-loop [89,90^●●,91]; (iv) i-tRF or internal-tRF, which are wholly contained within the mature tRNA’s span [78^●●]; (v) 3′-tRF, which are ~20 nt long and produced by cleavage at the T-loop [89,90^●●,91]; and (vi) tRF-1, which originate from cleavage of the precursor tRNA [90^●●,92]. As we showed, both nuclearly encoded and mitochondrially encoded tRNA produce tRF with characteristically different length and abundance profiles [78^●●].

Figure 2.

Some of the structural subtypes of tRF. Shown are the five subtypes that overlap the mature tRNA. Not shown is the sixth structural subtype (tRF-1) which overlaps the tRNA precursor.

During the last few years, molecules from these six subtypes have been under intensive studies in different organisms. These studies have been generating evidence that specific tRF are expressed as functional molecules, rather than degradation products. Their functions span various biological processes that extend well beyond translation [83,93–96].

There also remains the matter of how many possible genomic sources exist that could be producing tRF. Nominally, tRF can arise from the 610 nuclear tRNA or the 22 mitochondrial tRNA [97,98]. This statement is complicated by our recent reporting of 497 ‘tRNA-lookalikes’ [99]. These lookalikes are genomic sequences that best resemble one of the 632 known tRNA templates and appear to exhibit cell-type-specific transcription. The persistence of these lookalikes from primates to marsupials [100^●●] indicates that they are not a random event. It also suggests that there are potentially many more tRF-producing templates and, consequently, an even higher diversity of tRF sequences.

tRNA: the longer fragments - more than meets the eye

The expression of tRNA halves, 5′-tRH and 3′-tRH, is triggered by various molecular factors such as oxidative stress, heat/cold shocks, UV irradiation, and sex-hormone signaling pathways. These longer tRNA fragments are generated actively by the cell, that is they are not degradation products, and have functional consequences for translational regulation, cell proliferation, apoptosis, piRNA production, and possibly other mechanisms [67^●●,77^●●,85^●●,88,101].

Despite the undisputed contributions of NGS to biomedical research that have greatly accelerated ncRNA studies, the current standard RNA-seq methods do not fully capture all of the expressed RNAs. Indeed, because the adapter ligation step relies on the presence of 5′-P and 3′-OH at the respective termini of RNA molecules, ‘escapers’ can slip by undetected. ANG-generated 5′-tRH are one such escaper because they harbor a 2′,3′-cyclic phosphate (cP) at their 3′-ends [102]. Those cP-containing RNA cannot be ligated to a 3′-adapter and therefore remain uncaptured by standard RNA-seq, forming a hidden component of the transcriptome. Specific sequencing for cP-containing RNA can be achieved by our cP-RNA-seq [77^●●,103^●●] or RNA-seq using cP-specific RNA ligase [104]. In addition to the terminal phosphate states of RNA, internal post-transcriptional modifications of RNA can also affect the efficiency of cDNA amplification during the RNA-seq procedure.

tRNA: the shorter fragments - a regulatory group with a lot of diversity

The other three tRF subtypes that overlap the mature tRNA are considerably shorter than the two types of tRNA halves and have lengths around 20 nt. This is reminiscent of miRNA/isomiRs and begs the question whether tRF can also enter the RNA interference pathway through Argonaute loading. While a few examples have been reported of tRF acting like miRNA [105,106^●●] the situation appears to be more complicated. In fact, it was shown that only some of these shorter tRF can be loaded on Argonaute [76]; moreover, this process does not always depend on DICER [90^●●,106^●●].

The biogenesis of these shorter tRF is not known currently. Our analysis of all of the datasets of The Cancer Genome Atlas (TCGA), which represent 32 cancer types, uncovered a combined total of 23,377 5′-tRF, i-tRF and 3′-tRF [81^●]: 78% of these tRF belong to the i-tRF type. Looking across the TCGA datasets, we see that the expression profile of the typical tRF changes by cancer type [81^●], a dependence that makes it imperative to understand the regulatory roles of these molecules.

Generally speaking, the function of individual tRF is not known. This is a particularly notable point because of the large number of distinct tRF that have been found so far in different tissues. In recent work, we uncovered first evidence that tRF are associated with important cellular processes in multiple cancer settings [107^●●]. These processes include development, signaling, the proteasome, and metabolism. Interestingly, nearly half of the discovered tRF-mRNA associations involve tRF that arise from the 22 mitochondrial tRNA. Another intriguing finding was that the mRNA that are positively correlated with tRF have higher density in repetitive elements such as ALU, MIR, and ERVL [107^●●].

The task of discovering and quantifying tRF is complicated by several facts. The sequence of a given putative tRF may be such that it is shared by multiple isodecoders of the same isoacceptor or by multiple isoacceptors, shared by a tRNA and a tRNA-lookalike, or presence inside tRNA space and also elsewhere on the genome. Additionally, the multiple genomic instances of a tRF require careful bookkeeping in order to avoid misrepresenting tRF abundance. MINTmap [80^●●] is the first tool that considers all these matters and permits exhaustive and deterministic mining of tRF in the human genome while flagging tRF of ambiguous origin. Analogous tools for other genomes do not exist yet.

IsomiRs, tRF, human disease and health disparities

The newly identified isomiRs and tRF could have remained a relative curiosity had it not been for a number of notable properties that characterize them. Perhaps the most important consideration is whether they are degradation products. Through analyses of numerous public datasets we showed that both isomiRs and tRF are produced constitutively and exhibit the same expression profile in like datasets, in health and in disease [53,55,56^●●,78^●●,79,81^●,107^●●].

Moreover, we showed that the abundances of individual isomiRs and tRF (short and long) differ by tissue type and disease type/subtype, as well as by an individual’s sex, population group, and race [53,54^●,55,56^●●,77^●●,78^●●,81^●]. These differences translate into differences in the regulation of specific mRNA and pathways with the differences aligning with a patient’s sex, race, population origin, and probably age and other variables. For example, by analyzing diseases such as triple negative breast cancer and prostate cancer, we showed specific regulatory differences between White and Black/African American patients with these diseases [55,82^●●]. Similarly, we showed analogous differences that depend on a person’s sex in bladder, kidney, and lung cancers [107^●●]. In other words, isomiRs and tRF and their regulatory interactions are allowing us to uncover biology that is currently uncharacterized and is directly linked to health disparities.

At the same time, the attributes of these molecules and their multiple dependencies make them ideal candidates to serve as non-invasive disease biomarkers. Blood is typically the source biofluid of choice for determining whether a biomarker is present or not. This turns the biomarker question into a ‘many-to-one’ problem: multiple sources release DNA fragments, short ncRNA, long coding and non-coding RNA, and proteins into the bloodstream, generally encapsulated in exosomes. If the targeted DNA mutations or the targeted molecules are associated with multiple ailing tissues, being able to isolate them does not help pinpoint the molecules’ provenance. In contrast, the strong dependence of isomiRs and tRF on tissue type, tissue state and disease subtype can help address the limitations of previous methods. A new study comprehensively explored short RNA profiles in exosomes and revealed a diverse repertoire of miRNAs and tRF in these vesicles [108^●●].

Because of their properties, isomiRs and tRF are being explored actively as possible disease biomarkers [54^●]. Also, the dependence of these molecules’ profiles on sex, population, and race adds more power to their potential to form powerful, more personalized diagnostics. Lastly, because of the involvement of isomiRs and tRF in regulatory networks, both types of ncRNA represent possible novel therapeutic targets that can be tuned to a patient’s personal attributes.

Conclusions and outlook

Looking ahead it is reasonable to expect that in the next 25 years, most of the consequential ncRNA species will be fully characterized through development of specialized sequencing techniques. We will have a much more accurate view of the RNA-ome in human and other organisms as well as of the function of the various RNA in different tissues. Many important RNA molecules had remained hidden from us and beyond the reach of our technology. But this is now changing quickly and we expect that it will prove beneficial in advancing our understanding of the cell’s workings, in homeostasis and disease.

Moreover, the fast accumulating associations between these molecules and an individual’s attributes will undoubtedly herald a new era in precision medicine that is truly personalized. A necessary ingredient here will be the availability of data repositories that catalogue the myriad of RNA that are being discovered in different settings. Such efforts are already underway with the development of, for example, MINTbase [81^●], and piRBase [109] for tRF and piRNAs respectively. For miRNAs, the oldest database of its kind, miRBase [26], currently lists a single isoform per miRNA arm. To increase their value for biomedical researchers, it will be necessary to expand the score of such databases, or to develop new ones, in order to list the isomiRs that arise from different miRNA arms. It will also be important that such databases include information about isomiR expression in different tissue contexts, such as is done for tRF in MINTbase [81^●].

It is difficult, and imprudent, to speculate at this time how many distinct short ncRNA will be discovered when all is said and done. The discovery of tRNA halves with a cP modification at their 3′ termini is a good example of a category of abundant molecules that had gone unnoticed for nearly a decade [77^●●], because of technical limitations. Similarly, the i-tRF are a good example of a large category of functionally important molecules that also went unnoticed for a long time [78^●●], because of computational limitations and a lack of enough datasets that could be mined.

The emerging molecules are posing several considerable challenges that will need to be addressed as researchers go about designing experiments to determine the molecules’ function. We note here two such key challenges. The first is that of a scalable and inexpensive method for quantifying short RNA. As we showed in previous work, the currently available commercial qRT-PCR assays cannot accurately quantify the abundance of isomiRs, and, by extension, of tRF, piRNAs, and so on [110^●●]. In the absence of an effective commercial assay, and to assist us with our own experimental work, we proposed a first such approach [111^●]. The second challenge is that of selective silencing of a single short RNA. We are not aware of a technique that can effectively silence, for example, a single isomiR or tRF, while leaving untouched molecules with overlapping sequences from the same locus.

In closing, it is safe to say that a lot of work lies ahead. The work will be challenging but also exciting and rewarding. In the process, we will be able to chart new genomic territories, and get a better understanding of the cell’s inner workings, Sampling of these processes, which had been elusive for a long time, is slowly coming within reach. Nonetheless, to maximize both the scientific value and the societal benefit of what has yet to be uncovered, we need to remain open-minded and entertain possibilities that are not anticipated by the current state of scientific knowledge.