The Role of cis- and trans-Acting RNA Regulatory Elements in Leukemia

Simple Summary Alterations in primary RNA motifs and aberrant expression levels of non-coding RNA molecules have emerged as biomarkers of disease development and progression. Advances in antisense oligonucleotide (ASO) techniques and pharmacologic discoveries in targeting of RNA structures and RNA–protein interactions with small molecules open a new area in RNA therapeutics that may help in developing a next generation of anti-cancer drugs. Abstract RNA molecules are a source of phenotypic diversity and an operating system that connects multiple genetic and metabolic processes in the cell. A dysregulated RNA network is a common feature of cancer. Aberrant expression of long non-coding RNA (lncRNA), micro RNA (miRNA), and circular RNA (circRNA) in tumors compared to their normal counterparts, as well as the recurrent mutations in functional regulatory cis-acting RNA motifs have emerged as biomarkers of disease development and progression, opening avenues for the design of novel therapeutic approaches. This review looks at the progress, challenges and future prospects of targeting cis-acting and trans-acting RNA elements for leukemia diagnosis and treatment.


Introduction
Leukemia, a wide spectrum of blood cancers displaying abnormal proliferation and differentiation capacity of myeloid or lymphoid blood progenitors, is the most frequent type of cancer in children and one of the most common in adults [1]. Acute myeloid leukemia (AML) and acute lymphoblastic or lymphocytic leukemia (ALL) show rapid development and little or no cell differentiation. AML is primarily found in older adults, with a median age of 70 years at diagnosis. Highly heterogeneous clinically and genetically, AML is fatal in about~80% of elderly patients, and about 60% of people younger than 60 years old [2]. ALL is the most common form of pediatric leukemia, accounting for nearly 30% of all pediatric cancers. While most pediatric patients with ALL achieve remission, 30-35% of these therapies fail, and only 30-40% of adult patients with ALL achieve long-term, disease-free survival [3]. Myelodysplastic syndrome (MDS), myeloproliferative neoplasm (MPN), and chronic forms of myeloid and lymphocytic leukemia (CML, CLL), typically diagnosed in older patients, retain some functional blood cells and develop slowly, but are prone to progression into a hard-to-treat acute leukemia [4,5].
The genetics of adult and pediatric leukemia have been intensively studied [6][7][8][9]. Several studies performed a side-by-side comparison of pediatric and adult myeloid and lymphoblastic leukemia, focusing on protein coding genes with oncogenic and tumor-suppressor functions [10,11]. The research for pediatric B-ALL. However, the comparative analysis of splice isoforms in acute pediatric B-ALL lacking mutations in splicing factors genes and normal pro-B-cells identified thousands of aberrant local splice variations per sample [37].
High tissue-or context-specificity is another important characteristic of differential splicing in normal and malignant hematopoietic tissues. The analysis of alternatively expressed isoforms between aging hematopoietic stem cells (HSCs) and progenitor cells (HPCs) identified a significant divergence with only few isoforms of transcription and histone regulators being commonly upregulated [38]. Rojas et al. aimed to identify differentially spliced variants between two hematologic entities with a similar genetic background, 17 p deletion: primary plasma cell leukemia and multiple myeloma. The results of transcriptome analysis reveal a significant deviation between the two types of tumors. Interestingly, most of the differences were observed in the spliceosome machinery genes, which emphasizes the cell type-specificity of alternative splicing [39].

Alterations in Untranslated Regions (UTR) of mRNA
The untranslated regions in mRNA (5 UTR and 3 UTR) originate from pre-mRNA exons and flank a protein-coding sequence of mature messenger RNA on both sides of an open reading frame (ORF). The UTRs are rich in cis-acting elements and distinctive secondary structures (hairpins) that are recognized by regulatory ncRNA and RBPs. Similar to splicing, recurrent UTRs abnormalities were found in cancer and previously reviewed [23,40,41].

5 UTR Alterations in Leukemogenesis
Alterations in 5 UTRs can disrupt both translation efficiency and protein characteristics. For example, mutations in the 5 UTR of ANKRD26, the Ankirin Repeat Domain 26 Gene, lead to expression of N-terminally truncated protein and cause the autosomal-dominant form of inherited thrombocytopenia and increase predisposition to AML [42,43]. The rare cases of genetic predisposition to MDS/AML are linked to SNPs in various regions of the GATA2 gene, including 5 UTR, that cumulatively lead to GATA2 loss-of-function [44].
With the right sequence context in translation initiation sites (TIS), certain non-AUG start codons can generate expression comparable to a canonical, AUG start codon, whereas mutations in TIS change levels of expression [45]. Endogenous nucleotide repeats expansions upstream of coding-region and a shifts in ORFs is linked to production of abnormal peptides due the repeat-associated non-AUG translation (RAN) common for inherited neurodegenerative diseases [46,47]. A study of 17 patients with the family history of chronic lymphocytic leukemia (CLL) and 32 patients with early-onset B-cell CLL did not observe a pathological CAG repeats expansion [48]. The analysis of polymorphisms in thymidylate synthase 5 -UTR 28 bp tandem repeats found a lower blast counts in ALL patients with 2R2R allele, but no such genotype-dependent differences were observed in AML cases [49].
In the context of stress-related global repression of translation, the production of certain oncogenic proteins can increase due to the stress-induced activation of previously repressed upstream start codons [50]. Sendoel et al. demonstrated that during transformation of skin epithelial cells, certain cancer related mRNAs such as nucleophosmin (NPM1) exhibited increased ribosome occupancy in upstream CUG rather than in conventional AUG initiation sites of canonical ORFs. In addition to a selective generation of oncogenic proteins through unconventional start codons, researchers found a shift of transcriptome towards pathways of stemness and mediators of Wnt/β-catenin signaling [51,52]. These findings suggest that the adverse changes in the molecular-genetic profile occur before the early signs of transformation are phenotypically notable.

3 UTR Alterations in Leukemogenesis
Alternative cleavage and polyadenylation (APA) are a differential selection of AAUAAA polyadenylation sites in 3 UTR by APA factors, leading to the expression of different mRNA isoforms that code for the same protein [53,54]. APA is globally regulated in response to extracellular stimuli that Cancers 2020, 12, 3854 5 of 30 regulate proliferation and differentiation. The first example of 3 UTR shortening was described during T cell activation in response to changes in cell proliferation status [55]. Most fast-proliferating cells, including embryonic stem cells, express transcripts with shorter 3 UTR, though some transcripts, such as those encoding for cell adhesion molecules, may have extended 3 UTR [56,57]. The length of 3 UTR can determine the intracellular protein localization. For example, the long 3 UTR of CD47, a protein conveying antiphagocytosis through the "do not eat me" signal in leukemic cells, enables efficient cell surface expression of CD47, whereas the short 3 UTR primarily localizes CD47 protein to the endoplasmic reticulum [58].
A meta-data analysis of microarray data by Mayr and Bartel demonstrated that shorter mRNA isoforms in cancer cells display increased stability through the loss of microRNA-mediated repression and typically produce ten-fold more protein [59]. The bioinformatics study of alternative polyadenylation in 358 Pan-Cancer tumor and normal pairs across seven types of cancers identified that 91% of genes expressed in cancer have shorter 3 -untranslated regions (3 UTRs) to avoid microRNA-mediated repression [60]. A somatic mutation in 3 UTR, however, can create a new site for miRNAs recognition, causing downregulation of tumor suppressor genes in AML [61].
3 UTR shortening is associated with increased activity of oncogenes in blood and immune cells. For example, fusion transcripts of the Mixed Lineage Leukemia (MLL) gene that lack its native 3 UTR are associated with the increased activity of those fusions in leukemia cell lines and tumors compared to fusions that retain MLL 3 UTR [62]. Strongly proliferative mantle cell lymphoma (MCL) tumors have exceptionally high Cyclin D1 mRNA levels, expressing short Cyclin D1 mRNA isoforms with truncated 3 UTRs [63].
A study of 452 CLL cases and 54 patients with monoclonal B-lymphocytosis, a precursor disorder, comprised a comprehensive evaluation of recurrent mutations in non-coding regions and found recurrent alterations in the 3 region of NOTCH1, which cause aberrant splicing events, increase NOTCH1 activity, and result in a more aggressive disease [33]. Another study by Lee et al. investigated the oncogenic potential of mRNA processing events in 59 cases of CLL [64]. RNA sequencing revealed the widespread recurrent upregulation of truncated mRNAs and proteins that were caused by intronic polyadenylation. Truncated mRNAs predominantly represented tumor suppressors lacking full-length structure and functionality. Importantly, the role of these genes in cancer was underestimated before due to a lower mutation rate on a DNA level. Therefore, mis-splicing and aberrant polyadenylation can be a driving force of hematopoietic malignancies with few detectible genetic mutations.
Aberrant splicing in 3 UTR of splicing factor hnRNPA1 and reduction of its mRNA levels initiate a chain of mis-splicing events affecting oncogenes and tumor suppressors in pediatric B-ALL [37]. This finding suggests that aberrant splicing disturbing 3 UTRs may be a common mechanism of leukemogenesis for both adult and pediatric patients [65].

Prospective Therapeutic Value of Targeting Non-Coding Pre-mRNA and mRNA Sequences
Could these genetic alterations disrupting non-coding pre-mRNA regulatory sequences and mRNA UTRs have diagnostic or prognostic value in cancer? A functional analysis of alternative spicing mapping cancer-associated changes to changes in proteins indicates that mis-splicing impacts domains classically affected by somatic mutations in different genes and can be considered as an independent oncogenic process [66]. Therefore, detection of mutations in non-coding sequences disrupting pre-mRNA splicing, mRNA stability, and protein synthesis can have diagnostic or prognostic value. However, data variability should be taken into consideration while exploring alternative and aberrant splicing as a marker of disease development and progression. First, the tissue-specific expression patterns of differentially spliced pre-mRNAs and the adaptive nature of alternative splicing, which changes drastically with microenvironment and age, suggest that genetic analysis of samples with identical genetic background is preferable in order to decrease data inconsistency [67]. Clinically relevant phenotypes such as resistance to therapeutics or tumor repopulating capacity Cancers 2020, 12, 3854 6 of 30 would be the right starting point for identification of splice variants promoting clonal expansion [68]. The standardization of tissue sampling procedures is particularly important for long-term studies, where the occurrence of clonal mutations could change significantly upon treatments [69]. The genetic studies show that cells corresponding to relapse are present in a minor subpopulation at diagnosis [70]. Therefore, technical inability to detect mutations and the rapidity at which mutagenesis occurs may compromise the reliability of genetic testing. For example, the mis-spliced CD19 mRNA isoforms progressing to relapse were detected by Fisher et al. at diagnosis [30]. Another study, however, did not detect the genetic variants found at CART-19 relapse just one month before the disease reoccurred [31].
Although most aberrantly spliced mRNAs undergo nonsense-mediated decay (NMD), the successfully processed and translated messengers can produce atypical, tumor associated neopeptides. As discussed above, alteration in 5 UTRs of mRNAs can also increase the production of cancer-specific protein isoforms from non-canonical TIS. Hematologic malignances, especially AML, often reveal antigens not expressed by normal cells. That leukemia associated antigens are targeted by αβ and γδ T cells, NKT and NK cells that are proven to be functional against AML in combination with effector ligands and cytokines (perforin, TRAIL, IFN-γ, IFN type I, and IL12) [71,72]. If presented on MHC class I or II of a cell, those neopeptides work as tumor associated antigens (TAAs) and mediate tumor immunogenicity [73]. Seen as foreign by the adaptive immune system, neoepitopes, identified by various approaches, typically associated with better treatment outcomes in solid tumors [74]. Computational analysis of WES from 91 CLLs allowed for prediction of 22 mutated HLA-binding peptides per leukemia. HLA binding was experimentally confirmed for ∼55% of such peptides. Further analysis of WES data on 2488 samples across 13 different cancer types estimated from dozens to thousands of putative neoantigens per tumor, suggesting that neoantigens are frequent in most tumors [75].
The large whole exome sequencing (WES) and RNA-seq studies identified widespread splicing alterations in around 30% of differentially expressed transcripts. Even though many of them are not cancer drivers, those aberrations can contribute to tumor immunogenicity [76]. Jayasinghe et al. bioinformatic analysis indicates that most splicing site-creating mutations (SCMs) were generated within the TP53 and GATA3 genes [77]. Tumors with SCMs expressed both T cell markers (PD-1, CD8A, and CD8B) and immune checkpoint blockade PD-L1 molecule, indicating that alternative splice forms induced by SCMs increase the overall immunogenicity of these cancers. The proposition that PD-L1 immunotherapy could be a potential treatment for samples containing SCMs requires further investigation with in vitro and in vivo models of leukemia [77].

Regulatory Non-Coding RNA Molecules
Several large-scale and single cell sequencing studies explored transcriptomes of normal and malignant hematopoietic cells [78][79][80][81][82]. RNA landscape of the normal human hematopoietic hierarchy, featuring 38,860 unique ncRNAs, 20,466 mRNAs, and 900 miRNAs, displays highly lineage-specific expression of all types of ncRNAs (long non-coding RNA (lncRNA), long intervening ncRNAs (lincRNAs), pseudogenes, antisense transcripts (AS), retained introns, miRNA, and small nucleolar RNAs (snoRNAs)) [79]. The ncRNA expression in leukemia cells is also vastly lineage-specific, often exhibiting pleotropic, context-and concentration-dependent effects on cell physiology. Nevertheless, certain ncRNA loss-or gain-of-function is strongly associated with tumorigenesis and genes encoding those ncRNAs are known as tumor suppressors and oncogenes similar to protein-coding genes [24]. Trans-acting ncRNAs regulate gene expression in distal genomic regions while cis-acting RNA molecules attenuate gene expression of the locus of their origin or nearby (not to be confused with the internal cis-acting RNA motifs discussed above).
Research strategies elucidating the role of ncRNAs in leukemia can be summarized as follows: (i) identification of highly up-or downregulated ncRNA common for certain histological and cytogenetic subtypes of leukemia by analyzing either primary tumors and body fluids, or previously published arrays such as The Cancer Genome Atlas (TCGA) database; (ii) evaluating ncRNAs as potential Cancers 2020, 12, 3854 7 of 30 biomarkers of leukemia in a relationship with white cell blood count, overall survival (OS), event-or disease-free survival (EFS, DFS), minimal residual disease (MRD), and risk of relapse; (iii) mechanistic studies of ncRNA function in a cell through interaction with DNA, RNA, and protein targets. Finally, a large body of work has been dedicated to understanding the role of ncRNA in chemoresistance and developing anti-ncRNA targeted therapies.

Long Non-Coding RNA
Long non-coding RNAs are primary RNA transcripts over 200 nucleotides in length, which are named and categorized based on their genomic origin. Relative to protein-coding sequences, lncRNAs are defined as (i) sense-overlapping, antisense-overlapping, or both (ii) bi-directional, transcribed from sense and anti-sense DNA strands of neighbor genes; (iii) intronic, when transcribed from distal introns; and (iv) intervening/intergenic (lincRNA), not overlapping with annotated coding genes [83,84]. The current version of LncBook lists 270,044 lncRNAs, but only 1867 lncRNA are experimentally validated [85,86]. Long ncRNA expression and processing are similar to protein-coding genes such as promoter conservation and lncRNA splicing. Typically lacking long ORFs, lncRNAs do not produce fully functional proteins. However, lncRNAs with conserved regions comprise three times more ORFs with evidence of translation than non-conserved sequences. In addition, the conserved regions of intergenic lncRNAs, such as CYRANO, MALAT1, NEAT1 and MEG3, are significantly enriched in protein-RNA interaction motifs [85]. The specific, nuclear retention sequences predetermine lncRNA nuclear localization. If those motifs are excluded during splicing, lncRNA can be transported to the cytoplasm [87].
MEG3, are significantly enriched in protein-RNA interaction motifs [85]. The specific, nuclear retention sequences predetermine lncRNA nuclear localization. If those motifs are excluded during splicing, lncRNA can be transported to the cytoplasm [87].
Through binding with DNA, RNA, and proteins in the nucleus and cytoplasm, lncRNAs influence gene expression epigenetically, co-transcriptionally, and post-transcriptionally, acting as oncogenes [88][89][90][91][92][93][94] or tumor suppressors [95][96][97][98][99][100] in cancer, Figure 1, Table 1. with Polycomb complexes and other adapter proteins, form connections between transcriptional regulators and distal DNA sequences through DNA looping; lncRNAs transcribed from antisense to protein-coding genes DNA strands, e.g., AS-RBM15, PU.1-AS, regulate expression of these genes posttranscriptionally (d); both transcriptional and posttranscriptional mechanisms of action were described for some lncRNAs, e.g., HOTAIR, HOTAIRM1, UCA1, that regulate gene expression in their genomic locus (acting in cis), and distal genomic regions (acting in trans). (b) pre-mRNA splicing: cisacting pre-mRNA motifs are recognized by trans-acting RNA and protein factors during pre-mRNA splicing. Inherited or somatic mutations in splicing regulatory sequences of pre-mRNA cause mRNA (e) mRNA stability: cis-acting regulatory elements in 3 UTR determine mRNA stability; (f) miRNA sponging, endogenous competing lncRNA: HOTAIR and HOTAIRM1 sequester specific miRNAs; alterations in endogenous competing RNA influence miRNA levels. (g) protein levels: depletion of miR-20a, miR-125b, and miR206b by HOTAIRM1 increases mRNA stability and translation of autophagy regulators ULK1, E2F1, and DRAM2, and induces PML-RARA degradation. One of well-studied lncRNAs, X-inactive specific transcript (XIST) is a large, 17 kb, transcript involved in X-chromosome genes' inactivation. Several conserved repeats of XIST mediate recruitment of the epigenetic Polycomb Repressive Complexes (PRC), initiating gene silencing on X chromosome [106]. Deletion of Xist in the murine blood compartment induced highly aggressive MDS/MPN suggesting that Xist has a genome-wide impact and acts as a potent suppressor of myeloid blood malignancies [107].
HOX gene loci-associated cis-acting lncRNAs, HOX transcript antisense RNA (HOTAIR) and HOXA transcript at the distal tip (HOTTIP), program active chromatin through interaction with Polycomb and other adapter proteins and play oncogenic roles in leukemia [108,109]. Recently, Luo et al. investigated aberrant activity of HOTTIP in AML and showed that HOTTIP coordinates HOXA-driven topologically associated domain (TAD), including the expression of the posterior HOXA genes. HOTTIP also binds in trans with promoters of key hematopoietic regulators like PBX3, MYC, KIT, CD33, MEIS2, and RUNX1. In mice, Hottip displayed oncogenic properties leading to AML-like disease by altering the homeotic-hematopoietic gene-associated chromatin signature and transcription programs [91,101].
Oncogenic lncRNA HOTAIR sustains leukemia growth and proliferation by negative epigenetic regulation of p15 genes in the nucleus and by sponging miR-193a away from c-KIT mRNA in the cytoplasm [110,111]. Another example of intergenic trans-acting lncRNA enhancing oncogene expression through miRNA titration, or a competing endogenous RNA (ceRNA), is CCAT1. Often upregulated in M4-M5 subtypes of AML, CCAT1 inhibits monocytic differentiation and promotes proliferation by reducing miR-155 availability and consequently increases levels of c-MYC [112].
In addition to miRNA sponging, lncRNA are capable of altering protein synthesis by interfering with translational machinery. Daniel Tenen's group showed that the interplay between PU.1 sense and antisense RNAs, regulated from shared cis-regulatory DNA elements, is important for maintaining physiological dosage of PU.1 [102]. Originating from an intronic promoter, PU.1 anti-sense transcript (PU.1-AS) disrupts PU.1 translation between the initiation and elongation steps by selective binding with eIF4A initiation factor [102]. Therefore, elevated expression of PU.1-AS leads to downregulation of PU.1 and promotes myeloid leukemia [103]. Conversely, AS-RBM15, an anti-sense RNA transcribed in the opposite direction within exon 1 of the megakaryocytic regulator RBM15, promotes terminal differentiation of hematopoietic progenitors by enhancing RBM15 translation in a 5 cap-dependent manner. The overlapping region between AS-RBM15 RNA and 5 UTR of RBM15 mRNA functions as an enhancer of RBM15 protein synthesis in megakaryocytic leukemia [95].
A comprehensive genomic study of 5037 tumor samples and 935 cancer cell lines among 13 types of cancer, including leukemia, revealed both commonly expressed and cancer-type specific lncRNAs [138]. Compared to corresponding normal tissues, 15% of significantly upregulated and 11% of downregulated lncRNA were detected in several cancer types, with PCAT7, PVT1, and HOTAIR among the most commonly expressed lncRNAs. The somatic copy number alterations (SCNA) via SNP microarray showed that ovarian and lung cancers had the most of high-frequency (>25%) loss-or gain-of-function lncRNA SCNAs. Whereas AML displayed very few SCNAs, high expression of Breast Cancer Associated lncRNA8 (BCAL8) correlated with poor prognosis. Cancer-associated index SNPs were located in 11.7% of lncRNA loci, and roughly half of them were found in close proximity to protein-coding genes.
Gao et al. analyzed the impact of somatic mutations and lncRNA expression across 17 cancer types, and its connection with miRNA expression, methylation, and TF-lncRNA interaction [139]. The scientists found that lncRNA genes located on chromosomes 17 and 1 are more frequently involved in cancer, about 54% of lncRNA mutations occurred only in one cancer type, and only 0.27% were dysregulated in more than eight cancers allowing them to be classified as "common" for the given cohort. Importantly, most of those lncRNAs function as regulators of chromatin assembly and transcription and have a cancer biomarker potential for prediction of susceptibility to cancer, association with disease recurrence, and poor survival rates [140].

Circular RNA
Circular RNAs (circRNAs) are single-stranded RNA sequences covalently linked into circles that range from 100 nt to over 4 kb in size. Similar to miRNA and lncRNA, they comprise evolutionary conserved genomic regions. The biogenesis of circRNAs is linked to splicing and circularization, so-called back-splicing, of exonic, intronic, and other non-coding fragments of newly transcribed RNA. Similar to lncRNA, circRNAs can be translated to proteins and negatively regulate miRNA function by competing with their RNA targets.
The stable structure of circRNAs suggests a long-lasting effect on cellular physiology, making circRNAs suitable diagnostic and prognostic markers. In fact, circRNAs, most intensively studied in AML, were identified as potential biomarkers that can be applied at diagnosis, remission, or associated with resistance to therapy [141].
For example, circ-RNA microarray screening of 115 human samples revealed a strong association of hsa_circ_0004277 expression with AML development: hsa_circ_0004277 levels were significantly downregulated at diagnosis and normalized in remission [142]. circ-ANAPC7 was also proposed to be an additional marker to identify AML [143], but further studies with a larger number of AML samples and normal progenitor controls are required to confirm these observations. The analysis of 113 AML patients and 42 healthy donors identified that circular RNA originating from the Vimentin gene (circ-VIM) is significantly upregulated and associated with shorter survival in patients with non-acute promyelocytic leukemia and cytogenetically normal AML [144]. High levels of Vimentin itself, a type III intermediate filament that maintains cell integrity, is also associated with AML aggressiveness (e.g., higher count of white blood cells and low overall survival), especially in older patients [145]. Though not capable of carcinogenic transformation on their own, f-circular-RNA produced from fusion genes can promote leukemia development and resistance to therapies [146]. circPAN3 was shown to contribute to drug resistance through the circPAN3-miR-153-5p/miR-183-5p-XIAP axis [147]. Another promising circ-RNA marker is circ-PVT1. Similar to lncRNA-PVT1, one of the most common long non-coding RNAs, circ-PVT1 is upregulated in AML and ALL and promotes cell proliferation through supporting c-MYC expression by sponging let-7 family and miR-125 [148,149].
Cis-acting RNA motifs determine biogenesis and functions of circRNAs [153]. Typically referred to as the repetitive and non-repetitive long flanking introns of pre-mRNA, altered cis-acting elements can potentially abolish or increase expression of circRNAs. The genome-wide in silico search for genetic variants of human circRNAs and analysis of cancer datasets showed that chromosome 17 has a relatively large number of health-related genetic circRNA variants, chromosome 7 contains the highest number of complex mutations, and chromosomes 2 and 1 exhibited the highest number of cancer-related variants. The circRNA-related genetic SNPs, insertions and deletions (INDEL) that might be common for multiple circRNAs have not yet been reported [154].

Short Non-Coding RNAs
The small and medium size, 18-200 nt, non-coding RNAs, e.g., small interfering RNAs (siRNAs), micro RNAs (miRNAs), PIWI-interacting RNAs (piRNAs), small nuclear RNA(snRNA), small nucleolar RNA (snoRNA), promoter-associated small RNAs (PASRs), transcription initiation RNAs (tiRNAs), telomere small RNAs (tel-sRNAs), centrosome-associated RNAs (crasiRNAs), and many others, compose an array of endogenous molecules regulating multiple processes in a cell at the transcriptional, co-transcriptional and posttranscriptional levels. Among all classes of short ncRNA identified to date, miRNAs role in cancer has been investigated most thoroughly [155]. miRNA Single or clustered genes encoding primary miRNA transcripts (pri-miRNAs) ultimately processed into short,~22 nucleotide sequences, are dispersed throughout the genome and mostly conserved among species. Transcribed by RNA Polymerase II, pri-miRNAs undergo processing by Drosha complex in the nucleus. The processed long miRNA precursors (pre-miRNAs) are exported to the cytoplasm by exportin 5 and cleaved into double-stranded short precursors of miRNAs. After a double-stranded miRNA is loaded into RISC complex, one of the RNA strands, the passenger, is removed, allowing the seed sequence of miRNA to pair with mRNA targets. The main characteristic of miRNA gene silencing pathways is that the single-stranded miRNAs facilitate translational repression and mRNA destabilization through imperfect base-pairing typically with the 3 UTRs.
The first evidence of miRNA gene dysfunction provoking a blood malignancy was reported in 2002 by Calin et al. [156]. The polycistronic RNA encoding for the precursor of miR-15a-1 and miR-16b-1 was missing in 70% of B-cell chronic lymphoblastic leukemia with translocation at 13q14. Several powerful oncogenes promoting CLL such as Cyclin D1, MCL1, and anti-apoptotic factor BCL2, were identified as the downstream targets of miR-15a-1 and miR-16b-1 [157].
Another vivid example of tumor suppressor miRNAs are miR-145 and miR-146a, which are dysregulated in 5-q syndrome, a subtype of MDS characterized by severe anemia, variable neutropenia, and atypical megakaryocytes. The deletion of 1.5 Mb DNA on the long arm of chromosome 5 (del5q) leads to miR-145 and miR-146a loss-of-function and a subsequent upregulation of Toll-interleukin-1 receptor domain-containing adaptor protein (TIRAP) and tumor necrosis factor receptor-associated factor-6 (TRAF6), triggering phenotypical and functional features of MDS [158].
Another miRNA family playing an essential role in AML, CLL, and lymphomas is miR-29 (isoforms miR-29a, miR-29b, and miR-29c). However, miR-29s role in blood and other malignancies is dual as they can act as oncogenes or tumor-suppressors in different histological types of tumors [160]. The contextand dose-dependent roles were reported for several miRNAs in various cancers [24,161]. For example, miR-125b overexpression is shown to induce either myeloid or lymphoid leukemia depending on the time course and expression levels of miR-125b [162,163]. Narayan et al. demonstrated that forced expression of miR-155 to high levels (>50-fold above controls) displayed antitumor activity in different types of AML (MLL-AF9, MLL-ENL, and HoxA9/Meis1). Conversely, moderate upregulation of miR-155 was associated with alternative target selection, repression of myeloid differentiation genes, and with leukemic phenotypes in vitro and in vivo [164]. MiR-126 regulates quiescence and self-renewal in normal and malignant human hematopoietic stem cells with distinct outcomes [165,166]. Surprisingly, both overexpression and knockout of miR-126 promote leukemogenesis in AE9a-induced mouse model [167].
Aberrant expression of miRNA in various subtypes of myeloid and lymphoid leukemia was extensively investigated, and thoroughly reviewed [168][169][170][171]. In addition, miRNAs detected in body liquids and peripheral blood mononuclear cells from adult and pediatric leukemia patients were evaluated as biomarkers. For example, low levels of tumor suppressor miR-206 in serum of pediatric AML patients were associated with upregulated Cyclin D1 and unfavorable prognosis [172]. By examining miRNA expression in normal blood cells, de novo and relapsed pediatric ALL, Rzepiel et al. found that miR-128-3p and miR-222-3p expression correlates with minimum residual disease (MRD). However, the routine methods of MDR detection were more sensitive and technically reliable [173]. Since miR-150 was identified as one of the most abundant miRNAs in chronic lymphoblastic leukemia, several studies reported both high and low miR-150 levels correlating with poor clinical outcomes in CLL patients. Interestingly, cellular and serum levels of miR-150 were associated with opposite clinical prognoses: low cellular and high serum miR-150 levels were associated with the disease burden [174], indicating that some other cells could possibly be releasing miR-150. The discrepancies between studies evaluating circulating miRNAs can be explained by tissue specificity e.g., serum, plasma, or other body liquids may contain different levels of the same miRNA, and normalization methods used in quantitative polymerase chain reaction analysis.
Similar to other classes of ncRNA, abnormal miRNA expression and processing in cancer are caused by structural and functional changes in the human genome: chromosomal rearrangements, deletions, amplifications, and deregulated epigenetic and transcriptional control of gene expression. Although copy number alterations (CNAs), amplification, and deletions are powerful genetic mechanisms of miRNA deregulation, they are not quite common for AML. By studying 113 cases of AML, Ramsingh et al. show that only 18% of patients have CNAs involving miRNA genes, while multiple alterations in epigenetic and transcriptional regulators are in charge of miRNA abnormal expression [175].
Germline variants in miRNA genes may have a profound effect on miRNA transcription and maturation [176,177]; however, there are lower numbers of SNPs in miRNA genes than in other regions of the human genome, and the polymorphisms mostly affect the regulatory pri-miRNA and pre-miRNA sequences rather than seed motifs [178][179][180]. Sequencing analysis of miRNAs that are dysregulated in CLL identified mutations in the primary precursor of miR-16-1-miR-15a that alter the processing of these miRNAs and can cause loss of function similar to a deletion [181]. Accordingly, somatic mutations within miRNA seed regions are rare genetic events [182,183].

Therapeutic Approaches for Targeting RNA Molecules
Traditionally, therapeutic approaches for targeting a primary RNA structures were based on introduction of complementary DNA or RNA oligonucleotides, or their chemical equivalents, into the target cells. Oligonucleotides can function through RNase H-mediated RNA degradation, RNA interference (RNAi), or through a non-degradative steric hindrance mechanism by replacing or repressing RNA-binding proteins [184].
Meant to silence gene expression by inducing degradation of target mRNAs, double-stranded siRNAs and single-stranded antisense oligonucleotides (ASOs/AONss) are designed to perfectly match the target sequence. Synthetic miRNAs are introduced into a cell either to replace downregulated endogenous miRNAs (RNA mimics) or block the endogenous miRNAs, which resembles an antisense approach. Dorrance et al. demonstrated a successful miR-126 targeting by the transferrin or anti-CD45.2 antibody-conjugated nanoparticles containing antagomiR-126 both in vitro, in CD34+ blasts sorted from primary elderly AML patients, and in vivo, using Mll PTD Flt3 ITD mouse model [185]. AntagomiR-126 treatments led to~80% decrease in miR-126 levels in CD34+ blasts and were accompanied with a significant reduction of long-term colony forming cells frequency and a depletion of quiescent CD34+ subfraction as examined by serial replating assays [185]. While multiple preclinical studies showed therapeutic potential of miRNA mimics and antagomirs in leukemia cell cultures and animal models, none of them seemed to move forward with the clinical trials [169]. The first miRNA mimic to treat solid tumors, MRX34, entered the clinic in 2013 [186]. MRX34 was designed to restore expression of vastly downregulated miR-34a, which directly regulates at least 24 known oncogenes. At some point, the trial was stopped due to life-threatening immune responses in several patients, but, ultimately, the study was competed using dexamethasone premedication and dose-escalation protocols. Overall, MRX34 demonstrated an acceptable safety for most of the patients and showed the evidence of antitumor activity in a subset of patients with refractory tumors [187].
The 18 clinical trials of anti-sense therapies in chronic and acute leukemias targeted exclusively transcriptional regulators, mostly BCR-ABL. While some studies reported a significant improvement in survival for particular groups of patients [188], AONs stability, the targeted delivery to tissues, immunogenicity, and off-target effect remain major obstacles for oligonucleotide-mediated therapies [189]. The proof of concept studies using structurally stable, resistant to nucleases double-stranded LNA GapmeRs, e.g., against lnc-THADA4-1 in Juvenile Myelomonocytic Leukemia (JMML) [132], and antisense double-stranded DNA oligonucleotides (ADO) against BCR-ABL in CML [190], suggest that RNase H-mediated RNA degradation is a potentially effective therapeutic strategy, that requires further validation in vivo.
Delivery efficiency remains one of the important problems in nucleotide-based therapies. Therapeutic molecules can be trapped in endosomes, lysosome or disposed through exocytosis and, therefore, remain inactive [191,192]. Delivering RNA therapeutics to the specific cell types is another challenge. Most of the delivery technologies, including advanced, non-immunogenic lipid nanoparticles (LNPs) loaded with modified RNA, cannot distinguish between various cell types causing off-target effect and reducing desirable outcomes. Dan Peer's group developed a modular platform for targeted RNAi therapeutics named ASSET (Anchored Secondary scFv Enabling Targeting), which coats the LNPs with monoclonal antibodies [193]. Recently, Veiga et al. utilized ASSET platform and mRNA loaded LNPs for targeted gene expression in Ly6c+ inflammatory leukocytes, and achieved a selective protein expression in vivo [194].
Several commercially viable AON-based therapies are currently FDA approved, and are aimed to treat cytomegalovirus (CMV) retinitis, common in people with a compromised immune system, and hereditary conditions such as Duchenne muscular dystrophy (DMD) and spinal muscular atrophy (SMA) [195]. In the inherited degenerative diseases, AON-based therapies demonstrate partial or full restoration of protein functions by modulating the altered splicing and translation [196].
A combined high-throughput screening of antisense oligonucleotides and small molecules identified compounds promoting exon 51 skipping in dystrophin pre-mRNA [197]. Similar screens identified small molecules inducing desirable splicing phenotype for SMA and enhancement of the survival motor neuron (SMN) protein levels, improving motor functions in mice [198]. Interestingly, RNA-seq analysis indicated that compounds were quite selective and did not have a widespread effect on the transcriptome. This discovery opened a new perspective in targeting of RNA primary and secondary structures by chemical compounds as well as inhibiting RNA-protein interactions in human disease [199]. Prior to this, the interaction of small molecules with RNA were extensively studied in viruses. For example, small molecules were shown to interfere with the HIV transactivation response and Rev response element [200].
Velagapudi et al. investigated oncogenic non-coding RNA targeting by known anti-cancer drugs [201]. The team described a small molecular microarray-based approach, AbsorbArray, which allows for unmodified compounds, including FDA approved chemotherapeutics, to be probed for binding to RNA motif libraries in a high-throughput format. The primary screening identified that topoisomerase inhibitors bind the Dicer site of pre-miR-21 and inhibit miR-21 biogenesis. In vitro, these compounds, e.g., mitoxantrone, reduced mature miR-21 levels and modulated miR-21-mediated invasive phenotype. Importantly, the chemical crosslinking and a pull-down assay (Chem-CLIP) studies confirmed physical interaction between pre-miR-21 and the small molecule. Among different classes of compounds, topoisomerase inhibitors, kinase inhibitors, and splicing modulators were key classes that bound RNA [201].
The high-throughput methods for investigating chemical compounds targeting RNA molecules and mechanisms of drugs targeting RNA-protein interactions were recently reviewed. Anita Donlic and Amanda Hargrove placed a unique emphasis on the specifics of RNA structural elements or RNA-mediated interactions that enable disease-related functions in mammalian systems as well as the phenotypic changes observed upon treatment with targeted ligands [202]. Zhu et al. provide a comprehensive overview of the commercialized RNA-mediated therapies and those that are under clinical investigation [203]. A recent review by Peng Wu discusses the selective strategies for targeting RNA-binding proteins, and the high-throughput screening approaches to identify inhibitors of RNA-protein interactions [204]. A common theme of these and similar articles highlights the importance of understanding the principles of RNA-ligands efficient design and producing libraries of more specific RNA-binding chemotypes. For more effective pre-clinical assessment, RNA and RBP inhibitor testing systems should include cellular assays investigating interactions and metabolism of full-length molecules in a cell and animal models.

Concluding Remarks
Once defined as architects of eukaryotic complexity and the dark matter of cancer genomes [23,205], ncRNA molecules could represent important yet challenging therapeutic targets due to their pleotropic and context-dependent effect. The dual role of posttranscriptional regulators acting as oncogenes and tumor suppressors, however, is not limited to RNA molecules, but RBPs as well. Therefore, understanding RNA metabolism in living systems and selecting ribonucleoprotein targets that are best suited for therapies is as important as understanding their structural characteristics.
Another level of RNA network complexity lays in the abundance and variety of ncRNA interactions with mRNA and other ncRNA molecules. The multifaceted ncRNAs acting as transcriptional, co-, and posttranscriptional regulators indicate the importance of understanding the circuitous architecture of the RNA network. Although the selective targeting of upregulated oncogenic RNA molecules may seem a step towards personalized medicine, in most clinical settings only a limited number of patients respond to targeted therapies that address a single genetic abnormality [2]. Thus, targeting key elements of regulatory modules or common structural elements affecting multiple targets could be a more effective strategy against genetically heterogeneous blood cancers.
Understanding the functional significance of the somatic point mutations and genomic variants located in non-coding and untranslated regions of the genome is also a challenge since they can influence the expression of distal genes at both transcriptional and posttranscriptional levels. Annotation of twenty-three million regulatory SNPs that are involved in a wide range of processes, including proximal and distal transcriptional and posttranscriptional regulation of gene expression, indicates that roughly half of them are involved in RBP-and miRNA-mediated posttranscriptional regulation [206]. A global high-resolution search for protein RNA-binding domains led to the observation that mutations causing monogenic diseases,~10,000 human diseases including sickle-cell anemia, were enriched in genomic regions encoding for unconventional RNA-protein interactions [207]. Therefore, the role of cisand trans-acting RNA regulatory elements and RBPs in human disease might be larger than currently known.
The concept of RNA-targeting therapeutics using ASO, siRNA, miRNA and other synthetic RNA has been proven to be effective in some degenerative diseases. The efficient and safe targeted delivery of RNA therapeutics into specific tissues will be key for expanding those approaches to other clinical indications including cancer. Recent discoveries in the chemical targeting of RNA motifs and identification of small molecules disrupting RNA-protein and RNA-RNA interactions open a new area in RNA therapeutics that may help in developing a next generation of anti-cancer drugs.
Funding: This study was supported in part by the NIH NCI grants CA191550 and CA243167 (V.S.S.).