Long non-coding RNAs in hematological malignancies: translating basic techniques into diagnostic and therapeutic strategies

SAGE is the first high-throughput sequencing method of analyzing transcriptomes. SAGE can comprehensively profile gene expression by generating short-stretches of unbiased cDNA sequence called SAGE tags via restriction enzyme digestion [25, 26]. In addition, SAGE-tags are concatenated before cloning and sequencing to allow both identification and quantification of transcripts including lncRNAs, which are not possible by using individual microarrays. Nonetheless, the efficiency and specificity of this technique are not compelling, and hence modifications are still needed [27].

Tiling arrays were originally used for both identification and characterization of lncRNAs. Briefly, it allows the global transcription mapping of genomic regions in order to provide comprehensive alternative splicing analysis, and discovery of novel transcriptional sites and polymorphisms [28‐30]. This microarray-based technique shares identical biophysical principle with the traditional genomic microarray through hybridizing labeled cDNA or RNA to probes (small nucleotide polymers) immobilized on a glass slide (substrate). However, instead of probing for known sequences or predicted genes as in conventional microarray, tiling arrays intensively probe for specific sequences that exist in a contiguous region. The tiled oligonucleotides can be designed to cover either specific chromosomal regions or even the whole genome [30, 31]. Despite being able to identify and quantify the expression level of transcripts, tiling arrays were practically replaced by NGS technologies owing to the noise generated through cross hybridization and the weak binding to probes by transcripts with repetitive sequences (a fact that hinders studies in this direction).

CAGE is one of the most advanced techniques used in molecular genomics. Generally, this NGS-based method allows generating a snapshot of the 5′ ends of mRNAs [32]. CAGE mainly relies on the isolation and sequencing of small cDNA sequence-tags that are initiated from the 5′ end of RNA transcripts. Akin to SAGE, this technique also requires concatenation of cDNA-tags before cloning and sequencing. However, CAGE is able to identify the location of transcripts. CAGE has several advantages over SAGE and RNA-seq in being able to identify transcriptionally active promoter regions and poll-II-derived transcription start sites (TSS). Nevertheless, this method is limited to the detection of 5′-capped transcripts, but not able to analyze non-capped or circular RNAs [32, 33].

Currently, RNA-seq is a prevailing standard method for identifying novel lncRNAs owing to the low expense of sequencing (per base) and the single-nucleotide resolution [34]. RNA-seq encompasses translating RNAs into cDNA after fragmentation, and ultra-high-throughput sequencing in NGS platforms such as Illumina. In addition, high sequencing depths are needed to detect low-expressing lncRNAs [35]. For lncRNA detection, RNA-seq experiments can be performed with rRNA-depletion to enrich the reads of mRNAs and lncRNAs. Moreover, both poly(A)⁺ and poly(A)⁻ RNA fractions should be targeted in order to identify all types of lncRNA [35].

Considering the existing information of different approaches, it is hard to select the best method for identifying lncRNAs. However, several technical specifications and application parameters can be considered for method selection. For example, hybridization-based methods like tiling arrays are restricted to detecting transcripts that correspond to existing known genomic sequence while NGS-based methods like RNA-seq can provide substantially more informative data such as the precise location of transcript boundaries down to single-base resolution, the single-nucleotide polymorphisms in the transcribed regions, etc. [34, 36]. Moreover, the capacity to distinguish different isoforms and to detect differential allelic expression is rather limited in hybridization-based methods while such capacity is much higher in methods based on Sanger sequencing or NGS. From a technical perspective, methods based on hybridization or Sanger sequencing require larger amounts of RNA, whereas NGS-based RNA-seq requires relatively smaller amounts of RNA and are comparatively inexpensive. Most importantly, NGS-based methods have higher sensitivity, specificity, and reproducibility in both biological and technical replicates than hybridization-based methods [36, 37]. Although NGS-based RNA-seq provides more advantages than hybridization-based methods, there are still some limitations in detecting and tracing the expression of rare RNA isoforms even with NGS-based methods. Hence, further improvement is desperately required for better understanding of the transcriptomes.

To predict the putative function of lncRNAs, a common approach termed as “guilt by association” has been utilized to correlate specific lncRNAs with diverse cellular and physiological processes across different cell or tissue types [38]. By carrying out gene expression profiling, the co-expression model between certain sets of protein-coding genes or pathways and a given lncRNA of interest can be identified systematically. This allows a genome-wide understanding of lncRNAs and their co-expressed coding genes that are presumably co-regulated [38].

The lncRNA-mRNA co-expression profiles can be set up in different cellular pathways after gathering information from high-throughput profiling approaches. For instance, Hung et al. investigated the lncRNAs co-expressed with cell cycle-related genes by an ultrahigh-resolution tiling microarray [39]. In the study, a gene module map was constructed to show the correlation of gene sets co-expressed with target lncRNA versus the Gene Ontology Biological Processes data set. Moreover, a similar module map with curated gene sets of different signaling pathways from the Molecular Signatures Database (MSigDB c2 collection) was created and further verified to be enriched for cell cycle-associated data sets. Besides, gene set enrichment analysis has been used to evaluate the genes positively or negatively co-regulated with a specific lncRNA of interest [40]. After pathway analysis based on a database called Kyoto Encyclopedia of Genes and Genomes, highly correlated biological processes with related gene sets and pathways were linked to the target lncRNA.

Recently, a few web tools have been developed using enrichment strategies to predict lncRNA functionality [41]. The contexts in a newly established platform (decodeRNA) are based on matching lncRNA, microRNA (miRNA), and mRNA expression profiles from The Cancer Genome Atlas [41]. This platform offers information about ncRNA-pathway associations and the related genes that contribute to the ncRNA-pathway associations. In addition, a number of bioinformatics tools and databases are currently available, which allow extensive study of lncRNA expression profiles, localization, conservation, and structures in silico [42]. With the great assistance of all these bioinformatics resources, hypotheses for potential functions of a targeted lncRNA can be proposed and then investigated by further loss-of-function approaches.

Following the identification of lncRNAs, it is imperative to determine whether these non-coding transcripts indeed possess biological functions or not. To ascertain the physiological roles of lncRNAs in the cell, experimental studies with perturbation of lncRNA expression are necessary in order to reveal the contribution of an lncRNA to particular phenotypes (e.g., cell cycle, cell proliferation, apoptosis, differentiation, etc.). Although gain-of-function studies may reveal the significance of trans-acting regulatory roles for lncRNAs, loss-of-function approaches still represent the standard and most common strategy to investigate the function of a gene in reverse genetics [43]. With extensive prior experience of mRNA knockdown and great advancement of genome manipulation technologies, there are already a variety of choices of knockdown or knockout methods available for lncRNAs. In this review, the most common genetic manipulation approaches will be discussed in the next section and their conceivable readouts related to hematological disorders will be discussed in the section Functional targets for probing biological effects of lncRNAs in blood cancer cells.

Toward this end, RNA interference (RNAi) has been utilized extensively to knockdown lncRNAs with many successful examples of loss-of-function studies [20]. RNAi approaches generally make use of transcripts 20–40 nt in length and complementary to the target RNA transcript. Upon binding, the subsequently formed duplexes will then be degraded via cellular machinery [44]. These approaches have been extensively applied mainly because they are relatively fast and easy to use. In essence, they are cost-effective and can be specifically engineered to target an RNA sequence in a precise manner. Akin to protein-coding mRNAs, lncRNA expression can be downregulated in targeted cells by means of two typical RNA silencing-mediated strategies. One strategy is to transfect target cells with small interfering RNAs (siRNAs) that target the transcript of interest in a transient fashion while another is to introduce short hairpin RNAs (shRNAs) that are stably expressed. The siRNA strategy can lead to intense downregulation of targeted lncRNA and allow a gradient of knockdown by utilizing different doses of siRNAs. On the contrary, the shRNA strategy provides sustained knockdown of the lncRNA target, and hence is more suitable for experiments investigating the prolonged effects of targeted lncRNA depletion [45].

Although RNAi approaches are widely used with many successful examples, concerns have been raised about the effectiveness of these strategies for the depletion of nuclear and enhancer-associated lncRNAs. It has been argued that these lncRNAs predominantly localize in the nucleus and this impedes their susceptibility to the RNAi machinery, which is primarily located in the cytoplasm [46]. Besides, the low expression level of lncRNAs and their more structured nature may also hinder the RNAi-based methods. Consequently, several siRNA sequences are usually screened to find out a more potent one for the effective knockdown of a specific lncRNA.

Alternatively, some other oligo-mediated knockdown approaches have also been used, such as antisense oligonucleotides (ASOs) or “gapmers,” which are synthetic single-stranded nucleic acid derivatives that have higher stability against degradation and are more accessible to nuclear RNA sequences [47]. ASOs are able to efficiently degrade nuclear lncRNAs via a mechanism dependent on ribonuclease H, leading to depletion of nascent transcripts. In short, ASOs can form a DNA-RNA hybrid upon binding to the target RNA and then promote the cleavage of RNA by ribonuclease H [48, 49].

A recent systematic in vitro study compared the efficacy of ASOs and RNAi against seven lncRNAs with different predominant subcellular locations. The results revealed that nuclear lncRNAs were knocked down much better by ASOs, whereas RNAi showed relatively greater silencing effect on cytoplasmic lncRNAs [46]. Indeed, the data showed that ASOs were also powerful in suppressing cytoplasmic RNA level and this indicates that ASOs target not only nuclear-localized RNAs but also nascent RNA transcripts in the cytoplasm. In conclusion, if the cellular localization pattern of a target lncRNA is not known, ASOs would give a better chance of success in comparison to RNAi-based knockdown. However, these oligo-based methods still share some drawbacks with RNAi, including incomplete knockdown, unpredictable off-target effects, and transient inhibition effects. All these impose limitations on loss-of-function analysis of lncRNAs. Nevertheless, given the relatively straightforward manner of operation, RNAi- and ASO-mediated knockdown strategies remain a helpful and valuable tool for the initial investigation of lncRNA functionality.

To address the limitations of RNAi and ASOs, programmable nuclease-directed genome-editing methods provide a powerful alternative approach to characterizing the functional roles of lncRNAs both in vitro and in vivo [50, 51]. Particularly, clustered regulatory interspaced short palindromic repeats-CRISPR-associated endonuclease 9 (CRISPR-Cas9) has already been widely utilized over the last few years because of its simplicity, efficiency, and robustness. This genome-editing tool is now able to be carried out at the DNA level and in an efficient and rapid manner to achieve partial or total deletion of lncRNA loci (Fig. 2a) [52], or to block the expression of lncRNA by interrupting the promoter region of lncRNAs (Fig. 2b) [53]. In short, the system makes good use of an endonuclease called Cas9, which is directed by a specially designed single-guide RNA (sgRNA) to the desired site and then performs a site-specific cutting at a target gene (i.e., DNA sequence) of interest in order to achieve gene editing.

Although CRISPR-Cas9 system has facilitated various genomic studies, there are some constraints when targeting lncRNAs. A popular way to knock out a protein-coding gene by using CRISPR-Cas9 is via frameshift mutations that can be easily induced at an ORF by Cas9-mediated cleavage followed by a non-homologous end-joining repair. However, it may not be applicable to the generation of a knockout effect for non-protein-coding genes because of their non-coding nature as well as our limited knowledge about their functional mechanisms. In general, most sequences particularly responsible for the molecular functions of lncRNA transcripts have not yet been characterized. As predicting the functionally active part of lncRNAs is now impossible, Cas9-induced insertion/deletion is unlikely to influence the activities of lncRNAs since their functional domains may still be retained. In addition, some lncRNAs exert their functions by the act of transcription per se instead of the lncRNA transcript [54, 55]. In that case, genetic manipulation should specifically target regulatory regions controlling the transcription, which are often poorly annotated or remain largely unknown, leading to more challenges when studying such type of lncRNAs. Therefore, more comprehensive approaches tailored to lncRNA are necessary and will be discussed below.

Involvement in altering cell proliferation is one of the basic phenomena of any functional molecules including lncRNAs. Several studies implied that lncRNAs could significantly alter leukemic cell proliferation via various signaling pathways. In some leukemia cases, lncRNA regulates the target gene through cis-regulation. A recent example showed that lncRNA NALT, located upstream of the NOTCH1 gene, promoted cell proliferation in T leukemia cell lines via transcriptional activation of the NOTCH signaling pathway [86]. Another example is UCA1, one of the most common lncRNAs, which is highly expressed in acute myeloid leukemia (AML) and enhances cell proliferation by suppressing the expression of cell cycle regulator p27^kip1 [87]. On the other hand, tumor suppressor lncRNA BM742401 has been identified in chronic lymphocytic leukemia (CLL) [88]. The functional activity of lncRNA BM742401 mainly relies on epigenetic alteration, showing specific inverse correlation with methylation status. Overexpression of lncRNA BM742401 inhibits CLL cell proliferation through caspase-9-dependent intrinsic pathway, which suggests that lncRNA BM742401 can function as tumor suppressor in CLL [88].

Table 2 lists cell proliferation-associated lncRNAs together with their effects on cell proliferation in different types of leukemia. In AML, CCAT1 and NEAT1 modulate cell proliferation by regulating miRNA-mediated pathways [89, 90], whereas MEG3 and UCA1 are involved in cell cycle-related pathways [87, 91]. TUG1 increases cell proliferation through targeting Aurora kinase A (AURKA) [92] while CASC15 expression limits cell proliferation by regulating transcription factor SOX4 expression [93]. In acute lymphoblastic leukemia (ALL), overexpressed linc-PINT decreases cell proliferation via apoptosis activation and cell cycle arrest [94]; however, upregulated NALT promotes cell proliferation through interacting with NOTCH signaling pathway [86]. A unique lncRNA, PVT1, has been involved in acute promyelocytic leukemia (APL) through promoting cell proliferation by MYC [95]. In addition to acute leukemia, MEG3 also plays a role in regulating cell proliferation in chronic myeloid leukemia (CML) through acting as miRNA sponge [96]. HULC and PLIN2 are involved in PI3K/AKT and GSK-3β/β-catenin signaling pathways, respectively, to promote cell proliferation in CML [97, 98].

Most recently, several approaches have been commonly used to characterize lncRNA involvement in hematopoietic cell proliferation (Table 2). siRNA/shRNA- and ASO-mediated knockdown systems are still the major strategies used to investigate the effect of target lncRNA in combination with MTT cell proliferation assay, Cell Counting Kit-8 (CCK-8) assay, or MTS cell proliferation assay (Table 2). The efficiency of all these methods is similar and is able to detect certain numbers of cells in 96-well plate format (3 × 10³ to 1 × 10⁴ cells). After the incubation period, absorbance is measured by simple spectrophotometry and finally data analysis carried out.

Functionally related to cell proliferation, numerous lncRNAs have been reported to participate in cell cycle regulation through their diverse roles as regulators at epigenetic, transcriptional, and post-transcriptional levels [99]. A well-studied lncRNA related to cell cycle regulation is ANRIL, which is transcribed antisense to the INK4b-ARF-INK4a locus encoding three important cyclin-dependent kinase (CDK) inhibitors (p15INK4b, p14ARF, and p16INK4a) [100]. It has been revealed that lncRNA ANRIL functions in trans by recruiting polycomb repressive complexes PRC1 and PRC2 to the INK4b-ARF-INK4a locus, and this results in the silencing of the gene cluster by H3K27-trimethylation [101]. A detailed study using leukemia patient samples revealed that a polymorphism located at the ANRIL locus showed strong association with ALL phenotype, which might be caused by the transcriptional changes of ANRIL and the resulting altered expression of CDKN2A/B that encodes CDK inhibitors.

Another example of cell cycle-associated lncRNA is MALAT1, which is related to numerous human cancers and whose activity has been implicated in acute monocytic leukemia [102] and multiple myeloma [103, 104]. A study examined the cell cycle-dependent expression pattern of MALAT1 by utilizing cell cycle synchronization to sort the cells into specific cell-cycle phases [105]. Results showed that the expression level of MALAT1 was higher during G1/S and M phases when compared to G1 and G2 stages. Further study demonstrated that MALAT1 was required for proper expression of B-Myb, which regulates cell cycle-related proteins in G2/M phase [105]. Another study also showed that MALAT1 interacted with a nuclear protein hnRNP C that is functionally involved in G2/M phase [106]. Other lncRNAs including HOTAIR [107, 108], HOXA11-AS [40], lncRNA-HEIH [109], MEG3 [110], ncRNA_CCND1 [111], and PANDA [39, 112] are reported with their significant roles in regulating the expression of cell cycle regulators and respective cell cycle stages (Table 3).

There are three widely used approaches for analyzing cell cycle progression by flow cytometry [113]. The first one is univariate analysis generally based alone on cellular DNA content stained with either propidium iodide (PI) or 4′,6′-diamidino-2-phenylindole (DAPI). This approach reveals the cell cycle distribution of cells in three major cell cycle stages (G1, S, and G2/M) and is also able to detect apoptotic cells with fractional DNA content. The second approach relies on the bivariate analysis of DNA content and proliferation-associated proteins (e.g., cyclins A, B, D, and E). Such method can distinguish G0 cells from G1 cells, and identify mitotic cells. It can also help to relate the expression profile of other intracellular proteins to the cell cycle stage. The last approach is based on the detection of 5′-bromo-2′-deoxyuridine (BrdU) incorporation to label the DNA-replicating cells. By time-lapse measurement of BrdU-labeled cells, their rate of progression through different cell cycle phases can be estimated.

A major cause of blood malignancies is the dysregulation of hematopoiesis, which is a complex and highly ordered process requiring tight coordination of cell lineage specification and differentiation under the precise control of specific gene expression. Emerging discoveries implicated the involvement of lncRNAs in both normal and malignant hematopoiesis mainly by their influence on the expression of key regulators at particular differentiation stage. Table 4 displays some key examples of hematopoietic differentiation-related lncRNAs. In general, it is more target-oriented for studying the involvement of a given lncRNA in a particular lineage of the hematopoietic system since there have been some cues provided by the analysis of high-throughput data during the identification and annotation of lncRNA (as well as the “guilt by association” process). Therefore, a specific hematopoietic lineage can be focused, and gain- or loss-of-function experiments can then be carried out to examine the effects of lncRNA depletion or ectopic expression on the targeted phenotype, which is usually the differentiation toward a particular lineage.

For instance, HOTAIRM1 is a myelopoiesis-associated regulatory lncRNA identified by tiling array and is transcribed antisense to the HOXA genes [114]. HOTAIRM1 appeared to be restricted to myeloid lineage according to the initial array data, and the subsequent cDNA panel as well as qPCR experiments confirmed its myeloid lineage specificity. In addition, a marked increase of HOTAIRM1 expression responding to all-trans retinoic acid (ATRA) treatment in both K562 and NB4 cells further showed a prominent association with induced granulocytic differentiation. The subsequent functional study revealed that HOTAIRM1 depletion compromised HOXA1 and HOXA4 expression during retinoic acid-induced myeloid differentiation and hence attenuated the transcription of myeloid differentiation genes such as CD11b and CD18 [115]. In addition, numerous lncRNAs have been reported to have potential roles in regulating hematopoietic lineage differentiation (Table 4), which include LINC00173 and NEAT1 (granulocytic) [24, 116], lncRNAp53int1, lnc-MC and lnc-DC (monocytic) [117‐119], and EGO (eosinophilic) [120]. PU.1 AS is a negative regulator of the master hematopoietic transcription factor PU.1 [121].

The colony-forming cell (CFC) assay is a technique utilized to assess the proliferation and differentiation ability of hematopoietic progenitors by their ability to form colonies in a semi-solid medium. When cultured with appropriate cytokines, CFCs can divide and differentiate into a colony of more mature cells, which can in turn be identified by light microscopy. The morphology and number of colonies generated by a fixed number of input cells can provide preliminary information about the capability of progenitor cells to grow and differentiate. However, this assay is useful for assessing myeloid (including granulocytic, monocytic, erythroid, and megakaryocytic lineages), but not lymphoid, differentiation. A study employing the use of CFC assay uncovered the function of LINC00173, which showed a very specific expression in mature granulocytes [24]. By bioinformatics analysis of a collective dataset from RNA-seq and microarray platforms, it was hypothesized that LINC00173 played a potential role during granulopoiesis. In the study, LINC00173 was knocked down in CD34⁺ human hematopoietic stem and progenitor cells (HSPCs) by shRNAs, which then resulted in defective granulocytic differentiation in vitro. The cells were examined by leukocyte peroxidase staining and cell surface marker CD66b. In addition, methylcellulose-based and collagen-based colony-forming assays were further carried out, and the results revealed a reduction of myeloid colony formation after suppression of LINC00173, whereas erythroid colony formation was unaffected. The findings suggested that LINC00173 was not only involved in granulocytic differentiation but also in the development of early myeloid progenitor cells [24].

Despite the availability of therapeutics for treating hematological malignancies, development of drug resistance is still a stumbling block for effective treatment. For example, imatinib, a selective BCR-ABL1 kinase inhibitor, is highly effective in treating CML. Yet, some patients still develop imatinib resistance and the underlying mechanism of imatinib resistance is still far from complete. Mechanistic studies have identified that imatinib resistance acquired in CML can result from point mutations in the BCR-ABL kinase domain [122], overexpression of multidrug resistance protein 1 (MDR1) [123], or microenvironmental protection of leukemic cells [124]. Besides, most recent studies demonstrate that diverse functional molecules, including lncRNAs, are immensely involved in the development of chemotherapeutic resistance in different cancers including hematological malignancies. Several lncRNAs associated with imatinib resistance have been identified and partially characterized (Table 5): UCA1, SNHG5, HOTAIR, and MEG3 [123, 125‐127]. (Table 5 is at the end of the document.)

It has been shown that overexpressed lncRNA UCA1 modulates CML-imatinib resistance by acting as a competitive endogenous RNA (ceRNA) against miR-16, and this in turn stimulates MDR1 expression and alters the drug efflux system [123]. Besides, lncRNA SNHG5 also acts as a ceRNA against miR-205-5p and enhances ABCC2 expression, which is the core regulator of imatinib resistance [125]. It has also been identified that lncRNA HOTAIR promotes MDR1 expression through activating PI3K/Akt pathways [126]. Along with these oncogenic and drug-resistance-regulating lncRNAs, there is one anti-oncogenic and anti-drug-resistance-regulating lncRNA named MEG3. LncRNA MEG3 has been characterized in CML as being capable of decreasing the expression of MDR1, MRP1, and ABCG2 by suppressing miR-21 [127]. Taking together all these evidences mentioned above, lncRNAs might be considered a promising novel target for enhancing the therapeutic efficacy of imatinib in CML patients. However, further studies are still required to discover the whole mechanistic scenario of lncRNA-mediated drug resistance.

To investigate the correlation between lncRNAs and drug resistance, different approaches are commonly used as described below. The prerequisite of functionally characterizing drug resistance-associated lncRNAs is to generate in vitro cell line resistant to anti-cancer drugs. This is done by continually treating the cells with increasing doses of an anti-cancer drug for a certain period of time [115, 123, 125, 127]. In some cases, specimens can also be collected from patients developing drug resistance. After acquiring resistant cells, knockdown or overexpression of lncRNAs related to targeted drug resistance is carried out. Then, cell proliferation, cell differentiation, cell cycle, and apoptosis can be analyzed by using MTT or CCK-8 assay as well as flow cytometry. The results will be used to determine whether the lncRNAs can enhance drug resistance or re-sensitize the resistant cells. The final step of such study would be to delineate the mechanism of lncRNA involvement in drug resistance by using advanced techniques such as RNA immunoprecipitation and chromatin immunoprecipitation assay [123, 128, 129]. The selection of mechanistic characterization approach depends on the interaction of lncRNAs with other bioactive molecules such as DNA, other RNAs, or proteins. There are other ways to develop chemoresistance, e.g., via the transfer of lncRNAs associated with drug resistance to other cells by exosomes or drug resistance mediated by bone marrow microenvironment. Different supplementary techniques need to be applied accordingly, including the isolation of exosomes secreted by chemoresistant cancer cells [129], and co-culture of anti-cancer drug-resistant cells and/or exosomes (macrovesicles) with bone marrow stromal cells.

Angiogenesis is the process of generating new blood vessels from a pre-existing vascular network. In normal physiological condition, angiogenesis is strongly maintained by a precise balance between stimulatory and inhibitory signals [130‐132]. However, in case of imbalance, abnormal blood vessels can be formed and incorporated during the progression of different diseases including cancers, atherosclerosis, rheumatoid arthritis, corneal neovascularization and ischemic disease, and cancer metastasis [132]. Relevant research has identified exosomes as a crucial factor to modulate the formation of new blood vessels. It is well known that both normal physiological cells and cancerous cells can secrete exosomes. Exosomes have intercellular communication power between neighboring cells as well as distant cells [133], and they transport different bioactive molecules including proteins, peptides, lipids, miRNAs, and lncRNAs that act as regulators in recipient cells in a paracrine or endocrine manner [134‐137].

Advanced research has identified the functional involvement of exosomes in CML-mediated angiogenesis. Briefly, exosomes secreted by CML cells contain different functional molecules such as miR-92a, a member of miR-17-92 cluster (angiogenesis stimulator), which can be internalized into human umbilical vein endothelial cells (HUVECs) and modulates their tube formation by activating Src signaling pathway [138, 139]. Moreover, CML-released exosomes downregulate a target of miR-92a (named intergenic a5) in HUVECs and consequently promote cell migration and tube formation [139]. Furthermore, these exosomes facilitate tube-formation capability of HUVECs under hypoxic condition rather than normoxic condition through downregulation of an angiogenesis inhibitor called EFNA3 [140]. It has also been observed that CML-secreted exosomes enhance IL8 expression and modulate angiogenesis [141]. Taking all evidences together, it can be clearly stated that CML-secreted exosomes are able to modulate formation and regulation of new blood vessels via transporting functional biomolecules like miRNAs, which can further contribute to the development of drug resistance.

The significance of angiogenesis in blood cancers, such as leukemia, has been noticed since it plays an important role in both hematopoiesis and leukemogenesis [142]. Several miRNAs have been identified and characterized in relation to irregular angiogenesis onset and progression in blood cancer, but lncRNA involvement in angiogenesis is still very poorly known. Despite this fact, angiogenesis-associated lncRNAs have been characterized in different solid cancers as listed in Table 6. For example, several lncRNAs, namely OR3A4, MALAT1, PVT1, and CASC2, are critically involved in the initiation and progression of angiogenesis in gastric cancer through interacting with different functional molecules such as VEGFA, VEGF-C, MMP9, STAT3, and FAK [143‐146]. On the other hand, TUG1 and XIST promote angiogenesis in glioblastoma via the interaction of miRNAs (miRNA-299 and miR-137) with their respective protein-coding genes (VEGFA, FOXC1, and ZO-2) [63, 147]. In glioma, HULC and H19 promote angiogenesis via up-expressing ESM-1 and VASH2 through PI3K/Akt/mTOR signaling pathway and miRNA-29a respectively [148, 149] while HULC and MVIH promote angiogenesis in hepatocellular carcinoma by upregulating SPHK1 and inhibiting PGK1 secretion respectively [150, 151]. TUG1 and CRNDE stimulate angiogenesis in hepatoblastoma through regulating VEGFA expression (by downregulation of miR-34a-5p), and mTOR signaling pathway respectively [152, 153]. Besides, HOTAIR promotes cell growth and angiogenesis in nasopharyngeal carcinoma by upregulating VEGFA and Ang2 by GRP78 [56], whereas MALAT1 promotes angiogenesis in neuroblastoma by upregulating FGF2 expression [154]. MIAT promotes ocular angiogenesis in diabetes through upregulating VEGF by miRNA-150-5p [155]. Instead of enhancing angiogenesis, one lncRNA named MEG3 suppresses tumor cell proliferation and angiogenesis in pituitary adenomas through suppressing VEGF signaling pathway [156].

LncRNAs can potentially be attractive biological markers for diagnostic and prognostic application because their expression patterns are much more tissue-specific or disease type-specific when compared with those of protein-coding genes [160]. Such specificity and sensitivity enable lncRNAs to be a good indicator or predictor of disease stage that can be reflected by their differential expression level with reference to normal tissue. Nowadays, biomarker screening in patient extracellular fluids is one of the most promising methods for early diagnosis with the additional merit of its non-invasiveness. It has been found that the alteration of lncRNA levels associated with tumorigenesis can also be detected in body fluids like blood plasma and urine, which can readily be collected from patients and hence makes lncRNA a good potential biomarker [161].

Prostate cancer antigen 3 (PCA3) is the best-known example of lncRNA applied in clinical diagnosis since it demonstrated higher diagnostic performance than the conventional detection of prostate-specific antigen (PSA) serum level. PCA3 is a prostate-specific lncRNA and is overexpressed in > 90% of primary prostate tumors compared to benign tissues [162], which can be detected in patient urine samples and is undetectable in other tumor types [163]. PCA3 detection has been used in urine-based molecular diagnostic test that was approved by the Food and Drug Administration (FDA) in the USA and is now widely applied to prostate cancer detection [164, 165].

Indeed, lncRNAs also provide diagnostic and prognostic value in hematological diseases. A clinical research study revealed a subset of lncRNAs called B-ALL-associated long RNAs (BALR) by microarray analysis. It has been found that BALR are able to predict the cytogenetic subtypes of B-acute lymphoblastic leukemia (B-ALL) among the most prevalent abnormalities: mixed lineage leukemia (MLL) rearrangement, E2A-PBX1 translocation, and TEL-AML1 fusion. Upon further clinicopathologic data analysis, a high expression level of BALR-2 was found to be associated with poor survival and reduced responses to prednisolone used in treatment of B-ALL [166]. This finding indicates the possibility of using lncRNA to sub-classify disease and predict treatment response.

Another example is lncRNA PRAL, which was downregulated in primary multiple myeloma (MM) cells, especially in MM cells with del(17p). Analysis of survival curves revealed a significantly shorter disease-free survival and overall survival in MM patients with low PRAL expression [167]. In the same study, a potential correlation between PRAL and bortezomib sensitivity via the action on miR-210 was also demonstrated. All these data implied that PRAL expression might be utilized to predict the progression and prognosis of MM patients as well as the efficacy of bortezomib treatment.

Overall, lncRNAs could conceivably serve as effective biomarkers for disease diagnosis and prognosis owing to their high specificity and relative ease of sampling. In addition, their differential expression can be quite readily detected by common molecular biology techniques such as microarray assay, real-time qPCR, and RNA-seq. It is expected that by combining different lncRNA candidates or together with conventional biomarkers, the clinical judgment of medical professionals can be greatly facilitated in the future. Nonetheless, PCA3 is the only lncRNA that has been recommended as an FDA-approved biomarker to date. The utility of lncRNAs as clinical biomarkers is still in its infancy and the possible use of most known disease-associated lncRNAs was not confirmed yet. Therefore, more studies investigating the precise relationship between lncRNAs and disease pathology are needed, and standardization of the detection approach is also required in order to promote the use of lncRNAs in clinical settings.

On the other hand, the unique specificity of lncRNAs also makes them remarkable therapeutic targets by reducing off-target side effects. With the advances in oligonucleotide-based therapeutics, any particular member of human transcriptome, which cannot be targeted by small molecules or antibody-drugs, can be reached by nucleic acid-based drugs [168, 169]. In particular, ASOs have been subjected to clinical trials and some have already been clinically approved by the FDA [170‐172]. This promising approach is also emerging as a feasible treatment by targeting lncRNAs, which may provide a wider range of therapeutic options due to their functional diversity.

As a proof of concept, lncRNA MALAT1 as a therapeutic target has been tested in vivo in an animal model and has demonstrated therapeutic efficacy. In a mouse MMTV (mouse mammary tumor virus)-PyMT breast cancer model, enhanced cell adhesion and cystic differentiation and reduced migration were shown upon ASO-targeted knockdown of MALAT1 [173]. Another example is lncRNA Ube3a-ats, which has been targeted in mice by the administration of ASOs and which offers a potential therapeutic target for Angelman syndrome [174]. Although this approach is hopeful, a great challenge for clinical application is to validate the efficient delivery and long-lasting effects in human subjects. Hence, more preclinical studies and clinical studies in humans are still required. Nevertheless, since nucleic acid-targeting therapeutic strategies are evolving swiftly in terms of improved toxicity and pharmacokinetics, the identification and evaluation of targeted lncRNA in diseases may be plausibly translated into clinical applications in the near future.