MicroRNAs and new biotechnological tools for its modulation and improving stress tolerance in plants

Summary MicroRNAs (miRNAs) modulate the abundance and spatial–temporal accumulation of target mRNAs and indirectly regulate several plant processes. Transcriptional regulation of the genes encoding miRNAs (MIR genes) can be activated by numerous transcription factors, which themselves are regulated by other miRNAs. Fine‐tuning of MIR genes or miRNAs is a powerful biotechnological strategy to improve tolerance to abiotic or biotic stresses in crops of economic importance. Current approaches for miRNA fine‐tuning are based on the down‐ or up‐regulation of MIR gene transcription and the use of genetic engineering tools to manipulate the final concentration of these miRNAs in the cytoplasm. Transgenesis, cisgenesis, intragenesis, artificial MIR genes, endogenous and artificial target mimicry, MIR genes editing using Meganucleases, ZNF proteins, TALENs and CRISPR/Cas9 or CRISPR/Cpf1, CRISPR/dCas9 or dCpf1, CRISPR13a, topical delivery of miRNAs and epigenetic memory have been successfully explored to MIR gene or miRNA modulation and improve agronomic traits in several model or crop plants. However, advantages and drawbacks of each of these new biotechnological tools (NBTs) are still not well understood. In this review, we provide a brief overview of the biogenesis and role of miRNAs in response to abiotic or biotic stresses, we present critically the main NBTs used for the manipulation of MIR genes and miRNAs, we show current efforts and findings with the MIR genes and miRNAs modulation in plants, and we summarize the advantages and drawbacks of these NBTs and provide some alternatives to overcome. Finally, challenges and future perspectives to miRNA modulating in important crops are also discussed.


Summary
MicroRNAs (miRNAs) modulate the abundance and spatial-temporal accumulation of target mRNAs and indirectly regulate several plant processes. Transcriptional regulation of the genes encoding miRNAs (MIR genes) can be activated by numerous transcription factors, which themselves are regulated by other miRNAs. Fine-tuning of MIR genes or miRNAs is a powerful biotechnological strategy to improve tolerance to abiotic or biotic stresses in crops of economic importance. Current approaches for miRNA fine-tuning are based on the down-or up-regulation of MIR gene transcription and the use of genetic engineering tools to manipulate the final concentration of these miRNAs in the cytoplasm. Transgenesis, cisgenesis, intragenesis, artificial MIR genes, endogenous and artificial target mimicry, MIR genes editing using Meganucleases, ZNF proteins, TALENs and CRISPR/Cas9 or CRISPR/Cpf1, CRISPR/dCas9 or dCpf1, CRISPR13a, topical delivery of miRNAs and epigenetic memory have been successfully explored to MIR gene or miRNA modulation and improve agronomic traits in several model or crop plants. However, advantages and drawbacks of each of these new biotechnological tools (NBTs) are still not well understood. In this review, we provide a brief overview of the biogenesis and role of miRNAs in response to abiotic or biotic stresses, we present critically the main NBTs used for the manipulation of MIR genes and miRNAs, we show current efforts and findings with the MIR genes and miRNAs modulation in plants, and we summarize the advantages and drawbacks of these NBTs and provide some alternatives to overcome. Finally, challenges and future perspectives to miRNA modulating in important crops are also discussed.

Background
Plants are constantly challenged by numerous adverse conditions that modulate their evolution (Crisp et al., 2016). Plant responses to stresses involve a broad regulation of numerous genes, interfering with several agronomic traits, such as plant growth and productivity (Hackenberg et al., 2015;Yi et al., 2015). Understanding these mechanisms is important for the development of biotechnological tools to improve desirable agronomic traits (Teotia et al., 2016;Zhang, 2015).
MicroRNAs (miRNAs) are short (21-24 nucleotide) RNAs that modulate the amount and spatial-temporal accumulation of target mRNAs and indirectly interfere with several plant pathways. These molecules are derived from noncoding RNAs arising from the gene expression of miRNAs (MIR genes). MIR gene transcription can be activated by numerous transcription factors that can be regulated by another set of miRNAs. In addition, MIR genes or their transcripts are often regulated by other miRNAs.
Several types of abiotic or biotic stresses alter MIR gene expression profiles, modulating the accumulation of miRNAs and consequently of the targeted mRNAs (Ferdous et al., 2015;Hackenberg et al., 2015). In plants, miRNA targets form a broad group of genes involved in numerous biological processes, including development and defence responses to pathogens, insects and environmental stresses (Hackenberg et al., 2015;Yi et al., 2015). The modulation of target gene expression by miRNAs may occur at the transcriptional (site-specific DNA methylation mediated by miRNAs) and posttranscriptional (by mRNA degradation, translational inhibition or RNA deadenylation) levels (Borges and Martienssen, 2015). Stress-associated regulatory networks involving the activity of miRNAs are poorly understood, and unravelling such mechanisms is further complicated by the fact that one miRNA may regulate several genes and that some genes may be regulated by multiple miRNAs.
The understanding of the complexity of MIR genes has rapidly increased as a consequence of next-generation sequencing (NGS) and degradome sequencing technologies, as well as the exploitation of bioinformatics tools, open-source miRNA databases and stem-loop real-time RT-PCR. The constitutive, tissue-specific, or stress-or senescence-induced overexpression of specific MIR genes has been shown to improve desired agronomic traits in different plant species (Dong and Pei, 2014;Hajyzadeh et al., 2015;Ni et al., 2012;Trumbo et al., 2015;Zhang, 2015). In addition, artificial MIR genes, endogenous and artificial target mimicry, Meganucleases, ZNFs, TALENs, CRISPR/Cas9, CRISPR/ Cpf1 or CRISPR/Cas13a systems, and pri-miRNA or mature miRNA topical delivery have been shown to be useful for modulating miRNA accumulation.

Plant small RNA biogenesis: a brief overview
RNA interference (RNAi) in plants is a well-known mechanism that controls genome stability and epigenetic reprogramming, plant development, reproduction, defence responses and several other biological processes by regulating gene expression. Small interfering RNAs (siRNA) and miRNA are the two main small RNA (sRNA) classes involved in this regulation. The sRNAs are derived from single-or double-strand RNA (ss-or dsRNA) intermediates that form hairpin-like precursors and are subsequently processed by four DICER-LIKE proteins (DCL1 to 4; reviewed by Borges and Martienssen, 2015). miRNAs originate from species-or family-specific noncoding RNAs transcribed from introns, exons or intergenic regions. MIR genes are generally transcribed by RNA polymerase II, originating the primary miRNA transcripts (pri-miRNAs) that contain imperfect and self-complementary foldback regions. Following 5 0 m 7 Gcap addition and 3 0 polyadenylation, pri-miRNAs are converted into miRNA precursor sequences (pre-miRNAs) by DCL1 and other associated proteins (Borges and Martienssen, 2015). Pre-miRNAs are exported from nuclear Cajal bodies to cytoplasmatic P-bodies by HASTY protein and then processed by DCL1-4 to generate miRNA duplexes of typically 21-24 nucleotides in length. Different DCLs can process a single pre-miRNA, producing miRNA molecules of distinct sizes, for example, DLC1 and DCL4 of 21 nucleotides, DCL2 of 22 nucleotides and DCL3 of 24 nucleotides in length. These duplex miRNAs are 2 0 -O-methylated at both 3 0ends by HUA ENHANCER 1 (HEN1), protecting the miRNAs from the uridylation and degradation initiated by the nucleotidyl transferases HEN1 SUPPRESSOR 1 (HESO1) and UTP:RNA uridylyltransferase 1 (URT1) (Tu et al., 2015). Then, mature miRNAs (miRNA-5p or miRNA-3p strand) are loaded onto Argonaute (AGO) proteins and incorporated into a RNA-induced silencing complex (RISC), while miRNA star (miRNA*) molecules are most often degraded, although under certain circumstances these products might also be loaded onto AGO and complexed into a RISC. The miRNA-associated AGO scans RNA molecules for sequence or near-perfect sequence complementarity and promotes posttranscriptional gene silencing (PTGS) by catalysing the endonucleolytic cleavage, translation inhibition or deadenylation of the RNA target. Furthermore, Arabidopsis 24-nucleotide miRNAs associate with AGO4, AGO6 and AGO9 to mediate the transcriptional gene silencing (TGS) of target genes through RNAdirected DNA methylation (RdDM) (Borges and Martienssen, 2015; Figure 1). In addition, miRNA compartmentalization has been described in plants. For example, in Arabidopsis, miR390, AGO7 and TAS3 transcripts are complexed into cytoplasmic siRNA bodies that function in the biogenesis of TAS3-derived trans-acting siRNA (Mart ınez de Alba et al., 2015).
Posttranscriptional gene silencing can be amplified when target RNAs are cleaved and recognized by RNA-dependent RNA polymerase (RDR) proteins, resulting in the de novo formation of dsRNA molecules, which are processed by DCLs into secondary sRNAs. These secondary sRNAs accumulate in the cytoplasm and can move cell-to-cell through plasmodesmata, constituting the systemic silencing signal (Borges and Martienssen, 2015).

The role of miRNA in plant abiotic or biotic stress responses
Abiotic stresses, such as water deficit, salinity, low temperature, high temperature, heavy metal exposure, nutritional deprivation and high light intensity, as well as biotic stresses, such as viruses, bacteria, fungi, nematodes and insects, are major constraints to crop production worldwide. Throughout their evolution, plants have improved and developed mechanisms to respond and adapt to stressful conditions, including pathways in which miRNAs play a critical role in promoting stress tolerance (Shriram et al., 2016). MIR genes are up-or down-regulated in response to abiotic (reviewed by Ferdous et al., 2015;Hackenberg et al., 2015) and biotic stresses (Gupta et al., 2014) in numerous species, including soya bean, sugarcane, rice, maize, wheat and tomato. Studies on the expression or accumulation of these miRNAs have provided several lines of evidence to better understand the regulatory networks associated with defence mechanisms against different types of stresses. From these findings, several biotechnological tools have been applied for fine-tuning these networks and improving tolerance to stresses in important crops (Table 1).
The overexpression, up-or down-regulation or knock-in of transcribed MIR gene sequences has confirmed the involvement of miRNAs in stress responses in different plant species. For example, MIR gene overexpression with constitutive promoters (e.g. Cauliflower mosaic virus (CaMV) 35S, maize ubiquitin 1 and rice actin 1) has produced desirable agronomic traits, such as drought, cold, heat and salinity tolerance and resistance to pathogens (Table 1). Additional examples of patented inventions using miRNA overexpression include improved resistance to cyst nematodes (miR164 and miR396; WO2012058266 A1 and WO2012149316 A2), tolerance to drought (miR166; CN102250903A) and salinity (miR397; WO2007103767 A2), artificial MIR genes (US8536405 B2, WO2009079548 A3) and target mimicry (EP2873735 A1, WO2012056401A1).
However, strong constitutive overexpression often causes undesirable pleiotropic effects because miRNAs are frequently involved in the regulation of a number of miRNAs (Ferdous et al., 2017). The overaccumulation of certain miRNAs alters the expression of essential target genes involved in plant development, which may produce undesirable phenotypes (Trumbo et al., 2015). Greater target specificity can be achieved with overexpression driven by tissue-specific (Niu et al., 2016) or stress-inducible (Gao et al., 2015) promoters. Additional strategies include the overexpression of target mRNAs that are resistant to specific miRNAs (Guan et al., 2013), the expression of artificial target mimics that can cancel out the effect of endogenous miRNA activity (Sharma et al., 2016) (Table 1). More recently, genome-editing technologies based on the CRISPR/Cas9 or CRISPR/Cpf1 system have revealed new insights into miRNA fine-tuning, which has Figure 1 Overview of the miRNA pathway in Arabidopsis thaliana. In brief, MIR genes are transcribed by RNA polymerase II in the nucleus, and their primary transcripts are processed by the addition at 5 0 terminus of a 7-methyl guanosine cap, 3 0 polyadenylated and RNA spliced. Then, the transcripts fold into primary miRNAs (pri-miRNAs), which are quickly processed by DICER-LIKE 1 (DCL1), supported by zinc-finger protein SERRATE (SE), a dsRNAbinding protein that interacts with DCL1 (HYL1) and other molecular factors, generating the precursor miRNAs (pre-miRNAs) in Cajal bodies. Then, pre-miRNAs are nucleus-cytoplasm transported by HASTY (HST) to P-bodies. Next, the pre-miRNAs are processed by DCL1 to DCL4 originating mature miRNA duplexes of 21-24 nucleotides (nt) in length, which are immediately methylated by HUA ENHANCER1 (HEN1) at the 3 0 terminus. The double-strand miRNA is separated, and single-strand mature miRNA is integrated into RNA-induced silencing complex (RISC), while the strand complementary to the mature miRNA (miRNA star) is degraded or "reactivated" and integrated into the RISC. Mature miRNAs of 21-23 nt in length associated with HEN SUPPRESSOR 1 (HESO1), UTP:RNA uridylyltransferase 1 (URT1), Heat Shock Protein 90 (HSP90), Argonaute proteins 1 to 10 (AGO1 to 10) and other molecular factors are integrated into the RISC, which successively scans all cytoplasmic messenger RNAs (mRNA) mainly based on sequence homology. MiRNAs act in posttranscriptional gene silencing (PTGS) by mediating target mRNA cleavage, decapping or deadenylation by AGO1 to 10 proteins or engage in translation and elongation repression. Additionally, miRNAs of 24 nt in length are complexed with AGO4 and directed to the nucleus, acting on RNA-directed DNA methylation (RdDM), subsequently inducing transcriptional gene silencing (TGS). Thus, MIR genes can undergo TGS driven mainly by 24-nt miRNAs, and this methylation is stabilized and maintained by heterochromatic siRNAs (hetsiRNAs). SMALL RNA DEGRADING NUCLEASE 1 to 3 (SDN1 to 3) functions in the turnover of miRNA-5p or miRNA-3p before it is directed to the RISC. Finally, miRNAs produced and accumulated in single cells can be systematically moved (cell-to-cell) by the vascular system comprising phloem and xylem and spreading to adjacent cells (reviewed by Borges and Martienssen, 2015). Oryza sativa

Osa-miR393a
Agrostis stolonifera Altered phenotype (fewer amount but longer tillers, and reduced stomata density and denser cuticles) and improved tolerance to multiple stresses (drought, heat and salinity, and increased uptake of potassium)

Arabidopsis thaliana
Regulation of leaf morphology, ABA hypersensitivity and ABA-associated phenotypes and improves of drought stress tolerance Song et al. (2013Song et al. ( , 2012 Arabidopsis thaliana been employed in the editing (Zhou et al., 2017) or transcriptional regulation (Lowder et al., 2015;Tang et al., 2017) of MIR genes. Next, we present and discuss the advantages and limitations of each of these strategies.
Biotechnological approaches to fine-tuning of miRNA activity

Transgenesis, cisgenesis and intragenesis
The transference of one or more MIR genes between noncrossable plant species (transgenesis) or between crossable or the same species (cisgenesis and intragenesis) has been successfully performed in several plant species (Table 1). In addition, conceptually, any element used in T-DNA can be engineered to enhance transgene expression, or the same native sequence can be used to maintain the original cisgene features. Unlike cisgenesis, intragenesis employs hybrid sequences (optimized genes and/or additional promoter and terminator sequences) from sexually compatible species (Holme et al., 2013). For example, strong constitutive (mostly CaMV 35S), strong endogenous (Lu et al., 2010;Yang et al., 2013) or native tissue-specific or stressinduced (Niu et al., 2016) promoters were exploited to overexpress MIR genes (Table 1). However, the manipulation of MIR genes using any of these strategies, especially overexpression driven by a strong promoter, has also resulted in undesirable traits, such as pleiotropic phenotypes (Table 1). This result can be explained by the involvement of miRNAs in diverse and complex regulatory networks. Strong overexpression impairs the finetuning of several biological pathways, which might be overcome using specific promoters (e.g. tissue-specific, stress-induced or developmental stage-specific promoters).

Artificial MIR genes
Currently, various strategies exploit PTGS to down-regulate or uncover the functions of specific genes (Figure 2a). These approaches are primarily based on the accumulation of siRNAs derived from dsRNAs using a virus-induced gene silencing model (VIGS) and by the constitutive overexpression of an engineered target gene (full or partial length) in the antisense or sense/ antisense orientation. However, these approaches produce a diverse set of siRNAs that might potentially silence nontarget genes (resulting in off-target effects). To overcome this problem, an artificial MIR gene (amiRNA) strategy was developed to produce specific miRNAs and effectively silence target genes . These amiRNAs have a conserved secondary foldback structure similar to that of a typical pre-miRNA. Nonetheless, the original miRNA-5p:miRNA-3p sequence is replaced by an engineered miRNA targeting a specific mRNA. Thus, amiRNAs can be engineered to target any mRNA with higher specificity compared to strategies based on dsRNA overexpression or siRNA accumulation. Pre-amiRNA processing typically results in a single amiRNA targeting for a known sequence, thus helping to avoid off-target effects. In addition, the Figure 2 Constitutive or transient expression of (a) artificial MIR genes to accumulate artificial miRNA (amiRNA) and the depletion of the target mRNA . In brief, the amiRNA gene under the control of a typical promoter is transcribed by RNA polymerase II in the nucleus, and primary transcripts are similarly processed to canonical miRNAs, originating primary amiRNA (pri-amiRNA). The pri-amiRNA is processed by DCL1, SE and HYL1, originating precursor amiRNA (pre-amiRNA), which are processed again preferentially by DCL1, resulting in the increased accumulation of 21-nt amiRNA duplexes. However, it is not yet clear whether DCL2-4 also acts on this pre-amiRNA. Niu et al. (2006) showed that DCL1 development has a major effect on pre-amiRNA. However, DCL1 knockout plants (dcl1) also showed the accumulation of these 21-nt amiRNAs, although comparatively low accumulation was observed. This finding may suggest that DCL2-4 can also act in pre-amiRNA processing. Then, these 21-nt amiRNAs are methylated by HEN1, which subsequently undergoes nucleus-cytoplasm transport by HST to P-bodies and becomes integrated into the RISC, where it will act in a manner similar to canonical miRNAs on PTGS. (b) Target mimicry strategy to deplete specific miRNAs (Peng et al., 2018;Zhang et al., 2017). The constitutive or transient expression of the target mimic gene driven by a specific promoter is transcribed by RNA polymerase II in the nucleus. The primary transcripts are processed with the addition at 5 0 terminus of a 7-methyl guanosine cap, 3 0 polyadenylation and RNA splicing, originating primary mimic mRNA, which is then transported to the cytoplasm, thereby depleting the miRNAs. systemic movement of these mature amiRNAs is restricted in some cases (Carlsbecker et al., 2010), and the production of secondary siRNAs from pre-amiRNA sequences is very limited (Manavella et al., 2012). Furthermore, amiRNAs are stable and inheritable. The main drawbacks of this strategy are the selection of backbone or pre-amiRNA sequences for effective silencing without any off-target effects (Carbonell et al., 2015). For major target specificity, mature amiRNAs must have low sequence similarity with nontarget genes . In addition to the control of endogenous gene expression and the study of the role of new MIR genes, an amiRNA strategy has also successfully been used to knock out genes from insect pests, nematodes, viruses and other phytopathogens (Kis et al., 2016;Wagaba et al., 2016). Similarly, Arabidopsis expressing two amiRNAs targeting the mRNA of the P69 and HC-Pro proteins showed resistance to Turnip yellow mosaic virus and Turnip mosaic virus, respectively (Niu et al., 2006). Agrawal et al. (2015) showed that tobacco overexpressing an engineered amiRNA was resistant to Helicoverpa armigera.
Ju et al. (2017) developed a new overexpression system for artificial and endogenous miRNAs and siRNAs based on delivery from a viral satellite DNA vector for functional analysis in plants. This viral system generated promising results for the overexpression of endogenous or artificial MIR gene, siRNAs and transacting siRNAs in Nicotiana benthamiana.

Endogenous and artificial target mimicry
Similar to the overexpression of MIR genes, the negative regulation of the accumulation of some specific miRNAs allows the achievement of desirable agronomic characteristics. Endogenous target mimicry (eTMs) is another natural mechanism involved in modulating miRNA accumulation, controlling several biological processes in plants (Karak€ ulah et al., 2016;Lin et al., 2015). These eTMs are innumerous long noncoding RNA (lncRNA) or circular noncoding RNAs (circRNAs) transcribed from genome and differential expressed usually in response to stresses or other adverse conditions (Karak€ ulah et al., 2016;Liu et al., 2017). It acts as natural SPONGES mainly for quick fine-tune of miRNAs in plant response or adaptation to a new condition. To reproduce this mechanism, an artificial short tandem target mimic (STTM) strategy was developed to modulate miRNAs accumulation and improve desirable agronomic traits. This strategy is based on the transient or constitutive overexpression (driven by constitutive, tissue-specific or induced promoters) of an engineered lncRNA with high nucleotide sequence identity to target mRNA  Figure 2b). Similarly, the overexpression of synthetic circRNA is also suggested as potential alternative for miRNA modulation in plants. These STTMs contain two or more conserved binding site for the specific target miRNA, but with three nucleotide mismatches at the miRNA-cleavage site, which prevent its cleavage, whereas the miRNA remains hybridized and biologically inactivated. Thus, STTM sequester miRNAs from the endogenous target mRNA resulting in its up-regulation (Franco-Zorrilla et al., 2007). Several STTMs targeting the MIR genes in model and crop plants have been recently engineered and constitutively expressed as transgenes for the comprehensive functional analysis of miRNAs (Peng et al., 2018;Zhang et al., 2017). In addition, STTMs have been optimized to enhance lossof-function phenotypes caused by artificial single target mimics. Similarly, miRNA SPONGES are another synthetic transcript produced from transgenes with activity similar to STTMs. These RNAs SPONGES contain multiple miRNA binding sites in tandem repeated and are often used to inhibit whole families of related miRNAs in both plant and animal systems Thomson and Dinger, 2016). For example, Jiang et al. (2018) improved tomato resistance to Phytophthora infestans using a STTM strategy to silence miR482b from Solanum lycopersicum.

MIR gene editing using endonucleases
Meganucleases and zinc-finger proteins (ZNFs) were the first endonucleases engineered for plant genome editing. Meganucleases recognize double-strand DNA sequences of 12-42 nucleotides in length in a highly specific manner, consequently restricting the number of targets. In contrast, ZNFs consist of two modules of tandem repeat DNA-binding domains flanking the FokI nuclease catalytic domain (Osakabe and Osakabe, 2015). Each of these domains recognizes a unique nucleotide triplet, while each module has specificity for ≥24 nucleotides. Similar to ZNFs, transcription activator-like effector nucleases (TALENs) comprise two modules of tandem repeat DNA-binding motifs flanking a FokI motif. However, each TALEN DNA-binding motif recognizes a single nucleotide. Recently, clustered regularly interspaced short palindromic repeats/CRISPR-associated protein-9 nuclease (CRISPR/Cas9), CRISPR/Cpf1 or CRISPR/Csm1 systems, a new nuclease class guided by RNA (guide RNA), have been optimized for plant genome editing (Osakabe et al., 2016; Granted patent US9896696B2; Wang et al., 2018a). Both nucleases cause double-strand breaks (DSBs) at the target site, and during DSB repair, the insertion or deletion (indels) of nucleotides may occur. In addition to indels, nucleotide-specific editing can be achieved using engineered donor DNA.
CRISPR/Cas9 nonhomologous end joining (NHEJ; Figure 3a) can be achieved by the introduction of indels at pre-miRNA sequences or the miRNA processing sites of MIR genes, which impedes or retards miRNA biogenesis (Chang et al., 2016;Zhou et al., 2017). Similarly, indel insertion in target genes can interfere with miRNA-target mRNA pairing and lead to the subsequent failure of mRNA cleavage into RISC. In addition, homology-directed repair (HDR) (Figure 3b) and homology and recombination-directed repair (HRDR; Figure 3c) can be achieved by the full deletion or knock-in of MIR genes or their promoter sequences .
However, the knock-down or knock-in of MIR genes is challenging compared to the modification of protein-encoding genes due to the complexity of the regulatory networks, the reduced length for gRNA design or targeting, and the fact that miRNA is encoded within intron sequences, which hinders MIR gene editing (Barrangou et al., 2015;Basak and Nithin, 2015). Additionally, the length of MIR genes reduces the number of possible gRNA/Cas9 targets, thus reducing the odds of finding a target near the mature miRNA (Jacobs et al., 2015). In contrast, CRISPR/Cas9-mediated knock-down can be more efficient considering homologous, orthologous or paralogous MIR genes and pri-miRNA regions (Barrangou et al., 2015).
Typical binary vectors for the CRISPR/Cas9 system using a transgenic approach basically contain one selection marker, a Cas9 nuclease sequence that is codon-optimized to monocots or dicots and flanked by two nuclear localization signals, and gRNAs under the control of a specific promoter (Figure 3a-c). These vectors can also contain donor DNA fragments in tandem repeats, which are used as repair moulds of DSB in HDR and HRDR strategies (Figure 3b, c). For the CRISPR/Cas9 or Cpf1 system using a transgene-free strategy (without transgene integration into the plant genome), the nuclease and gRNA are synthetized  Figure 4). Jacobs et al. (2015) showed that CRISPR/Cas9 successfully targeted soya bean miR1514 and miR1509 using biolistic delivery for the transient expression of the CRISPR/Cas9 vector harbouring Cas9 nuclease and gRNAs. Similarly, Li et al. (2016a) induced specific mutations in the miR156 recognition site of the rice ipa1 gene (ideal plant architecture 1), which improved several traits related to plant architecture. Zhou et al. (2017) successfully generated mono-and biallelic mutations in several MIR genes of T0 rice lines, resulting in loss of function for target miRNAs.
In addition to MIR genes, the CRISPR/Cas9 system can be used for editing promoter or enhancer sequences (Chang et al., 2016;Zhou et al., 2017). A challenge yet to be overcome is the disruption or alteration of transcription start sites, cis-regulatory elements (e.g. TATA-or GC-box) or other binding sites of transacting factors (e.g. ABRE and DRE motifs) to achieve precise miRNA expression that improves tolerance without pleiotropic effects. Furthermore, CRISPR/Cas9 technology has recently been used to modulate gene expression through the activation or transcriptional repression of target genes. To this end, a deactivated Cas9 nuclease (dCas9), lacking the two domains needed for DNA double-strand cleavage (D10A/H840A), has been successfully used ( Figure 5). The dCas9 is incapable of cleaving any DNA but is successfully guided to the promoter sequence of desirable MIR genes. In contrast, it can be fused to other functional domains, for example dCas9:VP64 (quadruple tandem repeat of the Herpes simplex virus VP16-activation domain), dCas9:SRDX (synthetic transcriptional repressor pco-dCas9-3X) or dCas9:SET (methyltransferase domain of the H3K9me3 writer) and dCas9:AT (acetyltransferase domain), which act as transcriptional activators (Chavez et al., 2015), repressors  or epigenetic modifiers (O'Geen et al., 2017), respectively. The transcriptional modulation starts by dCas9 (or dCfp1) guided by gRNA to sequences immediately upstream of the transcriptional start site (TSS) of an MIR gene. In addition, the fused MS2-p65-HSF1 activation domains are simultaneously overexpressed, which will interact with the stem-loop of gRNA and recruit additional transcriptional factors to this promoter, improving its transcriptional level (Lowder et al., 2015). Several promising results already have been obtained using this approach, allowing further expansion of the applications of this technology (Lowder et al., 2018;Park et al., 2017). Tang et al. (2017) used the CRISPR/Cpf1 system to demonstrate the efficient transcriptional repression of miRNA159b using deactivated Cpf1 (without the domain of DNA cleavage) fused to the SRDX transcriptional repressor domain.
Furthermore, CRISPR/Cas13a use novel nuclease type (class II type VI-A endoribonuclease) also guided by gRNA to targeting and cleaving single-stranded RNA (ssRNA or mRNA). LwaCas13a from Leptotrichia wadei contain two nucleotide-binding domains (29 HEPN) associated with different RNase activity. It has been successfully established in mammalian and plant cells to knockdown of any exogenous or endogenous RNA (e.g. immunity against viral RNA, and single or multiple knock-down) and thus enable numerous approaches with RNA biology Aman et al., 2018). In contrast to usual RNAi system, CRISPR/Cas13a system also has activity in nuclear RNAs and greater target specificity. In addition, point mutation in HEPN domains abolished its nuclease activity (dead LwaCas13a or dCas13a), expanding the possibilities of its use (East-Seletsky et al., 2017). However, CRISPR/Cas13a has not yet been established for pre-, pri-miRNA or mature miRNA editing in plants. On the other hand, LwaCas13a or dCas13a has great potential to edit its transcripts by knock-down (e.g. miRNA binding, cleavage and degradation) or loss of function (e.g. affecting miRNA/mRNAbinding sites, both in miRNA and in mRNA). Similarly, eTMs modulation using CRISPR/Cas13a allows to increase the accumulation of specific miRNAs. In other context, dCas13a can be fused with a deaminase domain (e.g. ADAR2 domain to adenosine-toinosine deaminase, or dCMP domain to cytidine-to-uridine deaminase) and used to edit polymorphisms/mutations in pri-, pre-miRNA or miRNA sequences for various biological purposes (Cox et al., 2017). The LwaCas13a or dCas13a expression driven by tissue-specific or induced promoters, or in viral vectors can allow more precise and consistent modulation of this target RNA. At least, several orthologous Cas13 nucleases are being characterized from other bacterial species, such as PspCas13b from Prevotella sp. which showed higher levels of RNA knock-down compared to LwaCas13a (Cox et al., 2017).
Topical delivery of pri-miRNA or mature miRNA RNAi technology using transgene-free approach was recently optimized from topical delivery (foliar sprays) of the nanostructured and stabilized dsRNA molecules in model or crop plants to pathogens control or insect pest management (Joga et al., 2016;McLoughlin et al., 2018). Carrier nanoparticle (e.g. biopolymers of chitosan, silicon, carbon and clay nanosheets), ribonucleoprotein particle (e.g. peptide transduction domain-dsRNA binding domain) and cross-linkers (e.g. tripolyphosphate, dextran sulphate and poly-D-glutamic acid) were successfully optimized to improve the delivery and internalization of highly integrated RNA in plant cell (Cunningham et al., 2018). However, use of this technology for the delivery of pri-miRNA or mature miRNA aiming the modulation of endogenous genes to improve agronomic traits or cross-control of insect pest or pathogens is still being established in plants. The higher stability and internalization potential of the pri-miRNA compared to mature miRNA provides possibilities for manipulation of the transcriptional profile of adult plants without the use of transgenics. In respect to cross-control of insect pests, delivery of structured pri-miRNA can prevent its processing in the host plant (avoiding plant off-target modulation and pleiotropic effects) and the acquisition by the insect in this host plant results in natural delivery. These structured pri-miRNAs are viroid-like engineered molecules flanked by pH-dependent ribozymes domain, which are not processed by the RNAi machinery of plant, but are efficiently processed into insect digestive tract and cells (patent application from INPI under number: BR102017006904-4; Maria Fatima Grossi-de-Sa, personal communication, 20 February 2019). Currently, the cost of large-scale production of dsRNA is the major bottleneck; however, there are already some private companies that supply these molecules, nanoparticles and stabilizing compounds.

Epigenetic memory
Plants are frequently exposed to different and/or concomitant stresses, and these conditions trigger defence responses that minimize the negative effects of additional stress. The defence responses of plants are enhanced by retaining "molecular memories" of previous stress events through epigenetic mechanisms (Crisp et al., 2016). This epigenetic memory allows subsequent defence or adaptation responses to be more efficient upon exposure to the same stress. In some cases, this epigenetic information can be transmitted from generation to generation (Crisp et al., 2016;Kinoshita and Seki, 2014;Liu et al., 2015). The siRNA or miRNA-directed DNA methylation and histone modifications, which include methylation, phosphorylation, acetylation, ubiquitylation and sumoylation, are the main epigenetic modifications at genomic regions that contain regulatory sequences (e.g. transcription regulatory sequences), protein coding genes or MIR genes (Xie and Yu, 2015). The 24-nt miRNAs are considered one of the primary modes for the epigenetic modulation of the genome, which in turn modulates the expression of several genes or MIR genes involved in different biological processes (Crisp et al., 2016). Thus, epigenetic modifications in the genome from biotic or abiotic stress events act indirectly inhibiting or activating the transcription of innumerable Figure 4 CRISPR/Cas9 or CRISPR/Cpf1 system to target genes or generate MIR gene knock-down or knock-in using similar NHEJ, HDR or HRDR strategies, but transgene-free. Above, CRISPR/Cas9 vectors for in vitro (e.g. Escherichia coli) production of guide RNAs (gRNA), free donor DNA fragment and Cas9 or Cpf1 protein flanked by one or two nuclear localization signals (NLS) (Liang et al., 2017). After purification, the gRNA, nuclease protein and donor DNA fragment are mixed and assembled in microcarrier particles (tungsten or gold) and delivered into plant cells (e.g. embryogenic callus, apical meristem or protoplasts) using a biolistic particle delivery system.

Figure 3
Genome editing using CRISPR/Cas9 system. (a) The CRISPR/Cas9 system to target promoter sequences or generate MIR gene knock-down using nonhomologous end joining (NHEJ) strategy. Above, a typical CRISPR/Cas9 NHEJ binary vector carrying a T-DNA that contains a selection marker gene under control of a constitutive promoter, Streptococcus pyogenes Cas9 wild-type gene (SpCas9WT) codon-optimized to monocots or dicots and containing one or two nuclear localization signals under control of a constitutive promoter, one or two RNA guide RNAs (gRNA1 or gRNA2) and an RNA scaffold in tandem under control of the U6 RNA polymerase III promoter containing a guanine (G) extra at the 3 0 end. Below, an overview of the NHEJ strategy in plants containing T-DNA from the CRISPR/Cas9 binary vector integrated into the genome (constitutive expression) or by transient expression (e.g. in protoplasts) via biolistic approach or the type III secretion system of Agrobacterium tumefaciens (Chang et al., 2016;Zhou et al., 2017). Below, the complex gRNA:RNA scaffold is transcribed, associated with Cas9 nuclease in the nucleus and directed to the target sequence in genomic DNA. The gRNAs match the target sequence (promoter sequence or MIR gene) and mediate its cleavage by Cas9 nuclease next to the protospacer adjacent motif (PAM), generating a double-strand DNA break (DSB). After DNA cleavage, the damage is corrected by the DNA repair mechanism of the plant cell, but errors (insertion or deletions of any nucleotides, named indels) can be inserted in the repaired DSB sequence, resulting in indels within the transcription start site, cis-regulatory elements or other binding sites of trans-acting factors, leading to the up-or down-regulation of MIR gene expression. In addition, indels in miRNA processing sites prevent the biogenesis of these molecules. (b) CRISPR/Cas9 system to target promoter sequences or generate a MIR gene single knock-in using a homology-directed repair (HDR) strategy. Above, a typical CRISPR/Cas9 HDR binary vector carrying a T-DNA, similar to NHEJ, but containing only one gRNA and one to three copies of the donor DNA fragment engineered to match the target DNA sequence. These donor DNA fragments are flanked by the target sequences of the gRNAs at the 5 0 and 3 0 terminus, homology arms that flank the target site and mutated target sequence containing amino acid substitutions plus 4-6 synonymous substitutions, which prevent the edited DNA from being paired by the gRNA and cleaved by Cas9 nuclease. These donor DNA molecules can be delivered by transgene integration into the genome and released by the gRNA or by a biolistic approach using free donor DNA (donor DNA fragment without gRNA target), which will serve as a repair template for damaged DNA. Below, the complex gRNA:RNA scaffold is transcribed, associated with Cas9 nuclease in the nucleus and directed to the target gene sequence in the genome. The gRNAs match the target sequence (promoter sequence or MIR gene) and mediate its cleavage by the Cas9 nuclease next to the PAM sequence, generating DSB. After DNA cleavage, this damage is corrected by the DNA repair mechanism using the free donor DNA fragment as the repair template based on the homology sequence, resulting in nucleotide exchange and amino acid substitution Zhao et al., 2016). (c) CRISPR/Cas9 system to target promoter sequences or generate simultaneous MIR gene double knock-in using a homologous recombination-directed repair (HRDR) strategy. Above, a typical CRISPR/Cas9 HRDR binary vector carrying a T-DNA, similar to NHEJ and HDR, containing two gRNAs and donor DNA fragments engineered to simultaneously edit two interspaced target sites Zhao et al., 2016). In brief, DSB is repaired by the DNA repair mechanism using the free donor DNA fragment as the repair template. In this case, the donor DNA fragment is integrated into the target site by homologous recombination. Interkingdom cross-talk by the horizontal transference of miRNAs siRNA uptake from oral delivery and successful modulation of endogenous target through the process of RNAi was showed by first time in Caenorhabditis elegans. The horizontal transfer of miRNA until recently was filled of controversies both in the passive context (voluntary transfer during interaction, or by uptake and secretion during feeding in the cell of host organism) and active (release or ingestion of miRNAs by the host or parasite organism as a form of defence or counter-defence). Transgenic plants overexpressing dsRNA molecules and consequently accumulating high amount of siRNAs that target insect pest or nematodes genes were successful developed and displayed improved resistance to these parasite organisms (Zotti et al., 2018). However, it has been observed the requirement of dsRNA delivery with longer length for successful processing by the RNAi machinery of insect pest and efficient knock-down of its target genes (Maria Fatima Grossi-de-Sa, personal communication, 20 February 2019). This suggests that although RNAi machineries exhibit innumerable interkingdom similarities, they also have important specificities. In addition, it is known that any type of RNA outside its natural cellular context is easily and rapidly degraded or inactivated. The presence of numerous types of exoand endonucleases prevents the occurrence of any aberrant event in the host cell, both for endogenous and exogenous RNA. On the other hand, the successful delivery of at least some dsRNA, siRNA, pri-miRNA or mature miRNA molecules may occur; however, will they in fact act on the modulation of endogenous genes? In fact, it has been scientifically proven in the last years that the uptake of miRNAs and the regulation of target genes in host organisms are possible during compatible or incompatible interactions (Wang et al., 2018b;Zhang et al., 2012). In addition, high stability of pre-and pri-miRNA suggests their greater probability of being delivered successfully in the cells of parasite or parasitized organisms, but they will in fact be properly processed by the RNAi machinery of these organisms. It is believed that in some cases is possible, since that these molecules are compatible with RNAi machinery of the recipient organism. Unfortunately, sensitive methods, appropriate approaches and efficient protocols needed for elucidation of the pri-miRNA or mature miRNA interkingdom cross-talk are yet very limited. The NGS technologies helped by powerful bioinformatics tools have enabled to access more easily and precisely of the uptake and secretion of miRNA during host-parasit interaction. In addition, these approaches also allow to observe the regulation of endogenous genes by these miRNA (Dickinson et al., 2013;Chen et al., 2013). Furthermore, can these delivered miRNAs act as effector molecules modulating host defence mechanisms or in counterdefence? Weiberg et al. (2013) showed that Botritys cinerea-siRNAs are successful delivered in Arabidopsis thaliana cells during pathogenesis, binding to AGO1 and selectively silencing host immunity genes. Similarly, Zhang et al. (2016b) showed that in response to infection with Verticillium dahliae, cotton plants increase accumulation of Gh-miR166 and Gh-miR159, which were exported to the fungal hyphae for specific silencing of virulence genes. In this same context, can miRNA cross-talk also play a role in epigenetic feedback? In plants, there is still no concrete evidence for this, but some studies show the presence of numerous miRNAs in milk of human or animals. In this context, Perge et al. (2017) speculate that milk-derived miRNAs secreted in exosomes might be involved in the epigenetic reprogramming of the children. This would suggest that this mechanism may indeed also occur in plants. It is believed that this mechanism would be acquired evolutionarily from an interaction between individuals and that would act both in reprogramming of host defence and parasite counter-defence.
In respect to eTMs cross-talk, can they also act as exogenous SPONGES in parasite or parasitized organisms? On the one hand, eTMs are single-strand RNA extremely susceptible to the extracellular environment or to ribonucleases. However, could indeed its successful delivery trigger the regulation of endogenous miRNAs? Despite possible delivery success of eTMs, they have usually single binding site for sequence-specific miRNAs, thus restricting their action spectrum in other nontarget organisms. Unlike, circRNAs are more stability and can in fact be more active Figure 5 Transcriptional modulation (activation or repression) of MIR genes using CRISPR/dCas9 system. Above, a typical CRISPR/dCas9 binary vector carrying a T-DNA that contains a selection marker gene driven by a constitutive promoter; dead Cas9 (dCas9) gene codon-optimized to monocots or dicots and containing one or two nuclear localization signals (NLS) fused in tandem with dCas9:VP64, dCas9:SDRX, dCas9:SET or dCas9:AT functional domains, and also driven by a constitutive promoter; one RNA guide (gRNA) and an RNA scaffold in tandem driven by U6 RNA polymerase III promoter containing a guanine (G) extra at the 3 0 end; and MS2, p65 and HSF1 activation domains fused in tandem and containing one internal NLS driven by a constitutive promoter (Lowder et al., 2018;Tang et al., 2017). Below, an overview of the dCas9:VP64, dCas9:SDRX, dCas9:SET or dCas9:AT coupled with guide RNA and MS2:p65:HSF1 is directed to promoter region of MIR gene controlling the MIR gene transcription. in this cross-talk. Until then, it is known that any exogenous RNA captured by the cell is rapidly destroyed by the RNAi machinery and a secondary siRNA-based regulatory signal is then amplified, making this receptor cell even more prepared to destroy new molecules of this invasive RNA. For example, plant viruses efficiently bypass this primary defence mechanism of host by encoding suppressor proteins of the RNAi pathway. Thus, it is believed that successful modulation of targets trigger by secreted miRNAs or eTMs, which potentially act as effectors in cross-talk interaction, can be acquired during successive interaction and improved by evolutionary mechanisms.

Challenges and future perspectives
Novel MIR gene-based NBTs have emerged as a powerful strategy to improve several biological or agronomics traits, such as plant tolerance to abiotic and biotic stresses. Although these advances have been remarkable, its advantages and drawbacks need to be discussed critically for the search for best alternatives ( Table 2). The constitutive overexpression of specific MIR genes improves agronomic traits in several plant species. However, pleiotropic phenotypes were also observed in these transgenic events in consequence of the wide acting network of the miRNAs. The overexpression driven by tissue-or stage-specific and abiotic and biotic stress-or senescence-inducible promoters are powerful approaches to overcome these undesirable effects (Ferdous et al., 2017). The characterization of cis-regulatory elements in promoter sequences of canonical MIR genes can provide knowledge of the transcription factors associated with stress responses and how the MIR genes are regulated by abiotic or biotic stresses. From this, it is possible to better choose the type of promoter to be used, in some cases can opt for the canonical promoter sequence. In addition, artificial MIR genes are a great alternative usually more specific and precise compared to other RNAi strategies in modulation of target mRNA. However, its efficiency is also related to the choice of the backbone and the promoter sequence used for driven its expression. Thus, use of backbone from canonical MIR genes and expression also driven by specific promoters or topical delivery using carrier nanoparticles are some promising alternatives. On the other hand, viral vector-mediated miRNA delivery may be a usual alternative, but is limited to proofs of concept, since they present some limitations that prevent their use in commercial scale. Unlike the increase in miRNA accumulation, eTM, circRNA and artificial STTM are strategies successfully used to decrease miRNAs cargo in cell and efficiently improve desirable traits. In addition, eTMs modulation showed themselves better than STTM and expression driven by specific promoters can allow its modulation in adequate time and space. Furthermore, prospection of new eTMs or circRNAs, as well as delivery system based on viral vectors, nanoparticles or Agrobacterium tumefaciens, provides new alternatives for transgene-free plants. Genomeediting technologies have been successfully optimized in plants including modulation of MIR genes and miRNAs. Meganucleases, ZFNs and TALENs although present higher target specificity compared to the new nucleases used in plant genome editing, this high specificity and the limited range of currrenlty available modules restrict its uses to editing few MIR genes. In addition, the assembly of these modules becomes a more expensive and onerous activity compared to the new editing technologies based on CRISPR. The CRISPR/Cas9 NHEJ has successfully allowed the insertion with high specificity of transgene-free indels or knockout gene in several plant genomes. In addition, its high efficiency of editing, high percentage of homozygous mutant already in T0 or T1 generation, easy design and cloning of the binary vector, associated with efficient nanoparticle or biolistic-mediated CRISPR ribonucleoprotein delivery, provides a powerful alternative for transgene-free genomic editing. Similarly, CRISPR/Cas9 HDR or HRDR also allows transgene-free genome editing, deletion of full MIR gene sequence, repair of MIR gene sequence and repair of gene/ mRNA becoming resistant to specific miRNAs. However, both HDR and HRDR approaches are limited to low repair or recombination efficiency driven by synthetic donor DNA, added to critical design and onerous construction of binary vector. These drawbacks can be overcome by adjustment of the amount of DNA donor delivered, best selection of totipotent tissues for ribonucleoprotein delivery, use of the new or improved nucleases, and use of transgenic approach with later elimination of the transgene by Mendelian segregation. However, for example, the elimination of the transgene by segregation is not viable in sugarcane due to its vegetative propagation. In addition, efficient transcriptional modulation of MIR genes mediated by CRISPR technology using the deactivated nucleases has allowed to improve agronomic traits. Although it may be a transgenicdependent technology with potential of off-target transcriptional modulation, nucleases expression driven by tissue-specific or induced promoter, topical delivery of CRISPR ribonucleoprotein and Agrobacterium tumefaciens-mediated transient delivery can overcome these drawbacks. Unlike for these nucleases of DNA, the Cas13a acts with high specificity directly on the RNA molecules. Although it does not yet have many results in plant systems, the current results obtained in mammalian cells are quite promising. In this way, it allows the knock-down of pre-, pri-miRNA, mature miRNA, eTMs, circRNAs and mRNA, both cytoplasmic and nuclear. In addition, Cas13a can be associated with possibility of tissue-specific and site-specific RNA editing, transgene-free approach by topical delivery of CRISPR ribonucleoprotein or Agrobacterium tumefaciens-mediated transient delivery. Another transgene-free approach is based on topical delivery of linear or structured pre-miRNA and mature miRNA.
Although not yet fully optimized in plants, its main advantages are the efficient delivery using nanoparticles, high RNA internalization rate in plant cell, low cytotoxic effects and possibility of trans-kingdom cross-talk from topical delivery of structured amiRNAs in plants targeting pathogens or insect pest genes. In addition, the possibility of topical delivery of plant miRNAs associated with important traits in crops (acting as enhancers of phenotype or traits). Furthermore, studies on the interkingdom mobility of small RNAs (siRNA or miRNA cross-talk) may provide evidence to improve the understanding of pathogen-plant, nematode-plant or insect-plant interactions. The elucidation of miRNA cross-talk in the context of defence response, counterdefence and plant adaptation based on epigenetic mechanisms can assist in the development of NBTs. Finally, a better understanding of regulation and expression profile that MIR genes associated with abiotic or biotic stress tolerance of the intrinsic features of RNAi machinery in different kingdoms and exploitation of the advantages offered by NBTs may lead to practical biotechnological applications to improve agronomic traits in several crops worldwide. In addition, this knowledge enables the development of new biotechnological products with greater practicality, reduced generation time and low cost.