Long non-coding RNAs in plants: emerging modulators of gene activity in development and stress responses

Main conclusion Long non-coding RNAs modulate gene activity in plant development and stress responses by various molecular mechanisms. Abstract Long non-coding RNAs (lncRNAs) are transcripts larger than 200 nucleotides without protein coding potential. Computational approaches have identified numerous lncRNAs in different plant species. Research in the past decade has unveiled that plant lncRNAs participate in a wide range of biological processes, including regulation of flowering time and morphogenesis of reproductive organs, as well as abiotic and biotic stress responses. LncRNAs execute their functions by interacting with DNA, RNA and protein molecules, and by modulating the expression level of their targets through epigenetic, transcriptional, post-transcriptional or translational regulation. In this review, we summarize characteristics of plant lncRNAs, discuss recent progress on understanding of lncRNA functions, and propose an experimental framework for functional characterization. Electronic supplementary material The online version of this article (10.1007/s00425-020-03480-5) contains supplementary material, which is available to authorized users.


Introduction
Pervasive transcription of genomes contributes to the large number of non-coding RNAs. Long non-coding RNAs (lncRNAs) are typically defined as transcripts of more than 200 nucleotides length and without any protein coding potential (Quinn and Chang 2016;Budak et al. 2020). Since discovery of thousands of lncRNAs based on genomewide survey, the functional relevance of lncRNAs has been debated. They have been suggested to be 'transcriptional noise' (Hüttenhofer et al. 2005) rather than having specific biological functions (for review, see Kung et al. 2013). It is now becoming clear that lncRNAs represent a highly heterogeneous class of molecules that can be distinguished based on their biogenesis and functions, and by their position relative to other genomic features such as protein-coding genes or transposons   (Table 1).
Most lncRNAs are located within intergenic regions although intronic lncRNAs and natural antisense lncRNAs have been reported. Specialized groups of plant lncRNAs produced by RNA polymerase IV or V are important scaffolding components in the RNA directed DNA methylation (RdDM) pathway (Chekanova 2015). Several features of lncRNAs, including transcript length, expression level and specificity, biogenesis, post-transcriptional processing and degradation, are not only different from those of proteincoding mRNAs, but also heterogeneous among the lncR-NAs. Even though large numbers of lncRNAs have been identified via next generation sequencing (NGS), microarray and comparative genomics, only a small portion of lncRNAs have been functionally characterized. LncRNAs can regulate mRNA expression via cis and/or trans mechanisms, act as signals and decoys of miRNAs or RNA binding proteins, provide specificity for target molecules such as histone modifying enzymes, and function as scaffolds stitching together large molecular machinery (Wang and Chang 2011). In Communicated by Gerhard Leubner.
terms of the layers of regulation, lncRNAs can affect target gene activity at almost all levels of regulation, including chromatin, transcriptional, post-transcriptional, translational, and post-translational levels (Fatica et al. 2014;Lucero et al. 2020). In plants, lncRNAs have been shown to participate in regulation of developmental processes, biotic and abiotic stress responses, in addition to acting as modulators of the basic cellular machinery. Comparative analysis of lncRNAs in many plant species has deepened our understanding of conservation and evolution of lncRNAs. Transposable elements contributed significantly to the origin and diversification of lncRNAs in plants (Kapusta and Feschotte 2014). Many identified and experimentally verified lncRNAs have been curated and deposited into databases, making them accessible for functional studies [see, e.g., EVLncRNAs (Zhou et al. 2018(Zhou et al. , 2019, Supplemental table S1]. In this review, we summarize the characteristics and recent findings on plant lncRNA functions, and document the strategies and experimental approaches used in identification and analysis of plant lncRNAs.

Discovery and classification of lncRNAs
The first eukaryotic lncRNA, H19 with a length of 2.3 kb, was discovered in mouse in 1984 and is highly expressed during embryo development (Pachnis et al. 1984). Both H19 and its neighboring protein coding gene Igf2 are imprinted. H19 and Igf2 are maternally and paternally expressed, respectively, and form the H19/IGF2 cluster ( Fig. 1a) (Keniry et al. 2012;Nordin et al. 2014). Subsequently, many lncRNAs such as Xist, Airn, MALAT1, and HOTAIR were discovered and characterized in animals through genetic, molecular, and functional studies (Fatica et al. 2014). The first identified plant lncRNA, Enod40, was isolated as an early marker for nodule organogenesis in Medicago plants (Crespi et al. 1994). Enod40 was found to trigger changes in subcellular localization of the nuclear RNA binding protein MtRBP1 (Crespi et al. 1994;Campalans et al. 2004). Since then, plant lncRNAs have been identified as regulators of miRNA activity (Franco-Zorrilla et al. 2007), epigenetic regulation (Swiezewski et al. 2009;Wu et al. 2020) and modulation of chromatin structure Kim and Sung 2018). Furthermore, the two antisense lncRNAs LAIR (LRK Antisense Intergenic RNA) and MAS (MAF4 antisense RNA) were found to interact with WDR5 (a component of the COM-PASS-like complex) thereby regulating flowering time in rice and Arabidopsis, respectively (Wang et al. 2018;Zhao et al. 2018).
Based on their genomic position and orientation relative to their neighboring or overlapping protein coding genes, lncRNAs can be classified into intronic lncRNAs, intergenic lncRNAs (lincRNAs), natural antisense lncRNAs, and sense lncRNAs (Ariel et al. 2015;Fig. 1b). LincRNAs can be further classified based on the genomic features with which they are associated, such as promoters, enhancers, and transposable elements (Fig. 1b).
Enhancer-associated lncRNAs (eRNAs) are usually less than 2000 nt in length and bidirectionally transcribed from corresponding enhancers, as shown in animal model systems (Shlyueva et al. 2014). These eRNAs often lack polyA tails and are degraded by the exosome when they are released from RNA polymerase II (RNA pol II, Shlyueva et al. 2014). Bidirectional transcripts are not typically detected in enhancers or promoters of Arabidopsis and other plants, most likely due to rapid degradation (Thieffry et al. 2020 and references therein). Most eRNAs are functionally uncharacterized. Data from non-plant model systems suggest roles of eRNAs in mediating changes in chromatin status, though it has also been suggested that they represent products of 'accidental' RNA pol II activity at enhancers (Shlyueva et al. 2014). Transposable elementassociated lncRNAs (TE-lncRNAs) overlap with transposons that provide lncRNAs with distinct characteristics and chromatin environment. Transposons such as ALU elements promote nuclear localization of human lncRNAs (Lubelsky and Ulitsky 2018;Carlevaro-Fita et al. 2019). The evolutionary origins and functional diversification of lncRNAs are also influenced by transposable elements (Kapusta et al. 2013). Last but not least, many lncRNAs act as precursors  (Huang et al. 2017). Altogether, lncRNAs comprise a highly heterogeneous class of biomolecules that reflect differences in their biogenesis, functionality and turnover. In the following, we aim to provide an overview on characteristics of plant lnRNAs, pointing toward their distinct origins and mechanisms of action.

Abundance and size of lncRNA transcripts
LncRNAs have been identified in a wide range of plant species including Arabidopsis, rice, and maize. The number of lncRNAs identified varies depending on the technology used for identification in each species, and large-scale analyses have reported between 6480 ) and 6510 ) lncRNAs in Arabidopsis (Table 2). LncRNAs are usually shorter than protein-coding mRNAs, and they contain less exons. Some lncRNAs contain open reading frames (ORFs) with the potential of producing small peptides (Lin et al. 2020). While it is not known whether functional peptides are formed, small ORFs encoded in lncRNAs have been shown to affect growth of human cells (Chen et al. 2020).

Expression specificity and functionality
LncRNAs are typically expressed in a more tissue-specific manner than mRNAs of protein-coding genes. In Arabidopsis, ~ 32% of lncRNAs display organ-specific expression that could be verified by experimental methods such as qRT-PCR . High expression specificity of lncRNAs makes them potentially suitable as markers for tissues and developmental stages. Partly, the apparent specificity could also be attributed to the generally low expression level of lncRNAs, as well as limitations in detection by standard mRNA-sequencing protocols.

Biogenesis, splicing, and regulation of lncRNAs
As protein-coding mRNAs, biogenesis of most lncRNAs depends on RNA pol II-mediated transcription and cotranscriptional splicing. For instance, cold responsive lncRNA SVALKA is transcribed by RNA pol II, and it tightly regulates expression of C-REPEAT/DRE BINDING FACTOR 1 (CBF1) (Kindgren et al. 2018). Additional factors or other RNA polymerases also contribute to the biogenesis of lncRNAs (Liu et al. 2015). Arabidopsis lncRNA AtR8 is transcribed by RNA pol III and involved in the hypoxic stress response (Wu et al. 2012). A subset of lncR-NAs are produced by the plant-specific RNA pol IV or pol V (Liu et al. 2015). These lncRNAs can play a role in the RdDM pathway, in which RNA pol IV-transcribed lncR-NAs interact with INVOLVED IN DE NOVO 2 (IDN2), which then recruits a SWI/SNF chromatin remodelling complex to silence the activity of transposable elements (TEs) or genes by facilitating access of nucleosomes to DNA methylases (Zhu et al. 2013). Additionally, components of the miRNA pathway contribute to lncRNA biogenesis. For example, processing of a subset of lincRNAs requires SERRATE (SE), CAP BINDING PROTEIN20 (CBP20), and CAP BINDING PROTEIN80 (CBP80) . DICER-like proteins may also play roles in processing of plant lincRNAs (Ma et al. 2014). Consequently, these plant lncRNAs are usually processed into 24 nt het-siRNA by DCLs (e.g., DCL3) to methylate target genomic loci (e.g. TEs).
During RNA processing, lncRNAs are typically stabilized by capping and polyadenylation in the nucleus. A subset of lncRNAs in mammalians, such as MALAT1, are processed by RNase P, do not possess polyA tails and, instead, have a specialized 3′ end structure (Wilusz et al. 2008). In humans, non-polyadenylated lncRNAs (i.e., sno-lncRNAs) that are flanked by snoRNAs and protected by RNA binding proteins have also been identified (Yin et al. 2012). Among the non-polyadenylated lncRNAs, a specialized form of RNAs called circRNAs, such as circ-SEP3 in Arabidopsis (Conn et al. 2017), join their heads with tails covalently in a process called back-splicing that is mediated by the spliceosome machinery (Chen 2016). CircRNAs may regulate splicing of their cognate mRNAs, as was shown for circSEP3 and its target SEPALLATA3 (SEP3) (Conn et al. 2017). Differential polyadenylation, linked with changes in preferential subcellular localization, in response to stress has been described for rice and Arabidopsis lncRNAs (Di et al. 2014;Yuan et al. 2016Yuan et al. , 2018. In mammalians, ~ 13% of lncRNAs are transcripts that are derived from divergent transcription in promoters of protein-coding genes (Grzechnik et al. 2014). These divergent transcripts are associated with histone modification (e.g. H3K56ac), RNA pol II Tyr1 phosphorylation and chromatin remodeling factors (e.g. SWI/SNF). Furthermore, the directionality of these divergent lncRNAs is determined by the asymmetry of U1 snRNP and polyadenylation signals (Quinn and Chang 2016). However, divergent transcription does not appear to occur in the majority of genes in Arabidopsis thaliana (Hetzel et al. 2016;Thieffry et al. 2020). In addition to the RNA polymerase machinery, transcription factors (TFs) and chromatin environment (e.g., histone modification and DNA methylation) also contribute to the regulation of lncRNA expression (Quinn and Chang 2016).
Data from humans suggest that splicing efficiency of lncRNAs is lower than that of mRNAs, possibly due to lower binding of splicing factors and the presence of weaker splicing-related motifs (Melé et al. 2017). Low sequencing depth and limitation of RNA-seq assembly methods may also contribute to this observation, since RACE-seq of lncRNAs detected as many alternative splicing events in  Zhu et al. (2015) lncRNAs as in mRNAs (Lagarde et al. 2016). Systematic tissue-specific interrogation of lncRNA transcripts with a higher sequencing depth and transcriptomics of specific cellular compartments, such as the nucleus, will help us to get a better overview on the lnRNA expression/abundance, the repertoire of lncRNA splice forms and other aspects of lnRNA biogenesis in plants in the future.

Structure of lncRNAs
LncRNAs possess secondary structures which may be necessary for their functionality. There are usually two types of functional sites in lncRNAs: interacting sites which are necessary for sequence-specific interactions with RNA binding proteins, and structural sites which confer the identity of secondary and/or tertiary structures directing interacting partners (Fabbri et al. 2019). For example, COOLAIR participating in vernalization has a multi-way junction motif and two right-hand turn motifs (Hawkes et al. 2016), which are very conserved secondary structures in the Brassicaceae family. However, it is still unknown which proteins interact specifically with these motifs.

Subcellular localization of lncRNAs
mRNAs are usually exported into cytosol for translation. By contrast, after processing lncRNAs can reside in nucleus or get exported to cytosol or other subcellular locations and organelles, such as mitochondria, as demonstrated by RNA FISH and ribosome profiling (Carlevaro-Fita and Johnson 2019). Data from animal model systems showed that lncR-NAs are generally prone to be more enriched in the nucleus than in the cytoplasm compared to mRNAs (Derrien et al. 2012). Sequence elements within lncRNAs as well as RNA binding proteins contribute to the nuclear or cytosolic localization of lncRNAs, which reflects their cellular roles and functionality (Carlevaro-fita and Johnson 2019). For example, human lncRNAs containing ALU repeats are more prone to be retained in nucleus because of binding of specific splicing factors such as HETEROGENEOUS NUCLEAR RIBONUCLEOPROTEIN K (HNRNPK; Lubelsky and Ulitsky 2018). Some cytosolic lncRNAs are associated with mono-and poly-ribosomal complexes (see, e.g. Bazin et al. 2017;Hsu et al. 2016), and some of these lncRNAs could eventually contribute to biogenesis of small peptides. A set of nuclear lncRNAs are bound by chromatin, and this localization can be stabilized by U1 snRNP (U1 small nuclear ribonucleoprotein particle) in mammals (Yin et al. 2020).
Chromatin-associated lncRNAs potentially influence TF binding or the functionality of enhancers (Shlyueva et al. 2014). While these data from animal model systems indicate intricate mechanisms underlying the subcellular distribution of lncRNAs, less is known on plant lncRNAs. Many identified lncRNAs (e.g., COOLAIR, DRIR) in plants are localized to and act in nucleus. For example, cold-induced COOLAIR coats the FLC locus in nucleus and acts in FLC repression by changing the histone modification status (e.g., H3K36me3) dynamics (Rosa et al. 2016;Wu et al. 2020).
On the other hand, there are also cytoplasm localized cis-Natural Antisense Transcripts (cis-NATs) overlapping with protein coding genes and some of them could impact the translation of mRNAs (Deforges et al. 2019). In sum, the different types of subcellular localization point to various molecular mechanisms of action of lncRNAs in transcriptional and posttranscriptional control of gene expression.

Decay of lncRNAs
In terms of turn-over of lncRNAs, the half-lives of lncRNAs are typically shorter than those of mRNAs, which reveals complex regulation of lncRNA metabolism in plants (Szabo et al. 2020). LncRNAs are less efficiently synthesized and rapidly degraded (Mukherjee et al. 2017). Like mRNAs, plant lncRNAs can be degraded by both 3′-5′ exonucleolysis via the nuclear exosome and 5′-3′ exonucleolysis via exonucleases such as XRN2 and XRN3 (Kurihara et al. 2012). In mutants of exosome subunits, a set of specialized lncRNAs similar to CUTs (Cryptic unstable transcripts) and PROMPTs (Promoter upstream transcripts) emerged from TSSs of mRNAs (Chekanova et al. 2007;Chekanova 2015;Thieffry et al. 2020). Data from humans suggest that exosome-regulated lncRNAs modulate the activity of enhancers, resolving deleterious R-loop structures by the exosome (Pefanis et al. 2015;Nair et al. 2020). Similar to mRNAs, the quality of plant lncRNAs is also surveilled by the nonsense-mediated mRNA decay (NMD) pathway (Kurihara et al. 2009;Kirn et al. 2009;Drechsel et al. 2013). Interestingly, the up-frameshift (upf) mutants, defective in a component of the NMD pathway, accumulate high levels of transcripts derived from antisense transcription and intergenic regions (Kurihara et al. 2009). This suggests extensive regulation of lncRNA stability via several molecular regulatory pathways.

Functions and molecular mechanisms of lncRNAs in plants
The recently established lncRNA database EVLncRNAs collected 1543 experimentally validated lncRNAs from 77 species, including 428 lncRNAs from 44 plant species such as Arabidopsis and rice (Zhou et al. 2018(Zhou et al. , 2019. Despite limited functional characterization of most lncRNAs, studies so far have uncovered a wide range of possible functions and molecular mechanisms mediated by plant lncRNA activities (Datta and Paul 2019) (Fig. 2a).  . COOLAIR, including two short and long isoforms with polyA tails, is a class of natural antisense transcripts originating from the 3′ end of the FLC locus (Swiezewski et al. 2009). COOLAIR activity is regulated by 3′ processing factors FCA, FY, FPA, CstF64, and CstF77 (polyadenylation cleavage factors), and PRP8 (the spliceosome component) (Liu et al. 2010;Marquardt et al. 2014). However, detailed molecular mechanisms of COOL-AIR repressing FLC are still unknown, although the increasing level of histone demethylase FLD has been shown to contribute to H3K4me2 demethylation of FLC (for review, see Wu et al. 2020). COLDAIR is transcribed from the second FLC intron and acts as signal of early vernalization by recruiting the H3K27me3 writer CURLY LEAF (CLF), an enzymatic component of the PRC2 complex and a homolog of EZH2 in animals, to repress FLC (Fig. 2b) (Heo and Sung 2011;Kim et al. 2017). COLDWRAP is a lncRNA associated with the promoter of FLC, which also interacts with CLF to form an intragenic chromatin loop and to confer FLC repression (Kim and Sung 2018). Furthermore, a non-polyadenylated antisense transcript (ASL, for Antisense Long) is produced from the FLC locus. The function of ASL is still unknown but the expression level of ASL is downregulated in an rrp6l mutant (one of the exosome components, rrp6l1 rrp6l2 double mutant) (Shin and Chekanova 2014 (Ding et al. 2012a, b;Fan et al. 2016;Henriques et al. 2017;Severing et al. 2018;Fang et al. 2019). FLINC regulates ambient temperature-mediated flowering. T-DNA insertion mutants of FLINC flowered earlier due to upregulated FT expression while the underlying mechanism is not known (Severing et al. 2018). The circadian-regulated FLORE is a lncRNA antisense to CDF5 and is involved in promoting of photoperiodic flowering by repression of several CDFs and consequently activation of FT (Henriques et al. 2017). In sum, the different examples indicate interesting functions for lncRNAs in the environment-dependent modulation of flowering time, providing model systems for studying how gradual changes in environmental factors trigger a defined developmental decision at the transcriptional or posttranscriptional level.  (Gao et al. 2016). The long intergenic rice lncRNA XLOC_057324 is highly expressed in reproductive organs, and T-DNA insertion mutant analysis suggests roles in control of flowering and plant fertility (Zhang et al. 2014).

Modulation of reproductive organ development
Other functions of lincRNAs include specific processes directly related to plant fertility. BcMF11 is specifically expressed in pollen and is necessary for male fertility and pollen development in Brassica campestris ssp. chinensis b COLDAIR recruits PRC2 complex to deposit H3K27me3 marks at target gene FLC and thereby drives repression of FLC. c APOLO recognizes target gene by R-loop formation and decoys PRC1 protein. d ASCO can hijack splicing factor NSR to regulate alternative splicing of target genes. e ELENA1 evicts FIB2 from the FIB2-MED19a complex and contributes to activation of PATHOGENESIS-RELATED GENE 1 (PR1). f miR2118 targets PM1T to produce many phasiRNAs. g MIKKI acts as a target mimic to sequester miR171 away from its target ◂ (Song et al. 2013). SUPPRESSOR OF FEMINIZATION (SUF) is a lncRNA antisense to MpFGMYB, an important regulator of female sexual tissue differentiation in liverwort (Marchantia polymorpha). The suf loss of function mutant created by Cas9-based deletion displayed male-tofemale sexual conversion, probably due to failure to repress MpFGMYB in male tissues in the absence of SUF (Hisanaga et al. 2019). The intronic lncRNA AG-incRNA4 from the second intron of the floral homeotic AGAMOUS (AG) gene in Arabidopsis is expressed in leaves and interacts with the PRC2 complex component CLF to deposit H3K27me3 histone marks onto the AG locus, thereby contributing to repression of AG expression in leaves. Knockdown of AG-lincRNA4 resulted in AG activation in leaves by lowering the H3K27me3 level at the AG locus. Consequently, the corresponding mutant showed phenotypes resembling those of ectopic AG expression (Wu et al. 2018). LDMAR was identified in rice through map-based cloning and regulates photoperiod-sensitive male fertility via RdDM (Ding et al. 2012a, b;Zhou et al. 2012).
Small RNAs, including het-siRNAs, phase-siRNAs, and miRNAs, play a critical role in development and stress responses. For example, miR396-mediated regulation of HaWRKY6 plays a role in protection of damage caused by high temperature in sunflower and affects plant growth (Giacomelli et al. 2012). Identification of ncRNA-W6 (ncW6) in the promoter of HaWRKY6 revealed another layer of regulation of the gene by a non-coding RNA. ncW6 derives from a transposon of the MITE family and is able to form a hairpin structure that is processed into 24 nt het-siR-NAs by DCL3 to trigger DNA methylation in the flanking regions of HaWRKY6. DNA methylation changes chromatin structure of the HaWRKY6 locus and promotes the formation of a loop encompassing the whole locus to enhance transcription of HaWRKY6. The level of DNA methylation, and consequently, the formation of the loop and the expression level of HaWRKY6 are regulated in a tissuespecific manner (Gagliardi et al. 2019). Another lncRNA, PMS1T, identified by map-based cloning in rice, contributes to photoperiod-sensitive male sterility by producing phase-siRNAs in a miR2118-dependent manner (Fan et al. 2016) (Fig. 2f). Ef-cd is an antisense RNA in the OsSOC1 locus and positively regulates OsSOC1 activity by deposition of H3K36me3, thereby reducing the time-span that is needed to reach plant maturity without yield penalty (Fang et al. 2019). Together, these findings highlight important functions for lncRNAs in reproductive growth via different molecular mechanisms. Since many uncharacterized lncR-NAs are associated with genomic loci that encode developmental control genes, these will provide interesting targets of future research.

Response to abiotic and biotic stresses
As sessile organisms, plants must cope with various kinds of abiotic and biotic challenges. Plants have evolved intricate signaling cascades and molecular networks to combat these stresses. Under phosphate starvation conditions, Arabidopsis plants express the lncRNA Induced by Phosphate Starvation 1 (IPS1). IPS1 acts as an endogenous target mimic to sequester and repress miR399, a repressor of PHOS-PHATE2 (PHO2), which encodes a ubiquitin-conjugating E2 enzyme. Repression of PHO2 enhances phosphate uptake and accumulation (Fig. 2g) (Franco-Zorrilla et al. 2007).

ELF18-INDUCED LONG-NONCODING RNA1 (ELENA1)
is a 589-nt lincRNA conferring immunity of Arabidopsis. Plants with a reduced expression level of ELENA1 by an artificial miRNA are more sensitive to the bacterial pathogen Pseudomonas syringae pv. tomato DC3000 and show downregulation of several immunity marker genes, including PATHOGENESIS-RELATED GENE 1 (PR1). In contrast, overexpression of ELENA1 activates immune genes such as PR1. ELENA1 exerts its role via interacting with components of Mediator (Fig. 2e) (Seo et al. 2017). The lncRNA DROUGHT-INDUCED LNCRNA (DRIR) in Arabidopsis positively regulates salt and drought response. Plants overexpressing DRIR showed enhanced salt and drought tolerance and displayed higher survival rates under salt and drought stress conditions (Qin et al. 2017). Many other stress response-related lncRNAs have been identified, but their molecular mechanisms of action are yet to be investigated (see, e.g. Zhu et al. 2014;Wang et al. 2017).

Functions in other biological processes
LncRNAs have been shown to participate in diverse biological processes, such as leaf development, auxin signaling, and photomorphogenesis. TWISTED LEAF (TL) is a rice lncRNA antisense to OsMYB60 and required for maintaining leaf blade flattening by regulating the expression of its sense mRNA (Liu et al. 2018). The auxin responsive Arabidopsis lncRNA APOLO plays a role in fine-tuning the transcription of its neighboring PINOID (PID) gene, an important regulator of auxin polar transport, via formation of a chromatin loop involving the promoter of PID. The expression level of APOLO determines the chromatin environment in the promoter region of PID affecting histone modifications and the level of DNA methylation, and consequently the formation of the chromatin loop and the expression level of PID (Fig. 2c)  ). In addition to these cis effects, APOLO also regulates target loci in trans by formation of R-loop (DNA-RNA duplexes) mediated by short sequence complementarity and thereby decoying PRC1 to target loci to modulate their chromatin status . Furthermore, the photomorphogenesis-related lncRNA HID1 (HIDDEN TREASURE1) represses the transcriptional activity of its target gene PHYTOCHROME INTERACTING FAC-TOR 3 (PIF3). HID1 forms a large nuclear complex with as yet unknown proteins and modulates the chromatin structure in the PIF3 promoter, consequently repressing hypocotyl elongation of Arabidopsis seedlings (Wang et al. 2014).
LncRNAs function in basic nuclear regulatory processes by interacting with proteins. For example, nuclear speckles are nuclear domains enriched with splicing-related factors and located in interchromatin regions of nucleoplasm (Spector and Lamond 2011). It was shown that Arabidopsis ASCO-lncRNA competes for the NUCLEAR SPECKLE RNA-binding proteins (NSRs) and sequesters NSRs to modulate the alternative splicing pattern of target genes (Fig. 2d) (Bardou et al. 2014). LncRNAs are also components of the telomerase molecular machinery. For example, lncRNA AtTR is the RNA subunit of telomerase, which interacts with TELOMERASE REVERSE TRANSCRIPTASE (TERT) to maintain the integrity and stability of telomeres (Michal et al. 2019;Song et al. 2019). This indicates roles of lncR-NAs in genome integrity and genome functions beyond biological functions in development or environmental response, which emphasize the need for multiscale experimental methodologies to characterize lncRNA functions.

Experimental methodologies for functional characterization of lncRNAs
Similar to protein-coding genes, functions of lncRNAs can be investigated using forward and reverse genetics approaches. However, functional analysis of lncRNAs is hampered by the need to distinguish functions of the lncRNA transcript from that of its genomic locus. This is because lncRNAs are often produced from DNA genomic regions with other functions, e.g., loci of protein coding genes (in the case of intronic or antisense lncRNAs) or enhancers (e.g., in the case of eRNAs). Also RNAi-based knockdown of lincRNA activities can have side effects that are not related to the functions of lincRNAs, for instance, RNAi-mediated DNA methylation is possible to change the functionality of the genomic regions in other aspects (e.g., affecting enhancer activity). Finally, not the lincRNA transcript itself, but the process of transcription may exert a regulatory function (Gowthaman et al. 2020).
In plants, a small set of lncRNAs has been identified by map-based cloning and functionally characterized, such as LDMAR (Ding et al. 2012a), PMS1T (Fan et al. 2016), Ef-cd (Fang et al. 2019) and Iw1 (Huang et al. 2017). However, reverse genetics (e.g., based on T-DNA mutagenesis populations, RNAi, overexpression) is most commonly used for studies of lncRNA functions, because the vast majority of lncRNAs were identified by high throughput technologies. Every method used to perturb lncRNA functions has disadvantages. For example, T-DNA insertions or CRISPR/ Cas9-based deletions in intergenic regions may not only inhibit lncRNA expression, but also affect other functions of the DNA sequences, such as TF binding sites or regulatory elements within lincRNA loci, thereby altering the expression of nearby protein coding genes. When studying antisense, sense, or intronic lncRNAs, these approaches can also have side effects, such as modifying splicing of the associated protein-coding genes. The RNAi technology on the other hand is known to be prone to off-targeting, and may cause RdDM, thereby confounding functional interpretation of the target lncRNAs. Thus, a combination of different Here, we propose a workflow for functional investigation of plant lncRNAs (Fig. 3). When a candidate lncRNA is identified, the first task to perform a comprehensive inspection of the sequence and structure of the lncRNA. Rapid amplification of cDNA ends (RACE) can be used to obtain the full length transcript(s) of the lncRNA. Searching publicly available datasets, such as cap analysis of gene expression (CAGE) and polyA site sequencing (PAS-seq) (Shepard et al. 2011), and performing RNA-seq will give clues about the general structure as well as alternative splicing patterns of the lncRNA locus of interest. Northern blotting and quantitative RT-PCR (qRT-PCR) are standard approaches for investigation of the expression profiles of lncRNAs. GREEN FLUORESCENT PROTEIN (GFP) reporter imaging can be used to study dynamic lncRNA promoter activity. RNA-FISH allows study of the activity and localization of lncRNAs to the level of individual genomic loci (Rosa et al. 2016). Recent studies showed that some lncRNAs could translate into small peptides, and thus it is necessary to distinguish whether the lncRNA of interest functions as noncoding RNA or as small peptide. Several bioinformatics and experimental approaches can be employed for this purpose, such as CPC2 to test for coding potential test (Kang et al. 2017). Additionally, lncRNAs should be queried in protein databases including Pfam (Finn et al. 2016) and Uniprot (The UniProt Consortium 2017) to know whether they have potential homologous proteins. Ribosome footprints based on ribosome profiling are indicative of open reading frames, which are used to discriminate lncRNAs from protein coding genes (Lander 2014;Hsu et al. 2016;Bazin et al. 2017). Loss/gain-of-function mutants are generated to investigate functionality of the lncRNA. Since every technique has its own limitations (see above), it is necessary to use multiple different approaches such as T-DNA mutagenesis, RNAi, overexpression with constitutive and tissue-specific promoters, and CRISPR/Cas9-based mutagenesis combined with mutant complementation. A large number of publicly available T-DNA insertion lines are available for both Arabidopsis and rice. Analysis of independent mutant alleles and, importantly, transgenic mutant complementation (in trans) can be used to validate the functionality of lncRNAs (see, e.g. Fang et al. 2019). When a lncRNA has multiple isoforms, generating mutants for each isoform can distinguish the roles of individual isoforms. CRISPR/Cas9-based mutagenesis usually creates small indels in the target site ), which might not influence the functionality of the lncRNA. This can be overcome by introducing a pair of single guide RNA (sgRNA) to induce a larger indel in the corresponding lncRNA locus. Use of multiple such pairs of sgRNAs covering the entire lncRNA can help to dissect functional regulatory sites of the lncRNA. In these experiments, potential side effects arise from mutagenizing other functional DNA elements that reside within the lncRNA locus. Therefore, the target lncRNA locus should be evaluated carefully by taking into account existing information on TF binding sites or chromatin structure. In all types of mutant analyses, the phenotypic analyses should be complemented by monitoring changes in expression of the protein-coding genes flanking the lncRNA locus of interest. Especially for studying trans mechanisms of lncRNAs, (inducible) ectopic expression or artificial miRNA technology can be used for validation.
Functional lncRNAs typically interact with DNA, RNA, and proteins. The in vivo or in vitro approaches developed for investigating the RNA-protein (e.g., RIP and CLIP) (Cao et al. 2019), RNA-DNA (e.g., ChIRP) (Chu et al. 2012), and RNA-RNA (e.g., RAP-RNA) (Engreitz et al. 2014) interactions can be used to identify the molecular partner(s) interacting with lncRNAs. The subcellular localization of lncRNAs is also important, since it may provide clues on functions. For example, single molecule RNA FISH analysis revealed that COOLAIR and FLC transcripts are mutually exclusively expressed (Rosa et al. 2016). It is important to further develop in vitro and in vivo experimental methods to screen and validate the interaction between lncRNAs and their partner molecules. For example, a trimolecular fluorescence complementation (TriFC) system has been used to demonstrate lncRNA-protein interaction by tagging a lncRNA with the MS2 system (MS2 sequence and phage MS2 coat protein fused to YFP-N) and co-transfecting it together with the YFP-C tagged RNA-binding protein into tobacco leaves via Agrobacterium (Seo et al. 2019). Finally, we envision that efficient novel experimental and computational methods will be developed for investigation of the functionality of lncRNAs in plants at the level of single cells or subcellular compartments.

Conclusions and perspectives
Mounting evidence shows involvement of lncRNAs in wide ranges of biological processes, including development and stress responses. Efficient computational methods are urgently needed to predict functional lncRNAs for experimental validation. LncRNAs act in cis or in trans to regulate the function of their target genes through diverse mechanisms that involve interactions with DNA, RNA or proteins. Many plant lncRNAs (e.g., COLDAIR) function epigenetically to modulate the expression of their target genes by modifying histone modification status and chromatin organizations. However, despite the diversity of molecular mechanisms and functions, our understanding of most plant lncRNAs is still elusive and unclear. There are at least a couple of reasons. Firstly, the effects of lncRNAs might only be observed under specific conditions given that expression of most lncRNAs is highly tissue/condition-specific. Secondly, lncRNAs represent a heterogeneous group of RNA molecules in plants. Several subclasses of lncRNAs (e.g., enhancer RNAs) are largely coupled with regulatory DNA sequences (e.g., TFBSs), which makes it difficult to assess their (if any) specific functions. Development of technologies is imperative to understand molecular mechanisms of lncRNAs . Large-scale functional screens of lncRNAs by CRISPR/Cas9-based mutagenesis have been established in human and flies, although only a small percentage of lncRNAs showed context-specific phenotypic changes ). A similar system has yet to be developed for plant lncRNAs although large-scale mutagenesis populations have been created in several plant species by transformation of sgRNA libraries targeting protein-coding genes (Jacobs et al. 2017;Lu et al. 2017;Meng et al. 2017;Zhang et al. 2019;Liu et al. 2020;Bai et al. 2020). Finally, we need to investigate how we can effectively utilize the knowledge on beneficial lncRNAs in breeding programs to develop novel plant germplasm and elite crop varieties. An excellent example for this is provided by Ef-cd that promotes early maturity without yield penalty probably due to better nitrogen utilization and photosynthesis in rice. It functions like a dominant gene as plants homozygous or heterozygous for Ef-cd showed better agronomic performance compared to plants without Ef-cd. It thus is valuable for rice breeding. Fang et al. (2019) have developed molecular markers completely linked with Ef-cd, which can be used to identify new early maturity rice germplasm containing Ef-cd and to introgress Ef-cd into elite rice cultivars to further improve their maturity and agronomic performance based on markerassisted selection. For LDMAR and PMS1T, base editing can be used to change the unfavorable alleles into favorable ones as single nucleotide polymorphisms seem to be the cause for changes in fertility. These examples show that utilizing knowledge on plant lncRNA functions can open new possibilities for plant breeding research, thereby improving crop quality and performance.
Author contribution statement LC conceived the topic of the article. LC, KK, QHZ contributed to the writing of the manuscript. All authors read and approved the final version of the manuscript.
Funding Open Access funding enabled and organized by Projekt DEAL.

Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.