2.1. Translation-associated RNA cleavage

RNA cleavage is an essential prerequisite for the activity of RDR6, the key initiator of alternative RdDM, and complete and normal transcripts are immune to this process. For example, Creasey et al. suggested that miRNA-mediated RNA cleavage can trigger the production of TE-associated siRNAs, termed epigenetically activated siRNAs (easiRNAs), generated in the DNA methylation-deficient Arabidopsis mutants (Creasey et al., Reference Creasey, Zhai, Borges, Van Ex, Regulski, Meyers and Martienssen2014). Although several miRNA-independent pathways, such as mRNA splicing (Dalakouras et al., Reference Dalakouras, Lauter, Bassler, Krczal and Wassenegger2019; Kanno et al., Reference Kanno, Chiou, Wu, Lin, Matzke and Matzke2023; Oberlin et al., Reference Oberlin, Sarazin, Chevalier, Voinnet and Marí-Ordóñez2017), RNA processing (Dadami et al., Reference Dadami, Moser, Zwiebel, Krczal, Wassenegger and Dalakouras2013; Elvira-Matelot et al., Reference Elvira-Matelot, Bardou, Ariel, Jauvion, Bouteiller, Le Masson, Cao, Crespi and Vaucheret2016) and RNA surveillance pathways (Martínez de Alba et al., Reference Martínez de Alba, Moreno, Gabriel, Mallory, Christ, Bounon, Balzergue, Aubourg, Gautheret, Crespi, Vaucheret and Maizel2015; Moreno et al., Reference Moreno, de Alba, Bardou, Crespi, Vaucheret, Maizel and Mallory2013; Szádeczky-Kardoss et al., Reference Szádeczky-Kardoss, Csorba, Auber, Schamberger, Nyikó, Taller, Orbán, Burgyán and Silhavy2018), have been proposed to initiate the RNA cleavage required for alternative RdDM, these mechanisms appear to act only on specific transcripts and do not broadly explain the targeting specificity of alternative RdDM.

Previous studies have identified GC content as a key factor in ensuring robust transgene expression and preventing entry to the alternative RdDM pathway (Brule & Grayhack, Reference Brule and Grayhack2017; Parret et al., Reference Parret, Besir and Meijers2016; Sidorenko et al., Reference Sidorenko, Lee, Woosley, Moskal, Bevan, Merlo, Walsh, Wang, Weaver, Glancy, Wang, Yang, Sriram and Meyers2017). This finding is further supported by evidence that TEs tend to use codons that are suboptimal for the host translational machinery, resulting in weaker translational activities compared to endogenous genes (Kim et al., Reference Kim, Wang, Lei, Li, Fan and Cho2021). A similar reduction of translational activity was observed in the transcripts of a retrotransposon, known as EVADE (Oberlin et al., Reference Oberlin, Rajeswaran, Trasser, Barragán-Borrero, Schon, Plotnikova, Loncsek, Nodine, Marí-Ordóñez and Voinnet2022). Kim et al. also reported significant differences in GC3 content (GC content at the 3rd nucleotide position of a codon) between genes and TEs across diverse eukaryotic organisms, suggesting that codon usage and the resulting translational efficiency may serve as a universal mechanism for host genomes to discriminate self from non-self RNAs (Kim et al., Reference Kim, Wang, Lei, Li, Fan and Cho2021). Given that viruses, as being foreign to their hosts, typically exhibit codon usage patterns distinct from that of host (Belalov & Lukashev, Reference Belalov and Lukashev2013; Gaunt & Digard, Reference Gaunt and Digard2022; Jitobaom et al., Reference Jitobaom, Phakaratsakul, Sirihongthong, Chotewutmontri, Suriyaphol, Suptawiwat and Auewarakul2020; Plant & Ye, Reference Plant and Ye2022), it is plausible that translational activity, determined by codon composition, contributes broadly to recognition and silencing of foreign genetic elements.

Translational inefficiency not only reduces protein output but also can trigger RNA quality control mechanisms that result in RNA cleavage, a critical prerequisite essential for RDR6 activity in alternative RdDM. One such mechanism is the no-go RNA decay (NGD) pathway, which targets transcripts where ribosomes stall during translation (Inada & Beckmann, Reference Inada and Beckmann2024; X. Li et al., Reference Li, Zhou and Li2025; Monaghan et al., Reference Monaghan, Longman and Cáceres2023; Müller et al., Reference Müller, Becker, Denk, Hashimoto, Inada and Beckmann2025). These stalls can be caused by strong secondary structures, rare codons, disome formation (ribosome collisions) or other features that impede ribosome progression. In response, the NGD machinery induces endonucleolytic cleavage near the stall site, generating RNA fragments that can potentially serve as substrates for RDR6. Consistent with this notion, Oberlin et al. was able to detect cleaved RNA ends at where ribosome is stalled in EVADE RNA (Oberlin et al., Reference Oberlin, Rajeswaran, Trasser, Barragán-Borrero, Schon, Plotnikova, Loncsek, Nodine, Marí-Ordóñez and Voinnet2022), and Kim et al. found signals for frequent RNA cleavage and stacked ribosomes enriched in TEs of Arabidopsis (Kim et al., Reference Kim, Wang, Lei, Li, Fan and Cho2021). This positions NGD as a potential contributor to the initiation of siRNA production and silencing via alternative RdDM.

Emerging evidence suggests that NGD operates in plants, particularly in controlling non-native RNAs like that of viruses. For example, recent studies have shown that NGD factors, such as PELOTA1 and HBS1, play a direct role in suppressing viral infection in plants (Ge et al., Reference Ge, Cao, Qiao, Cui, Li, Shan, Gong, Zhang, Li, Wang, Zhou and Li2023; Pouclet et al., Reference Pouclet, Gagliardi and Garcia2023). In Arabidopsis, these proteins recognise conserved A-rich motifs (G_{1 – 2} A_{6 – 7}) in potyvirus genomes and trigger endonucleolytic cleavage of viral RNAs, limiting their accumulation. This mirrors the observation that TEs are typically AT-rich, a feature associated with suboptimal translational efficiency (Boissinot, Reference Boissinot2022; Kim et al., Reference Kim, Wang, Lei, Li, Fan and Cho2021). Moreover, a recent study suggests that aberrant RNAs arise from ribosome stalling prior to the full establishment of transgene silencing (Kramer et al., Reference Kramer, Ratnayake, Edwards, Lowrey, Klaas, Sidorenko, Rowan, Michelmore, Meyers and Slotkin2025). These findings together highlight NGD as a targeted antiviral mechanism, linking translational surveillance to gene silencing and reinforcing the idea that inefficient translation may mark transcripts as non-self or aberrant.

The siRNA pathway in plants offers a strategic advantage over simple RNA degradation by acting more like an adaptive immune system. It enables sequence-specific recognition of foreign or aberrant RNAs and converts these into heritable silencing signals through RdDM. Unlike transient RNA decay, siRNA-mediated silencing can be amplified, systemic and long-lasting, ensuring robust and persistent defence. While there is rich, albeit circumstantial, evidence suggesting that NGD may act as a trigger for the initial RNA cleavage required to initiate the alternative RdDM pathway, direct empirical evidence linking NGD to siRNA biogenesis remains to be established. Once siRNAs are produced and DNA methylation is established at target loci, the canonical RdDM machinery can further strengthen and maintain the silenced state.

2.2. Localisation to siRNA bodies

RDR6 has been shown to localise to stress granules (SGs), cytoplasmic compartments formed particularly under environmental stress conditions (Kim et al., Reference Kim, Wang, Lei, Li, Fan and Cho2021; Kumakura et al., Reference Kumakura, Takeda, Fujioka, Motose, Takano and Watanabe2009; Moreno et al., Reference Moreno, de Alba, Bardou, Crespi, Vaucheret, Maizel and Mallory2013; Tan et al., Reference Tan, Luo, Yan, Liu, Aizezi, Cui, Tian, Ma and Guo2023; Wen et al., Reference Wen, Hu, Pi, Zhang, Duan, Li, Li, Zhao, Yang, Zhao, Liu, Su, Li and Zhang2024). SGs themselves are dynamic cytoplasmic structures that temporarily store and sort untranslated mRNAs during stress, helping cells modulate gene expression and maintain homeostasis (Ren et al., Reference Ren, Zhao and Zou2025; Zhao & Li, Reference Zhao and Li2025). In Arabidopsis, RDR6-containing SGs, also known as siRNA bodies, appear to form through liquid–liquid phase separation (LLPS) mediated by SGS3. Deletion of the LLPS domain in SGS3 disrupts its ability to localise to these granules, which in turn prevents the recruitment of RDR6 and impairs siRNA biogenesis (Kim et al., Reference Kim, Wang, Lei, Li, Fan and Cho2021; Tan et al., Reference Tan, Luo, Yan, Liu, Aizezi, Cui, Tian, Ma and Guo2023) (Figure 2).

Figure 2.

RNA pathways initiating siRNA biogenesis. RNA cleavage is a critical prerequisite for entry into the siRNA pathway and can be initiated through multiple mechanisms. Ribosome stalling, often caused by suboptimal codons, can induce RNA cleavage and promote RNA localisation to siRNA bodies. m6A-modified RNAs are commonly linked to RNA destabilisation and stress granule localisation, yet the contribution of m6A-binding ECT (EVOLUTIONARILY CONSERVED C-TERMINAL DOMAIN) family proteins to this process remains unclear. Additional RNA modifications, including 5^′ NADylation and 3^′ uridylation, catalysed by DXO1 (DECAPPING AND EXORIBONUCLEASE PROTEIN 1) and TUTases (TERMINAL URIDYLYLTRANSFERASES), respectively, are closely associated with RNA degradation and siRNA biogenesis. Blue circle, m7G cap; black circle, m6A RNA methylation; yellow circle, NAD+ cap.

The formation of siRNA bodies adds an additional layer of specificity and selectivity in the system as not all transcripts are equally guided to these compartments. Studies have shown that RNAs with reduced translational activity are more strongly enriched in SGs (Helton et al., Reference Helton, Dodd and Moon2025; Khong et al., Reference Khong, Matheny, Jain, Mitchell, Wheeler and Parker2017; Khong & Parker, Reference Khong and Parker2018; Kim et al., Reference Kim, Wang, Lei, Li, Fan and Cho2021; Matheny et al., Reference Matheny, Rao and Parker2019, Reference Matheny, Van Treeck, Huynh and Parker2021). This selective enrichment of poorly translating RNAs in SGs appears to be evolutionarily conserved as it has been observed across diverse eukaryotes, including yeast, humans and plants. Although the precise biochemical mechanisms guiding ribosome-depleted RNAs to SGs remain unclear, these findings support the model that translational inefficiency is a key determinant in directing transcripts to the alternative RdDM pathway by funnelling them into cellular compartments where essential silencing components are localised.

In addition to reduced translation, N6-methyladenosine (m⁶A) RNA modification has been shown to promote the localisation of specific transcripts to SGs (Di Timoteo et al., Reference Di Timoteo, Giuliani, Setti, Biagi, Lisi, Santini, Grandioso, Mariani, Castagnetti, Perego, Zappone, Lattante, Sabatelli, Rotili, Vicidomini and Bozzoni2024; Fan et al., Reference Fan, Wang, Lei, Li, Chu, Yan, Wang, Wang, Yang and Cho2023; Li et al., Reference Li, Liu, Guo, Zhang, Chen, Liu, Cheng, Deng, Qiu, Zhang, Goh, Wang and Peng2024; Song et al., Reference Song, Chen, Wang, Cheng and Shyh-Chang2024; Wu et al., Reference Wu, Su, Zhang, Zhang, Wong, Ma, Shao, Hua, Shen and Yu2024). A recent study in Arabidopsis demonstrated that m⁶A-marked RNAs are preferentially enriched in these compartments, and this localisation depends on m⁶A reader proteins that facilitate LLPS (Fan et al., Reference Fan, Wang, Lei, Li, Chu, Yan, Wang, Wang, Yang and Cho2023; Wu et al., Reference Wu, Su, Zhang, Zhang, Wong, Ma, Shao, Hua, Shen and Yu2024). This suggests that m⁶A may serve as a licensing mark, guiding transcripts into the silencing pathway. While a direct role of m⁶A in promoting siRNA biogenesis remains to be confirmed, these observations support a model in which m⁶A facilitates the selective targeting of transcripts for silencing through compartmentalisation.

TE RNAs in plants are marked with high levels of m⁶A, reflecting a strategic layer of post-transcriptional regulation by the host. For example, in Arabidopsis, the retrotransposon Onsen is methylated by m⁶A upon heat activation, and this methylation restricts its activity by retaining its RNA in SGs (Fan et al., Reference Fan, Wang, Lei, Li, Chu, Yan, Wang, Wang, Yang and Cho2023). Similarly, TEs from other eukaryotes, such as LINE-1 elements in humans, are also modified by m⁶A (Barter & Cho, Reference Barter and Cho2025; Hwang et al., Reference Hwang, Jung, Mun, Lee, Park, Baek, Moon, Kim, Kim, Choi, Go, Tang, Choi, Choi, Cha, Park, Liang, Kim, Han and Ahn2021; Wei et al., Reference Wei, Yu, Yang, Liu, Gao, Huang, Dou, Liu, Zou, Cui, Zhang, Zhao, Liu, He, Sepich-Poore, Zhong, Liu, Li, Kou and He2022; Xiong et al., Reference Xiong, Wang, Lee, Li, Chen, Liao, Hasani, Nguyen, Zhu, Krakowiak, Lee, Han, Tsai, Liu and Li2021). Beyond TEs, many studies have identified m⁶A RNA methylation on viral RNAs as well (Y. Li et al., Reference Li, Chen and Sun2025; Manners et al., Reference Manners, Baquero-Perez and Whitehouse2019; Zhang et al., Reference Zhang, Peng and Wang2023), suggesting that this epitranscriptomic mark broadly functions to label non-native RNAs. However, because the same RNA modification also regulates endogenous mRNAs, m⁶A-mediated differentiation may rely on additional molecular signatures. For instance, transposon RNAs are often methylated at multiple sites throughout their sequence, whereas typical mRNAs usually exhibit a single methylation peak near stop codons (Fan et al., Reference Fan, Wang, Lei, Li, Chu, Yan, Wang, Wang, Yang and Cho2023). Such differences in the number and position of m⁶A marks may help distinguish non-native RNAs. Altogether, m⁶A might act as a host-encoded marker to selectively flag foreign RNAs, guiding them into silencing pathways and reinforcing genomic defence at the RNA level.

The localisation of RNAs to siRNA bodies represents a crucial selective step in siRNA production. This selective guidance relies on key features including reduced translational activity and m⁶A RNA methylation, which together help funnel target RNAs into the alternative RdDM pathway. Despite growing evidence supporting the importance of SG formation in RNA silencing, major knowledge gaps still remain. Notably, the precise molecular mechanisms by which m⁶A modifications and translational repression coordinate to drive RNA partitioning into SGs are still unclear. Additionally, whether similar principles apply universally across different organisms and RNA types warrant further investigation. Addressing these questions will be critical to fully understanding how cells maintain genomic stability through epitranscriptomic and translational control.

2.3. RNA decay versus RNA silencing

So far, we have examined the functional impacts of translational inactivation on the initiation of siRNA biogenesis, focusing on RNA cleavage and localisation to siRNA bodies, both of which confer RNA specificity to the silencing pathway. We also discussed NGD as a potential mediator of translation-associated cleavage of non-native RNAs. However, as a host RNA quality control mechanism, NGD primarily promotes RNA degradation rather than siRNA production (Deragon & Merret, Reference Deragon and Merret2025; Szádeczky-Kardoss et al., Reference Szádeczky-Kardoss, Gál, Auber, Taller and Silhavy2018; Wu et al., Reference Wu, Fu, Ren, Liu, Zhang and Ruan2023; You et al., Reference You, He, Hang, Zhang, Cao, Guo, Chen, Cui and Mo2019; Zhang et al., Reference Zhang, Zhu, Liu, Hong, Xu, Zhu, Shen, Wu, Ji, Wen, Zhang, Zhao, Wang, Lu and Guo2015). In this section, we review current knowledge on the interplay between RNA decay and RNA silencing in plants.

The relationship between RNA decay and siRNA biogenesis in plants is often antagonistic, with RNA decay acting as a primary RNA quality control mechanism that can override RNAi (Christie et al., Reference Christie, Brosnan, Rothnagel and Carroll2011; Kim, Reference Kim2023). Both pathways target aberrant or excessive RNAs in processing (P) bodies and siRNA bodies, respectively, but RNA decay typically acts first to prevent the activation of silencing responses. This prioritisation is crucial because RNA silencing, once triggered, can lead to the unintended silencing of functional endogenous genes. Thus, RNA decay serves to suppress inappropriate RNA silencing activity by rapidly degrading faulty RNAs before they can be recognised by the siRNA machinery. For instance, studies in Arabidopsis have shown that mutants deficient in DECAPPING 2 (DCP2) or EXORIBONUCLEASE 4 (XRN4), the P body components, accumulate aberrant RNAs that are subsequently processed into siRNAs, triggering gene silencing (Elvira-Matelot et al., Reference Elvira-Matelot, Bardou, Ariel, Jauvion, Bouteiller, Le Masson, Cao, Crespi and Vaucheret2016; Gregory et al., Reference Gregory, O’Malley, Lister, Urich, Tonti-Filippini, Chen, Millar and Ecker2008; Gy et al., Reference Gy, Gasciolli, Lauressergues, Morel, Gombert, Proux, Proux, Vaucheret and Mallory2007; Li & Wang, Reference Li and Wang2018; Martínez de Alba et al., Reference Martínez de Alba, Moreno, Gabriel, Mallory, Christ, Bounon, Balzergue, Aubourg, Gautheret, Crespi, Vaucheret and Maizel2015; Sorenson et al., Reference Sorenson, Deshotel, Johnson, Adler and Sieburth2018; Souret et al., Reference Souret, Kastenmayer and Green2004; Thran et al., Reference Thran, Link and Sonnewald2012). These findings illustrate how RNA decay pathways act to suppress RNA silencing under normal conditions, ensuring that RNA silencing remains a specialised response reserved for transgenes, viruses or highly abnormal transcripts, rather than ordinary endogenous mRNAs.

Although structurally complete and intact, RNAs modified at the 5^′ and 3^′ ends, such as nicotinamide adenine dinucleotide (NAD⁺) capping and 3^′ uridylation, can be directed into RNA stability or silencing pathways. At the 5^′ end, some RNAs bear non-canonical caps like NAD⁺ instead of the typical m⁷G cap. In Arabidopsis, NAD⁺-capped RNAs are recognised as aberrant and are preferentially targeted for degradation, limiting their potential to engage in RNA silencing (Carpentier et al., Reference Carpentier, Receveur, Cadoudal and Merret2025; Kwasnik et al., Reference Kwasnik, Wang, Krzyszton, Gozdek, Zakrzewska-Placzek, Stepniak, Poznanski, Tong and Kufel2019; Wang et al., Reference Wang, Li, Zhao, You, Le, Gong, Mo, Xia and Chen2019). Conversely, other studies suggest that NAD⁺-capped RNAs can be recruited to RDR6 and processed to siRNAs (Pan et al., Reference Pan, Li, Huang, Zhong, Wu, Wang, Zhang, Cai, Guo, Chen and Xia2020; Yu et al., Reference Yu, Willmann, Vandivier, Trefely, Kramer, Shapiro, Guo, Lyons, Snyder and Gregory2021). At the 3^′ end, uridylation, the addition of uridines by terminal uridylyl transferases (TUTases), can similarly mark RNAs for degradation or processing into small RNAs. In plants, uridylation of viral RNAs or transgene-derived transcripts promotes their degradation and enhances silencing (de Almeida et al., Reference de Almeida, Scheer, Gobert, Fileccia, Martinelli, Zuber and Gagliardi2018; Joly et al., Reference Joly, Garcia, Hily, Koechler, Demangeat, Garcia, Vigne, Lemaire, Zuber and Gagliardi2023; Scheer et al., Reference Scheer, de Almeida, Ferrier, Simonnot, Poirier, Pflieger, Sement, Koechler, Piermaria, Krawczyk, Mroczek, Chicher, Kuhn, Dziembowski, Hammann, Zuber and Gagliardi2021; Wang et al., Reference Wang, Kong, Wang, Wang, Zhong, Lao, Dong, Zhang, Huang, Mo, Yu and Ren2022). These modifications act as molecular tags that distinguish normal RNAs from those targeted for silencing, thereby ensuring precise regulation of gene expression and defence. However, the mechanisms by which specific RNA modifications are selected to channel RNAs into either degradation or silencing pathways remain incompletely understood, and further research is needed to elucidate the determinants governing these RNA fates.