ALBA proteins facilitate cytoplasmic YTHDF-mediated reading of m6A in Arabidopsis Authors: Marlene Reichel, Mathias Due Tankmar, Sarah Rennie, Laura Arribas-Hernández, Martin Lewinski, Tino Köster, Naiqi Wang, Anthony A Millar, Dorothee Staiger, Peter Brodersen Published: 2024 ABSTRACT N6-methyladenosine (m6A) exerts many of its regulatory effects on eukaryotic mRNAs by recruiting cytoplasmic YT521-B homology-domain family (YTHDF) proteins. Here, we show that in Arabidopsis thaliana, the interaction between m6A and the major YTHDF protein ECT2 also involves the mRNA-binding ALBA protein family. ALBA and YTHDF proteins physically associate via a deeply conserved short linear motif in the intrinsically disordered region of YTHDF proteins and their mRNA target sets overlap, with ALBA4 binding sites being juxtaposed to m6A sites. These binding sites correspond to pyrimidine-rich elements previously found to be important for m6A binding to ECT2. Accordingly, both the biological functions of ECT2, and its binding to m6A targets in vivo, require ALBA association. Our results introduce the YTHDF-ALBA complex as the functional cytoplasmic m6A-reader in Arabidopsis, and define a molecular foundation for the concept of facilitated m6A reading, which increases the potential for combinatorial control of biological m6A effects. Introduction N6-methyladenosine (m6A) occurs widely in eukaryotic mRNAs. It is introduced into pre-mRNA during transcription in adenosines in DRACH/GGAU (D = A/G/U, R = A/G, H = A/C/U) motifs by a deeply conserved RNA polymerase II-coupled methyltransferase complex (Balacco and Soller, 2019). m6A is required to complete embryogenesis in vertebrates and plants (Geula et al, 2015; Zhong et al, 2008). It is also important for yeast sporulation (Clancy et al, 2002) and for sex determination and neuronal development in insects (Haussmann et al, 2016; Lence et al, 2016). Many developmental functions of m6A rely on cytoplasmic RNA-binding proteins (RBPs) specialized for m6A recognition, or “reading”, via a YTH domain (Arribas-Hernández et al, 2018; Ivanova et al, 2017; Kontur et al, 2020; Lasman et al, 2020). These YTH domain family (YTHDF) m6A readers contain the YTH domain at the C-terminus, preceded by a long intrinsically disordered region (IDR) (Patil et al, 2018). Higher plants encode an expanded family of YTHDF proteins with, for instance, 11 members in Arabidopsis thaliana (Arabidopsis) (Fray and Simpson, 2015; Scutenaire et al, 2018). They are called EVOLUTIONARILY CONSERVED C-TERMINAL REGION 1-11 (ECT1-11) with reference to the deeply conserved YTH domain at the C-terminus (Ok et al, 2005), following an IDR of more variable length and sequence. ECT2 and ECT3 are crucial for post-embryonic development (Arribas-Hernández et al, 2018; Arribas-Hernández et al, 2020), as they stimulate cell division in primordial cells (Arribas-Hernández et al, 2020). Thus, double knockout of ECT2 and ECT3 causes the slow formation and aberrant morphology of leaves, roots, stems, flowers, and fruits, and these phenotypes are generally exacerbated by additional knockout of ECT4 (Arribas-Hernández et al, 2020). The developmental role of the m6A-ECT module is conserved in plants, because the knockout of tomato and rice ECT genes also causes delayed development (Ma et al, 2022; Yin et al, 2022). Three features of the molecular functions of ECT proteins that promote growth during organogenesis have been defined. First, they are deeply conserved, because the sole YTHDF protein encoded by the liverwort Marchantia polymorpha that diverged from higher plants ~450 million years ago (Magallón et al, 2013; Su et al, 2021) can functionally replace Arabidopsis ECT2 when expressed in primordial cells in ect2 ect3 ect4 mutants (Flores-Téllez et al, 2023). Second, ECTs interact with the major cytoplasmic poly(A)-binding proteins PAB2/4/8 (Song et al, 2023; Tankmar et al, 2023). This interaction is mediated by a conserved tyrosine-rich motif in the IDR of ECT2 and is required for developmental functions of ECT2 (Tankmar et al, 2023). Third, most Arabidopsis ECT paralogues across phylogenetic subclades retain the ability to complement ect2 ect3 ect4 mutants upon ectopic expression in primordial cells (Flores-Téllez et al, 2023). For the three Arabidopsis ECT proteins unable to perform this basal function, the divergence can at least in part be ascribed to differences in their N-terminal IDRs (Flores-Téllez et al, 2023), including the loss of the PAB2/4/8-interacting motif (Tankmar et al, 2023). Thus, the molecular properties of the IDRs of ECT proteins are central to understand their biological functions. At least three distinct molecular properties of IDRs in RBPs are expected to contribute to their functions. First, IDRs often mediate self-assembly such that above a critical concentration, they condense into a phase distinct from the aqueous solution (Wiedner and Giudice, 2021). This is also the case for plant ECT proteins (Arribas-Hernández et al, 2018; Scutenaire et al, 2018), and negative feedback regulation of important stress-related m6A-containing mRNAs may indeed rely on ECT-mediated phase separation (Lee et al, 2023; Wu et al, 2024). Second, the IDR may influence RNA-binding activity, either by stabilization of the RNA-bound conformation of the globular RNA-binding domain (Stowell et al, 2018), or through direct RNA-binding activity, as is well-described in the case of Arg-Gly-Gly (RGG) repeats in IDRs (Chong et al, 2018). The non-RGG-containing IDR of ECT2 also has direct RNA-binding activity. Recently reported in vitro binding assays establish that ECT2 only has appreciable m6A-binding activity when its YTH domain is combined with elements of the N-terminal IDR (Seigneurin-Berny et al, 2024), consistent with the observation that deletion of the IDR from ECT2 strongly reduces RNA-binding capacity in vivo (Tankmar et al, 2023), and with observations of ECT2 crosslinks to target mRNAs specific to the IDR in crosslinking-immunoprecipitation-sequencing (CLIP-seq) data (Arribas-Hernández et al, 2021a). Third, short linear motifs (SLiMs) may be used to mediate direct binding to other proteins (Holehouse and Kragelund, 2024), including other RBPs and regulators of the rate of translation and mRNA decay, as in the example of the ECT2-PAB2/4/8 interaction (Tankmar et al, 2023). The ALBA (acetylation lowers binding affinity) family of proteins was found in mRNA interactome capture screens to be a prominent group of mRNA-associated RBPs in Arabidopsis (Marondedze et al, 2016; Reichel et al, 2016). The ALBA superfamily of proteins contains an archaeal and two eukaryotic families. Proteins in the Sac10b archaeal family (Aravind et al, 2003) exhibit acetylation-sensitive DNA-binding activity and have histone-like properties (Bell et al, 2002; Forterre et al, 1999; Wardleworth et al, 2002; Xue et al, 2000), but may also have RNA chaperone functions (Zhang et al, 2020). The two eukaryotic families group around two distinct subunits of RNaseP/MRP complexes, Rpp20 or Rpp25 (Aravind et al, 2003). Plants encode ALBA proteins belonging to both eukaryotic families. The Rpp20-related forms are short and contain only the ~95 amino acid globular ALBA domain, while the Rpp25-related forms are long and contain ~200-300 amino acid C-terminal extensions, often IDRs with many RGG repeats (Goyal et al, 2016). The sequence similarity within the eukaryotic families is limited, and in most cases, it is not clear whether the ALBA proteins are mRNA-binding or have other RNA-related functions. mRNA-binding ALBA proteins have been studied in the parasitic protist Trypanosoma brucei, where short and long forms are required for translational regulation of many mRNAs during the transition between mammalian and insect hosts, in particular for growth after commitment to differentiation into the insect-specific form (Bevkal et al, 2021; Mani et al, 2011). A requirement of ALBA proteins for growth is recurrent in several plant species (Honkanen et al, 2016; Magwanga et al, 2019), first observed in the liverwort M. polymorpha where the sole long RGG-repeat-containing ALBA protein is necessary for the development of root-like structures called rhizoids (Honkanen et al, 2016). Arabidopsis encodes three short ALBA proteins in the Rpp20 group, ALBA1-3, and three long ALBA proteins in the Rpp25 group, ALBA4-6 (Goyal et al, 2016). Single knockouts of ALBA1 and ALBA2 cause defective root hair development, but no overall growth defects (Honkanen et al, 2016). In contrast, combined knockout of ALBA4-6 leads to slow seedling development, including defective root growth (Tong et al, 2022). A similar defect in root growth was also observed in cotton upon RNAi-mediated knockdown of ALBA genes (Magwanga et al, 2019), further supporting the idea that ALBA proteins stimulate tissue growth in plants. Nonetheless, the molecular basis for their growth-promoting function has not been defined. In this study, we show that ALBAs and ECT2 associate via a deeply conserved SLiM in the IDR of ECT2 to form an efficient m6A reader complex in Arabidopsis. The mRNA target sets of ALBA proteins overlap significantly with those of m6A-ECT2/3, and ALBA4 binding sites in 3′-UTRs are juxtaposed to m6A sites. Finally, ALBA proteins facilitate the association of ECT2 with m6A-modified transcripts and are necessary for the biological functions of m6A-ECT2/3. Thus, our results uncover a mechanism for facilitated m6A reading by YTHDF-interacting RBPs with binding sites in close proximity to m6A. Results The N8 IDR-element of ECT2 is required for normal growth of leaf primordia We previously showed that a 37-amino acid residue region in the N-terminal IDR of ECT2, N8, is required for full activity in promoting the growth of leaf primordia (Tankmar et al, 2023). Since deletion of the N8-encoding region from an ECT2-mCherry gDNA transgene caused a decrease, not abolishment, of the complementation frequency of the ect2-1 ect3-1 ect4-2 (henceforth, te234) triple knockout mutant (Arribas-Hernández et al, 2018; Tankmar et al, 2023), we first sought to corroborate the importance of N8 by independent means. To this end, we used CRISPR-Cas9 in the ect3-1 ect4-2 genetic background to generate a chromosomal in-frame ECT2 deletion matching almost exactly ΔN8 (ect2-5, Fig. 1A; Appendix Fig. S1). The resulting ect2-5 ect3-1 ect4-2 mutant exhibited slow emergence of the first true leaves, albeit less pronounced than te234 (Fig. 1B,C). These results verify that deletion of N8 causes partial loss of ECT2 function. We also confirmed that the ECT2-5 protein accumulated to levels similar to the wild-type protein (Fig. 1D), excluding the possibility that the partial loss of ECT2 function in ect2-5 mutants is due to decreased dosage.Figure 1The N8 IDR-element of ECT2 is required for growth promotion, RNA association, and interaction with ALBA proteins.(A) Schematic representation of wild type and mutant ECT2 proteins. The MobiDB (Di Domenico et al, 2012) track (top) displays regions predicted to be structured or disordered. (B) Images of representative seedlings of the indicated genotypes taken at 7 days after germination (DAG). (C) Quantification of first true leaf size in seedlings of the indicated genotypes 7 DAG. 50 seedlings were measured for each genotype (n = 50). The boxes show the interquartile range (25th–75th percentile), with the central line marking the median. Whiskers extend 1.5 times the interquartile range. Statistical differences between the indicated genotypes were calculated using a Student’s t-test. NS not significant. (D) Protein blots of total lysates prepared from 12-day-old seedlings of the indicated genotypes, were probed with ECT2-specific antisera (Arribas-Hernández et al, 2018). Arrows indicate the positions of the ECT2WT protein and the ECT2-5 protein containing the N8-like deletion. The asterisk indicates an unspecific band. Ponceau staining serves as the loading control. (E) Results of an in vivo UV crosslinking-ECT2-mCherry-immunoprecipitation experiment, followed by PNK-labeling of precipitated RNA with γ-32P-ATP. Left panel, autoradiogram of 32P-radiolabelled RNA-protein complexes purified from plants expressing ECT2WT-mCherry, ECT2ΔN8-mCherry or the aromatic cage mutant ECT2W464A-mCherry. Molecular weight marker positions and the location of the verified ECT2-mCherry-RNA complexes (Arribas-Hernández et al, 2021a) are indicated. The presence of several bands of unequal intensity is due to partial proteolysis of the ECT2 IDR during immunoprecipitation and differential labeling efficiency of the different RNPs (Arribas-Hernández et al, 2021a). Right panels, mCherry immunoblots of the immunoprecipitated (top) and total fractions (input, bottom). Samples were pools of three independent lines for each genotype. (F, G) Volcano plots showing the differential abundance of proteins co-purified with ECT2-mCherry variants (RFP-trap) measured by mass spectrometry of immunopurified fractions (IP‐MS). All ECT2-mCherry variants were expressed in the te234 mutant background. Diagrams above each plot indicate the proteins compared. Statistical significance was determined using empirical Bayes statistics with Benjamin–Hochberg adjusted P values. Experiments were done in biological triplicates. The data underlying the plot in (F) have previously been published (Tankmar et al, 2023). (H) Co-immunoprecipitation assay using mCherry immunoprecipitation from 10-day-old seedlings expressing the indicated ECT2-mCherry variants (see (A)), followed by immunoblot analysis with mCherry- and ALBA1-specific antibodies. Seedlings from three independent transgenic lines were pooled in this experiment. (I) Volcano plots showing the differential abundance of proteins co-purified with ALBA4-GFP as determined by IP-MS from total lysates prepared from 7-day-old seedlings. Statistical significance was calculated using empirical Bayes statistics with Benjamini–Hochberg adjusted p values. The experiment was done in biological triplicates. Source data are available online for this figure. The N8 IDR-element of ECT2 is required for growth promotion, RNA association, and interaction with ALBA proteins. (A) Schematic representation of wild type and mutant ECT2 proteins. The MobiDB (Di Domenico et al, 2012) track (top) displays regions predicted to be structured or disordered. (B) Images of representative seedlings of the indicated genotypes taken at 7 days after germination (DAG). (C) Quantification of first true leaf size in seedlings of the indicated genotypes 7 DAG. 50 seedlings were measured for each genotype (n = 50). The boxes show the interquartile range (25th–75th percentile), with the central line marking the median. Whiskers extend 1.5 times the interquartile range. Statistical differences between the indicated genotypes were calculated using a Student’s t-test. NS not significant. (D) Protein blots of total lysates prepared from 12-day-old seedlings of the indicated genotypes, were probed with ECT2-specific antisera (Arribas-Hernández et al, 2018). Arrows indicate the positions of the ECT2WT protein and the ECT2-5 protein containing the N8-like deletion. The asterisk indicates an unspecific band. Ponceau staining serves as the loading control. (E) Results of an in vivo UV crosslinking-ECT2-mCherry-immunoprecipitation experiment, followed by PNK-labeling of precipitated RNA with γ-32P-ATP. Left panel, autoradiogram of 32P-radiolabelled RNA-protein complexes purified from plants expressing ECT2WT-mCherry, ECT2ΔN8-mCherry or the aromatic cage mutant ECT2W464A-mCherry. Molecular weight marker positions and the location of the verified ECT2-mCherry-RNA complexes (Arribas-Hernández et al, 2021a) are indicated. The presence of several bands of unequal intensity is due to partial proteolysis of the ECT2 IDR during immunoprecipitation and differential labeling efficiency of the different RNPs (Arribas-Hernández et al, 2021a). Right panels, mCherry immunoblots of the immunoprecipitated (top) and total fractions (input, bottom). Samples were pools of three independent lines for each genotype. (F, G) Volcano plots showing the differential abundance of proteins co-purified with ECT2-mCherry variants (RFP-trap) measured by mass spectrometry of immunopurified fractions (IP‐MS). All ECT2-mCherry variants were expressed in the te234 mutant background. Diagrams above each plot indicate the proteins compared. Statistical significance was determined using empirical Bayes statistics with Benjamin–Hochberg adjusted P values. Experiments were done in biological triplicates. The data underlying the plot in (F) have previously been published (Tankmar et al, 2023). (H) Co-immunoprecipitation assay using mCherry immunoprecipitation from 10-day-old seedlings expressing the indicated ECT2-mCherry variants (see (A)), followed by immunoblot analysis with mCherry- and ALBA1-specific antibodies. Seedlings from three independent transgenic lines were pooled in this experiment. (I) Volcano plots showing the differential abundance of proteins co-purified with ALBA4-GFP as determined by IP-MS from total lysates prepared from 7-day-old seedlings. Statistical significance was calculated using empirical Bayes statistics with Benjamini–Hochberg adjusted p values. The experiment was done in biological triplicates. Source data are available online for this figure. N8 is necessary for full RNA association of ECT2 We next conducted in vivo UV crosslinking and immunoprecipitation (CLIP) experiments to test whether RNA association was affected by the deletion of N8. We quantified crosslinked RNA immunoprecipitated with ECT2WT-mCherry or ECT2ΔN8-mCherry by polynucleotide kinase (PNK)-mediated radiolabeling, using the previously described assay conditions that allow assignment of the radiolabeled species as ECT2-mCherry-RNA complexes with different sizes resulting from cleavage of the IDR in the lysis buffer (Arribas-Hernández et al, 2021a). These experiments revealed a reproducible reduction in RNA association of ECT2ΔN8-mCherry compared to ECT2-mCherry, albeit less pronounced than the reduction obtained with the m6A-binding deficient ECT2W464A-mCherry mutant (Arribas-Hernández et al, 2021a) (Fig. 1E). Thus, N8 is involved in RNA association, either directly or through interaction with other RBPs whose presence may enhance the affinity of ECT2 for m6A-containing mRNAs. N8 is necessary for interaction with ALBA proteins To test whether N8 is required for the association of ECT2 with other RBPs, we used comparative immunoprecipitation-mass spectrometry (IP-MS) with stable transgenic lines expressing comparable amounts of either ECT2WT-mCherry or ECT2ΔN8-mCherry in the te234 background ((Tankmar et al, 2023), Appendix Fig. S2A). We also included three lines of ECT2ΔN-mCherry lacking the entire N-terminal IDR ((Tankmar et al, 2023), Fig. 1A) as an additional negative control. All immunopurifications were done in the presence of RNaseA to recover RNA-independent interactors. These experiments revealed that the family of ALBA proteins, in particular ALBA1/2/4/5, were prominent interactors of ECT2 (Fig. 1F; Appendix Fig. S2C), and that the interaction was strongly dependent on N8 (Fig. 1G; Dataset EV1). Because of these qualities of ECT-ALBA co-purifications, and because the mRNA-binding capacity of ALBA proteins may be of interest in connection with the requirement of N8 for ECT2 binding to RNA in vivo, we focused further experiments on the ALBA protein family. We used three different approaches to verify the ALBA-ECT interaction and its dependence on N8. First, we raised an antibody specific for ALBA1 (Appendix Fig. S2B) and used it to confirm that ALBA1 enrichment is reduced, but not abolished, upon deletion of N8 (Fig. 1H). We also included two larger IDR deletion mutants in this experiment, ECT2ΔN-mCherry and ECT2ΔN.2-mCherry lacking the ~200 amino acid residues proximal to the YTH domain ((Tankmar et al, 2023), Fig. 1A,H). ALBA1 levels were not detectable in immunopurified fractions of these two mutants (Fig. 1H), perhaps suggesting that additional determinants of ALBA interaction are located in the IDR outside of the N8 region. Second, an inspection of IP-MS data with HA-ECT2 and with tagged versions of the two YTHDF paralogs ECT3 (ECT3-Venus) and ECT1 (ECT1-TFP) (Tankmar et al, 2023), both of which have m6A-binding capacity (Arribas-Hernández et al, 2018; Arribas-Hernández et al, 2021a; Arribas-Hernández et al, 2021b; Flores-Téllez et al, 2023; Lee et al, 2023), revealed enrichment of ALBA proteins over the negative controls (Appendix Fig. S2C). Third, comparative IP-MS analysis carried out with ALBA4-GFP, and free GFP revealed a clear enrichment of several ECT proteins, including ECT1-8 and ECT11, in the ALBA4-GFP purified fractions (Fig. 1I; Appendix Fig. S2D; Dataset EV1). These results indicate that ALBA and ECT proteins physically associate in vivo and that the ECT2-ALBA association involves the N8 region of the ECT2 IDR. We also take particular note of the combination of two properties. First, deletion of N8 causes reduced RNA binding of ECT2 in vivo. Second, ECT interactors of ALBAs include ECT1 and ECT11 which have m6A-binding capacity but not the function of ECT2 required for leaf formation (Flores-Téllez et al, 2023). Hence, our results suggest that the ALBA-ECT interaction mediates a molecular property common to all ECT proteins, perhaps m6A-binding. AlphaFold3 modeling highlights a conserved SLiM in N8 as key for the interaction of ECT2 with ALBA domains and RNA Because many proteins, in addition to ALBA1/2/4/5, lose enrichment in immunopurified ECT2 fractions upon deletion of N8 (Fig. 1G), we sought to further narrow the region in the IDR of ECT2 required for ALBA interaction. We noticed that a SLiM within N8 is conserved in all YTHDF protein clades of flowering plants as well as in YTHDFs from early-diverging clades of land plants such as bryophytes, including M. polymorpha YTHDF (Fig. 2A; Appendix Fig. S3). Since the N8 region is required for full association of ECT2 with both mRNA and ALBA proteins in vivo, we hypothesized that the N8 element might mediate an interaction between the three molecules, perhaps via the conserved SLiM. Thus, we used AlphaFold3 (Abramson et al, 2024) to query whether a complex composed of an ALBA-domain dimer (Wardleworth et al, 2002), an ECT2 fragment spanning the YTH domain plus the SLiM-containing proximal part of the IDR, and an m6A-containing 10-nt RNA could be modeled. Interestingly, AlphaFold3 generated a model of high confidence overall (Figs. 2B–D and EV1A,B). The model features several interactions between the N8-SLiM and the YTH domain, and situates the SLiM centrally between the YTH domain, the ALBA domains, and the m6A-containing RNA (Figs. 2B and EV1A). Because these properties offer straightforward explanations for the reduced ALBA- and RNA-association of ECT2ΔN8 in vivo, we devoted further efforts to the study of the SLiM and refer to it as the YTHDF-ALBA Interaction Motif (YAIM) in the remainder of this report.Figure 2The ECT2-ALBA interaction is mediated by a conserved short linear motif in the N8 element of the ECT2 IDR.(A) Logo representation of sequence conservation in the N8 region of the IDR of YTHDF proteins in the DF-A, -B, -C, -D, and -E clades of 34 land plant species distributed along all main phylogenetic clades, from liverworts and mosses to flowering plants (Dataset EV7). Fern DF-D proteins and the entire gymnosperm- and fern-specific DF-F clade were excluded from the analysis because they do not have an apparent YAIM motif (see Appendix Fig. S3). The logo (Schneider and Stephens, 1990) was generated using the Weblogo tool (Crooks et al, 2004), and sequences were aligned with MUSCLE (Madeira et al, 2024). (B) AlphaFold3 model of the complex between ECT2 (YTH domain plus a YAIM-containing fragment of the N-terminal IDR), two ALBA5 subunits (ALBA domains only), and a 10-nt RNA [5’-AAA(m6A)CUUCUG-3’]. The YAIM is accentuated in space-fill mode (magenta, C; blue, N; red, O), all other protein elements in cartoon mode, and the RNA in stick mode. (C) Same view of the model as in panel (B) but colored according to the predicted local distance difference test (pLDDT) score calculated by Alphafold3 to indicate model confidence on a local per-residue basis (Abramson et al, 2024). (D) 2D plot generated by AlphaFold3 showing the Predicted Aligned Error (PAE) indicating the Expected Position Error (EPE) in Ångströms (white-green scale) in the relative positions of each pair of residues in the complex (Abramson et al, 2024). The location of subunits and structural elements along the axes is indicated. An additional view of the complex is provided in Fig EV1. (E) Schematic representation of the ECT2YAIM mutant with alanine substitutions in the YTH-ALBA Interaction Motif (YAIM) highlighted in red. (F) Categorized leaf size (s) distribution of 9-day-old primary transformants of te234 mutants expressing wild type or mutant versions of ECT2-mCherry as indicated. Statistical differences between the indicated genotypes were calculated using pairwise Fisher exact tests with Holm-adjusted p values. NS not significant. (G) Anti-mCherry immunoblot from total lysates of 9-day-old seedlings of transgenic lines expressing either a fully complementing ECT2WT-mCherry transgene (Arribas-Hernández et al, 2018) or the ECT2YAIM-mCherry construct (L1-L3, three independent lines), or without any ECT2 transgene (–), all in the te234 mutant background. Dashed lines indicate that lanes have been removed for presentation purposes. Ponceau staining is used as a loading control. (H) Results of an in vivo UV crosslinking-ECT2-mCherry-immunoprecipitation experiment, followed by PNK-labeling of precipitated RNA with γ-32P-ATP. Top panel, autoradiogram of 32P-radiolabelled RNA-protein complexes purified from plants expressing ECT2WT‐mCherry, the aromatic cage mutant ECT2W464A-mCherry, or ECT2YAIM-mCherry. Molecular weight marker positions and the location of the verified ECT2-mCherry-RNA complexes (Arribas-Hernández et al, 2021a) are indicated. The presence of several bands of unequal intensity is due to partial proteolysis of the ECT2 IDR during immunoprecipitation and differential labeling efficiency of the different RNPs (Arribas-Hernández et al, 2021a). Middle and bottom panels, immunoblots against mCherry showing the ECT2-mCherry proteins in the IP (middle) and total lysates (input, bottom). Samples were pools of three independent lines for each genotype. (I) Volcano plot showing a differential abundance of proteins detected by mass spectrometry in mCherry immunoprecipitates from te234 seedlings expressing either ECT2YAIM-mCherry or ECT2WT-mCherry. Statistical significance was determined using empirical Bayes statistics with Benjamini–Hochberg adjusted p values. The experiment was done in biological triplicates. Source data are available online for this figure.Figure EV1The YTH-ALBA Interacting Motif (YAIM) is central for ECT2-ALBA interaction (supporting data).(A) Alternative view of the AlphaFold3 model shown in Fig. 2B of the complex between ECT2 (YTH domain plus a YAIM-containing fragment of the N-terminal IDR), two ALBA5 subunits (ALBA domains only), and the 10-nt RNA [5′-AAA(m6A)CUUCUG-3′]. The YAIM is accentuated in space-fill mode (Magenta, C; Blue, N; Red, O), all other protein elements in cartoon mode, and the RNA in stick mode (supports Fig. 2B). (B) Same view of the model as in panel (A), but colored according to the predicted local distance difference test (pLDDT) score calculated by Alphafold3 to indicate model confidence on a local per-residue basis (Abramson et al, 2024) (supports Fig. 2C). (C) Silver staining of aliquots of immunopurified fractions used for LS-MS/MS analysis of differential enrichment in ECT2-mCherry and ECT2YAIM-mCherry purifications (supports Fig. 2I). (D) Co-immunoprecipitation analysis of the ECT2-ALBA1 interaction. Nine-day-old seedlings from three independent transgenic lines expressing each ECT2-mCherry variant were pooled prior to mCherry immunoprecipitation and analysis by western blot. Source data are available online for this figure. The ECT2-ALBA interaction is mediated by a conserved short linear motif in the N8 element of the ECT2 IDR. (A) Logo representation of sequence conservation in the N8 region of the IDR of YTHDF proteins in the DF-A, -B, -C, -D, and -E clades of 34 land plant species distributed along all main phylogenetic clades, from liverworts and mosses to flowering plants (Dataset EV7). Fern DF-D proteins and the entire gymnosperm- and fern-specific DF-F clade were excluded from the analysis because they do not have an apparent YAIM motif (see Appendix Fig. S3). The logo (Schneider and Stephens, 1990) was generated using the Weblogo tool (Crooks et al, 2004), and sequences were aligned with MUSCLE (Madeira et al, 2024). (B) AlphaFold3 model of the complex between ECT2 (YTH domain plus a YAIM-containing fragment of the N-terminal IDR), two ALBA5 subunits (ALBA domains only), and a 10-nt RNA [5’-AAA(m6A)CUUCUG-3’]. The YAIM is accentuated in space-fill mode (magenta, C; blue, N; red, O), all other protein elements in cartoon mode, and the RNA in stick mode. (C) Same view of the model as in panel (B) but colored according to the predicted local distance difference test (pLDDT) score calculated by Alphafold3 to indicate model confidence on a local per-residue basis (Abramson et al, 2024). (D) 2D plot generated by AlphaFold3 showing the Predicted Aligned Error (PAE) indicating the Expected Position Error (EPE) in Ångströms (white-green scale) in the relative positions of each pair of residues in the complex (Abramson et al, 2024). The location of subunits and structural elements along the axes is indicated. An additional view of the complex is provided in Fig EV1. (E) Schematic representation of the ECT2YAIM mutant with alanine substitutions in the YTH-ALBA Interaction Motif (YAIM) highlighted in red. (F) Categorized leaf size (s) distribution of 9-day-old primary transformants of te234 mutants expressing wild type or mutant versions of ECT2-mCherry as indicated. Statistical differences between the indicated genotypes were calculated using pairwise Fisher exact tests with Holm-adjusted p values. NS not significant. (G) Anti-mCherry immunoblot from total lysates of 9-day-old seedlings of transgenic lines expressing either a fully complementing ECT2WT-mCherry transgene (Arribas-Hernández et al, 2018) or the ECT2YAIM-mCherry construct (L1-L3, three independent lines), or without any ECT2 transgene (–), all in the te234 mutant background. Dashed lines indicate that lanes have been removed for presentation purposes. Ponceau staining is used as a loading control. (H) Results of an in vivo UV crosslinking-ECT2-mCherry-immunoprecipitation experiment, followed by PNK-labeling of precipitated RNA with γ-32P-ATP. Top panel, autoradiogram of 32P-radiolabelled RNA-protein complexes purified from plants expressing ECT2WT‐mCherry, the aromatic cage mutant ECT2W464A-mCherry, or ECT2YAIM-mCherry. Molecular weight marker positions and the location of the verified ECT2-mCherry-RNA complexes (Arribas-Hernández et al, 2021a) are indicated. The presence of several bands of unequal intensity is due to partial proteolysis of the ECT2 IDR during immunoprecipitation and differential labeling efficiency of the different RNPs (Arribas-Hernández et al, 2021a). Middle and bottom panels, immunoblots against mCherry showing the ECT2-mCherry proteins in the IP (middle) and total lysates (input, bottom). Samples were pools of three independent lines for each genotype. (I) Volcano plot showing a differential abundance of proteins detected by mass spectrometry in mCherry immunoprecipitates from te234 seedlings expressing either ECT2YAIM-mCherry or ECT2WT-mCherry. Statistical significance was determined using empirical Bayes statistics with Benjamini–Hochberg adjusted p values. The experiment was done in biological triplicates. Source data are available online for this figure. The YTH-ALBA Interacting Motif (YAIM) is central for ECT2-ALBA interaction (supporting data). (A) Alternative view of the AlphaFold3 model shown in Fig. 2B of the complex between ECT2 (YTH domain plus a YAIM-containing fragment of the N-terminal IDR), two ALBA5 subunits (ALBA domains only), and the 10-nt RNA [5′-AAA(m6A)CUUCUG-3′]. The YAIM is accentuated in space-fill mode (Magenta, C; Blue, N; Red, O), all other protein elements in cartoon mode, and the RNA in stick mode (supports Fig. 2B). (B) Same view of the model as in panel (A), but colored according to the predicted local distance difference test (pLDDT) score calculated by Alphafold3 to indicate model confidence on a local per-residue basis (Abramson et al, 2024) (supports Fig. 2C). (C) Silver staining of aliquots of immunopurified fractions used for LS-MS/MS analysis of differential enrichment in ECT2-mCherry and ECT2YAIM-mCherry purifications (supports Fig. 2I). (D) Co-immunoprecipitation analysis of the ECT2-ALBA1 interaction. Nine-day-old seedlings from three independent transgenic lines expressing each ECT2-mCherry variant were pooled prior to mCherry immunoprecipitation and analysis by western blot. Source data are available online for this figure. The YAIM is required for ECT2-ALBA interaction and ECT2 function We next generated a YAIM mutant of ECT2 containing several alanine substitutions (Fig. 2E). The ECT2YAIM-mCherry mutant exhibited a reduced te234 complementation frequency similar to ECT2ΔN8-mCherry (Fig. 2F), despite the fact that protein levels in several independent transgenic lines were similar to those obtained with an ECT2WT-mCherry transgene (Fig. 2G). These observations demonstrate the in vivo importance of the YAIM for ECT2 function. At the molecular level, the ECT2YAIM-mCherry mutant also exhibited defects closely resembling those of ECT2ΔN8-mCherry: less RNA could be crosslinked and immunoprecipitated with ECT2YAIM-mCherry than with ECT2WT-mCherry (Fig. 2H), and ALBA1/2/4/5 were depleted in ECT2YAIM-mCherry immunopurifications relative to ECT2WT-mCherry (Figs. 2I and EV1C; Dataset EV1). We also used the ALBA1 antibody to verify reduced association with ECT2YAIM-mCherry compared to ECT2WT-mCherry (Fig. EV1D). Taken together, we conclude that the YAIM is required for ALBA association and for full target RNA-binding of ECT2 in vivo, as predicted by the AlphaFold3 model of the (ALBA5)2-ECT2-RNA complex. We note, however, that ALBA1/2/4/5 were not the only proteins depleted from ECT2-mCherry purifications upon mutation of the YAIM, perhaps suggesting that the primary function of the YAIM is to mediate ALBA- and RNA-interaction, and that the ECT2-ALBA-RNA complex generates a platform required for interaction with multiple other proteins. A model for concerted m6A-ECT-ALBA function in vivo The results presented so far suggest that ECTs and ALBAs act in concert to bind to m6A-sites in mRNA targets. A basic prediction of this hypothesis is that ECT2 and ALBAs are expressed in the same cells. Examination of expression patterns using fluorescent protein fusions expressed under the control of endogenous promoters showed that ALBA1, ALBA2, and ALBA4 are indeed expressed in mitotically active cells of root and leaf primordia, as is ECT2 (Fig. 3A,B). The tight co-expression of ECT2 and ALBA proteins was also evident from analysis of published root single-cell mRNA-seq data (He et al, 2023; Shahan et al, 2022) (Appendix Fig. S4). Further assessment of the subcellular localization by confocal microscopy indicated that ALBA1, ALBA2, ALBA4, and ALBA5 localize to the cytoplasm (Fig. 3C), as do ECT2, ECT3 and ECT4 (Arribas-Hernández et al, 2018; Arribas-Hernández et al, 2021b; Arribas-Hernández et al, 2020).Figure 3The expression patterns and subcellular localizations of ECTs and ALBAs overlap.(A) Fluorescence microscopy of 5-day-old seedlings co-expressing ECT2-mCherry and ALBA1-TFP (top panel), ECT2-mCherry and ALBA2-TFP (middle panel), or ECT2-mCherry and ALBA4-Venus (bottom panel). (B) Confocal microscopy images of mCherry, TFP and Venus fluorescence in root tips of plants co-expressing ECT2-mCherry and ALBA1-TFP (top) or ECT2-mCherry and ALBA4-Venus (bottom). (C) Confocal images of GFP fluorescence and DAPI staining in root tips of plants expressing ALBA1-GFP, ALBA2-GFP, ALBA4-GFP, ALBA5-GFP, and 35S-GFP. Source data are available online for this figure. The expression patterns and subcellular localizations of ECTs and ALBAs overlap. (A) Fluorescence microscopy of 5-day-old seedlings co-expressing ECT2-mCherry and ALBA1-TFP (top panel), ECT2-mCherry and ALBA2-TFP (middle panel), or ECT2-mCherry and ALBA4-Venus (bottom panel). (B) Confocal microscopy images of mCherry, TFP and Venus fluorescence in root tips of plants co-expressing ECT2-mCherry and ALBA1-TFP (top) or ECT2-mCherry and ALBA4-Venus (bottom). (C) Confocal images of GFP fluorescence and DAPI staining in root tips of plants expressing ALBA1-GFP, ALBA2-GFP, ALBA4-GFP, ALBA5-GFP, and 35S-GFP. Source data are available online for this figure. The model further predicts that ALBAs and ECTs share a significant overlap in mRNA target sets, that they have juxtaposed binding sites around m6A sites in those target mRNAs, and that at least some direct mRNA targets associate less with ECTs in vivo in the absence of ALBA proteins. We previously demonstrated the feasibility of using TRIBE (Target Identification of RNA-binding Proteins by Editing) (McMahon et al, 2016) and iCLIP (Individual Nucleotide-Resolution Crosslinking and Immunoprecipitation) (König et al, 2010) to address such predictions using transcriptome-wide analyses in vivo (Arribas-Hernández et al, 2021a; Arribas-Hernández et al, 2021b; Meyer et al, 2017). In TRIBE, the catalytic domain (cd) of the A-I RNA-editing enzyme ADAR is fused to the RNA-binding protein of interest, and targets are identified by mRNA-seq as mRNAs containing sites significantly more edited in cells expressing the ADARcd fusion compared to a free ADARcd control (Arribas-Hernández et al, 2021a; McMahon et al, 2016). TRIBE can also be used to estimate differential protein-mRNA association between two conditions based on quantitative changes in editing proportions in target mRNAs. For example, many shared ECT2/3 target mRNAs are more highly edited by ECT3-ADARcd in the absence of ECT2, indicating that the two proteins compete for the same binding sites in vivo (Arribas-Hernández et al, 2021b). In iCLIP, target mRNAs are identified by co-purification with the protein of interest after covalent crosslinking in vivo, and binding sites are deduced from the position of frequent reverse transcription termination events at crosslink sites (König et al, 2010). We, therefore, set out to test predictions on shared and interdependent ECT-ALBA target binding in vivo using combined iCLIP and TRIBE analyses focused on ALBA4 (long form), ALBA2 (short form), and ECT2. Identification of mRNA targets of ALBA4 using iCLIP We first aimed to identify direct mRNA targets and binding sites of ALBA4 via iCLIP. To this end, we used transgenic lines expressing ALBA4-GFP under the control of the endogenous ALBA4 promoter in the alba4-1 alba5-1 alba6-1 (henceforth, alba456) mutant background (Appendix Fig. S5A,B), verified to carry T-DNA-induced knockout mutations in all three ALBA genes by RT-qPCR (Appendix Fig. S5C) and western blot (Appendix Fig. S5D) analyses. Initial immunoprecipitation tests with or without prior UV crosslinking and followed by polynucleotide kinase (PNK) labeling established that RNA-protein complexes were specifically purified with ALBA4-GFP after UV crosslinking (Fig. 4A). We therefore prepared and sequenced libraries from RNA immunopurified with ALBA4-GFP or GFP alone after crosslinking in vivo (Appendix Fig. S6A–D), using the recently developed iCLIP2 protocol (Buchbender et al, 2020; Lewinski et al, 2024). This effort identified 379,670 high-confidence replicated sites for ALBA4-GFP, corresponding to 7744 genes (henceforth referred to as ALBA4 iCLIP2 targets). We further defined a “strong” set by filtering low scores, resulting in 63,695 sites mapping to 7509 genes. In the GFP-only samples, only 81 sites in 13 genes were detected (Fig. 4B; Dataset EV2). Thus, nearly all ALBA4 iCLIP2 targets are strong candidates for bona fide ALBA4 target mRNAs.Figure 4The mRNA targets bound by ECT2/3 and ALBA2/4 overlap substantially.(A) Top, autoradiogram of 32P-labeled RNA-protein complexes obtained by PNK/γ-32P-ATP labeling of immunopurified material from ALBA4-GFP- or GFP-expressing plants. Immunoprecipitations were carried out with or without UV crosslinking and after precipitation with GFP-Trap beads (IP+). (IP−) indicates mock immunoprecipitation with RFP-Trap beads. Treatment of the precipitate with RNase I (+RNase) indicates the size of the precipitated protein. Marker positions and the location of the ALBA4-GFP RNA adducts are indicated. Bottom, immunoblots of input, supernatant (SN) after IP, and immunoprecipitated (IP) fractions, probed with GFP antibodies. Samples are pools of 3 independent lines for each genotype. (B) Number of called iCLIP peaks and associated genes for ALBA4-GFP, GFP alone, and ECT2-mCherry (Arribas-Hernández et al, 2021a). Strong ALBA4-GFP peaks are defined as those with a score higher than the median, per gene. (C) Scaled metagene profiles showing the enrichment along the gene body (5’UTR, CDS or 3’UTR) of ALBA4-GFP iCLIP2 peaks. (D) Overlap of ECT2 and ALBA4 iCLIP mRNAs. The overlap is highly significant (p = 0 according to permutation test based on random sampling of genes from transcriptome with matched expression patterns, see Methods). (E) Scatter plot of the editing proportions (E.P. = G/(A + G)) of potential and significant editing sites (E.S.) determined by comparing mRNA-seq data obtained from transgenic lines expressing ALBA2-FLAG-ADARcdE488Q or FLAG-ADARcdE488Q in the Col-0 background, both under the control of the ALBA2 promoter (seedlings, shoot tissue). Significance was determined using the hyperTRIBER pipeline (Rennie et al, 2021), specifying an adjusted p value <0.01 and log2 fold-change >1. (F) The same analysis as in (E), but carried out with roots of lines expressing ALBA4-FLAG-ADARcd or FLAG-ADARcd under the control of the ALBA4 promoter in the Col-0 background. (G) Overlap of ALBA4 targets identified using iCLIP2 and TRIBE analysis. The overlap is highly significant (p = 0, permutation test, as in D). (H) Overlap between ALBA4-TRIBE targets (roots) and ALBA2 HyperTRIBE targets (shoots). The overlap is highly significant (p = 0, permutation test, as in D). Most non-overlapping targets are expressed specifically in shoots or roots (Fig. EV2C). (I) Overlap between high-confidence ALBA4 targets, supported by iCLIP and TRIBE, and ECT2/3 targets, supported by ECT2/3 HyperTRIBE and ECT2 iCLIP. The overlap is highly significant (p = 0, permutation test, as in D). (J) Overlap between ALBA2 HyperTRIBE targets and ECT2/3 HyperTRIBE targets. The overlap is highly significant (p = 0, as in D). Source data are available online for this figure.Figure EV2Analysis of ALBA2 and ALBA4 target sets (supporting data).(A) Proportion of genes supported by ALBA4-TRIBE according to the number of iCLIP2 crosslink sites in the gene. Points represent true proportions, and intervals represent background distribution based on sampling genes of similar expression levels to the target genes from ALBA4 iCLIP2. Error bars are based on 2.5 and 97.5% quantiles and n = 177, 2282, 2855, 1299, 1132 is the number of iCLIP2 crosslink sites considered in each group (from left to right). The top panel represents the strong set of replicated ALBA4 iCLIP2 peaks and bottom panel represents the full set of replicated ALBA4 iCLIP2 peaks. (B) Support of ALBA4 iCLIP2 crosslink site-containing genes according to whether the gene is also supported by ECT2 iCLIP crosslink site-containing genes. (C) Scatter plot comparing expression for targets specific to either ALBA4-TRIBE (roots) or ALBA2 HyperTRIBE (aerial tissues). Values represent the averaged log2(TPM + 1) values across the Col-0 WT lines in each experiment. (D) Table showing defined strict and permissive gene sets for ECT2/3 and ALBA4. Venn diagrams provide visual representations of the ECT2/3 and ALBA4 strict and permissive gene sets. (E) Venn diagram for overlap between ALBA4 permissive and ECT2/3 permissive genes. Smaller Venn diagrams indicate overlap if either the ALBA4 or ECT2/3 permissive target sets were a randomly selected set of genes with a similar expression distribution to the true sets. (F) Venn diagrams for overlap between ECT2 iCLIP-derived target genes (strict) and ALBA4 iCLIP2 genes, where the ALBA4 set either contains or does not contain crosslink sites in its 3′-UTR. The mRNA targets bound by ECT2/3 and ALBA2/4 overlap substantially. (A) Top, autoradiogram of 32P-labeled RNA-protein complexes obtained by PNK/γ-32P-ATP labeling of immunopurified material from ALBA4-GFP- or GFP-expressing plants. Immunoprecipitations were carried out with or without UV crosslinking and after precipitation with GFP-Trap beads (IP+). (IP−) indicates mock immunoprecipitation with RFP-Trap beads. Treatment of the precipitate with RNase I (+RNase) indicates the size of the precipitated protein. Marker positions and the location of the ALBA4-GFP RNA adducts are indicated. Bottom, immunoblots of input, supernatant (SN) after IP, and immunoprecipitated (IP) fractions, probed with GFP antibodies. Samples are pools of 3 independent lines for each genotype. (B) Number of called iCLIP peaks and associated genes for ALBA4-GFP, GFP alone, and ECT2-mCherry (Arribas-Hernández et al, 2021a). Strong ALBA4-GFP peaks are defined as those with a score higher than the median, per gene. (C) Scaled metagene profiles showing the enrichment along the gene body (5’UTR, CDS or 3’UTR) of ALBA4-GFP iCLIP2 peaks. (D) Overlap of ECT2 and ALBA4 iCLIP mRNAs. The overlap is highly significant (p = 0 according to permutation test based on random sampling of genes from transcriptome with matched expression patterns, see Methods). (E) Scatter plot of the editing proportions (E.P. = G/(A + G)) of potential and significant editing sites (E.S.) determined by comparing mRNA-seq data obtained from transgenic lines expressing ALBA2-FLAG-ADARcdE488Q or FLAG-ADARcdE488Q in the Col-0 background, both under the control of the ALBA2 promoter (seedlings, shoot tissue). Significance was determined using the hyperTRIBER pipeline (Rennie et al, 2021), specifying an adjusted p value <0.01 and log2 fold-change >1. (F) The same analysis as in (E), but carried out with roots of lines expressing ALBA4-FLAG-ADARcd or FLAG-ADARcd under the control of the ALBA4 promoter in the Col-0 background. (G) Overlap of ALBA4 targets identified using iCLIP2 and TRIBE analysis. The overlap is highly significant (p = 0, permutation test, as in D). (H) Overlap between ALBA4-TRIBE targets (roots) and ALBA2 HyperTRIBE targets (shoots). The overlap is highly significant (p = 0, permutation test, as in D). Most non-overlapping targets are expressed specifically in shoots or roots (Fig. EV2C). (I) Overlap between high-confidence ALBA4 targets, supported by iCLIP and TRIBE, and ECT2/3 targets, supported by ECT2/3 HyperTRIBE and ECT2 iCLIP. The overlap is highly significant (p = 0, permutation test, as in D). (J) Overlap between ALBA2 HyperTRIBE targets and ECT2/3 HyperTRIBE targets. The overlap is highly significant (p = 0, as in D). Source data are available online for this figure. Analysis of ALBA2 and ALBA4 target sets (supporting data). (A) Proportion of genes supported by ALBA4-TRIBE according to the number of iCLIP2 crosslink sites in the gene. Points represent true proportions, and intervals represent background distribution based on sampling genes of similar expression levels to the target genes from ALBA4 iCLIP2. Error bars are based on 2.5 and 97.5% quantiles and n = 177, 2282, 2855, 1299, 1132 is the number of iCLIP2 crosslink sites considered in each group (from left to right). The top panel represents the strong set of replicated ALBA4 iCLIP2 peaks and bottom panel represents the full set of replicated ALBA4 iCLIP2 peaks. (B) Support of ALBA4 iCLIP2 crosslink site-containing genes according to whether the gene is also supported by ECT2 iCLIP crosslink site-containing genes. (C) Scatter plot comparing expression for targets specific to either ALBA4-TRIBE (roots) or ALBA2 HyperTRIBE (aerial tissues). Values represent the averaged log2(TPM + 1) values across the Col-0 WT lines in each experiment. (D) Table showing defined strict and permissive gene sets for ECT2/3 and ALBA4. Venn diagrams provide visual representations of the ECT2/3 and ALBA4 strict and permissive gene sets. (E) Venn diagram for overlap between ALBA4 permissive and ECT2/3 permissive genes. Smaller Venn diagrams indicate overlap if either the ALBA4 or ECT2/3 permissive target sets were a randomly selected set of genes with a similar expression distribution to the true sets. (F) Venn diagrams for overlap between ECT2 iCLIP-derived target genes (strict) and ALBA4 iCLIP2 genes, where the ALBA4 set either contains or does not contain crosslink sites in its 3′-UTR. mRNA target sets of ALBA proteins overlap significantly with those of ECT2/ECT3 We first noticed that ALBA4 iCLIP2 sites occurred in coding regions and, even more predominantly, in 3′-UTRs, with the 3′-UTR enrichment particularly apparent in the strong set (Fig. 4C). Importantly, more than 90% of ECT2 iCLIP targets are also ALBA4 iCLIP2 targets (Fig. 4D). Hence, ECT2 mainly binds to mRNAs that are also targeted by ALBA4. To corroborate this essential conclusion, we employed TRIBE to identify targets of both a long (ALBA4) and a short (ALBA2) ALBA protein family member by independent means. We used the improved variant HyperTRIBE relying on a hyperactive mutant (E488Q) of the ADARcd (Xu et al, 2018) for ALBA2, but had to proceed with TRIBE for ALBA4, because expression of the hyperactive ADARcd fused to ALBA4 was lethal (see Methods). In both cases, lines expressing comparable levels of free and ALBA-fused ADARcd were selected for mRNA-seq analysis (Appendix Fig. S7). Significantly differentially edited sites between fused and free ADARcd exhibited higher editing proportions in the ALBA2/4-FLAG-ADARcd fusions, as expected (Fig. 4E,F). These differentially edited sites defined 5272 target mRNAs for ALBA2 and 5995 for ALBA4 (Fig. 4E,F; Dataset EV3). Using these target sets, ALBA4 iCLIP2 targets and the previously defined ECT2/3 targets (Arribas-Hernández et al, 2021a; Arribas-Hernández et al, 2021b) for comparative analyses, we revealed the following three properties of ALBA2/4 and ECT2/3 target mRNAs and the relations between them. (1) The ALBA4 iCLIP2 target set is robust, because the overlap with ALBA4-TRIBE is significant (Fig. 4G). In particular, TRIBE support of ALBA4 iCLIP2 targets is prominent for those target mRNAs with multiple called iCLIP peaks (Fig. EV2A,B). (2) ALBA4 and ALBA2 target a common set of mRNAs (Fig. 4H) and differences between the two target sets can largely be explained by the tissue source used for the analysis (aerial tissues for ALBA2, roots for ALBA4) (Fig. EV2C). (3) The overlaps between the ECT2/3 target set and both the high-confidence set of ALBA4 targets supported by iCLIP2 and TRIBE and the set of ALBA2 HyperTRIBE targets are highly significant, as demonstrated by comparison to corresponding random target sets (Figs. 4I,J and EV2D–F; Appendix Fig. S8; Dataset EV4). We conclude that ECT2/3 and ALBA2/4 mRNA target sets significantly overlap, thus fulfilling a second key requirement of the model of concerted mRNA binding by ECT-ALBA modules. ALBA proteins bind to pyrimidine-rich elements in the vicinity of m6A We next analyzed the positions of ALBA4 binding sites in their targets using the iCLIP2 data. Metagene analysis normalizing for region length showed a peak in the density of ALBA4 binding sites in 3′-UTRs, if less pronounced than ECT2 binding sites and m6A- sites because ALBA4 binding sites also occur in coding regions as noted above (Fig. 5A). The RPS7A and TUBULIN ALPHA-5 genes provide illustrative examples of this close alignment of m6A, ECT2 and ALBA4 sites (Fig. 5B). Both ALBA4 iCLIP2 and ECT2 iCLIP peaks (Arribas-Hernández et al, 2021a) are enriched upstream of m6A sites determined by Nanopore direct RNA sequencing (Parker et al, 2020) (Fig. 5C,D), with ALBA peaks situated either at or slightly upstream of ECT2 peaks (Fig. 5E). Strikingly, the enrichment of ALBA4 peaks at m6A-sites was much more pronounced when considering peaks in ECT2 targets compared to non-targets. Indeed, the ALBA4 peak enrichment around m6A-sites in ECT2 non-targets showed a distribution similar to the location-matched background (Fig. 5F). These key observations demonstrate that the important prediction of a juxtaposition of ECT2 and ALBA4 binding sites on target mRNAs is fulfilled, and strongly suggest mutual dependence on target mRNA binding.Figure 5ALBA4 binds to pyrimidine-rich elements juxtaposed to m6A.(A) Scaled metagene profiles showing the enrichment along the gene body (5’UTR, CDS, or 3’UTR) of the called ALBA4 iCLIP2 peaks. ECT2 iCLIP peaks (Arribas-Hernández et al, 2021a) and Nanopore-determined m6A density (Parker et al, 2020) are shown for reference. (B) Representative examples of ECT2 and ALBA4 common targets showing the location of ALBA4 iCLIP2 and ECT2 iCLIP crosslink sites (Arribas-Hernández et al, 2021a), and m6A sites (Parker et al, 2020). (C) Number of ALBA4 iCLIP2 crosslink sites per 1000 Nanopore-derived m6A sites, as a function of distance from the m6A sites. (D) Number of ECT2 iCLIP crosslink sites per 1000 Nanopore-derived m6A sites, as a function of distance from the m6A sites. (E) Number of ALBA4 iCLIP2 crosslink sites per 1000 ECT2 crosslink sites, as a function of distance from the ECT2 crosslink sites. (F) Number of ALBA4 iCLIP2 crosslink sites per 1000 nanopore-derived m6A sites, as a function of distance from the m6A site and according to whether containing genes are also targets of ECT2 or non-ECT2 targets. For each set, a matched background set was defined as positions on similarly expressed genes with a similar metagene distribution to the true set. (G) Number of the indicated motifs (selected from (Arribas-Hernández et al, 2021a)) per 1000 nanopore-determined m6A sites (top), ECT2 iCLIP crosslink sites (middle) or ALBA4 iCLIP2 crosslink sites (bottom). For each set, a matched background set was defined as positions on similarly expressed genes with a similar metagene distribution to the true set. ALBA4 binds to pyrimidine-rich elements juxtaposed to m6A. (A) Scaled metagene profiles showing the enrichment along the gene body (5’UTR, CDS, or 3’UTR) of the called ALBA4 iCLIP2 peaks. ECT2 iCLIP peaks (Arribas-Hernández et al, 2021a) and Nanopore-determined m6A density (Parker et al, 2020) are shown for reference. (B) Representative examples of ECT2 and ALBA4 common targets showing the location of ALBA4 iCLIP2 and ECT2 iCLIP crosslink sites (Arribas-Hernández et al, 2021a), and m6A sites (Parker et al, 2020). (C) Number of ALBA4 iCLIP2 crosslink sites per 1000 Nanopore-derived m6A sites, as a function of distance from the m6A sites. (D) Number of ECT2 iCLIP crosslink sites per 1000 Nanopore-derived m6A sites, as a function of distance from the m6A sites. (E) Number of ALBA4 iCLIP2 crosslink sites per 1000 ECT2 crosslink sites, as a function of distance from the ECT2 crosslink sites. (F) Number of ALBA4 iCLIP2 crosslink sites per 1000 nanopore-derived m6A sites, as a function of distance from the m6A site and according to whether containing genes are also targets of ECT2 or non-ECT2 targets. For each set, a matched background set was defined as positions on similarly expressed genes with a similar metagene distribution to the true set. (G) Number of the indicated motifs (selected from (Arribas-Hernández et al, 2021a)) per 1000 nanopore-determined m6A sites (top), ECT2 iCLIP crosslink sites (middle) or ALBA4 iCLIP2 crosslink sites (bottom). For each set, a matched background set was defined as positions on similarly expressed genes with a similar metagene distribution to the true set. Because we previously showed that several sequence motifs are enriched around ECT2 binding sites (Arribas-Hernández et al, 2021a), we went on to study whether any of these motifs were enriched at ALBA4 binding sites. We included 6 motifs identified as enriched around ECT2 iCLIP sites in our previous study (Arribas-Hernández et al, 2021a). This analysis revealed that uridine- or pyrimidine-rich motifs in the immediate vicinity of m6A/ECT2 binding sites are strongly enriched precisely at ALBA4 crosslink sites (Fig. 5G), suggesting that these sequences may be ALBA4 binding sites in vivo. Deep learning supports pyrimidine-rich elements in the vicinity of m6A as determinants of ALBA4-ECT2 binding One potential pitfall of this conclusion is that the photochemical properties of nucleobases result in a bias of UV-induced RNA-protein crosslinks to occur at uridines (Angelov et al, 2023; Hafner et al, 2021) such that iCLIP sites can be located at nearby uridines if the actual binding site lacks this nucleotide. For example, many miCLIP sites obtained by UV crosslinking of an m6A-specific antibody to RNA in vitro map to uridines surrounding the uridine-depleted major m6A consensus site (DRACH) (Arribas-Hernández et al, 2021a). Therefore, we employed neural networks to identify sequence elements that distinguish m6A sites bound by ECT2/ALBA4 from m6A sites not bound by these proteins. We first collected Arabidopsis m6A sites from multiple published sources and curated a compendium of 41,883 non-overlapping m6A sites which have properties highly consistent with the smaller set of sites identified by Nanopore direct RNA sequencing (Parker et al, 2020) (Fig. EV3A–F; Dataset EV5, see Methods). The high quality of these sites is supported by their strong enrichment in single-nucleotide resolution m6A sites recently obtained with the m6A-specific allyl chemical labeling and sequencing method (m6A-SAC-seq (Wang et al, 2024)) (Fig. EV3G). Of these, 16,406 sites were annotated as ECT2-positive and 22,866 were ALBA4-positive (Fig. 6A). Although there was a large overlap between the two proteins, there was a sizable set of bound sites unique to each protein (Fig. EV3H), allowing analysis of sequence features of sites bound by both proteins and uniquely bound sites. We then used sequences surrounding all sites for input into a neural network trained simultaneously on two binary outputs: whether ECT2 was bound or unbound, and whether ALBA4 was bound or unbound (Fig. 6A). This model performed well when predicting the presence of ALBA4 or ECT2 at m6A sites on gene sets excluded during model training (“held-out set”; average AUC = 0.74 (ECT2) and 0.76 (ALBA4), based on fivefold cross validation), with predicted binding probabilities clearly distinguishing between bound and unbound sites (Fig. 6B). As expected, predicted binding probabilities for the two proteins correlated (PCC = 0.71, Fig. 6C). Importantly, some differences between the two suggested that the model had learned specific sequence patterns relevant to each protein. To investigate this, we leveraged the filters learned in the first convolutional layer, since these represent motifs identified de novo by the model. We converted the sequences of the highest-scoring instances into position weight matrices and fit a generalized linear model predicting motif presence additively from the network-predicted ECT2 and ALBA4 binding probabilities (see Methods). From this model, the coefficient for each protein (motif score) can be interpreted as the effect of that protein controlling for the other (Fig. 6D). This analysis identified the uridine-/pyrimidine-rich motifs UAUUUU and UUUACUUU as determinants of both ECT2-bound and ALBA4-bound m6A sites (Fig. 6D). Indeed, the UAUUUU and UUUACUUU motifs were highly enriched at ALBA4 iCLIP sites and located just upstream of ECT2 iCLIP sites (Fig. 6E), thus providing independent experimental evidence that these motifs act as ALBA4 binding sites. This conclusion is particularly important because it provides a simple molecular explanation for our previous machine learning-based finding that uridine- or pyrimidine-rich motifs are important for the distinction between m6A sites bound or not by ECT2 (Arribas-Hernández et al, 2021a): juxtaposed m6A sites and uridine-/pyrimidine-rich elements provide the context required for binding of the ECT-ALBA module.Figure 6Neural network analysis identifies U-rich motifs in the vicinity of m6A as determinants of ALBA4-ECT2 binding.(A) Strategy for deep learning. m6A sites were annotated according to the presence or absence of either ECT2 or ALBA4, and a convolutional neural network was trained, which takes sequences surrounding m6A as input and predicts the probability of ECT2 and ALBA4 binding. (B) Boxplots showing predicted binding probabilities from the network, split according to protein and binding status. From left to right: n = 25,477, 16,406, 19,017, 22,866, and boxplots show {minima, 25th percentile, median, 75th percentile, maxima}. (C) Scatter plot of the predicted ALBA4 binding probabilities against the ECT2 binding probabilities from the network. Counts depict the density of sites. (D) Output-specific enrichment scores for de novo motifs learned by the convolutional neural network, calculated using a generalized linear model for predicting motif presence from the predicted presence of ECT2 and ALBA4 at m6A-centered sequences using model. Colored circles indicate interesting motifs determined as specific to ALBA4 (yellow), ECT2 (blue), or both (green). (E) Enrichment of motif sets indicated in D around ALBA4 iCLIP2, ECT2 iCLIP, and nanopore-derived m6A sites (Parker et al, 2020). Gray shows location-matched background positions.Figure EV3A deep learning model to derive an augmented set of single-nucleotide-resolution m6A sites by integration of miCLIP and Nanopore data.(A) Strategy for the m6A deep learning model. Each Nanopore m6A site was paired with a location-matched negative (see Methods for definition of location-matching), and sequences ±100 bp from the site were extracted and used as input to a convolutional neural network tasked with predicting the presence or absence of m6A in the center of the input sequence. AUC area under the curve. (B) Strategy for correcting the exact position of m6A sites based on the deep learning model. For each site and all individual positions within a ±50 bp window, a ±100 bp sequence was extracted and used as input to the m6A-network to predict the presence of m6A. The position with the strongest prediction was chosen as the corrected position for that site. (C) Reference base distribution of m6A sites as determined by Nanopore, miCLIP, or the corrected positions. (D) Enrichment of nanopore-determined m6A sites around the augmented set of m6A positions, before and after correction. (E) Enrichment of RRACH motifs per 1000 augmented m6A sites, before and after correction. (F) IGV view of a representative transcript, AUTOPHAGY 8F (At4g16520), showing the positions of m6A sites experimentally determined by either nanopore or miCLIP, and the positions resulting from the correction. (G) Enrichment of m6A sites identified by m6A-SAC-seq (Wang et al, 2024) around miCLIP sites before and after correction, and around nanopore-derived sites. (H) Annotation of m6A sites according to ECT2 and ALBA4 crosslink sites within 100 nt. Venn diagrams show overlap between the two proteins for both positives (bound) and negatives (non-bound) separately. Neural network analysis identifies U-rich motifs in the vicinity of m6A as determinants of ALBA4-ECT2 binding. (A) Strategy for deep learning. m6A sites were annotated according to the presence or absence of either ECT2 or ALBA4, and a convolutional neural network was trained, which takes sequences surrounding m6A as input and predicts the probability of ECT2 and ALBA4 binding. (B) Boxplots showing predicted binding probabilities from the network, split according to protein and binding status. From left to right: n = 25,477, 16,406, 19,017, 22,866, and boxplots show {minima, 25th percentile, median, 75th percentile, maxima}. (C) Scatter plot of the predicted ALBA4 binding probabilities against the ECT2 binding probabilities from the network. Counts depict the density of sites. (D) Output-specific enrichment scores for de novo motifs learned by the convolutional neural network, calculated using a generalized linear model for predicting motif presence from the predicted presence of ECT2 and ALBA4 at m6A-centered sequences using model. Colored circles indicate interesting motifs determined as specific to ALBA4 (yellow), ECT2 (blue), or both (green). (E) Enrichment of motif sets indicated in D around ALBA4 iCLIP2, ECT2 iCLIP, and nanopore-derived m6A sites (Parker et al, 2020). Gray shows location-matched background positions. A deep learning model to derive an augmented set of single-nucleotide-resolution m6A sites by integration of miCLIP and Nanopore data. (A) Strategy for the m6A deep learning model. Each Nanopore m6A site was paired with a location-matched negative (see Methods for definition of location-matching), and sequences ±100 bp from the site were extracted and used as input to a convolutional neural network tasked with predicting the presence or absence of m6A in the center of the input sequence. AUC area under the curve. (B) Strategy for correcting the exact position of m6A sites based on the deep learning model. For each site and all individual positions within a ±50 bp window, a ±100 bp sequence was extracted and used as input to the m6A-network to predict the presence of m6A. The position with the strongest prediction was chosen as the corrected position for that site. (C) Reference base distribution of m6A sites as determined by Nanopore, miCLIP, or the corrected positions. (D) Enrichment of nanopore-determined m6A sites around the augmented set of m6A positions, before and after correction. (E) Enrichment of RRACH motifs per 1000 augmented m6A sites, before and after correction. (F) IGV view of a representative transcript, AUTOPHAGY 8F (At4g16520), showing the positions of m6A sites experimentally determined by either nanopore or miCLIP, and the positions resulting from the correction. (G) Enrichment of m6A sites identified by m6A-SAC-seq (Wang et al, 2024) around miCLIP sites before and after correction, and around nanopore-derived sites. (H) Annotation of m6A sites according to ECT2 and ALBA4 crosslink sites within 100 nt. Venn diagrams show overlap between the two proteins for both positives (bound) and negatives (non-bound) separately. Binding to target mRNA in vivo involves mutual ALBA-ECT dependence We next assessed whether ALBA proteins are necessary for mRNA target association of the wild type ECT2 protein. Initially, we used the CLIP-PNK assay with ECT2-mCherry expressed in wild type, or the alba1-2 alba2-2 alba4-1 alba5-1 (henceforth alba1245) or alba456 mutant backgrounds, carrying T-DNA insertion alleles in the corresponding ALBA genes (Appendix Figs. S2B and S5, see Methods). These experiments showed that ECT2-mCherry was associated with less RNA in the alba mutants compared to wild type, with the clearest effects (~2.5-fold reduction) observed in alba456 mutants (Figs. 7A and EV4A). We next used ECT2 HyperTRIBE to estimate the relative target mRNA binding in wild type and in alba1245 mutants by differential editing. We chose this method both to gain sensitivity and to assess directly whether mRNAs that associate less with ECT2 in vivo in alba1245 mutants are, in fact, dual ECT2/ALBA targets. We selected five independent lines expressing ECT2-ADAR in both wild type and alba1245, and performed mRNA-seq of root tissues to provide the raw data for analysis of differential editing. Positions exhibiting significant differential editing according to the hyperTRIBER package (Rennie et al, 2021) were strongly biased in the direction of lower editing in alba1245, although these results were potentially biased by the expression of the ECT2-ADAR fusion protein not being balanced between the two conditions (Fig. EV4B,C,H). For this reason, we developed a highly robust alternative statistical modeling approach, correcting the editing proportions for mRNA levels of ADAR to obtain a smaller, high-confidence set of significantly differentially edited sites between the two backgrounds (see Methods). As a further control, we also performed differential editing analysis using only those replicates whose ECT2-ADAR expression was nearly perfectly matched as judged by both mRNA-seq read densities and protein blots, resulting in a smaller set of sites which overlapped significantly with the set from the robust modeling approach (Fig. EV4F,G). Overall, these analyses converged on the same conclusion: editing proportions by ECT2-FLAG-ADARcdE488Q in ECT2/ALBA4 dual mRNA targets tended to be higher in wild type than in alba1245 mutants, indicating that ALBA proteins facilitate target mRNA binding of ECT2 in vivo (Fig. 7B,D). Because the structural model of the ALBA-ECT2 interaction suggests that RNA association by the ALBA domain may also be enhanced by ECT proteins, we did the reciprocal experiment with the short ALBA2 protein. Thus, we expressed ALBA2-FLAG-ADARcdE488Q either in wild type or ect2-3 ect3-2 ect4-2 (Gte234) mutant backgrounds and carried out analysis of differential editing proportions as above. We found that editing proportions by ALBA2-FLAG-ADARcdE488Q of ALBA2-ECT2 dual targets were higher in wild type than in Gte234 mutants (Figs. 7C,E and EV4D,E,I), indicating that there is mutual ALBA-ECT dependence for mRNA target association in vivo. Taken together, our CLIP-PNK assays and TRIBE-based assessment of target mRNA association in vivo support the conclusion that the ALBA domain acts as a unit with the YTH domain to facilitate m6A-binding.Figure 7ALBA proteins are required for ECT2 target mRNA binding and biological function.(A) Results of an in vivo UV crosslinking-ECT2-mCherry immunoprecipitation experiment, followed by PNK-labeling of precipitated RNA with γ-32P-ATP. Top panel, autoradiogram of 32P-radiolabelled RNA-protein complexes purified from plants expressing ECT2WT‐mCherry in the indicated genetic backgrounds. Molecular weight marker positions and the location of the verified ECT2-mCherry-RNA complexes (Arribas-Hernández et al, 2021a) are indicated. The presence of several bands of unequal intensity is due to partial proteolysis of the ECT2 IDR during immunoprecipitation, and differential labeling efficiency of the different RNPs (Arribas-Hernández et al, 2021a). Middle and bottom panels, mCherry immunoblots of the immunoprecipitated (middle) and total fractions (input, bottom). Samples were pools of three independent lines for each genotype. (B) Scatter plot showing the editing proportions (E.P. = G/(A+G)) of ECT2-FLAG-ADARcdE488Q-catalyzed editing sites between Col-0 WT and alba1245. Green, sites whose change in editing proportions is statistically significant and that are located in dual-bound mRNAs. Yellow, sites whose change in editing proportions is statistically significant but that are located in mRNAs not targeted by both ECT2 and ALBA4 (non-dual bound). Light green/light yellow, candidate sites whose change in editing proportions is not statistically significant. (C) Scatter plot showing the editing proportions of ALBA2-FLAG-ADARcdE488Q-catalyzed editing sites between Col-0 WT and ect2-3 ect3-2 ect4-2 (Gte234). Color scheme as in (B). (D) Quantification of the tendency of sites differentially edited by ECT2-FLAG-ADARcdE488Q between Col-0 and alba1245 to be less highly edited in alba1245. Left, histogram showing the fraction that sites in dual-bound targets comprise of either less highly edited sites in alba1245 (down) or more highly edited sites in alba1245 (up). The histogram also illustrates the fraction of editing sites in dual-bound targets relative to all editing sites for comparison. Right, boxplot showing the median log2 differential editing proportions for editing sites either in dual-bound mRNA targets (TRUE, n = 10,272) or in other mRNAs (FALSE, n = 20,343), indicated p value is based on two-sample t-test and boxplots show {minima, 25th percentile, median, 75th percentile, maxima}. (E) Quantification of the tendency of sites differentially edited by ALBA2-FLAG-ADARcdE488Q between Col-0 and Gte234 to be less highly edited in Gte234. Analogous to the analyses presented in (D) for ECT2-ADAR in Col-0 vs alba1245 (n = 8248 and n = 20,024 for dual-bound and other mRNAs, respectively). (F) Representative photographs of seedlings and rosettes of the indicated genotypes at three different time points were given in days after germination (DAG) in soil. (G) Trichome branching is sorted by the number of spikes in the indicated genotypes. Branches were counted on at least 150 trichomes on each of at least six plants for each genotype (n = ∼1000). Data were fitted to a proportional odds model in R for statistical analyses (see Methods) with Bonferroni-corrected p values. NS not significant. Source data are available online for this figure.Figure EV4Mutual ECT-ALBA dependence for target mRNA binding.(A) Autoradiogram (top) of RNA-protein complexes from ECT2‐mCherry expressed in Col-0, alba1245, or alba456 after UV crosslinking and immunoprecipitation with RFP-trap beads. Marker positions and the location of the ECT2-mCherry-RNA adducts are indicated. Immunoblots against mCherry show the precipitated protein in the IP (middle) and input (bottom). Samples were pools of three independent lines for each genotype. (B) Western blot of the independent lines selected for HyperTRIBE analysis of ECT2 in Col-0 and alba1245. Ponceau staining was used as a loading control. (C) Detected number of transcripts per million of ECT2-FLAG-ADAR in lines used for HyperTRIBE. Light bars indicate lines selected for the two-sample analysis. (D) Western blot of the independent lines selected for HyperTRIBE analysis of ALBA2 in Col-0 and gte234. Ponceau staining was used as a loading control. (E) Detected number of transcripts per million of ALBA2-FLAG-ADAR in lines used for HyperTRIBE. (F) Overlap of individual sites detected as significant in ECT2-HT in Col-0 vs. alba1245 according to three different approaches—HyperTRIBER pipeline (no correction), HyperTRIBER pipeline (using lines indicated in C) and developed a robust model. (G) As in (F) but counts depict containing genes for significant sites. (H–I) The proportion of sites significantly differentially edited by ECT2- (H) or ALBA2- (I) -FLAG-ADARE488Qcd fusions in Col-0 vs. the indicated mutant background (alba1245 in H; Gte234 in I), which are targets of ALBA2, ALBA4, ECT2, or both ALBA4 and ECT2 (dual-targets). UP/DOWN categories are defined according to whether the log2 fold-change was negative (lower in the mutant background, DOWN) or positive (higher in the mutant background, UP). Source data are available online for this figure. ALBA proteins are required for ECT2 target mRNA binding and biological function. (A) Results of an in vivo UV crosslinking-ECT2-mCherry immunoprecipitation experiment, followed by PNK-labeling of precipitated RNA with γ-32P-ATP. Top panel, autoradiogram of 32P-radiolabelled RNA-protein complexes purified from plants expressing ECT2WT‐mCherry in the indicated genetic backgrounds. Molecular weight marker positions and the location of the verified ECT2-mCherry-RNA complexes (Arribas-Hernández et al, 2021a) are indicated. The presence of several bands of unequal intensity is due to partial proteolysis of the ECT2 IDR during immunoprecipitation, and differential labeling efficiency of the different RNPs (Arribas-Hernández et al, 2021a). Middle and bottom panels, mCherry immunoblots of the immunoprecipitated (middle) and total fractions (input, bottom). Samples were pools of three independent lines for each genotype. (B) Scatter plot showing the editing proportions (E.P. = G/(A+G)) of ECT2-FLAG-ADARcdE488Q-catalyzed editing sites between Col-0 WT and alba1245. Green, sites whose change in editing proportions is statistically significant and that are located in dual-bound mRNAs. Yellow, sites whose change in editing proportions is statistically significant but that are located in mRNAs not targeted by both ECT2 and ALBA4 (non-dual bound). Light green/light yellow, candidate sites whose change in editing proportions is not statistically significant. (C) Scatter plot showing the editing proportions of ALBA2-FLAG-ADARcdE488Q-catalyzed editing sites between Col-0 WT and ect2-3 ect3-2 ect4-2 (Gte234). Color scheme as in (B). (D) Quantification of the tendency of sites differentially edited by ECT2-FLAG-ADARcdE488Q between Col-0 and alba1245 to be less highly edited in alba1245. Left, histogram showing the fraction that sites in dual-bound targets comprise of either less highly edited sites in alba1245 (down) or more highly edited sites in alba1245 (up). The histogram also illustrates the fraction of editing sites in dual-bound targets relative to all editing sites for comparison. Right, boxplot showing the median log2 differential editing proportions for editing sites either in dual-bound mRNA targets (TRUE, n = 10,272) or in other mRNAs (FALSE, n = 20,343), indicated p value is based on two-sample t-test and boxplots show {minima, 25th percentile, median, 75th percentile, maxima}. (E) Quantification of the tendency of sites differentially edited by ALBA2-FLAG-ADARcdE488Q between Col-0 and Gte234 to be less highly edited in Gte234. Analogous to the analyses presented in (D) for ECT2-ADAR in Col-0 vs alba1245 (n = 8248 and n = 20,024 for dual-bound and other mRNAs, respectively). (F) Representative photographs of seedlings and rosettes of the indicated genotypes at three different time points were given in days after germination (DAG) in soil. (G) Trichome branching is sorted by the number of spikes in the indicated genotypes. Branches were counted on at least 150 trichomes on each of at least six plants for each genotype (n = ∼1000). Data were fitted to a proportional odds model in R for statistical analyses (see Methods) with Bonferroni-corrected p values. NS not significant. Source data are available online for this figure. Mutual ECT-ALBA dependence for target mRNA binding. (A) Autoradiogram (top) of RNA-protein complexes from ECT2‐mCherry expressed in Col-0, alba1245, or alba456 after UV crosslinking and immunoprecipitation with RFP-trap beads. Marker positions and the location of the ECT2-mCherry-RNA adducts are indicated. Immunoblots against mCherry show the precipitated protein in the IP (middle) and input (bottom). Samples were pools of three independent lines for each genotype. (B) Western blot of the independent lines selected for HyperTRIBE analysis of ECT2 in Col-0 and alba1245. Ponceau staining was used as a loading control. (C) Detected number of transcripts per million of ECT2-FLAG-ADAR in lines used for HyperTRIBE. Light bars indicate lines selected for the two-sample analysis. (D) Western blot of the independent lines selected for HyperTRIBE analysis of ALBA2 in Col-0 and gte234. Ponceau staining was used as a loading control. (E) Detected number of transcripts per million of ALBA2-FLAG-ADAR in lines used for HyperTRIBE. (F) Overlap of individual sites detected as significant in ECT2-HT in Col-0 vs. alba1245 according to three different approaches—HyperTRIBER pipeline (no correction), HyperTRIBER pipeline (using lines indicated in C) and developed a robust model. (G) As in (F) but counts depict containing genes for significant sites. (H–I) The proportion of sites significantly differentially edited by ECT2- (H) or ALBA2- (I) -FLAG-ADARE488Qcd fusions in Col-0 vs. the indicated mutant background (alba1245 in H; Gte234 in I), which are targets of ALBA2, ALBA4, ECT2, or both ALBA4 and ECT2 (dual-targets). UP/DOWN categories are defined according to whether the log2 fold-change was negative (lower in the mutant background, DOWN) or positive (higher in the mutant background, UP). Source data are available online for this figure. Inactivation of ALBA and ECT genes causes similar developmental phenotypes We finally characterized mutants in ALBA genes to assess whether they exhibit phenotypes characteristic of reduced m6A-ECT function. As previously reported, single alba mutants (Appendix Fig. S2) did not show obvious developmental phenotypes (Tong et al, 2022). In contrast, alba123 mutants and, in particular, alba1245 and alba456 mutants showed pleiotropic developmental defects, including slower growth, defects in leaf morphology, and delayed flowering (Figs. 7F and EV5A,B). Similar observations on the smaller stature of alba456 mutants have been reported by others (Tong et al, 2022). Although some of these phenotypes are reminiscent of phenotypes displayed by ect2 ect3 ect4 mutants, they are not identical. We, therefore, assessed a quantifiable phenotype seen consistently in mutants with defects in m6A-ECT function: increased branching of leaf epidermal hairs (trichomes, (Bodi et al, 2012)) that can be detected even in single ect2 and ect3 mutants (Arribas-Hernández et al, 2018; Scutenaire et al, 2018; Wei et al, 2018). We found that alba1245 and alba456 mutants showed increased trichome branching (Fig. 7G), with a phenotypic strength intermediate between ect2 or ect3 single mutants and the te234 triple knockout mutant. Importantly, ECT2 protein levels in alba1245 and alba456 mutants were only slightly lower than in wild type (Fig. EV5C), excluding the trivial possibility that the phenotypic similarity between composite alba and ect mutants is due to drastically reduced ECT protein levels in alba1245 and alba456. We conclude that the developmental defects of composite alba mutants are consistent with defective m6A-ECT function, as predicted by the model of m6A-ECT interaction facilitated by ALBA proteins.Figure EV5Phenotypic analysis of alba1245 and alba456 mutants.(A) Leaf profiles of alba mutants at day 17 after germination. (B) Delayed flowering phenotype of higher-order alba and ect mutants (5-week-old, germinated directly on soil). (C) Protein blot of total lysates prepared from 10-day-old seedlings of the indicated genotypes, probed with ECT2-specific antisera (Arribas-Hernández et al, 2018). The arrow indicates the position of the ECT2WT protein. Ponceau staining serves as the loading control. Source data are available online for this figure. Phenotypic analysis of alba1245 and alba456 mutants. (A) Leaf profiles of alba mutants at day 17 after germination. (B) Delayed flowering phenotype of higher-order alba and ect mutants (5-week-old, germinated directly on soil). (C) Protein blot of total lysates prepared from 10-day-old seedlings of the indicated genotypes, probed with ECT2-specific antisera (Arribas-Hernández et al, 2018). The arrow indicates the position of the ECT2WT protein. Ponceau staining serves as the loading control. Source data are available online for this figure. Discussion Our results on the ALBA-ECT interaction and target binding in vivo provide strong support for the conclusion that the YTH domain of major plant YTHDF proteins is insufficient for full m6A binding in vivo, because it requires facilitation by ALBA proteins. In the following paragraphs, we discuss how this new understanding of the m6A-YTH interaction impacts the thinking of m6A-mediated genetic control in plants and other eukaryotes. Functional implications of recognition of m6A by the ALBA-YTHDF module The discovery that m6A reading in plants involves YTHDF-m6A binding modulated by a third player, the ALBA proteins, introduces increased potential to integrate information into combinatorial control of biological effects of m6A. A key determinant of those effects is the fraction of m6A target mRNAs bound by YTHDF, in turn, determined by the stoichiometry of m6A modification in mRNA, and YTHDF concentration and affinity for m6A-sites. Since we now understand that the affinity is not a constant, but must be tunable via, for instance, ALBA concentration and post-translational modification, we envision that plants have evolved to take advantage of this combinatorial potential to generate a gradient of m6A outputs that matches the cellular environment measured by multiple environmental and developmental sensors. Conservation of the ALBA-YTHDF unit and generality of RBP-assisted m6A-YTH interaction It is an important observation that Arabidopsis YTHDF proteins both with and without the molecular properties required to complement organogenesis defects of te234 mutants (Flores-Téllez et al, 2023) retain the conserved YAIM and interact with ALBA4. This observation further supports the generality of ALBA-assisted m6A-binding among Arabidopsis YTHDF proteins. Thus, it is a pertinent question how widespread this phenomenon is. The YAIM is deeply conserved in land plant YTHDF proteins, strongly suggesting that the ALBA-YTHDF unit is conserved over the 500 million years of land plant evolution. In fungi and animals, the YAIM is not conserved, and fungal and animal ALBA-family proteins are so divergent in sequence that conservation of the details of their molecular functions cannot be assumed. In addition, Trypanosoma brucei where ALBA proteins clearly perform functions in mRNA control (Bevkal et al, 2021; Mani et al, 2011) does not encode YTHDF proteins, providing an example that the two families do not always have linked functions in eukaryotes. These observations raise two immediate questions. First, given the deep conservation of the YTH domain, it is of interest how the m6A-YTHDF interaction is made efficient in organisms where ALBA proteins are unlikely to assist binding directly as in plants. We see two possible answers. Either other, as yet unidentified classes of RBPs evolved to facilitate m6A reading by YTHDF proteins, or the YTHDF proteins evolved to read m6A independently of other RBPs. In the latter case, comparative structure-function studies between, for instance, Arabidopsis and human YTHDF-m6A-RNA interactions should reveal the probably subtle structural features that may allow ALBA-independent efficient m6A-interaction. In this context, a YTH-proximal element in the IDRs of mammalian YTHDF proteins is of particular interest for at least two reasons. First, its location relative to the YTH domain is reminiscent of the YAIM described here for plant YTHDFs. Second, it is predicted by AlphaFold (Jumper et al, 2021) to engage in YTH domain interactions, perhaps via disorder-to-order transition upon RNA-binding to stabilize the RNA-bound form (Sikorski et al, 2023), as observed for the Schizosaccharomyces pombe YTH domain protein Mmi1 (Wang et al, 2016). This may also be the core function of the YAIM in ECT2, assisted in vivo by ALBA proteins, as deletion of a small YAIM-containing region of ECT2 reduces its m6A-binding capacity in vitro (Seigneurin-Berny et al, 2024). The existence of non-ALBA facilitators of YTHDF-m6A binding in other organisms should not be entirely discarded, however. The mammalian IGF2BP/IMP/ZBP family of RBPs has been suggested to act as m6A readers based on multiple lines of evidence, including m6A-dependent target mRNA association and the similar positions of m6A sites and IGF2BP2 CLIP sites in 3′-UTRs of target mRNAs (Huang et al, 2018). Because the m6A mapping methodology used at the time had limited resolution, it is possible that m6A sites are, in fact, adjacent to IGF2BP2 CLIP sites, particularly since the IGF2BP/IMP/ZBP recognition element (CAUH) defined in previous transcriptome-wide studies (Hafner et al, 2010) is not identical to the DRACH m6A consensus site. The slight off-set between IGF2BP CLIP site and m6A distributions (Huang et al, 2018) is indeed reminiscent of the 3′-UTR distributions of m6A sites and ALBA4 iCLIP sites observed here, and the identification of IGF2BP2 as a prominent interactor of YTHDF1/2/3 in IP-MS experiments (Zaccara and Jaffrey, 2020) is more easily reconciled with a function in facilitated m6A binding by YTHDFs than direct m6A binding competing with YTHDFs. Thus, in light of our results on the ALBA-YTHDF-m6A module in plants, it may be appropriate to consider whether facilitated m6A-reading by YTHDF proteins could have evolved independently in several eukaryotic lineages, and, for mammals in particular, whether a function as a facilitator of m6A reading might explain many of the results originally interpreted to reveal a direct reader function of the IGF2BPs (Huang et al, 2018). A second important question concerns molecular functions fulfilled by ALBA proteins independently of YTHDF proteins. Such functions are anticipated for a number of reasons. First, while most ALBA4 mRNA binding sites in 3′-UTRs appear to be linked to m6A sites, binding sites in open reading frames were even more numerous and were found in mRNAs with no evidence of m6A modification or ECT2/3 binding. Indeed, ALBA proteins have been found to play a role in heat adaptation via the regulation of Heat Shock Factor-encoding mRNAs, primarily with binding sites in open reading frames (Tong et al, 2022). Second, even the YTHDF-linked ALBA functions may involve properties in addition to assisted m6A-binding, because many ECT2-associated proteins were depleted in the immunoaffinity-purified fraction of the ECT2YAIM mutant defective in ALBA interaction. Finally, we note that while this report identifies a molecular role of the ALBA domain, it does not address the function of the C-terminal IDR of long ALBA proteins, expected to be of considerable biological importance given the stronger phenotypes of alba456 compared to alba123 mutants, as reported here and by others (Tong et al, 2022). Finally, we note that while this report clearly demonstrates the role of ALBA proteins in the facilitation of m6A reading, it does not exclude the possibility that other RBP families co-purifying with ECTs could also play a role in this process. Methods Reagents and tools tableReagent/resourceReference or sourceIdentifier or catalog number Experimental models Arabidopsis thaliana Col-0N/AN/AE. coli DH5α cellsInvitrogen18265-017E. coli BL21 cellsNEBC2530HAgrobacterium tumefaciens strain GV3101(Koncz and Schell, 1986)N/A Recombinant DNA pCAMBIA3300-U vector(Nour-Eldin et al, 2006)N/ApGEM-T EasyPromegaA137ApGGD003Addgene48835pGGE000Addgene48860pGGF003Addgene48844pGGZ001Addgene48868His6-SUMO pET24Addgene29711pDONR/ZeoThermo Fisher12535035pMDC111(Curtis and Grossniklaus, 2003)N/ApMDC164(Curtis and Grossniklaus, 2003)N/A Antibodies anti-ECT2(Arribas-Hernández et al, 2018)N/Aanti-ALBA1This studyN/Aanti-mCherryAbcamab183628Goat anti-Rabbit IgGSigma-AldrichA6154-1ML Oligonucleotides and other sequence-based reagents PrimersThis studyDataset EV6 Chemicals, enzymes, and other reagents KAPA Hifi Hotstart Uracil+ ReadyMixRocheKK2801Phusion High-Fidelity DNA PolymeraseNEBM0530LBsaI-HFNEBR3733ST4 DNA-LigaseThermo Scientific15224017KOD Hot Start DNA PolymeraseSigma-Aldrich71086Gateway BP Clonase IIThermo Fisher11789020Gateway LR Clonase™ II Plus enzymeThermo Fisher12538120MS-agarPhytoTech LabsM407TryptoneBD211705Yeast extractBD212750AgarBD214010Glufosinate ammoniumSigma45520AmpicillinServa13398.02KanamycinServa26898.03IPTGThermo FisherR0392Ni-NTA agarose beadsmacherey-nagel745400.25Gravity columnBio-Rad7326008ImidazoleThermo ScientificJ17525His6-tagged ULP1 proteaseKind gift from Prof. Birthe Kragelund, UCPH.N/AComplete protease inhibitor EDTA-freeSigma-Aldrich11873580001NuPAGE™ LDS Sample BufferThermo FisherNP00074–20% Criterion™ TGX™ Precast gelBio-Rad5671094Amersham Protran Premium nitrocellulose membraneGE Healthcare Life SciencesGE10600002TRIzolSigmaT9424 - 200 MLRQ1 RNase-Free DNasePromegaM6101RNaseOutTM Recombinant RNase InhibitorInvitrogen10777019QIAgen RNeasy mini kitQiagen74104RNeasy MinElute Cleanup KitQiagen74204SuperScript® III Reverse TranscriptaseInvitrogen18080093RNaseOutTMThermo Fisher10777019SensiFAST SYBR No-ROX MixBiolineBIO-98005RFP-trap agarose beadsChromotekrta-20GFP-trap agarose beadsChromotekgta-20Sepharose 4B beadsSigma-Aldrich4B200T4-polynucleotide kinaseThermo FisherEK0032Turbo DNaseInvitrogenAM2238Ambion RNase IInvitrogenAM2294T4 RNA ligase 1 high concentrationNEBM0437MRibolockThermo FisherEO0381ɣ-32P- ATPHartmann AnalyticSRP-301-504–12% NuPAGE Bis-Tris gelInvitrogenNP0321BOXProteinase KRoche3115887001Phenol:chloroform:isoamyl alcohol (25:24:1)Sigma-AldrichP3803-400MLMyONE Silane beadsThermo Fisher37002DProNex beadsPromegaNG2001RLT bufferQiagen79216RNA Clean and Concentrator-5 kitZymo ResearchR1015Ultra low-range GeneRulerThermo FisherSM1212Page Ruler prestained protein ladderThermo Fisher266160.45 μm filterFisher Brand15216869MG-132Sigma-AldrichM7449PMSFCarl Roth6367.1Plant Protease Inhibitor cocktailSigma-AldrichP9599DTTSigma-AldrichD0632-10GSDSJT Baker4095-02TrisAppliChemA1086.1000HClVWR20252.290NaClVWR chemicals27810.295MgCl2Merck1-05833.1000MOPSFisher BioReagentsBP308-500EDTASigma-Aldrich27285-1KG-RIgepal (NP-40)SigmaI8896-100DMSOCarl Roth4720.1β-MercaptoethanolSigma-AldrichM6250Tween-20SigmaP9416-100ML Software Rotor-Gene 6000 series softwareQIAGENClustalW(Thompson et al, 1994)AlphaFold3(Abramson et al, 2024)R https://www.R-project.org/ FastQC https://www.bioinformatics.babraham.ac.uk/projects/fastqc/ AWK, Mawk 1.3.4Flexbar (3.5.0)(Roehr et al, 2017)STAR(Dobin et al, 2013)umi_tools (1.0.1) https://umi-tools.readthedocs.io/en/latest/index.html samtools (1.14)(Li et al, 2009)PureCLIP (1.3.1)(Krakau et al, 2017)bedtools (2.27.1)(Dale et al, 2011; Quinlan and Hall, 2010)Salmon(Patro et al, 2017)hyperTRIBER(Rennie et al, 2021)R-package eulerr(Larsson and Gustafsson 2018; Larsson, 2022)Tensorflow(Abadi et al, 2016)Keras https://github.com/keras-team/keras R-package universalmotif(Tremblay, 2024a; Tremblay, 2024b)IGV (Integrative Genomics Viewer)(Robinson et al, 2011)R-package DESeq2(Love et al, 2014)R-package tidyverse(Wickham et al, 2019)R-package GenomicRanges(Lawrence et al, 2013)R-package BSgenome https://bioconductor.org/packages/release/bioc/html/BSgenome.html R-package INLA(Rue et al, 2009) https://www.r-inla.org/ Other QIAGEN Rotor-Gene-Q real-time PCR machineQiagenCL-3000 UVP cross-linkerAnalytik JenaTyphoon FLA7000GE HealthcareHiLoad Superdex™ 200 10/300 GL prep grade columnGE HealthcareHPLC ÄKTA Purifier systemGE HealthcareNextSeq sequencerIllumina Reagents and tools tableReagent/resourceReference or sourceIdentifier or catalog number Experimental models Arabidopsis thaliana Col-0N/AN/AE. coli DH5α cellsInvitrogen18265-017E. coli BL21 cellsNEBC2530HAgrobacterium tumefaciens strain GV3101(Koncz and Schell, 1986)N/A Recombinant DNA pCAMBIA3300-U vector(Nour-Eldin et al, 2006)N/ApGEM-T EasyPromegaA137ApGGD003Addgene48835pGGE000Addgene48860pGGF003Addgene48844pGGZ001Addgene48868His6-SUMO pET24Addgene29711pDONR/ZeoThermo Fisher12535035pMDC111(Curtis and Grossniklaus, 2003)N/ApMDC164(Curtis and Grossniklaus, 2003)N/A Antibodies anti-ECT2(Arribas-Hernández et al, 2018)N/Aanti-ALBA1This studyN/Aanti-mCherryAbcamab183628Goat anti-Rabbit IgGSigma-AldrichA6154-1ML Oligonucleotides and other sequence-based reagents PrimersThis studyDataset EV6 Chemicals, enzymes, and other reagents KAPA Hifi Hotstart Uracil+ ReadyMixRocheKK2801Phusion High-Fidelity DNA PolymeraseNEBM0530LBsaI-HFNEBR3733ST4 DNA-LigaseThermo Scientific15224017KOD Hot Start DNA PolymeraseSigma-Aldrich71086Gateway BP Clonase IIThermo Fisher11789020Gateway LR Clonase™ II Plus enzymeThermo Fisher12538120MS-agarPhytoTech LabsM407TryptoneBD211705Yeast extractBD212750AgarBD214010Glufosinate ammoniumSigma45520AmpicillinServa13398.02KanamycinServa26898.03IPTGThermo FisherR0392Ni-NTA agarose beadsmacherey-nagel745400.25Gravity columnBio-Rad7326008ImidazoleThermo ScientificJ17525His6-tagged ULP1 proteaseKind gift from Prof. Birthe Kragelund, UCPH.N/AComplete protease inhibitor EDTA-freeSigma-Aldrich11873580001NuPAGE™ LDS Sample BufferThermo FisherNP00074–20% Criterion™ TGX™ Precast gelBio-Rad5671094Amersham Protran Premium nitrocellulose membraneGE Healthcare Life SciencesGE10600002TRIzolSigmaT9424 - 200 MLRQ1 RNase-Free DNasePromegaM6101RNaseOutTM Recombinant RNase InhibitorInvitrogen10777019QIAgen RNeasy mini kitQiagen74104RNeasy MinElute Cleanup KitQiagen74204SuperScript® III Reverse TranscriptaseInvitrogen18080093RNaseOutTMThermo Fisher10777019SensiFAST SYBR No-ROX MixBiolineBIO-98005RFP-trap agarose beadsChromotekrta-20GFP-trap agarose beadsChromotekgta-20Sepharose 4B beadsSigma-Aldrich4B200T4-polynucleotide kinaseThermo FisherEK0032Turbo DNaseInvitrogenAM2238Ambion RNase IInvitrogenAM2294T4 RNA ligase 1 high concentrationNEBM0437MRibolockThermo FisherEO0381ɣ-32P- ATPHartmann AnalyticSRP-301-504–12% NuPAGE Bis-Tris gelInvitrogenNP0321BOXProteinase KRoche3115887001Phenol:chloroform:isoamyl alcohol (25:24:1)Sigma-AldrichP3803-400MLMyONE Silane beadsThermo Fisher37002DProNex beadsPromegaNG2001RLT bufferQiagen79216RNA Clean and Concentrator-5 kitZymo ResearchR1015Ultra low-range GeneRulerThermo FisherSM1212Page Ruler prestained protein ladderThermo Fisher266160.45 μm filterFisher Brand15216869MG-132Sigma-AldrichM7449PMSFCarl Roth6367.1Plant Protease Inhibitor cocktailSigma-AldrichP9599DTTSigma-AldrichD0632-10GSDSJT Baker4095-02TrisAppliChemA1086.1000HClVWR20252.290NaClVWR chemicals27810.295MgCl2Merck1-05833.1000MOPSFisher BioReagentsBP308-500EDTASigma-Aldrich27285-1KG-RIgepal (NP-40)SigmaI8896-100DMSOCarl Roth4720.1β-MercaptoethanolSigma-AldrichM6250Tween-20SigmaP9416-100ML Software Rotor-Gene 6000 series softwareQIAGENClustalW(Thompson et al, 1994)AlphaFold3(Abramson et al, 2024)R https://www.R-project.org/ FastQC https://www.bioinformatics.babraham.ac.uk/projects/fastqc/ AWK, Mawk 1.3.4Flexbar (3.5.0)(Roehr et al, 2017)STAR(Dobin et al, 2013)umi_tools (1.0.1) https://umi-tools.readthedocs.io/en/latest/index.html samtools (1.14)(Li et al, 2009)PureCLIP (1.3.1)(Krakau et al, 2017)bedtools (2.27.1)(Dale et al, 2011; Quinlan and Hall, 2010)Salmon(Patro et al, 2017)hyperTRIBER(Rennie et al, 2021)R-package eulerr(Larsson and Gustafsson 2018; Larsson, 2022)Tensorflow(Abadi et al, 2016)Keras https://github.com/keras-team/keras R-package universalmotif(Tremblay, 2024a; Tremblay, 2024b)IGV (Integrative Genomics Viewer)(Robinson et al, 2011)R-package DESeq2(Love et al, 2014)R-package tidyverse(Wickham et al, 2019)R-package GenomicRanges(Lawrence et al, 2013)R-package BSgenome https://bioconductor.org/packages/release/bioc/html/BSgenome.html R-package INLA(Rue et al, 2009) https://www.r-inla.org/ Other QIAGEN Rotor-Gene-Q real-time PCR machineQiagenCL-3000 UVP cross-linkerAnalytik JenaTyphoon FLA7000GE HealthcareHiLoad Superdex™ 200 10/300 GL prep grade columnGE HealthcareHPLC ÄKTA Purifier systemGE HealthcareNextSeq sequencerIllumina Reagents and tools table Plant material and growth conditions All lines used in this study are in the Arabidopsis thaliana Col-0 ecotype. The following mutant and transgenic lines mentioned have been previously described: ect2-1 ect3-1 ect4-2 (te234) (Arribas-Hernández et al, 2018), ect2-1 ECT2W464A-mCherry (Arribas-Hernández et al, 2018), ect3-2 ECT3-Venus (Arribas-Hernández et al, 2018), ect2-1 HA-ECT2 (Tankmar et al, 2023). The alba1-1(GABI_560B06), alba1-2 (SALK_069210), alba2-1 (GABI_128D08), alba2-2 (SALKseq_069306), alba3-1 (SAIL_649_E11), alba4-1 (SALK_015940), alba5-1 (SALK_088909), and alba6-1 (SALK_048337) single mutants were obtained from the Arabidopsis Biological Resource Center (ARBC). Seeds were sterilized by immersing them in 70% EtOH for 2 min, followed by incubation in 1.5% NaOCl, and 0.05% Tween-20 for 10 min, after which the seeds were washed twice with H2O. The seeds were then spread on plates containing Murashige & Skoog (MS) medium (4.1 g/l MS salt, 10 g/l sucrose, 8 g/l Bacto agar). The plates were stratified in darkness at 4 °C for 2–5 days before transfer to Aralab incubators at 21 °C, with a light intensity of 120 μmol/m2 and a photoperiod of 16 h light/8 h dark. When needed, after 10 days of growth, seedlings were transferred to soil and kept in Percival incubators under identical settings. Generation of ect2-5 ect3-1 ect4-2 by CRISPR-Cas9 genome engineering For the targeted creation of an in-frame deletion mutant at the endogenous ECT2 locus, we employed the pKIR1.1 CRISPR-Cas9 system (Tsutsui and Higashiyama, 2017). Two plasmids, pKIR1.1-ect2-N8A and pKIR1.1-ect2-N8B, expressing sgRNAs were constructed by ligating oligonucleotides that target ECT2 into pKIR1.1, as described (Tsutsui and Higashiyama, 2017). The crRNAs were designed to yield a deletion resembling ECT2∆N8 as closely as possible. The plasmids were then transformed into ect3-1 ect4-2 mutants, and transformants were selected on MS-agar supplemented with 25 μg/mL hygromycin. After transfer to soil, plants with deletions in ECT2 were identified via PCR using primers spanning the deletion. Progeny from plants with deletions of the expected size, as confirmed by migration in a 1% agarose gel, were plated on MS supplemented with 25 μg/mL hygromycin. Hygromycin-sensitive plants, indicative of the absence of Cas9 and homozygosity of the deletion, were rescued and transferred to MS-agar for recovery. Subsequently, these plants were genotyped and Sanger sequenced for identification of in-frame deletions. Western blotting, utilizing antibodies raised against synthetic peptides in the ECT2 IDR outside the deleted region (Arribas-Hernández et al, 2018), was performed to confirm the in-frame deletion. Primers are listed in Dataset EV6. Construction of transgenic lines To generate the constructs pro(ALBA2):ALBA2-FLAG-TFP:ter(ALBA2), pro(ALBA4):ALBA4-VENUS:ter(ALBA4), pro(ALBA2):ALBA2-FLAG-ADAR:ter(ALBA2), pro(ALBA4):ALBA4-FLAG-ADAR:ter(ALBA4), pro(ECT2):ECT2YAIM-mCherry:ter(ECT2), PCR-amplified DNA fragments were pieced together by USER cloning (Bitinaite and Nichols, 2009) in all cases except for pro(ECT2):ECT2YAIM-mCherry:ter(ECT2) in which an appropriate dsDNA containing the YAIM-mutations was synthesized (Integrated DNA Technologies, gBlocks). As a template for PCR, we used plasmids containing wild-type pro(ECT2):ECT2-mCherry:ter(ECT2) (Arribas-Hernández et al, 2018) for ECT2-mCherry constructs, pro(ECT2):ECT2-FLAG-ADAR:ter(ECT2) for FLAG-ADAR constructs (Arribas-Hernández et al, 2021a), and pro(ECT3):ECT3-VENUS:ter(ECT3) for VENUS constructs (Arribas-Hernández et al, 2018). DNA fragments were amplified using dU-substituted primers and KAPA Hifi Hotstart Uracil+ ReadyMix (Bitinaite and Nichols, 2009). The amplified fragments were inserted into the pCAMBIA3300-U vector, a modified version with a double PacI USER cassette (Nour-Eldin et al, 2006). To clone pro(ALBA1):ALBA1-FLAG-TFP:ter(ALBA1), we made use of Greengate cloning. Briefly, PCR fragments were amplified using Thermo Scientific Phusion High-Fidelity DNA Polymerase (NEB) and ligated into entry vectors through BsaI-restriction cloning. The pro(ALBA1):ALBA1 gDNA fragment was subcloned into pGEM-T Easy by A-tailing (Promega) prior to BsaI-restriction cloning. The vectors containing pro(ALBA1):ALBA1 (in pGEM-T Easy), linker-TFP (pGGD003), ALBA1 3′UTR and downstream sequences (in pGGE000), and the D-AlaR cassette (pGGF003) were combined in a ‘Greengate reaction’ using BsaI-HF (NEB), T4 DNA-Ligase (Thermo Scientific), and pGGZ001 as the destination vector. pro(ALBA2):ALBA2-FLAG-TFP:ter(ALBA2), and pro(ALBA4):ALBA4-FLAG-Venus:ter(ALBA4) fusions were constructed by USER cloning with the primers listed in Dataset EV6. To clone ALBA1-GFP, ALBA2-GFP, ALBA4-GFP, and ALBA5-GFP used for confocal microscopy, we employed Gateway cloning. ALBA gene fragments, including 5′-regions, exons/introns to the gene’s end (excluding the stop codon), were amplified with attB1 and attB2 sites for Gateway cloning using KOD Hot Start DNA Polymerase. Purified amplicons were cloned into pDONR/Zeo via Gateway BP Clonase II (Thermo Fisher) and transformed into E. coli α-select cells. Subsequently, entry clones were recombined with the destination vectors pMDC111 and pMDC164, respectively (Curtis and Grossniklaus, 2003; Earley et al, 2006) via Gateway LR Clonase II (Thermo Fisher) to generate expression clones. All plasmids were verified through restriction digestion and sequencing before being transformed into respective plants using Agrobacterium-mediated floral dip (Clough and Bent, 1998). Primers are listed in Dataset EV6. Screening for te234 complementation Screening of primary transformants (T1s) expressing wild-type, deletion, or point mutant variants of ECT2-mCherry in the te234 background was done as previously described (Tankmar et al, 2023). In brief, primary transformants were selected on MS-agar plates containing glufosinate ammonium (7.5 mg/L (Sigma)) to select plants with the transgene and ampicillin (10 mg/l) to restrict agrobacterial growth. Nine days after germination, primary transformants were categorized according to the size(s) of the first true leaves: full complementation (s ≥ 1 mm), partial complementation (0.5 mm < s < 1 mm), or no complementation (s ≤ 0.5 mm). The complementation percentages were then determined by dividing the number of seedlings in each complementation category by the total number of transformants. Statistical analysis of complementation data Statistical significance of the different T1 complementation categories was determined using Fisher’s exact test, and the Holm–Bonferroni method was applied to address multiple testing. Student’s t-test was used to evaluate the significance of differences in leaf size between Col-0 WT, de34 (ect3-1 ect4-2), te234 (ect2-3 ect3-1 ect4-2), and the CRISPR-generated ect2-5 ect3-1 ect4-2. Analysis of trichome phenotypes Counts of trichomes with different numbers of branches and the statistical analysis of the raw data were done as described (Arribas-Hernández et al, 2018). Western blotting Western blotting was performed as described (Tankmar et al, 2023). Briefly, 100–300 mg of tissue were ground in liquid nitrogen and resuspended in 5 volumes of IP buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 10% glycerol, 5 mM MgCl2, and 0.1% Nonidet P40), supplemented with 1x protease inhibitor (Roche Complete tablets) and 1 mM DTT. The lysate was centrifuged at 13,000×g for 10 min and 4× LDS sample buffer (277.8 mM Tris-HCl pH 6.8, 44.4% (v/v) glycerol, 4.4% LDS, and 0.02% bromophenol blue) was added to a final concentration of 1× LDS. Subsequently, the samples were denatured at 75 °C for 10 min and run on a 4–20% Criterion™ TGX™ Precast gel in 1× Tris-glycine, 0.1% SDS buffer at 90–120 V for ~1 h on ice. The proteins were transferred onto an Amersham Protran Premium nitrocellulose membrane (GE Healthcare Life Sciences) in a cold transfer buffer (1 × Tris-glycine, 20% EtOH) at 80 V for 1 h on ice. The membrane was then blocked in 5% skim milk in PBS-T (137 mM NaCl, 2.7 mM KCI, 10 mM Na2HPO4, 1.8 mM KH2PO4, pH 7.4, 0.05% Tween-20) for 30 min. After blocking, membranes were probed with antibodies specific for ECT2 (Arribas-Hernández et al, 2018), ALBA1 (1:2000, see below), ALBA4 (1:1000, see below), or commercially available antibodies against mCherry (Abcam ab183628, 1:2000 dilution) at 4 °C overnight. Membranes were then washed three times in PBS-T, incubated with HRP-coupled goat anti-rabbit antibody and developed using chemiluminescence detection, as previously described (Arribas-Hernández et al, 2018). RNA extraction and qRT-PCR Total RNA was extracted from frozen and ground plant powder using TRIzol® (1 mL per 500 mg sample). 14 μg RNA was treated with 14 μL of RQ1 RNase-Free DNase (Promega) and 1 μL of RNaseOutTM Recombinant RNase Inhibitor (Invitrogen) following the manufacturer’s protocol. The RNA was then purified using the QIAgen RNeasy mini kit following the RNeasy column clean-up protocol. The RNA quantity and quality were determined via NanoDrop and agarose gel electrophoresis. cDNA was prepared using SuperScript® III Reverse Transcriptase (Invitrogen), with the addition of RNaseOutTM. For qRT-PCR, 0.4 μL 10 μM specific primer pairs (mixture of forward and reverse primers) was mixed with 10 μL SensiFAST SYBR (Bioline) mastermix and 9.6 μL of cDNA. All the qRT-PCR reactions were performed in three technical replicates, carried out on a QIAGEN Rotor-Gene-Q real-time PCR machine and analyzed with the Rotor-Gene 6000 series software (QIAGEN). CYCLOPHILIN (At2g29960) was used for normalization. Primers are listed in Dataset EV6. CLIP-PNK assays of ECT2-mCherry variants Twelve-day-old seedlings were UV-crosslinked with 2000 mJ/cm2 and ground into a fine powder in liquid nitrogen. Immunoprecipitation with RFP-trap beads (Chromotek), washes, DNase and RNase digestion, PNK-labeling, SDS-PAGE, membrane transfer, and autoradiography were performed as described in (Arribas-Hernández et al, 2021a). We used 20 μL of beads for 1 g of tissue in 1.5 mL of iCLIP buffer for every sample. Immunoprecipitation and LC-MS Immunoprecipitations of ECT1-TFP, ECT2-mCherry variants, and ECT3-Venus were performed as described by (Tankmar et al, 2023), while immunoprecipitations of ALBA4-GFP or GFP were performed as described by (Speth et al, 2014). Briefly, 7-day-old seedlings expressing ALBA4-GFP or GFP alone were harvested and ground into fine powder using liquid nitrogen. For each replicated, 0.5 g of ground plant tissue was homogenized in 1.5 mL IP buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 10% glycerol, 0.1% Triton-X100) supplemented with 2% (w/v) PVP40, Roche Complete Protease Inhibitor cocktail (1 tablet/50 mL), 100 μM MG-132, 1 mM PMSF and Sigma Plant Protease Inhibitor cocktail (1/30 v/v). Samples were centrifuged at 16,000 × g for 5 min at 4 °C, the supernatant was transferred to a new tube, and centrifugation was repeated for 10 min. The supernatant was again transferred to a new tube and filtered through a 0.45-μm filter. For Co-IP, 1 mL of cell extract at a concentration of 2 μg/μL was first added to 50 μL of sepharose beads for pre-clearing and incubated for 30 min at 4 °C with constant rotation. After centrifugation at 1000 × g for 1.5 min at 4 °C, the cell extract was added to 20 μL GFP-Trap beads and incubated for 2.5 h at 4 °C with constant rotation. The beads were washed 4x in Co-IP wash buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 10% glycerol, 0.05% Triton-X100, Roche Complete Protease Inhibitor cocktail (1 tablet/50 mL)) and proteins were eluted by addition of 40 μL 2x LDS sample buffer to the beads and incubation at 70 °C for 10 min. For control samples treated with nucleases, beads were washed once in Co-IP wash buffer (+10 mM MgCl2) after the IP. Beads were then resuspended in 100 μL Co-IP wash buffer (+10 mM MgCl2) and treated with 2 μL Turbo DNase (Thermo Fisher Scientific) and, optionally, 5 μL of a 1:50 dilution of RNase I (Ambion) for 10 min at 37 °C and 1200 rpm. Beads were then washed three times with Co-IP wash buffer, and elution was performed as described above. Mass spectrometry data was analysed as in (Tankmar et al, 2023). Protein expression of ALBA1 An ALBA1 (AT1g29250) cDNA was amplified from oligo(dT)-primed reverse transcription products of DNase-treated total RNA from Col-0 wild type using the primer set MT303-MT304. The resulting PCR product was ligated in a frame downstream of His6-SUMO in a pET24-derived vector containing His6-SUMO (Twist Bioscience). For recombinant protein expression, the plasmid encoding His-SUMO-ALBA1 was transformed into E. coli BL21 (DE3 7tRNA) codon plus. Cells were grown at 37 °C in LB medium supplemented with 35 μg/ml kanamycin, and expression was induced at OD600 ≈ 0.6 by the addition of 0.5 mM IPTG. Following induction, the cells were grown at 18 °C overnight and harvested by centrifugation. The cell pellet was resuspended in 20 mM Tris-HCl (pH 8), 10 mM imidazole, and 300 mM NaCl supplemented with 1 mM DTT and EDTA-free protease inhibitor (cOmplete; Roche). Cells were lysed once using a French press (20,000 psi). Crude lysate was cleared by centrifugation at 30,000 × g for 30 min at 4 °C and filtered through a 0.45-μm membrane. His-SUMO-ALBA1 was purified on Ni2+-NTA resin by incubation for 1 h at 4 °C, after which the beads were washed in wash buffer (20 mM Tris-HCl pH 8, 20 mM imidazole, 200 mM NaCl), and the bound protein was eluted in elution buffer (300 mM imidazole, 20 mM Tris-HCl pH 8, 300 mM NaCl). The eluted protein was dialyzed overnight into 20 mM Tris-HCl pH 8, 200 mM NaCl, 1 mM 2-mercaptoethanol followed by cleavage after the His6-SUMO tag with heterologously expressed His6-tagged ULP1 protease, a kind gift from Birthe Kragelund. Ni2+-NTA resin was used to bind the protease and impurities bound to the Ni2+-NTA resin in the first affinity purification, and ALBA1 was collected in the flowthrough. ALBA1 was further purified on a HiLoad Superdex™ 200 10/300 GL prep grade column (GE Healthcare) connected to an HPLC ÄKTA Purifier system (GE Healthcare). Eluates were monitored at A280, and purity was assessed by SDS-page analysis. Development of ALBA1 and ALBA4 antibodies The anti-ALBA1 and anti-ALBA4 antibodies were affinity-purified by Eurogentec from serum collected from rabbits immunized with recombinant ALBA1 protein or a 1:1 mix of the KLH-coupled ALBA4 peptides H-CGFNNRSDGPPVQAAA-OH and H-CNGPPNEYDAPQDGGY-NH2 (Eurogentech). The ALBA4 peptides were synthesized by Schafer-N Aps, Copenhagen, Denmark. Protein alignment and logo representation All YTHDF protein sequences from 36 plant species, or selected DF-A, -B, -C, -D, and -E protein sequences from 34 plant species (Dataset EV7) were aligned using MUSCLE (Madeira et al, 2024). The resulting alignments were trimmed from both ends using Jalview (Waterhouse et al, 2009), to leave the region comprising the ECT2 N8 motif. Short and under-represented amino acid insertions not contained in ECT2 were also removed for clarity (positions marked as blue arrowheads in Appendix Fig. S3). To build the N8 motif logo, we used the online tool Weblogo (Crooks et al, 2004). Structural modeling using AlphaFold3 The structural model of the ECT2-(ALBA5)2-RNA complex was generated by AlphaFold3 (Abramson et al, 2024) using default settings and the following sequence input: one molecule of ECT2 (gene model AT3g13460.1, amino acid residues 373–616), two molecules of ALBA5 (gene model AT1g20220.1, amino acid residues 18–114), one molecule of RNA (5′-AAA[m6A]CUUCUG-3′). ALBA4-GFP iCLIP experiments and library preparation iCLIP experiments were carried out based on the method previously employed for Arabidopsis GRP7-GFP (Meyer et al, 2017) and the optimized iCLIP2 protocol (Buchbender et al, 2020; Lewinski et al, 2024). Briefly, 7-day-old seedlings expressing ALBA4-GFP or GFP alone grown at 20 °C in LD (16 h light, 8 h dark) were crosslinked with 254 nm UV light at 2000 mJ/cm2, snap frozen, and ground into a fine powder in liquid nitrogen, and homogenized in iCLIP lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 4 mM MgCl2, 5 mM DTT, 1% SDS, 0.25% sodium deoxycholate, 0.25% Igepal) supplemented with Roche Complete Protease Inhibitor cocktail (1 tablet/50 mL). The lysate was cleared by centrifugation and filtration (0.45-μm pore) of the supernatant. After pre-clearing with 200 μL of sepharose beads for 1 h at 4 °C, RNP complexes were immunopurified with GFP-Trap beads (ChromoTek) for 4 h at 4 °C under constant rotation. We used 50 μL of beads for 3 g of tissue in 5 mL of iCLIP lysis buffer for every replicate. After washing four times with iCLIP wash buffer (2 M urea, 50 mM Tris-HCl pH 7.5, 500 mM NaCl, 4 mM MgCl2, 2 mM DTT, 1% SDS, 0.5% sodium deoxycholate, 0.5% Igepal, supplemented with Roche Complete Protease Inhibitor cocktail (1 tablet/50 mL)), and twice with PNK wash buffer (20 mM Tris-HCl, pH 7.4, 10 mM MgCl2, 0.2% Tween-20), RNP complexes attached to the beads were subjected to treatment with DNase (Turbo DNase [Ambion], 4 U/100 μL) and optionally RNase I (Ambion, 1 U/mL) at 37 °C for 10 min. Subsequently, RNA 3′-ends were dephosphorylated (PNK [Thermo Fisher] in buffer containing 350 mM Tris-HCl pH 6.5, 50 mM MgCl2, 25 mM DTT) for 20 min at 37 °C, followed by one wash with PNK wash buffer, one wash with high-salt buffer (50 mM Tris-HCl pH 7.4, 1 M NaCl, 1 mM EDTA, 1% Igepal, 0.1% SDS, 0.5% sodium deoxycholate) and two more washes with PNK wash buffer. The L3 linker was then ligated to the 3′-RNA ends (with NEB HC RNA Ligase in ligation buffer (200 mM Tris-HCl pH 7.8, 40 mM MgCl2, 40 mM DTT with RiboLock and PEG8000) at 16 °C and 1250 rpm for >16 h. Samples were then washed twice in high-salt buffer and once in PNK wash buffer before the RNA was radioactively labeled at the 5′-end by PNK-mediated phosphorylation using ɣ-32P- ATP (20 min at 37 °C). The labeled RNP complexes were subjected to SDS-PAGE (4–12% NuPAGE Bis-Tris gel with 1x MOPS buffer) and blotting on a nitrocellulose membrane (Protran BA-85). Pieces of membrane containing a size range of RNA species bound to the protein (a smear above the expected molecular weight localized by autoradiography) were excised and subjected to proteolysis (200 μg of Proteinase K [Roche] in 200 μL of PK buffer [100 mM Tris-HCl pH 7.4, 50 mM NaCl, 10 mM EDTA] for 20 min at 37 °C) to release RNA bound to small peptides. The RNA was then purified using phenol-chloroform (pH 7.0) and ethanol precipitation and used to prepare sequencing libraries following the iCLIP2 protocol (Buchbender et al, 2020): reverse transcription with SSIII (Invitrogen) and an RT oligo complementary to the L3 liker followed by RNA hydrolysis and cDNA clean-up with MyONE Silane beads (Thermo Fisher). A second adapter was then ligated to the 3’OH of the cDNAs (with NEB HC RNA Ligase in NEB ligation buffer plus 5% DMSO, 1 mM ATP, and 22.5% PEG8000) at 20 °C and 1250 rpm overnight. The adapter contains a bipartite unique molecular identifier (UMI) and an experimental barcode, allowing for PCR duplicate removal and sample multiplexing, respectively. After another MyONE Silane clean-up, the cDNA library was pre-amplified in a first PCR (6 cycles), followed by size selection with ProNex beads (Promega) to remove short cDNAs and primer dimers. The cDNA library was then amplified in a second PCR, followed by a second ProNex size selection to remove PCR primers and finally prepare the cDNA library for sequencing. The second PCR was carried out with 10 μL of cDNA and eight cycles for each replicate. Samples were multiplexed and sequenced in the NextSeq sequencer (NextSeq® 500/550 Mid Output Kit v2 (150 cycles)) at the Genomics Core Facility at IMB (Mainz, Germany). ALBA4-GFP iCLIP analysis All reads from iCLIP experiments were quality-checked after multiple processing steps with FastQC (0.11.9). The distribution of read counts assigned to sample barcodes was computed using awk (GNU awk 5.0.1). Reads were demultiplexed, sequencing adapters removed from 3′ ends and subsequently quality- as well as length-trimmed (--min-read-length 15 -q WIN -qf sanger –min-read-length 15) with Flexbar (3.5.0) while keeping the random UMI parts in the read id field (--umi-tags). A genome index was created using STAR (2.7.3a) using the Arabidopsis thaliana genome version TAIR10. The genome annotation from Araport (version 11) was specified to mark the location of splice junctions. Quality trimmed reads were then mapped using STAR and the created genome index, allowing only softclipping of 3′ ends (--alignEndsType Extend5pOfRead1) to preserve the position of the crosslinked nucleotide. PCR duplicates were removed using umi_tools (1.0.1) by considering the UMI tag in the read id field and the mapping coordinates. The uniquely mapped and deduplicated reads from each ALBA4-GFP and GFP replicate were merged together using samtools (1.14) and peak called with PureCLIP (1.3.1) in standard mode (-bc 0) to identify short and defined peak coordinates. In order to learn the HMM parameters, only the first two chromosomes were specified (-iv ‘Chr1;Chr2’), and the precision to store probabilities was set to long double (-ld). Clusters of directly adjacent called peaks were merged and reduced to the position with the highest reported PureCLIP score (1-nt resolution). Binding sites were defined as called peaks, extended by 4 nt (−4…0… + 4) in both directions with bedtools (2.27.1). Sites which reported crosslinks in only 1 out of 9 position were removed as they are considered artifacts. To confirm that the binding sites are supported by at least 2 replicates and a sufficient number of reads (reproducible binding sites), the coordinates of binding sites were overlapped with crosslink positions from every replicate (ALBA4-GFP and GFP independently). The distribution of crosslinks per binding sites was used to determine a reproducibility threshold. After defining a distribution quantile of 30% as the minimal filtering threshold, only binding sites above this threshold in at least two out of three replicates were kept. Due to the low amount of uniquely mapped reads the GFP control was not tested for reproducibility. Reproducible binding sites of ALBA4-GFP overlapping with binding sites from the GFP control were removed using bedtools and reported in browser extensible data (BED) format. Targets of ALBA4-GFP were defined as transcripts overlapping reproducible binding sites. Only the locations of representative gene models from Araport (version 11) were considered. For visual inspection data tracks were generated from uniquely mapped ALBA4-GFP and GFP-only reads using bedtools. Sample preparation for TRIBE and HyperTRIBE RNA extraction and library preparation was performed as previously described(Arribas-Hernández et al, 2021a). Total RNA was extracted from manually dissected root tips for ALBA4-FLAG-ADAR and apices (removing cotyledons) for ALBA2-FLAG-ADAR and ECT2-FLAG-ADAR of five independent lines (10-day-old T2 seedlings) with each of the lines being used as biological replicate. TRIBE/HyperTRIBE analyses for ALBA2 and ALBA4 vs. free ADAR controls For all TRIBE/HyperTRIBE experiments, reads were mapped to the TAIR10 genome using STAR (Dobin et al, 2013) (version 2.7.11) and transcripts quantified using Salmon (Patro et al, 2017) based on the Araport11 transcriptome (Cheng et al, 2017) augmented with the DNA sequence for the ADAR clone. The hyperTRIBER pipeline (Rennie et al, 2021) was employed in order to quantify all positions with at least one mismatch to the genome, filter candidate positions by mutation type (A-to-G or T-to-C for forward or reverse strands, respectively), and replicate agreement, and formally test these candidates using a generalized linear model-based approach for assessing the difference in editing proportions between free ADAR control samples vs. fusion samples, retaining positions with a log2FC >1, an adjusted p value <0.01 and a minimum editing proportion of 0.01. All sets were further annotated using the hyperTRIBER pipeline based on Araport11 gene annotations and prioritizing highly expressed transcripts in the control lines in the case of positions overlapping multiple transcripts. HyperTRIBE analysis for ECT2 in the alba1245 background and ALBA2 in the Gte234 background Unequal levels of ECT2-FLAG-ADAR or ALBA2-FLAG-ADAR expression between different genetic backgrounds in the same HyperTRIBE experiment could result in misinterpretation of results due to biased ADAR-driven editing patterns. This was supported by inspection of the initial results from the hyperTRIBER pipeline (Rennie et al, 2021) when comparing ECT2-FLAG-ADAR-expressing plants in the Col-0 vs alba1245 backgrounds. This preliminary analysis showed stronger editing in the direction of the samples with higher average ADAR expression, supported by western blots. To investigate further, we first re-ran the pipeline on only four lines (two per genetic background), selected such that the average number of reads mapping to ADAR was approximately equal between the two genetic backgrounds. Compared to the naïve analysis of all five lines per genotype, the significantly differently edited sites were visually less biased in the direction of high ADAR expression, indicating that unequal ADAR expression leads to spurious results if left uncorrected. Furthermore, we observed a pattern whereby sites on lowly expressed genes tended to exhibit a larger editing proportion. To robustly account for differences in ADAR expression as measured by mRNA-seq read counts, we formulated a Bayesian hierarchical model as follows. First, we split the samples into three groups according to the expression of the ADAR clone (ADAR_BIN) and binned expression levels into five groups (EXPR_BIN). Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Y}_{{ijkc}}$$\end{document}Yijkc denote the observed count of base G at the i-th position, with the j-th level of ADAR_BIN, the k-th level of EXPR_BIN, and under condition c. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Y}_{{ijkc}}$$\end{document}Yijkc is assumed to follow a Binomial distribution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Y}_{{ijkc}}\sim {\mbox{Binomial}}({n}_{{ijkc}},{p}_{{ijkc}})$$\end{document}Yijkc~Binomial(nijkc,pijkc) where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{{ijkc}}$$\end{document}nijkc represents the number of trials for each combination of position, ADAR_BIN level, EXPR_BIN level, and condition, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{{ijkc}}$$\end{document}pijkc is the probability of observing base G. Then the logit of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{{ijkc}}$$\end{document}pijkc is modeled as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\log \left(\frac{{p}_{{ijkc}}}{1-{p}_{{ijkc}}}\right)={{{{\rm{\eta }}}}}_{{ijkc}}$$\end{document}logpijkc1−pijkc=ηijkc where the linear predictor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\eta }}}}}_{{ijkc}}$$\end{document}ηijkc is given by:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\eta }}}}}_{{ijkc}}={{{{\rm{\beta }}}}}_{0}+{{{{\rm{\beta }}}}}_{j}+{{{{\rm{\gamma }}}}}_{k}+{{{{\rm{\delta }}}}}_{{jk}}+{u}_{{ic}}$$\end{document}ηijkc=β0+βj+γk+δjk+uicwhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\beta }}}}}_{0}$$\end{document}β0 is the intercept, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\beta }}}}}_{j}$$\end{document}βj is the effect of ADAR bin j, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\gamma }}}}}_{k}$$\end{document}γk is the effect of expression bin k, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\delta }}}}}_{{jk}}$$\end{document}δjk is the corresponding ADAR expression interaction and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${u}_{{ic}}\sim N\left(0,{{{{\rm{\tau }}}}}_{c}^{-1}\right)$$\end{document}uic~N0,τc−1 is a position-specific random effect with condition-specific precision parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\tau }}}}}_{c}$$\end{document}τc. The model was fit using the Integrated Nested Latent Laplace (INLA) framework. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${u}_{{iA}}$$\end{document}uiA and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${u}_{{iB}}$$\end{document}uiB denote the random effects for position i under conditions A and B, then the linear combination is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mbox{LC}}}_{i}={u}_{{iA}}-{u}_{{iB}}$$\end{document}LCi=uiA−uiB was computed from the posterior distribution of the fitted model. The mean \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\mu }}}}}_{{{\mbox{LC}}}_{i}}$$\end{document}μLCi and standard deviation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\rm{\sigma }}}}}_{{{\mbox{LC}}}_{i}}$$\end{document}σLCi of samplings from the fitted posterior were used to generate Z-scores \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Z}_{i}=\frac{{{{{\rm{\mu }}}}}_{{{\mbox{LC}}}_{i}}}{{{{{\rm{\sigma }}}}}_{{{\mbox{LC}}}_{i}}}$$\end{document}Zi=μLCiσLCi which were converted into p values and subsequently adjusted to a false discovery rate. Importantly, the list of significant genes from this analysis strongly overlapped with the smaller list of genes from the two-sample analysis described above (Fig. EV4F,G). Finally, position-specific corrected editing proportions from the fitted model were further estimated by assuming ADAR to be exactly to the center bin and used for producing scatter plots for all tested positions. Definitions of strict and permissive gene sets Strict sets: ALBA4, the intersection of iCLIP (strong) and ALBA4-TRIBE-associated gene sets. ECT2, intersection between ECT2/3 HyperTRIBE and ECT2 iCLIP (110 KDa) target sets (Arribas-Hernández et al, 2021b). Permissive sets: union instead of intersection between the above sets for ALBA4 and ECT2, respectively. Definition of matched background sets For sites of interest (iCLIP crosslink sites, m6A site positions, etc.), enrichment of surrounding features (motifs, sites of interest) was assessed by comparing the same features around matched background sets. These sets were calculated similarly to our previous publication (Arribas-Hernández et al, 2021a), where, briefly, each site of interest was paired with a site of similar proportion along the relevant transcript feature (5′-UTR, CDS, or 3′-UTR). To avoid expression bias, the pool of background genes was limited to those of the set of interest, or alternatively based on a set of genes selected to have a similar expression distribution. Venn diagrams and significance of overlaps Venn diagrams were generated using custom code and the R-package eulerr (https://CRAN.R-project.org/package=eulerr) (Larsson and Gustafsson 2018; Larsson, 2022). To assess the significance of overlaps between two sets of genes, a random set of genes of size equal to the number of genes in the first set was selected. To avoid expression bias—due to random genes being, on average, more lowly expressed than the sets of interest—the expression distribution of the random set was matched to that of the first set. We calculated the number of genes in the first set overlapping with the second set, as well as the number of genes for each of the 1000 random samples overlapping with the second set. The p value was calculated as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p-{\mbox{value}}=1-2\times \left(\frac{\left|{hsum}-\frac{1000}{2}\right|}{1000}\right)$$\end{document}p−value=1−2×hsum−100021000 where hsum is the number of cases where the number of genes in the random set overlapped more with the second set. In cases where there were zero instances where the random set had a better overlap with the second set than the first set, the p value was set to “<0.001”, indicating a high significance of overlap. This procedure was carried out using a custom script, which also returned a single random set of expression distribution matched genes. This random set was used in the Venn diagrams to provide a visual indication of the expected overlap by chance. To check for possible false positives in genes with fewer than 2–5 iCLIP sites, we overlapped the set with the ALBA4 HyperTRIBE data and looked for the percentage of support. We noted that genes with only a single, low-quality iCLIP site tended to be supported by ALBA4 HyperTRIBE to a similar level as random sets of expressed genes, providing justification for considering the more robust set for subsequent analyses. Metagene plots Metagene plots showing enrichment of features in 5′-UTR, CDS and 3′-UTR corrected for the size of the annotated region were generated using a strategy similar to what we previously reported (Arribas-Hernández et al, 2021a). Single-cell co-expression analysis of ECT2 We first obtained single-cell mRNA-seq root tip data (He et al, 2023; Shahan et al, 2022). To avoid bias due to differences in UMI count between ECT2-expressing (ECT2+) cells and non-expressing (ECT2-) cells, each ECT2+ cell was matched with an ECT2- cell of similar UMI count. For each gene G expressed within the range of 20–80% of the resulting total cells, counts of G+ and G− cells for each of the ECT2+ and ECT2- sets were used to perform a Fisher’s exact test, whereby a high odds ratio represents a high corresponding between ECT2 and the tested G, indicative of co-expression. Motif analysis We first considered the set of motifs previously defined on the basis of ECT2 iCLIP data (Arribas-Hernández et al, 2021a). Background sites for m6A (nanopore-derived (Parker et al, 2020)), ALBA4 iCLIP, and ECT2 iCLIP were generated following a similar strategy to what we previously reported (Arribas-Hernández et al, 2021a), ensuring that the distribution of site locations across gene features were identical for both the true set and the background set. We subsequently removed background sites which, by chance, overlapped with sites from the true sets (within 100 bp). For both the true sets and background sites, we calculated the number of motifs present per 1000 sites (a normalization allowing for comparability across different sets), for each position up to 100 bp from the site. Curation of m6A site set We first collected m6A sites for A. thaliana from multiple published sources (Parker et al, 2020; Tang et al, 2020). As nanopore-derived sites are not subjected to UV-bias, we trained a neural network to differentiate between the 20,858 m6A sites identified by nanopore (Parker et al, 2020) and a corresponding set of 20,715 location-matched negative sites (Fig. EV3A). The neural network used as input extracted sequence ±100 bp regions around all positions (R-packages BSgenome (Pagès, 2024) and AThaliana), which was converted from FASTA to one-hot encoded format (Pedregosa et al, 2011). As output, the network predicted the presence or absence of the m6A at the center point of the input sequence. The network was based on four blocks of 1D convolutional layer with ReLu activation, batch normalization, and max pooling of size 2. The output of these four blocks was flattened, run through a fully connected layer, and then passed into a fully connected output layer of output size 1 with a sigmoid activation function. The model was trained specifying the binary cross entropy loss function using Keras with a Tensorflow backend (Abadi et al, 2016) specifying binary cross entropy loss function. The model showed excellent performance, with AUC ranging from 0.85 to 0.92 over the five folds. This model enabled us to fine-adjust sites from other sets by systematically shifting their positions and selecting those with the highest probability (Fig. EV3B). Consequently, we augmented the smaller set of nanopore-derived positions with a broader set exhibiting properties highly consistent with nanopore-identified sites (Fig. EV3C–F). Notably, ~90,000 miCLIP-derived positions not only shifted to locations similar to nearby nanopore-defined sites, but also consolidated into fewer positions, indicating that many miCLIP-identified sites represent imprecise locations. Overall, our augmentation strategy yielded a compendium of 41,883 m6A sites in A. thaliana. Convolutional neural network-based de novo motif detection We annotated each of 41,883 m6A sites as bound or unbound according to an overlap of either ECT2 iCLIP or ALBA iCLIP sites within 100 nt. For each task (task 1: predicting the presence of ECT2 binding, task 2: predicting the presence of ALBA4 binding), we defined the positives as m6A sites bound by the relevant protein and the negatives as unbound m6A sites. For each m6A site, 300 nt of sequence was extracted on either side, creating 601 nt long sequences, which were embedded using one-hot encoding and passed as input in a convolutional neural network with two outputs—(1) presence or absence of ECT2 iCLIP and (2) presence or absence of ALBA iCLIP. The network architecture consisted of five blocks of 1D convolutional layers with RELU activation, a 0.2 drop-out layer, a batch normalization layer, and a max-pooling layer with pool size 2. Each convolutional layer had 64 filters, with a kernel size of 8 in the first layer and 6 thereafter. The output was then flattened into one dimension and passed through a separate connected layer of kernel 32 for each output, which was specified as a fully connected layer of size 1 using a sigmoid activation function. The network was trained using Keras with a Tensorflow backend (Abadi et al, 2016), specifying the binary cross entropy loss function for each output. Fivefold cross-validation strategy for machine learning models Sites were split into five sets of similar size. Since there are often multiple m6A sites on a single gene, and these sites often fall within overlapping windows, we separated training and test sets such that no gene was present in both sets. This has the effect of ensuring that sequences were fully separated between training and testing sets, so that performance on the testing sets could be fairly evaluated without being inflated by possible overtraining. Each testing set consisted of one of the five sets, and the training set the remaining sets combined. All predictions used in subsequent analyses were based only on sets held out of the training process. Modeling of RBP-specific motifs After fitting each fold weights for the 32 learned convolutional filters of length 8 from the initial layer (that is, the layer connecting to the one-hot encoded input sequence) were extracted, resulting in a total of 160 filters. For each of these filters individually, we scanned through all sequences from the training set, selected the top 5000 high-scoring positions, and used the resulting nucleotide frequencies at each of the 8 positions to derive a position weight matrix. These position weight matrices were then allocated a consensus name using the R-package universalmotif (Tremblay, 2024). For each motif, m6A-centered sequences were classified as containing or not containing the given motif within 150 bp of the methylation site. To detect RBP-specific binding motifs, a generalized linear model (glm) assuming a binomial-distributed response (logistic regression) was used to predict motif presence as the dependent variable, where the two predictors in the model were the probability of ECT2 binding from the neural network and the probability of ALBA4 binding from the neural network. In this way, the coefficient for ECT2 binding is interpreted as the strength of correspondence with that motif whilst controlling for binding of ALBA4, and vice-versa. Z-scores for each of the two proteins for all motifs were then extracted from the model and plotted as enrichment scores. Supplementary information Appendix Peer Review File Dataset EV1 Dataset EV2 Dataset EV3 Dataset EV4 Dataset EV5 Dataset EV6 Dataset EV7 Source data Fig. 1 Source data Fig. 2 Source data Fig. 3 Source data Fig. 4 Source data Fig. 7 EV and Appendix Figure Source Data Expanded View Figures Appendix Peer Review File Dataset EV1 Dataset EV2 Dataset EV3 Dataset EV4 Dataset EV5 Dataset EV6 Dataset EV7 Source data Fig. 1 Source data Fig. 2 Source data Fig. 3 Source data Fig. 4 Source data Fig. 7 EV and Appendix Figure Source Data Expanded View Figures