Zipcode RNA-Binding Proteins and Membrane Traf ﬁ cking Proteins Cooperate to Transport Glutelin mRNAs in Rice Endosperm

In rice ( Oryza sativa ) endosperm cells, mRNAs encoding glutelin and prolamine are translated on distinct cortical-endoplasmic reticulum (ER) subdomains (the cisternal-ER and protein body-ER), a process that facilitates targeting of their proteins to different endomembrane compartments. Although the cis - and trans -factors responsible for mRNA localization have been de ﬁ ned over the years, how these mRNAs are transported to the cortical ER has yet to be resolved. Here, we show that the two interacting glutelin zipcode RNA binding proteins (RBPs), RBP-P and RBP-L, form a quaternary complex with the membrane fusion factors N -ethylmaleimide-sensitive factor (NSF) and the small GTPase Rab5a, enabling mRNA transport on endosomes. Direct interaction of RBP-L with Rab5a, between NSF and RBP-P, and between NSF and Rab5a, were established. Biochemical and microscopic analyses con ﬁ rmed the co-localization of these RBPs with NSF on Rab5a-positive endosomes that carry glutelin mRNAs. Analysis of a loss-of-function rab5a mutant showed that glutelin mRNA and the quaternary complex were mis-targeted to the extracellular paramural body structure formed by aborted endosomal traf ﬁ cking, further con ﬁ rming the involvement of endosomal traf ﬁ cking in glutelin mRNA transport. Overall, these ﬁ ndings demonstrate that mRNA localization in plants co-opts membrane traf ﬁ cking via the acquisition of new functional binding properties between RBPs and two essential membrane traf ﬁ cking factors, thus de ﬁ ning an endosomal anchoring mechanism in mRNA localization. The cDNA sequences of RBP-P, RBP-L, NSF, Rab5a, and its mutant forms (Ra-b5a Q70L and Rab5a G45D ) were cloned into pSAT1-nEYFP-C1 or pSAT1-cEYFP-C1-B vectors for BiFC analysis. GFP- or RFP-fusion vector driven by double 35S promoters was constructed by replacing the N-terminal EYFP in pSAT1-nEYFP-C1 with GFP cDNA or RFP cDNA sequences. cDNAsequencesofNSFandRab5awerethenclonedintoaGFP-orRFP-fusion vector protein fusions. culture at 26°C for the BiFC ﬂ uorescence using a model no. confocal microscope (Leica). Negative controls using empty procedure.The localization of target proteins or complexes examiningatleast ﬁ vedifferentprotoplastcells.Tocon ﬁ rmtheinvolvement of Rab5a and the corresponding complexes in endocytic pathway, pro- toplastincubationwastreatedwithFM4-64endocytictracer(Invitrogen)at a ﬁ nal concentration of 10 m M for 15 to 30 min before


INTRODUCTION
Localization of mRNAs is a universal mechanism to efficiently drive protein targeting in eukaryotes and prokaryotes. The targeting of mRNAs facilitates the accumulation of the locally translated proteins to specific cellular compartments and, hence, is an essential mechanism in establishing cell polarity, patterning, and fate determination as well as protein sorting (Herbert and Costa, 2019;Hughes and Simmonds, 2019;Tian et al., 2019bTian et al., , 2020 mRNA localization occurs as a multistep process. After transcription, cis-acting elements (RNA zipcodes) are recognized and bound by trans-acting factors, mainly RNA binding proteins (RBPs) to form a primary mRNA-nucleoprotein (mRNP) complex. After export to the cytoplasm, the mRNP complex undergoes extensive remodeling with recruitment of new factors and detachment of others enabling cytoskeletal-based transport to the destination site (Blower, 2013;Weis et al., 2013;Tian and Okita, 2014).
Although extensive knowledge on mRNA localization has been acquired by studies in Drosophila melanogaster, yeast (Saccharomyces cerevisiae), and mammalian cells, only a few examples have emerged from higher plants. The best-defined model in plants is storage protein mRNA localization in developing rice (Oryza sativa) endosperm cells, where mRNAs encoding glutelin and prolamine are recognized by zipcode RBPs and transported to two distinct cortical endoplasmic reticulum (ER) subdomains, the cisternal-ER, and protein body-ER (PB-ER), respectively (Chou et al., 2019;Tian et al., 2019b). Translation of prolamine mRNAs on the PB-ER results in the assembly of prolamine intracisternal granules that form an ERderived protein body I (PB-I), while glutelin precursors are exported to the Golgi and then transported to protein storage vacuoles (PSVs) for processing and storage (Chou et al., 2019;Tian et al., 2019b). Although several cytoskeleton-associated RBPs required for mRNA localization have been identified (Doroshenk et al., 2009, information on how these mRNAs are transported to distinct ER subdomains remains elusive. Emerging evidence from fungal model systems reveals the intimate link of mRNA transport with membrane trafficking (Schmid et al., 2006;Jansen et al., 2014;Haag et al., 2015;Niessing et al., 2018). Several mRNAs from yeast, Candida albicans, and Ustilago maydis are co-transported with mobile ER or shuttling endosomes (Schmid et al., 2006;Jansen et al., 2014;Haag et al., 2015;Pohlmann et al., 2015;Niessing et al., 2018). ASH1 as well as other mRNAs are co-transported on tubular ER that moves to the emerging bud or daughter cell in yeast. This process is mediated by the RBPs She2p and She3p, with She2p having membrane binding properties and She3p serving as an adaptor protein linking the mRNP-cER to Myo4P protein (Schmid et al., 2006;Niessing et al., 2018). The cdc3 mRNA is transported on shuttling endosomes in the smut fungus, U. maydis, a process requiring localization of the RBP Rrm4 on the endosomes and the interaction of a membrane-associated linker protein Upa1 with Rrm 4 (Pohlmann et al., 2015;Niessing et al., 2018). Specific adaptor proteins appear to be needed to hitch mRNPs on endosomes for active transport over long distance. More recently, neuronal RNA granules have been shown to hitchhike on moving lysosomes using annexin11 as a tether (Liao et al., 2019). Although co-transport of mRNAs with membranous compartments was proposed to be a common mechanism in higher eukaryotes (Jansen et al., 2014), whether the mechanism is utilized by higher plants remains to be determined.
Previous studies suggested that endocytosis and membrane trafficking likely play a role in mRNA localization in plants. For example, the loss of function of the small GTPase Rab5a and its cognate guanine nucleotide exchange factor (GEF) resulted in defects in endocytosis and membrane trafficking and the mistargeting of glutelin proteins to the prolamine containing PB-I as well as to the extracellular paramural body (PMB) in rice endosperm cells (Fukuda et al., 2011;Wen et al., 2015). As storage protein targeting is regulated by their mRNA localization in rice endosperm cells, the mis-targeting of glutelin proteins in the mutant suggests a relationship between endosomal transport and glutelin mRNA localization in rice. The extracellular distribution of glutelin mRNAs within PMBs from a mutant expressing a defective GEF (Yang et al., 2018) further supports the possible involvement of endosomal trafficking in glutelin mRNA transport. However, direct evidence depicting the co-transport of glutelin mRNAs with shuttling endosomes and how endosomal trafficking are engaged in glutelin mRNA localization have yet to be established. Such mistargeting of glutelin mRNAs in rice lines expressing mutant Rab5a and GEF may simply be a consequence of pleiotropy.
Recent studies (Tian et al., , 2019a identified two RBPs, RBP-P and RBP-L, which contain two and three RNA recognition motif (RRM) domains, respectively. These RBPs specifically bind to the glutelin zipcode mRNA sequences and regulate glutelin mRNA localization. In this study, using these two glutelin zipcode RBPs as entry points, we identified their interacting partners, N-ethylmaleimide-sensitive factor (NSF) and the small GTPase Rab5a, which participate in endosomal membrane trafficking. The four proteins may form a quaternary complex carrying glutelin mRNAs for active transport on endosomes to the cortical ER membrane. The identification of these key linker proteins that enable endosome-mediated mRNA transport in rice endosperm cells provides new insights on how mRNAs can be distributed to specific locations in eukaryotes.

RBP-P Interacts with Membrane Fusion Factor NSF
Previous studies (Doroshenk et al., 2014;Tian et al., 2018Tian et al., , 2019a established that the glutelin zipcode RBPs RBP-P and RBP-L, which interact with each other, are essential for localization of glutelin mRNAs to the cisternal-ER, as mutations in these RBPs led to the mis-localization of glutelin mRNAs. To obtain additional insight into how glutelin mRNAs are transported to the ER, we performed immunoprecipitation-mass spectrometry (IP-MS) studies using affinity-purified anti-RBP-P antibody ( Figure 1A). Tandem mass spectrometry of a major polypeptide band observed in the IP generated with anti-RBP-P but not with control anti-GFP identified a major interacting protein as NSF.
To determine whether RBP-P interacts directly with NSF, yeast two-hybrid (Y2H) studies were performed. In such two-hybrid screening, the two proteins of interest were fused to the activating domain and the DNA binding domain of yeast GAL4 transcription factor, respectively. Interaction of these proteins restore GAL4 that, in turn, activate the transcription of His and Ade reporter genes. While no background interaction of RBP-P and NSF with a complementary empty vector was observed, yeast cells carrying both RBP-P and NSF genes activated the reporter genes and survived in selection medium lacking His and Ade ( Figure 1B). These results indicate that RBP-P and NSF interact under stringent binding conditions.
To further substantiate the interaction between RBP-P and NSF, bimolecular fluorescence complementation (BiFC) analyses using tobacco (Nicotiana tabacum) BY-2 cells was performed. In addition to supporting an interaction between protein pairs, BiFC also provides intracellular information on where this interaction occurs. RBP-P and NSF were fused to two complementary nonfluorescent fragments (n-and c-terminal halves) of enhanced yellow fluorescent protein (EYFP), nEYFP and cEYFP, respectively, and co-transformed into live BY-2 cells. If the two proteins interact, the two EYFP fragments are brought in close proximity to reform the native protein structure enabling emission of yellow fluorescence and, in turn, direct visualization of the protein complex in live cells (Kerppola, 2006;Miller et al., 2015). EYFP fragments lacking one of the putative protein partners were used as a negative control. As shown in Figures 2A to 2C, while no interaction was detected between RBP-P or NSF with the empty EYFP fragment controls, cells expressing both RBP-P and NSF fused to EYFP fragments emitted bright yellow fluorescence, indicating that RBP-P interacted with NSF. Closer examination reveals that RBP-P/NSF complexes were distributed to the cytoplasm as intensely bright clusters together with more loosely diffuse structures ( Figure 2C).
NSF is a soluble hexameric ATPase commonly found in the cytoplasm of eukaryote cells (Mastick and Falick, 1997;Zhao et al., 2007Zhao et al., , 2010Guo et al., 2017), which predominantly plays a major role as a chaperone in intracellular membrane fusion events. Through its interaction with the adaptor protein, soluble NSF attachment protein (SNAP), NSF binds to soluble NSF activating protein receptor (SNARE) complexes and utilizes the energy of Mg 21 -dependent ATP hydrolysis to disassemble the SNARE protein complex, and thus facilitating the recycling of SNARE proteins for further cycles of membrane fusion (Zhao et al., 2007;Chang et al., 2012;Ryu et al., 2015). During this process, SNAP serves as chaperone by stimulating the ATPase activity of NSF and dissociating from NSF after ATP hydrolysis (Zhao et al., (A) Precipitation of NSF by RBP-P antibody as revealed by IP-MS. Left representation, immunoblot (IB) analysis to test the IP reliability; right representation, silver-stained SDS-polyacrylamide gel of eluted samples from anti(a)-GFP and RBP-P IPs. Input, starting material of rice lysate; Ub, unbound fraction from IPs; B, bound fraction (eluted samples) from IPs. Blue asterisk (*) indicates a modified form of RBP-P. The bands indicated by red and black arrows were excised for MS analysis, and NSF was identified as a specific protein precipitated by anti-RBP-P but not by anti-GFP. (B) Interaction between RBP-P and NSF revealed by Y2H assay. Yeast colonies co-transfected with pGBK and pGAD constructs were labeled 1 through 4 (as described in the top tabletop). -, empty vector. Yeast cells carrying the corresponding genes were grown on SD/-Leu/-Trp medium as growth control and SD/-Leu/-Trp/-His/-Ade/13-AT selection medium to detect their interaction. Note that only the yeast cells carrying both NSF and RBP-P survived on the strict selection medium (bottom), suggesting that RBP-P interacts with NSF.

2007). This transient interaction between NSF and SNAP is only
detected under conditions where a nonhydrolyzable ATP is used as a substrate (Hanson et al., 1995;Barnard et al., 1997;Chang et al., 2012). Alternatively, the addition of EDTA to chelate Mg 21 and thus inhibit ATPase activity has been used to detect the transient interaction between NSF and SNAP (Hanson et al., 1995;Barnard et al., 1997;Chang et al., 2012;Li et al., 2018).
Based on the established properties of NSF, we performed coimmunoprecipitation (Co-IP) experiments to assess the formation of RBP-P/NSF complexes in rice endosperm cells. Rice seed lysates supplemented with 1 mM of ATP and 8 mM of MgCl 2 or EDTA were incubated with affinity-purified anti-RBP-P and NSF antibodies ( Figure 2D) or anti-GFP antibody, the latter used as a negative control. No proteins were captured by the control GFP antibody, revealing the reliability of the IP experiments. Interestingly, RBP-P and NSF were co-precipitated in the presence of Mg 21 -ATP but not when EDTA was added to the seed lysates ( Figure 2D). This result indicates that formation of the RBP-P/NSF complex in rice endosperm cells requires Mg 21 -ATP, a condition distinct from those complexes involving the stable interaction between NSF and SNAP (Hanson et al., 1995;Barnard et al., 1997;Chang et al., 2012;Li et al., 2018). The differences in binding properties, as well as the absence of SNAP in RBP-P IPs ( Figure 2D), indicate that the events of membrane fusion requiring NSF-SNAP are not required for formation of the RBP-P/NSF complex. On the other hand, the interaction between NSF with RBPs suggests that NSF may function in mRNA metabolism by its interaction with RBP-P.

RBP-P Indirectly Interacts with Rab5a through NSF
The interaction of the glutelin zipcode trans-factor RBP-P with NSF supports a close relationship between glutelin mRNA transport with membrane trafficking. Previous studies demonstrated that Rab5a, an evolutionarily conserved key GTPase involved in the biogenesis of early endosomes and membrane trafficking in the cytoplasm (Woodman, 2000;Saito and Ueda, 2009;Ito et al., 2018), is also required for glutelin mRNA localization (Doroshenk et al., 2010) and vesicular membrane transport between the Golgi and PSV in rice endosperm cells (Wang et al., 2010;Fukuda et al., 2011). Therefore, we investigated whether the abovementioned RBP-P/NSF complex is associated with Rab5a to regulate the endosomal transport of mRNAs.
We first performed a transient expression study to confirm the association of rice Rab5a (Figures 2E and 2F; Supplemental Figure 1) with endosomes. FM4-64 is a lipophilic dye that initially labels the plasma membrane and, subsequently, internalizes with membrane vesicles transported along the endocytic pathway (Vida and Emr, 1995;Ueda et al., 2001). In this experiment, BY-2 protoplasts expressing GFP-Rab5a were treated with FM4-64 for 15 min before observation. As shown in Figure 2F, red fluorescence derived from FM4-64 was evident on the plasma membrane and in internalized endosomes. GFP-Rab5a displayed a similar distribution pattern and co-localized with the internal FM4-64-labeled vesicles ( Figure 2F), an observation confirming that Rab5a is associated with endosomes.
We then studied the relationship of Rab5a with the abovementioned RBP-P/NSF complex. Co-IP studies were performed using affinitypurified antibodies to Rab5a, RBP-P, and NSF. All three proteins were found in IPs generated by anti-Rab5a, anti-RBP-P, and anti-NSF ( Figure 2D), suggesting that they form a multiprotein complex in rice endosperm cells. Protein complex formation apparently required active ATPase hydrolysis as all three protein interactions could only be simultaneously detected in the presence of Mg 21 and ATP and not when EDTA and ATP were included ( Figure 2D).
To further characterize this protein interactome and the formation of their complexes, BiFC (Figures 2G to 2J) studies were performed. Interaction between NSF and Rab5a was observed in the tested cells, and their complexes existed in the cytoplasm. By contrast, Rab5a does not interact with RBP-P ( Figure 2J) and, therefore, the association of RBP-P with Rab5a in the coimmunoprecipitates ( Figure 2D) is mediated through the interaction between NSF and Rab5a.
Rab GTPases are highly conserved small proteins and share two conserved regions that are required for conformational switching between active and inactive states of GTP hydrolysis ( Figure 2E; Supplemental Figure 1). The switch regions are disordered in an inactive conformation and adopt a well-defined conformation in the active stage (Stenmark and Olkkonen, 2001). Amino acid substitutions within the switch regions affects their conformational switching. While the GDP-fixed Rab5a (G45D) (D) SNAP-independent formation of a protein complex consisting of RBP-P, NSF, and Rab5a as revealed by Co-IP. IP studies were conducted with anti(a)-GFP, RBP-P, NSF, or Rab5a antibodies in the presence of ATP and MgCl 2 or with EDTA. GFP antibody was used as a negative control. Input, the starting rice lysate. Note that the association of RBP-P with NSF and Rab5a requires MgCl 2 and ATP, unlike the interaction of SNAP with NSF and Rab5a, which requires EDTA.
(E) Schematic structure of Rab5a in rice. The two switch regions are indicated as Switch 1 and Switch 2. The two mutation sites (G45D and Q70L) to generate the GDP-and GTP-fixed forms of Rab5a, respectively, are indicated in red.
(G) to (J) Rab5a interacts with NSF (I), but not to RBP-P (J), as shown by BiFC. mutation in switch 1 inhibits the conformation switching to an active GTPase stage and restricts Rab5a in a GDP-bound form, a Q70L mutation in switch 2 locks it in a constitutively active GTP-bound form (Lee et al., 2009). Only the GTP-bound form of Rab5 are membranebound during endosomal trafficking (Yuan and Song, 2020).
To investigate whether the NSF-Rab5a complex is associated with active endosomes, the interaction of NSF with the two structural forms of Rab5a containing G45D or Q70L was analyzed by BiFC ( Figures 2K and 2L). Although NSF directly interacts with both GDP-(Rab5a G45D ) and GTP-(Rab5a Q70L ) forms of Rab5a, their complexes are observed as two distinct populations. When bound to the GTP-fixed Rab5a (Q70L) form ( Figure 2L), the NSF complex was distributed as endosomal punctate structures. By contrast, the NSF complex constituted by GDP-fixed Rab5a (G45D) showed a diffuse distribution throughout the cytoplasm ( Figure 2K). These results are consistent with the view that the GTP-fixed Rab5a (Q70L) form is membrane-bound while the GDPfixed Rab5a (G45D) is mainly soluble.
The direct interaction between NSF and Rab5a provides an indirect link to RBP-P and, in turn, RNA localization. To confirm the association of the RBP-P/NSF complex to Rab5a-positive endosomes, we performed a three-way co-localization test consisting of RBP-P and NSF as BiFC interacting partners in BY-2 cells expressing Rab5a tagged with the red fluorescence protein (RFP; Figures 2M and 2N). Florescence analysis of BY-2 cells showed that RFP-Rab5a WT was distributed as diffuse signals throughout the cytoplasm and as punctate structures (endosomes) near the plasma membrane ( Figure 2M, middle). By contrast, RFP-Rab5a Q70L was present only as punctate structures ( Figure 2N, middle) indicating that the activated GTPase form of Rab5a is associated with endosomes. The fluorescence distribution pattern seen for RFP-Rab5a WT is consistent with membrane-free Rab5a being distributed in the cytoplasm and active Rab5a associated with endosomes (Yuan and Song, 2020).
As earlier seen in Figure 2C, BiFC-linked RBP-P/NSF complexes are distributed as fluorescent punctate structures in BY-2 cells ( Figures 2M and 2N). These RBP-P/NSF complexes co-localized with RFP-Rab5a-associated endosomes in the cytoplasm, especially with membrane-associated GTP-fixed (Q70L) Rab5a ( Figures  2M and 2N). To further confirm that the RBP-P/NSF complexes are associated with endosomes, we used FM4-64, which specifically labels these small membrane compartments ( Figure 2O). The fluorescent signals from the RBP-P/NSF complexes co-localized with endosomal compartments labeled by FM4-64, providing direct evidence that RBP-P/NSF complexes co-localize with endosomes. Consistent with the abovementioned Co-IP analysis ( Figure 2D) where RBP-P, NSF, and Rab5a may co-assemble into a multiprotein complex in rice endosperm cells, the BiFC results suggest that RBP-P is associated with endosomes through an NSFmediated interaction with Rab5a.
It should be noted that not all RBP-P/NSF complexes colocalized with the Rab5a-linked endosomes ( Figures 2M and  2N, open arrowheads). Similarly, the location of several Rab5aactive endosomes ( Figures 2M and 2N, open arrowheads) do not coincide with RBP-P/NSF complexes. The lack of overlap in the distribution of a population of RBP-P/NSF complexes and Rab5 endosomes suggests the multiple roles of these proteins in RNA localization and membrane trafficking.

RBP-L Is Involved in the RBP-P/NSF/Rab5a Complex
We had previously demonstrated that RBP-P co-assembles with RBP-L to form a protein complex that is essential for storage protein mRNA localization (Doroshenk et al., 2014;Tian et al., 2018;Tian et al., 2019a). To determine whether RBP-L is also involved in interacting with NSF or Rab5a, BiFC studies were conducted ( Figures  3A to 3F). Such analysis showed that RBP-L interacts directly with Rab5a but not with NSF ( Figures 3A to 3C). The RBP-L/Rab5a WT complex is observed as two distinct populations. One population is distributed as endosomal punctate structures with the bulk located close to the plasma membrane. The co-localization of RBP-L/ Rab5a WT with endosomes is supported by their close association with FM4-64-labeled endosomal compartments ( Figure 3D). A second population of RBP-L/Rab5a WT is viewed as a diffuse cloud around the nucleus. Similar to the RBP-P/NSF/Rab5a complex, the interaction of RBP-L with Rab5a was not dependent on the functional state of Rab5a as it interacts with both the GTP-bound Ra-b5a Q70L and GDP-bound Rab5a G45D ( Figures 3E and 3F), although the distribution patterns are distinct. RBP-L/Rab5a G45D is distributed mainly as a diffuse cloud around the nucleus and near the plasma membrane, whereas RBP-L/Rab5a Q70L is observed predominantly as some discrete endosomal punctate structures.
To further determine whether RBP-L is associated with the RBP-P/NSF/Rab5a complex, we performed Co-IP with paraformaldehyde (PFA)-fixed seed extracts to optimize the capture of any potential dynamic endosome-associated complexes formed in vivo. Such results showed that, irrespective of the antibody used in the initial IP, the RBP-P, RBP-L, NSF, and Rab5a were detected in IPs generated by all four antibodies ( Figure 3G). By contrast, SNAP is only present in IPs obtained with anti-NSF and anti-Rab5 but not in IPs generated by antibodies to RBP-P and RBP-L ( Figure 3G). These results are consistent with the Co-IP results depicted in Figure 2D where SNAP is present in IPs generated with antibodies to NSF and Rab5a in the presence of EDTA but not in the presence of MgCl 2 . Overall, these results are consistent with the view that NSF-Rab5a complexes exist as two separate populations. One NSF-Rab5a population together with SNAP participates in Rab5mediated endosomal fusion (Woodman, 2000;Zhao et al., 2007), where NSF and SNAP disassemble the SNARE complex formed by Rab5-mediated membrane fusion. A second NSF-Rab5a complex contains RBP-P and RBP-L and is independent of SNAP.
To further address whether RBP-P and RBP-L are attached to endosomes together, sequential IPs were performed ( Figure 3H). In this analysis, the RBP-P/NSF/Rab5a complexes precipitated by anti-RBP-P antibodies were next subjected to an additional IP using anti-RBP-L. All four proteins were detected in the second IP generated by anti-RBP-L, suggesting that RBP-P, RBP-L, NSF, and Rab5 co-assemble to form a quaternary protein complex.

The Quaternary Complex Attaches To Active Endosomes Carries Glutelin mRNAs
To investigate whether the quaternary complex binds glutelin mRNA, we performed RNA-IP analysis to detect the in vivo association of the complex with glutelin mRNAs ( Figure 3I). In this analysis, antibodies to RBP-P, RBP-L, NSF, and Rab5a were utilized to capture the associated RNA-protein complexes, and the RNA was subsequently isolated from the IP fractions and subjected to RT-PCR using specific primers to amplify glutelin transcripts. Compared with the negative empty-resin control and anti-GFP antibody, glutelin mRNAs were highly enriched in IPs generated by all four antibodies ( Figure 3I). The mRNA amount associated with NSF and Rab5a was lower than that of RBP-P and RBP-L, a result consistent with the roles of these proteins in membrane fusion events. Overall, these results support the view that the quaternary complex contains glutelin mRNAs.
Based on these results, a working model of cytosolic glutelin mRNA transport is proposed in Figure 4. The mRNP complex containing glutelin mRNA, RBP-P, and RBP-L is bound to Rab5aassociated endosomes through a four-way interactome, i.e., the direct interaction of RBP-P with RBP-L, of RBP-L with Rab5a, of NSF with RBP-P, and of Rab5a with NSF. The GTP-bound Rab5a is associated with endosomes (Yuan and Song, 2020), as suggested by distribution of this activated Rab5a form as endosomal punctate structures (Figures 2N and 3F). (G) Association of RBP-P and RBP-L with NSF and Rab5a as revealed by IP. The IP experiment was conducted with developing seeds treated with 1% PFA to preserve labile protein complexes. Input, the starting rice lysate; Control, GFP antibody; IB, immunoblot.
(H) RBP-P, RBP-L, NSF, and Rab5a form a ternary complex, as revealed by sequential IPs. The first IP was conducted with anti(a)-GFP (negative control) or RBP-P antibodies. Proteins captured by RBP-P antibody were subjected to a second IP using RBP-L antibody. Subsequent immunoblot analysis showed the presence of NSF and Rab5a together with RBP-P and RBP-L. Input, the starting rice lysate; FT, flow-through (unbound fraction); E, bound fraction; W1 and W2, washing fractions. (I) The RBP-P/RBP-L/NSF/Rab5a complex contains glutelin mRNAs based on RNA-IP. RNAs extracted from each IP generated by anti-RBP-P, anti-RBP-L, anti-GFP, or empty resin (-CT) were subjected to RT-PCR using glutelin-specific primers and resolved by agarose gel electrophoresis. GFP antibody (a-GFP) and empty resin (-CT) were used as negative controls. Input, PCR products using cDNA synthesized from total RNAs. Actin was used as the control gene to verify the specific binding of the complex to glutelin RNAs. Scale bars, 20 mm.

Loss of Function of Rab5a Results in Mis-targeting of Glutelin mRNAs
To provide evidence in support of this model, we analyzed a rab5a mutant EM960 (Fukuda et al., 2011) expressing a GDP-fixed (G45D) Rab5a ( Figure 5A). Similar to the phenotype shown in the EM956 mutant lacking Rab5a (Fukuda et al., 2011) or a mutant line expressing a defective Rab5a effector GEF (Wen et al., 2015), normal endosomal trafficking is disrupted in the endosperm cells of GDP-fixed rab5a mutant and leads to the formation of PMBs ( Figures 5B and 5C), an aborted endosome complex containing mis-sorted endomembrane proteins. These extracellular PMBs, which contain numerous electron-dense vesicles, are located in the space between the invaginating plasma membrane and the cell wall in the mutant endosperm cells (Figures 5B and 5C).
To investigate the co-localization of RBP-P, RBP-L, and NSF with Rab5a and the subcellular localization of their complex in rice endosperm cells, we performed double immuno-fluorescence labeling on thin sections of rice developing seeds using antibodies raised against each of the four proteins. Although the bulk of these proteins were evidently independent of Rab5, there was ample evidence for co-localization of RBP-P, RBP-L, and NSF with Rab5a. The co-localization of these proteins with Rab5a was apparent as punctate structures in the cytoplasm, particularly in the cortical region underneath the plasma membrane ( Figures 6A,  6C, and 6E), an intracellular location enriched in Rab5a-mediated endosome activity (Chavrier et al., 1990;Fischer von Mollard et al., 1994). To directly assess the co-localization of these proteins, the fluorescence intensity profiles of these proteins were quantified along a specific linear distance ( Figure 6, right panels). The fluorescence signals for the proteins examined overlapped substantially, indicating that RBP-P, RBP-L, and NSF co-localized to Rab5a-labeled endosomal compartments in rice endosperm cells. The independent distribution of RBP-P, RBP-L, and NSF with Rab5a was also evident in the BiFC/RFP double labeling ( Figures 2M and 2N), which is indicative of their roles in other cellular processes. This view is also supported by the Co-IP results ( Figures 2F and 3H) where IPs by antibodies to RBP-P, RBP-L, and NSF contained only a small proportion of the total Rab5a amounts.
The RBPs retain their co-localization with Rab5a in the rab5a mutant ( Figures 6B and 6D), an expected observation as RBP-L as well as NSF interact with both the GDP-and GTP-fixed Rab5a forms ( Figures 2K, 2L, 3E, and 3F). These protein complexes exist as punctate structures within the PMBs in the rab5a mutant, (Figures 6B, 6D, and 6F). Hence, disruption of membrane trafficking in the rab5a mutant displaces not only endosomal proteins to the extracellular PMBs but also RBP-mRNA complexes.
Transmission electron microscopy (TEM) analysis further confirmed the co-localization of RBP-P and RBP-L with Rab5a on endosomes in wild-type and aborted endosome vesicles within PMBs in rab5a mutant (Figure 7). In wild type, Rab5a-mediated endosomes were observed as electron-dense vesicles with an irregular shape likely due to endosomal fusion ( Figures 7A to 7E). Colocalization of RBP-P and RBP-L with Rab5a was observed on those endosomes, further suggesting that mRNP complexes, carrying glutelin mRNAs bound by RBP-P and RBP-L, are transported on endosomes. In the rab5a mutant line, normal endosomal trafficking is disrupted, resulting in the displacement of mRNAassociated endosomes to the extracellular PMBs ( Figures 7F to 7J).
To further address whether glutelin mRNAs are associated with the Rab5a-mediated endosomes, in situ RT-PCR on developing endosperm sections was performed to locate glutelin mRNAs. In this experiment, Rhodamine B dye was used to specifically stain the PB-ER (Muench et al., 2000) while glutelin mRNAs were labeled with Alexa488-dUTP by in situ RT-PCR using glutelinspecific primers (Figure 8). In wild type, glutelin mRNAs are localized on the cisternal-ER separate from PB-ER (Figure 8). In rab5a mutant cells, however, they were mainly distributed to the PMBs and with smaller amounts associated with the PB-ER. These results indicate that normal glutelin mRNA localization is disrupted in the rab5a mutant. This result resonates with the retention of the complexes formed by RBP-P, RBP-L, NSF, and Rab5a in PMBs (Figures 6 and 7), further supporting a Rab5adependent endosome transport of glutelin mRNAs. In rice endosperm cells, glutelin mRNA is bound by the scaffold complex of RBP-P and RBP-L, forming an mRNP complex. Through direct or NSF-mediated indirect interaction with membrane-associated Rab5a, RBP-L and RBP-P link the mRNP complex to endosomes for active transport via the cytoskeleton. Rab5a effectors (e, light orange) may be involved to stabilize the quaternary complex and modulate GTP-bound active Rab5a-derived endosomal trafficking on actin filaments in rice endosperm cells. Other unknown RBPs or factors, showed in light gray shapes, may also be involved to constitute the mRNP complex and define the linkage onto endosomes.
Taken all together, the results described here support the view that glutelin mRNAs are transported to the cisternal-ER membrane via their hitchhiking on endosomes. In the rab5a mutant, however, glutelin mRNA-containing mRNP complexes are not transported to their normal location but are mis-targeted to the PB-ER and displaced to the extracellular PMBs because of the disruption in normal endosomal trafficking.

DISCUSSION
The study of rice storage protein mRNA localization as a model has provided considerable information on why and how plant mRNAs are localized to specific ER compartments. Although zipcode ciselements of glutelin and prolamine mRNAs and several key RBPs have been identified (Hamada et al., 2003;Washida et al., 2009;Doroshenk et al., 2012;Washida et al., 2012;Tian et al., 2018Tian et al., , 2019a, the mechanism by which glutelin and prolamine mRNAs are transported to distinct subdomains of the ER membrane has yet to be established. Here, we report that rice endosperm cells employ Rab5a and NSF-proteins traditionally known for their essential roles in endosomal trafficking and membrane fusion events-as adaptor proteins linking mRNA-protein complexes to endosomes for active transport of glutelin mRNAs. Active transport of mRNAs along cytoskeleton networks by hitchhiking on motile, membrane-bound organelles has been reported in mammalian and fungi cells. In human axons, the Ras GAP SH3 domain binding protein1-labeled RNA granules were found to co-localize and co-traffic with moving lysosomes along microtubules (Liao et al., 2019). The membranous amyotrophic lateral sclerosis-associated phosphoinositide binding protein, annexin A11, functions as a molecular tether between RNA granules and lysosomes (Liao et al., 2019). In addition to mRNAs, neuronal precursor microRNAs are transported to the tip of growing axons by hitchhiking on late endosomes/lysosomes (Corradi et al., 2020).
Unlike the abovementioned examples in mammalian and fungi cells, higher plants apparently adapted the two membrane trafficking factors, NSF and Rab5a, to mediate transport of storage protein mRNAs. The linking of glutelin mRNP complexes to endosomes is mediated by the four-way interactions of NSF and Rab5a with the two glutelin zipcode RBPs, RBP-P and RBP-L. These interacting proteins likely form a quaternary complex enabling endosomal transport of glutelin mRNAs. Such proteinprotein interactions are accomplished through a gain in binding properties by Rab5a and NSF, and the glutelin zipcode mRNA binding proteins, RBP-P and RBP-L, which allow for the heterotypic interaction of an RBP with a membrane fusion factor (RBP-P/NSF and RBP-L/Rab5a). The interaction of an RBP with a membrane trafficking factor, i.e., among RBP-P with NSF and RBP-L and Rab5a, as well as between NSF and Rab5a, are unexpected findings as such interactions have not been reported in other organisms. It remains unclear whether the interactions of these membrane trafficking factors to RBPs and even between themselves is an inherent property or whether they are unique to higher plants. While Upa1-mediated endosomal mRNA transport is proposed as evolutionarily conserved in fungi (Müller et al., 2019), further investigation is needed to assess whether endosomal mRNA transport by the NSF-Rab5a-RBP machinery is a widespread phenomenon in transporting mRNAs among higher plants and other eukaryotic organisms.
It appears that the recognition of NSF and Rab5a with RBPs is highly selective. These non-random interactions highlight the diverse binding capabilities of NSF, Rab5a, as well as for RBP-P and RBP-L in rice endosperm cells. Both RBP-P and RBP-L contain RRMs, with the former having two RRM motifs and the latter having three RRM motifs. Due to the conservation of the RRM motifs, the recognition of NSF and Rab5a to RBP-P and RBP-L, respectively, is likely via the unique N-and C-terminal regions that flank the RRM motifs. Indeed, our preliminary results (Supplemental Figure 2) show that the N-terminal end of RBP-P is essential for its interaction with NSF. Through the selective recognition of NSF and Rab5a to other specific RBPs, these membrane trafficking factors could serve as the core components enabling endosome-coupled mRNA transport with the RBPs specifying the mRNA species.
NSF is a homo-hexamer with three domains: the N-terminal domain (NSF-N), required for SNAP-SNARE binding; the ATP binding domain 1, responsible for ATPase activity; and the ATP binding domain 2, responsible for hexamerization (Tagaya et al., 1993;Zhao et al., 2007). The NSF-N domain is the main protein-protein interaction site for NSF binding to other proteins. NSF-N domain is likely responsible for its interaction with the N terminus of RBP-P in rice as well (Supplemental Figure 2). Sequence alignment of NSF homologues from rice, Arabidopsis (Arabidopsis thaliana), Drosophila, yeast, and human show that the N-terminal region displays considerable sequence diversity among these species (Supplemental Figure 3). Further structural studies are required to investigate their binding mechanism and to determine whether this kind of interaction occurs in other species.
Rab5a interacts with a large number of proteins, including its regulators (activating effectors and inhibitors), linkers to molecular motors, membranous factors, components of membrane fusion complexes, protein kinases, and phosphatases (Hervé and Bourmeyster, 2018;Pylypenko et al., 2018). Many of these interacting proteins function as Rab5a effectors to activate endosomal transport and tightly control the specific functions of Rab5a in membrane composition and modification, cytoskeleton regulation, and intracellular trafficking (Hervé and Bourmeyster, 2018;Pylypenko et al., 2018). Although NSF has been reported to interact with several Rab5a effectors to drive membrane fusion during endosomal docking (McBride et al., 1999;Grosshans et al., 2006), no direct relationship between Rab5a with NSF or with a RBP had been established until now. These findings will stimulate future research efforts on identifying new alternative roles of Rab5a and NSF beyond membrane fusion.
Although the GTPase activity of Rab5a is not required for the apparent formation of the quaternary protein complex as the four proteins still co-localized in the G45D Rab5a line (Figures 6 and 7), Rab5 GTPase activity is required for proper mRNA transport. Endosome formation and membrane trafficking are dependent on Rab5 GTPase activity (Woodman, 2000;Zeigerer et al., 2012). Null Rab5a activity results in aborted endosomal transport in the mutant, which, in turn, disrupts glutelin mRNA transport on endosomes and causes its mis-localization to the PB-ER as well as displacement to the extracellular PMBs ( Figure 8). Therefore, the active GTPase activity of Rab5a is essential for the transport of glutelin mRNAs on endosomes. This also raises the question of whether NSF and RBP-L act as Rab5a effectors to modulate the role of Rab5a in mRNA transport and localization. While this hypothesis needs further examination, other effectors, functionally equivalent to Rabaptin-5 and EEA1 that interact with both NSF and Rab5a (McBride et al., 1999;Grosshans et al., 2006), may also be involved. These effectors, including the abovementioned GEF, may allow further stabilization of the linkage of mRNP complex to active endosomes and regulate endosomal mRNA transport via the cytoskeleton (Figure 4).
The identification and study of rice lines expressing mutations in Rab5a were instrumental in establishing its involvement in both RNA and membrane trafficking in developing rice endosperm. While Rab5 is an essential growth factor, loss of Rab5a only had a slight effect on rice growth and development as its activity is complemented by other Rab5 activities. Rice expresses three other Rab5 genes, Rab5b, Rab5c, and Rab5d (Supplemental Figure 4). Like Rab5a, Rab5c is a conventional type, while Rab5b and Rab5d are plant-specific type homologous to the Arabidopsis plant-specific Rab5F1/ARA6. During rice growth and development, the loss of Rab5a is offset by Rab5c and possibly by Rab5b and Rab5d. Although the other Rab5 isoforms can compensate for Rab5a, they are only able to partially fill this role as rab5a rice lines grow slower and flower later than normal, likely due to their significant lower expression compared with Rab5a (Supplemental Figure 5). At the grain filling stage where there is massive protein transport from the ER via the Golgi to the storage vacuole, the reduced expression of these other Rab5 isoforms fails to maintain normal membrane trafficking-resulting in the secretion of storage proteins, the formation of PMBs, and mislocalization of glutelin mRNAs.
Efforts to identify rice lines expressing defective NSF have not been productive, likely because of the importance of its ATPase activity in membrane fusion. Mutations in NSF were reported to induce severe defect and cell lethality in several organisms (Boulianne and Trimble, 1995;Golby et al., 2001;Mohtashami et al., 2001;Horsnell et al., 2002;Zhao et al., 2007). For example, Drosophila expresses two NSF isoforms, dNSF1 and dNSF2. dNSF1 is dominant in the adult central nervous system while dNSF2 is broadly distributed at the larval/adult stages of development, respectively (Boulianne and Trimble, 1995). While dNSF1 null mutants perish as pharate adults, dNSF2 deletion mediates a recessive lethal phenotype, which is not even rescued by the addition of a dNSF2 transgene (Golby et al., 2001;Mohtashami et al., 2001). In yeast, loss-of-function mutation of the NSF gene, Sec18, resulted in a dominant lethal phenotype (Horsnell et al., 2002). These studies suggest that NSF mutations are pleiotropic and cause severe growth problems. In Arabidopsis, even a subtle mutation of NSF caused severe abnormal Golgi morphology (Tanabashi et al., 2018). Given that NSF is coded by a single gene copy in the rice genome, mutations that affect NSF activity will likely confer a strong lethal phenotype in rice.
Although beyond the scope of this study, NSF mutations that affect binding to RBP-P would be a viable approach to obtain further insight on its role in mRNA trafficking. As discussed earlier, the NSF-N domain is likely responsible for its interaction with the N terminus of RBP-P. Selected residues in the NSF-N region can be replaced by amino acids that alter charge or conformation and then tested for its protein-interactive properties by Y2H analysis. Mutations in NSF that abolish its interaction with RBP-P but not with Rab5a would be potential sites for genetic alteration by clustered regularly interspaced short palindromic repeats (CRISPR) technology and, thereby, disrupting glutelin mRNA transport but maintaining normal function in membrane vesicle transport.
Overall, this study provides evidence on how glutelin mRNPs are able to hitchhike on trafficking endosomes in rice endosperm cells by exploiting the binding properties of RBPs and membrane trafficking factors, NSF and Rab5a. These findings will provide the basis for future research on membrane trafficking-mediated mRNA transport and the unique functions of NSF and Rab5a in this cellular process.

Plant Materials and Growth Conditions
The rice wild type (Oryza sativa 'Japonica' variety TC65) and rab5a mutant line EM960 (Fukuda et al., 2011) were potted in Sunshine Brand #1 soil (Sungro Horticulture) and grown in walk-in growth chambers with a diurnal cycle of 12-h light/12-h dark at 27°C and a lighting intensity of 400 to 700 mmol m 22 s 21 using a combination of metal halide and high-pressure sodium lamps.

Antibodies
Full-length cDNAs encoding RBP-P, RBP-L, NSF, SNAP, BiP, and GFP were cloned into pET30a for His-tagged fusion protein expression. The His- The PB-ER was stained by Rhodamine B dye (magenta), and glutelin mRNAs were labeled by in situ RT-PCR in the presence of Alexa-488-UTP (green). In wild-type (WT) endosperm cells (top), glutelin mRNAs are localized on the cisternal-ER adjacent to the PB-ER. In rab5a mutant (bottom), glutelin mRNAs are mis-targeted to the PMBs (asterisk) and PB-ER (arrows). Fluorescence intensity graphs on the very right show the relative position of magenta and green fluorescence signals in the regions indicated by rectangle in the Merge representations. X-and y axes represent the fluorescence intensity and position of the signals (pixels), respectively, evaluated by the software FUJI (ImageJ). Note the extensive overlap in the distributions between PB-ER and glutelin mRNAs in rab5a but not in wild type. Scale bar, 20 mm. tagged proteins were then purified and used to immunize New Zealand White rabbits for antibody production. Anti-Rab5a rabbit and mouse antibodies were generated from an earlier study by Fukuda et al. (2011). For immunofluorescence labeling, Alexa Fluor-488-labeled goat anti-mouse IgG antibody (cat. no. A32723; Invitrogen) and Alexa Fluor-594-labeled goat anti-rabbit IgG antibody (cat. no. A32740; Invitrogen) were used as secondary antibodies. For the immunogold labeling, Electron Microscopy (EM)-grade, 10-nm-gold-conjugated goat anti-mouse IgG (cat. no. 25,128;Electron Microscopy Science) and 112;Electron Microscopy Science) were used as secondary antibodies.

IP Analysis
Most IP experiments were conducted as described by Doroshenk et al. (2014) except that incubations with antibody were conducted in the presence or absence of 1 mM of ATP/GTP, 8 mM of MgCl 2 , or EDTA. To maximize capture of transient mRNP complexes on endosomes, developing seeds were treated with 1% (w/v) PFA. Detailed procedures for IP-MS, the identification of NSF protein, and PFA fixation for the enhanced IP are described below.

IP-MS Analyses
Affinity-purified RBP-P or GFP antibodies were prepared using immobilized metal affinity chromatography based on irreversibly oxidized Co(III)-IDA resin as previously described by Crofts et al. (2010) and cross linked to Protein A/G agarose resin using the Pierce Crosslink Immunoprecipitation Kit according to the manufacturer's instructions (Thermo Fisher Scientific). Protein extraction and Co-IP experiments were performed at 4°C unless indicated. Two g of dehulled mid-developing wild-type rice seed harvested 12 to 14 d after flowering were frozen in liquid nitrogen and ground to a powder. Proteins were extracted in 6 mL of IP buffer (20 mM of Tris-HCl at pH 7.5, 0.15 M of NaCl, 1 mM of EDTA, and 0.5% v/v Nonidet P-40) containing 13 protease inhibitor cocktail and 0.53 phosphatase inhibitor (Sigma-Aldrich). The crude extract was clarified by twice centrifuging at 12,000g for 10 min. The resulting supernatant was gently rotated overnight with agarose resin to eliminate nonspecific interactions. Seven-hundred microliters of the unbound protein fraction were added to columns containing RBP-P or GFP antibodies cross linked to Protein A/G resin or naked resin and incubated ;7 h with rotation. The columns were washed five times with IP buffer and once with 13 conditioning buffer, and bound proteins were eluted with 50 mL of IgG elution buffer (Thermo Fisher Scientific). The antibody-conjugated Protein A/G resin columns were regenerated by washing with 13 Coupling Buffer (Thermo Fisher Scientific) followed by IP buffer. Fresh, precleared rice-seed extract was added and incubated overnight. The columns were washed and proteins eluted, as above.
Eluted proteins from a total of five Co-IP experiments using anti-RBP-P, anti-GFP antibodies, or naked resin were pooled, then precipitated with addition of 100% (w/v) trichloroacetic acid (TCA) to a 20% (w/v) final concentration. The TCA precipitate was washed with acetone and then resuspended in 30 mL of SDS sample buffer containing 4 M of urea and 5% (v/v) b-mercaptoethanol. Protein samples were resolved on 10% SDS-PAGE gels and stained using a silver-nitrate-staining protocol (Chevallet et al., 2006). Because the protein profile of the minus antibody control IP looked very similar to the GFP co-immunoprecipitate, proteins from only the GFP and RBP-P co-immunoprecipitates were analyzed by mass spectrometry. Gel slices corresponding to similar positions within each lane were excised, in-gel-trypsin digested, and subjected to liquid chromatography-tandem MS as previously described by Doroshenk et al. (2009). Proteins were identified by searching the O. sativa National Center for Biotechnology Information non-redundant database (134,548 sequences) using the software Mascot (www.matrixscience.com) as previously described by Doroshenk et al. (2009).

IP with 1% PFA Fixation
Antibodies raised against RBP-L, RBP-P, NSF, or Rab5a were affinitypurified as mentioned above. Twenty microliters of resuspended protein A Magbeads slurry (25% [w/v] slurry; GenScript) in a 1.5-mL tube was washed twice with 13 PBS and then incubated overnight at 4°S under rotation with 30 mg of affinity-purified antibody (diluted to 500 mL by 13 PBS). Antibodies were cross linked to protein-A Magbeads using the Pierce Crosslink Immunoprecipitation Kit (Thermo Fisher Scientific).
Dehulled developing rice seeds were fixed with 1% (w/v) PFA in the modified phosphate-buffered saline buffer (1.76 mM of KH 2 PO 4 , 10 mM of Na 2 HPO 4 , 136 mM of NaCl, 2.6 mM of KCl, 5 mM of MgCl 2 , and 10% [v/v] glycerol) under vacuum for 30 min. The reaction was stopped by vacuuming in quenching buffer (0.333 M of Tris and 10% [v/v] glycerol) for 10 min. Ten to 20 aleurone-layer peels collected from the fixed seeds were ground to a fine powder using an ice-cold mortar and pestle in 800 mL of lysis buffer (50 mM of Tris-HCl at pH 7.5, 150 mM of NaCl, 10 mM of EDTA, 0.5% [v/v] NP-40, 0.1% [v/v] Triton X-100, 13 proteinase inhibitor cocktail, 100 mg/mL of cycloheximide, and 20 units/mL of RiboLock RNase Inhibitor [Thermo Fisher Scientific]), and then centrifuged at 1,000 g for 5 min to remove starch followed by centrifugation at 12,000g for 10 min. The clear supernatant was added to the prepared Magbeads for incubation overnight at 4°S with gentle rotation. After washing twice with lysis buffer, the bound fractions were eluted with an IgG elution buffer (Thermo Fisher Scientific) and neutralized by addition of 1:10 volume of 1 M of Tris-HCl at pH 8.8. The neutralized elution samples were analyzed by immunoblotting.

In Situ RT-PCR
In situ RT-PCR on developing rice seed sections was performed as previously described by Washida et al. (2009). The specific primers Glutelin-F 59-CCCTCAAGCATACAGGCGTG-39 and Glutelin-R 59-CGCTCTCTTGAT TGCACTTGTCC-39 were used in the PCR to amplify glutelin RNAs.

Construction of Vectors
Gene sequences of RBP-P, NSF, and their truncated forms were cloned into pGAD T7 and pGBK T7 vectors and used as preys and baits, respectively, in Y2H analyses for detection of protein-protein interaction. BiFC vectors of pSAT1-nEYFP-C1 and pSAT1-cEYFP-C1-B for N-and C-terminal EYFP fusion, respectively, were obtained from the Arabidopsis Biological Resource Center (https://www.arabidopsis.org). The cDNA sequences of RBP-P, RBP-L, NSF, Rab5a, and its mutant forms (Ra-b5a Q70L and Rab5a G45D ) were cloned into pSAT1-nEYFP-C1 or pSAT1-cEYFP-C1-B vectors for BiFC analysis. GFP-or RFP-fusion vector driven by double 35S promoters was constructed by replacing the N-terminal EYFP in pSAT1-nEYFP-C1 with GFP cDNA or RFP cDNA sequences. The cDNA sequences of NSF and Rab5a were then cloned into a GFP-or RFPfusion vector to obtain RFP protein fusions.

RNA-IP, Y2H, and BiFC Assays
Experimental procedures for these studies were performed as previously described by Doroshenk et al. (2014) and Tian et al. (2018Tian et al. ( , 2019a. Briefly, for RNA-IP analysis, developing rice seeds, collected 10 to 14 d after flowering, were subjected to 1% (w/v) PFA fixation and extracts prepared and used for IP as mentioned above. The elution samples obtained from IP were incubated at 70°C for 45 min to reverse the RNA-protein cross links followed by RNA extraction using TRIzol (Invitrogen), cDNA synthesis using M-MLV reverse transcriptase (Promega), and PCR with 20 to 25 cycles of amplification using glutelin-and ACTIN-specific primers (Glutelin-F and Glutelin-R as mentioned above for in situ RT-PCR, Actin-F 59-TCCATCTTG GCATCTCTCAG-39, and Actin-R 59-GTACCCGCATCAGGCATCT-39). The antibodies used in the RNA-IP were affinity-purified as aforementioned. In Y2H analysis, synthetic dropout (SD) growth media without Leu and Trp (SD/-Leu/-Trp) was used to screen positive transformants and selection media without Leu, Trp, His, and Ade (SD/-Leu/-Trp/-His/-Ade medium) supplemented with 3 mM of 3-amintriazole (3-AT; SD/-Leu/-Trp/-His/-Adenine/13-AT) or 40 mg/L of X-a-Gal to verify protein-protein interaction.
BY-2 suspension cells were used for BiFC analyses. In brief, BY-2 cells were treated with cell wall digestion buffer (1% [w/v] -Aldrich], 20 mM of KCl, 10 mM of CaCl 2 , 20 mM of 2-[N-morpholino]ethanesulfonic acid hydrate, and 0.5 M of Suc at pH 5.7) at room temperature for 3 h. After washing with W5 solution (154 mM of NaCl, 125 mM of CaCl2, 5 mM of KCl, and 5 mM of Glc at pH 5.8 to 6.0), the BY-2 protoplasts were subjected to PEG-mediated transformation with the vectors of pSAT1-nEYFP-C1 and pSAT1-cEYFP-C1-B as previously described by Tian et al. (2018). After culture at 26°C for 16 h, the BiFC fluorescence images were observed using a model no. SP-8 confocal microscope (Leica). Negative controls using empty vectors were also examined to check the reliability of the transformation procedure. The localization pattern of target proteins or complexes was determined by examining at least five different protoplast cells. To confirm the involvement of Rab5a and the corresponding complexes in endocytic pathway, protoplast incubation was treated with FM4-64 endocytic tracer (Invitrogen) at a final concentration of 10 mM for 15 to 30 min before observation.

Microscopy
Light microscopy was performed on 10-mm-thick sections of developing rice seed samples embedded in LR-white resin. The sections were positioned on X-tra slides (Leica), stained by 1% Toluidine blue and observed using a model no. BH-2 Light microscope (Olympus). Co-localization test of RBP-P, RBP-L, and NSF with Rab5a in rice endosperm cells was performed through double-immunolabeling using the rabbit anti-RBP-P, RBP-L, and NSF antibodies and mouse anti-Rab5a antibodies (see Antibodies) on 1-mm-thick LR-white sections as described in Fukuda et al. (2011), and observed under an Epi-Fluorescent Microscope (Leitz) with a model no. DFC425C camera (Leica). The fluorescence intensity of green and red signals was analyzed using the plot profile tool in the software FUJI (ImageJ). TEM analysis was performed as described in Tian et al. (2018).

Accession Numbers
Sequence data from this article can be found in the GenBank/European Molecular Biology Laboratory data libraries under National Center for Biotechnology Information accession numbers shown in the legends of Supplemental