Cytoskeleton members, MVBs and the ESCRT-III HvSNF7s are putative key players for protein sorting into protein bodies during barley endosperm development

Cereal endosperm is a short-lived tissue adapted for nutrient storage, containing specialized organelles, such as protein bodies (PBs) and protein storage vacuoles (PSVs), for the accumulation of storage proteins. PBs can be used as efficient biotechnological systems to produce high yields of stable recombinant proteins. During development, protein trafficking and storage require an extensive reorganization of the endomembrane system. Consequently, endomembrane-modifying proteins will influence the final grain quality, yield and recombinant protein production. Barley, a cereal crop of worldwide importance for the brewing industry, animal feed and to a lesser extent, human nutrition, has been identified as promising candidate for recombinant protein production. However, little is known about the molecular mechanism underlying endomembrane system remodeling during barley grain development. By using in vivo label-free quantitative proteomics profiling, we quantified 1,822 proteins across developing barley grains. Based on proteome annotation and a homology search, 95 proteins associated with the endomembrane system were identified, and 83 of these exhibited significant changes in abundance during grain development. Clustering analysis allowed characterization of three different development stages; notably, integration of proteomics data with in situ subcellular microscopic analyses showed a high abundance of cytoskeleton proteins associated with acidified protein bodies at the early development stages. Endosomal sorting complex required for transport (ESCRT)-related proteins and their transcripts are most abundant at early and mid-development. Specifically, multivesicular bodies (MVBs), and the ESCRT-III HvSNF7 proteins are associated with protein bodies (PBs) during barley endosperm development. Taken together, our proteomics results specifically identified members of the cytoskeleton, MVBs, and ESCRT as putative key players for protein sorting into PBs during barley endosperm development. These results present a comprehensive overview of proteins involved in the rearrangement of the endomembrane system during barley early grain development and will provide the basis for future work on engineering the endomembrane system to optimize nutrient content and to produce high yields of recombinant proteins.


Background
After differentiation, fully developed cereal endosperm can account for up to 75% of the grain weight and comprises four major cell types: aleurone, starchy endosperm, transfer cells, and the cells of the embryo surrounding region (Olsen, 2001). The starchy endosperm thereby functions as a storage site, as it accumulates starch and seed storage proteins (SSPs) (Olsen, 2004). The aleurone layer supports seed germination by mobilizing starch and SSP reserves in the starchy endosperm by releasing hydrolytic enzymes that help to degrade the stored nutrients in the endosperm (Olsen, 2004). Contrary to the persistent endosperm of cereals, the cellular endosperm of Arabidopsis thaliana supports the developing and growing embryo, resulting in a gradually depleted endosperm as the embryo grows. Finally, the massive A. thaliana embryo is only accompanied by a single peripheral layer, the aleurone layer, in mature seeds (Olsen, 2004). Consequently, A. thaliana cannot to be used as a model system to study the endomembrane system in grain endosperm.
In cereals, SSPs, which account for more than 50% of the grain protein content (Psota et al., 2007;Guo et al., 2016), accumulate in the outer layer of the endosperm, the subaleurone, and in the starchy endosperm, the latter in parallel with starch granules (Moore et al., 2016). In barley, for instance, globulins and prolamins comprise the major endosperm SSPs (Shewry and Halford, 2002).
The SSP trafficking routes depend on the cereal species, endosperm layer and development stage (Ibl and Stoger, 2012;Arcalis et al., 2014;Zheng and Wang, 2014). SSPs are produced by the secretory pathway and reach their final destinations by two main routes: soluble albumins and globulins travel through ER and Golgi to protein storage vacuoles (PSVs); and most prolamins are finally deposited in specific, ER-derived protein bodies (PBs) or/and in PSVs. Along with these main routes, other organelles are proposed to be involved in the SSP trafficking in cereal endosperm, e.g., multivesicular bodies (MVBs), precursor-accumulating (PAC) vesicles and/or dense vesicles (DVs) (Ibl and Stoger, 2012).
More recently, PBs have arisen as potential targets for recombinant protein production, especially pharmaceutical proteins (Tschofen et al., 2016). Indeed, these specialized storage organelles provide the ability to accumulate and store a high amount of (recombinant) proteins, subsequently offering high stability in planta and during harvest (Stoger et al., 2005;Hofbauer and Stoger, 2013). Interestingly, PBs are highly resistant to gastrointestinal digestion in vitro and in vivo, enabling PB bioencapsulation (reviewed in (Arcalis et al., 2014). In this context, barley (Hordeum vulgare), the fourth-most important cereal in terms of food production after maize, wheat, and rice (FAOSTAT 2016), holds great molecular farming potential. Barley is a self-pollinating, diploid crop and can be easily transformed (Horvath et al., 2000;Eskelin et al., 2009;Magnusdottir et al., 2013;Ritala et al., 2014). Recently, active mammalian antimicrobial LL-37 peptide was produced in barley endosperm (Holaskova et al., 2018). Further, recombinant anti-HIV-1 monoclonal antibody 2G12 was produced and accumulated within the endosperm of transgenic barley grain; 2G12, stored in ER-derived protein bodies, showed a comparable equilibrium and kinetic rate constants compared to recombinant 2G12, synthesized in Chinese hamster ovary cells (Hensel et al., 2015).
Depending on the protein trafficking, PB composition is affected thereby influencing the qualitative output. For instance, the SSP composition defines the malting purposes, most specifically for the brewing industry (Gupta et al., 2010). Thus, endomembrane-modifying proteins within the endomembrane system will have an influence in the final grain quality/yield and recombinant protein production.
In barley, microscopic analyses revealed that spherical PSVs, which are stable in the aleurone during development, underwent a dynamic rearrangement including fusion, rupture and degeneration in the subaleurone and starchy endosperm (Cameron-Mills, 1980;Ibl et al., 2014;Hilscher et al., 2016). Between 8 and 12 DAP, PSVs reduced their size and even degenerated in the starchy endosperm (Ibl et al., 2014). Simultaneously, PBs were found in distinct compartments in the subaleurone but also within the vacuoles, where fusion events of PBs were observed by live cell imaging (Ibl et al., 2014). In the starchy endosperm, PBs were tightly enclosed by vacuoles that finally degenerated and released the PBs (Ibl et al., 2014). Recent bioinformatic, proteomic and RT-qPCR analyses shed the first light on proteins possibly involved in the rearrangement of the endomembrane system in barley endosperm (Hilscher et al., 2016;Roustan et al., 2018;Shabrangy et al., 2018): ESCRT-III proteins were identified by a bioinformatic assay (Hilscher et al., 2016) and localization studies of recombinantly expressed HvSNF7a, HvVPS24 and HvVPS60a ESCRT III components in barley revealed different localization of these proteins within the endosperm (Hilscher et al., 2016). It was further suggested that the steady-state association of ESCRT-III may be influenced by cell layer-specific protein deposition or trafficking and remodeling of the endomembrane system in the endosperm (Hilscher et al., 2016). Additionally, the ER was identified to be most abundant in the starchy endosperm and ER rearrangements were characterized, including relocalization of HvPDIL1-1 during development (Roustan et al., 2018).
This work aimed to temporally map in situ the endomembrane system during barley endosperm development, using integrative cell biology experimental approaches. In this study, in vivo label-free proteomics approaches allowed the quantification of 1,822 proteins. Among these, 95 proteins could be associated with the endomembrane system. More specifically, seven out of eight proteins related to the ESCRT machinery showed specific expression patterns and transcript abundances associated with early and mid-development. In this context, confocal and transmission electron microscopy analyses located MVBs and HvSNF7 at the periphery of PBs and later within PBs, playing a putative role in protein sorting to PBs at mid-development.
Taken together, our proteomics results provide comprehensive developmental mapping of proteins potentially involved in the rearrangement of the endomembrane system during the 6 barley early grain-filling process and the specific identification of cytoskeleton members, MVBs, and ESCRT as putative key players for protein sorting into PBs during barley endosperm development.

Sample preparation for proteomics analyses
Total proteins were extracted from barley grains harvested at 6, 10, 12 and ≥ 20 DAP in three biological replicates. Stages were chosen as previously described (Shabrangy et al., 2018). Extraction was performed following an adapted phenol-phase extraction protocol as described in (Roustan et al., 2017). Subsequently, proteins were re-suspended in urea buffer (8 M urea, 100 mM ammonium bicarbonate) to measure protein concentration with a Bradford Assay prior protein content normalization and trypsin digestion. Following overnight digestion, peptides were desalted using a C18 solid phase extraction (Agilent Technologies, Santa Clara, USA). After solid-phase extraction, the corresponding eluates were dried in a vacuum concentrator.

Nano-HPLC and Orbitrap Elite tune methods
Mass spectrometry (MS) was performed as previously described (Roustan et al., 2018). Peptide pellets were resolved at a protein concentration equivalent of 0.1 µg/µL. A total of 0.5 µg of the mixture was separated on an EASY-Spray PepMap RSLC 7 μ × 0 c colu (Thermo Fisher Scientific, Waltham, Massachusetts, USA). After elution using a 150 min linear gradient at a 300 nL/min flow rate generated with an UltiMate 3000 RSLCnano system, the peptides were measured with an LTQ-Orbitrap Elite (Thermo Fisher Scientific, Waltham, Massachusetts, USA) as described in (Roustan et al., 2018). Mass analyzer settings were as follows: ion transfer capillary temperature 275 °C, full scan range 350-1800 m/z, FTMS resolution 60,000. Each FTMS full scan was followed by up to twenty data-dependent (DDA) CID tandem mass spectra (MS/MS spectra) in the linear triple quadrupole (LTQ) mass analyzer. Dynamic exclusion was enabled using list size 500 m/z values with an exclusion width ±10 ppm for 60 s. Charge state screening was enabled and unassigned and +1 charged ions were excluded from MS/MS acquisitions. For injection control, automatic gain control for full scan acquisition in the Orbitrap was set to 1 x 106 ion population, and the maximum injection time (max IT) was set to 200 ms. Multistage activation was enabled with neural losses of 24. 49, 32.66, 48.999, 97.97, 195.94, and 293.91 Da for the 10 most intense precursor ions. Prediction of ion injection time was enabled, and the trap was set to gather 5 x 103 ions for up to 100 ms. Orbitrap online calibration using internal lock mass calibration on m/z 371.10123 from polydimethylcyclosiloxane was used.

Data processing and protein identification
Raw files were processed with MaxQuant 1.5 (http://www.maxquant.org) and the Andromeda search algorithm (Cox and Mann, 2008;Cox et al., 2011;Tyanova et al., 2015) on the barley UniProt database (http://uniprot.org). Peptide identification was performed using the following settings: mass tolerance for precursors was set to 5 ppm and for fragment masses up to 0.8 Da. The maximum FDR was set to 0.01%. Three missed cleavages were allowed. The dynamic modifications allowed were: methionine oxidation (M) and protein Nterminal acetylation. The fixed modification allowed was: Carbamidomethyl (C). Label-free quantification was done at the MS1 level with at least two peptides per protein. PTXQC was used to assess data quality and statistical analysis was performed with Perseus 1.5 software (Bielow et al., 2016;Tyanova, 2016).
Protein annotation was performed with the MERCATOR tool (http://mapman.gabipd.org/de) (Lohse et al., 2014). Unknown proteins were identified by using BLAST at the UniProt homepage searching for the most identical cereal protein. Proteins were classified to compartment-specific proteins fu ctio al associated with a specific subcellular endomembrane pathway or organelle), a d as trafficking regulators (functional associated with several organelles) based on published data. In the final dataset, representative proteins were quantified in at least 9 of the 12 samples analyzed. Data were first Log2 transformed prior Z-scores (zero mean, unit variance) and were finally used to calculate the relative protein abundance. A one-ay a alysis of aria ce ANOVA a d Stude t's t-tests were performed with Perseus 1.5 software. Cluster analysis was performed with fuzzy-c means algorithm implemented in GProX (Rigbolt et al., 2011

Western blots
To confirm the MS results, we have used the same extraction protocol for the Western blots as for the MS analysis. For the analyses of the transgenic line p6U::SNF7.1-mEosFP, five transgenic and wild type (GP) grains were extracted by PBS buffer pH 6. After protein extraction, we measured the quantity of protein with a Bradford assay. Western blots were performed as previously described (Roustan et al., 2018
Besides bioinformatic analyses, the specificity of the primers was tested by PCR on the cDNAs. At least three biological replicates were used, and three technical replicates were performed for RT-qPCR. For the normalization studies of ESCRT transcripts, we used the following reference genes as described in Shabrangy et al. (2018): ARF (ADP-ribosylation factor), FBPA (fructose-bisphosphate aldolase), and SAM (S-adenosyl-L-methionine) for whole grains. Normalization was calculated as described in (Vandesompele et al., 2002;Shabrangy et al., 2018). For statistical analyses we perfor ed a Stude t's t-test (two-tailed distribution, twosample unequal variance (heteroscedastic)) using the Microsoft Excel software program.
Bimolecular Fluorescence Complementation (BiFC). A single colony of transformed Agrobacterium tumefaciens was inoculated in 5 mL of YEB-Medium (0.5% beef extract, 0.5% sucrose, 0.1% yeast extract, 0.05% MgSO4*7H2O) containing appropriate antibiotics and incubated at 28 °C overnight. In the morning, 1 ml of the pre-culture was taken and reinoculated into 5 ml of the same medium. Cells were collected by centrifugation at 5000 g for 5 min and the pellet was washed with 1 mL infiltration buffer (10 mM MES pH 5.7, 10 mM MgCl2, 100 µM acetosyringone). Washed cells were re-collected by centrifugation (5000 g, 5 min) and washed two additional times with 500 µl infiltration buffer and finally adjusted to an OD600 of 0.3. The resuspended bacteria containing the corresponding binary expression vectors for BiFC were mixed in a ratio of 1:1 and incubated for 3 h in darkness. Nicotiana benthamiana plants were cultivated in the greenhouse on soil, maintained at 60% humidity, with a 14 h light period and a 25 °C day/19 °C night temperature cycle. The bacterial solution was inoculated into the entire leaf area through the abaxial sides using a 1 mL syringe; two leaves per plant were infiltrated. After infiltration, the plants were kept in a tray with a hood at 25 °C. After 2-5 days, the detection of protein-protein interaction by BiFC was performed using confocal microscopy (Leica SP5 CLSM). The excitation wavelength was 514 nm (argon laser) and emission was detected between 525-600 nm and 680-760 nm for YFP and autofluorescence detection, respectively.
Transmission electron microscopy (TEM). The coat was removed for grains harvested at ≥ 20 DAP, chopped and immediately fixed in 4% (w/v) paraformaldehyde plus 2.5% (v/v) glutaraldehyde in 0.1 M sodium cacodylate, pH 7.4, overnight at 4 °C. After washing with sodium cacodylate buffer, the samples were immersed in OsO4 for one and a half hours at room temperature followed by washing with buffer. The chemically fixed samples underwent dehydration in an ethanol series (30%, 50%,70%, 95% ethanol for 10 min each, and two times 100% ethanol). Prior to infiltration with resin, the ethanol was replaced by 100% acetone. The dehydrated samples were infiltrated with low viscosity resin (Agar Scientific Ltd, Stansted, UK) as follows: 1/3 volume resin and 2/3 volume acetone for 15 min, 1/2 volume resin and 1/2 volume acetone for 30 min, 2/3 volume resin and 1/3 volume acetone for 3 h, and pure resin overnight. Subsequently, samples were transferred in resin-filled tubes and polymerized in an oven at 65 °C for two days.
Ultrathin sections (90 nm) of the grains were cut by using an ultramicrotome LEICA EM UC7 (Wetzlar, Germany) and a diamond knife type ultra 45° (DIATOME Ltd., Switzerland), collected on copper grids (300 mesh) and stained with 2.5% gadolinium triacetate for 30 min and 3% lead citrate for 8 min. Grids were analyzed at 120 kV with a TEM Zeiss Libra 120 equipped with a LaB6 (Lanthanum-Hexaboride) cathode. Images were acquired using a side port camera Morada G2, 11 MP (Soft Imaging System GmbH, Münster, Germany) and a bottom mount camera Sharp:eye TRS (2 × 2 KP) with iTEM software and processed using Adobe Photoshop CS5.

Electrolyte leakage (EL) assays
EL assays were performed as described previously (Shi et al., 2012;Jiang et al., 2017). Five grains from each development stage (6, 10, 12 and ≥ 20 DAP) were collected in 15 mL tubes containing 8 mL of deionized water, and the electrical conductivity (EC) was measured (S0). The samples were gently shaken at 22 °C for 15 min, and the resulting EC was measured again (S1). Then, the samples were incubated at 100 °C for at least 15 min and shaken at 22 °C for another 20 min, and the resulting EC was measured as S2. EL was calculated as follows:

Multivariate statistics unravel proteome reorganization across multiple biological processes during barley grain development
To identify proteins that are localized at the endomembrane system and/or are functionally associated with the rearrangement of the endomembrane system in barley, grains of different development stages including 6, 10, 12 and ≥ 20 DAP were harvested as previously described (Shabrangy et al., 2018). We used liquid chromatography-mass spectrometry (LC-MS/MS) to identify a total of 3,005 proteins. To validate our in vivo label-free quantification data obtained by MS, we compared the relative intensity of the heat shock protein 70 (HSP70) to a semiquantitative Western blot using an anti-HSP70-specific antibody. Supplemental Figure 2A shows that the expression pattern of HSP70 over the four development stages is in accordance with protein quantification via MS-based proteomics (Supplemental Table 2). Thus, our biochemical data confirmed the results of our MS-based study and previously published results (Kaspar-Schoenefeld et al., 2016). Additionally, measured protein abundances were highly reproducible with an a erage Pearso 's correlation coefficients of > 0.95 between biological replicates (Supplemental Figure 2B). The performed Principal Component Analysis (PCA) supports these results and shows that samples grouped and clustered in a stage-specific manner ( Figure 1A).
A total of 1,822 out of 3,005 were quantified in at least 9 out of the 12 analyzed biological samples (Supplemental Table 2). To determine the proteins that were significantly changed along the grain development, we applied a one-way ANOVA analysis, corrected with a permutation-based false discovery rate (FDR) (p < 0.05). Among the 1,822 quantified proteins, the abundance of 1,544 proteins was significantly changed (Supplemental Table 2), indicating that most proteins were distinctly regulated in their expression during early grain developmental stages. To assess the proteome dynamics during barley grain development, both PCA and a hierarchical bi-clustering analysis (HCA) were performed. Thereby identical clusters of stage-specific groups related to development stages of the barley grain were identified ( Figure 1A and 1B). An investigation of the PCA results shows that PC1 (76.7% of variance) separated the sample based on the different development stages, particularly between the early (6 DAP) and late development stages (≥ 20 DAP). PC2 (13.7% of variance) was rather more involved in the early development stage discrimination. Protein loadings on each principal component are indicated in Supplemental Table 2. As expected, SSPs were among the highest loadings on PC1. Interestingly, most of the detected changes occurred at ≥ 20 DAP ( Figure 1B).
To refine our protein expression pattern analysis, unsupervised clustering was performed with GproX software to partition the temporal profiles of 1,544 significantly changed proteins measured at all time points. Six clusters based on a fuzzy-mean clustering process (Rigbolt et al., 2011) were detected ( Figure 1C and Supplemental Table 2): Proteins presenting a higher expression level at 6 DAP than at ≥ 20 DAP belong to Clusters One, Five and Six. Those three clusters account for most of the significantly changed proteins (in total, 962 proteins). It is known that endosperm development involves cell division, cellular differentiation events and the deposition of SSPs between early, late and mid-development (Sabelli and Larkins, 2009;Zhang et al., 2016). These findings were confirmed by toluidinestained sections prepared at 6, 12 and ≥ 20 DAP that revealed fully cellularized endosperm including three aleurone cell layers at 6 DAP (Supplemental Figure 3A). Cluster Four accounts for 126 proteins and shows its highest protein abundance between 10 and 12 DAP, which corresponds to mid-stage, where the differentiation of aleurone, subaleurone, and starchy endosperm is finalized ( Figure 1C and Supplemental Figure 3A). Finally, the 456 proteins associated with Cluster Two and Three exhibit higher expression levels at ≥ 20 than at 6 DAP, corresponding to the end of mid-stage and beginning of the late stage, where the accumulation of storage reserves can be observed in the endosperm ( Figure 1C and Supplemental Figure 3). Using LC-MS/MS, all 20 identified SSPs showed significantly higher abundance at ≥ 20 DAP (Supplemental Figure 3B, C).
Toluidine staining of sections prepared at 6, 12 and ≥ 20 DAP confirmed protein accumulation starting at 12 DAP and increasing at ≥ 20 DAP in subaleurone as well as in the starchy endosperm (Supplemental Figure 3A). Additionally, Cluster Two and Three contain HvPDIL1-1 and HINs (Roustan et al., 2018;Shabrangy et al., 2018). From the development stage perspective, specific trends associated with biological processes could be observed. For instance, most of the proteins related to DNA, RNA, protein synthesis, photosynthesis, and starch degradation metabolisms are highly represented in Clusters One, Five and Six. Contrary, proteins related to SSPs, Late-Embryogenesis-Associated (LEA) proteins, starch synthesis, and fermentation were preferentially associated with Clusters Two and Three (Supplemental Table 2).

Identification of molecular regulators for the endomembrane system during barley grain development
We examined our proteomics dataset for proteins that are functionally associated with a specific subcellular endomembrane compartment (designated as compartment-specific proteins ) as well as for spatio-regulated trafficking regulators associated with several organelles within the plant cells (designated as trafficking regulators ). Based on a BLAST search, 95 proteins could be identified in total, where 60 compartment-specific proteins (secretory pathway, peroxisome, plasma membrane (PM), sorting, transport, degradation, and vacuolar processing) and 35 trafficking regulators were defined (dynamins, SNAREs, disulfide-generating enzyme and-carrier, ATPase, GTPase, GTPase-activating protein, RAB regulator, and GTP binding protein) within the plant cell (Table 1). The abundance of all identified endomembrane related proteins at 6, 10, 12, and ≥ 20 DAP were visualized by a heat map categorizing the protein expression pattern based on Pearson correlation (Figure 2).

16
Compartment-specific proteins and trafficking regulators were categorized in pink and blue, respectively ( Figure 2).
Proteins could be associated with the three main development stages (Figure 2, Supplemental Figure 4A): 39 proteins present a higher expression during the development stage I (green cluster), 15 proteins presented an expression peak at the development stage II (yellow cluster), and 40 proteins are associated to the development stage III (red cluster). Within each stage, compartment-specific proteins and trafficking regulators were categorized (Table 1, Supplemental Figure 4A, B). Taken together, our data identified molecular regulators for the endomembrane system in developing barley grain.

Development stage I contains proteins of high abundance associated with endocytosis and cytoskeleton, plasma membrane proteins, and ATPases
Among the 39 proteins identified in the development stage I, proteins related to the secretory pathway (ER, Golgi, Golgi -ER), plasma membrane, sorting pathway (endocytosis, ESCRT), transport (vesicle-mediated transport, cytoskeleton), degradation (Autophagy-related protein 3) and to vacuolar processing as well as trafficking factors (dynamins, SNAREs, disulfidegenerating enzymes and carriers, ATPase, GTPase, GTPase-activating protein and GTP binding protein) were enriched at 6 and 10 DAP ( Figure 3A).
More precisely, plasma membrane-associated proteins such as putative voltage-gated potassium channel subunit beta (F2CWF3), Membrane steroid-binding protein 1 (F2CS48), Nicastrin (M0UEQ6) and Proton pump-interactor 1 (A0A287RSX4) present decreasing abundance in concomitance to proteins associated to the endocytosis processes such as dynamins (A0A287GK50, A0A287N3M7, A0A287W654, A0A287MCV3), tubulin (A5CFY5, A5CFY9), actin (A0A287FFF9), myosin (A0A287MS88), EH2 (A0A287Y199) and CLC2 (A0A287XZU3). In line with the constant need of energy necessary for endocytosis and plasma membrane remodeling processes, three vacuolar-ATPase subunits were identified (A: M0XFC8, B1: F2DCK0, E: A0A287L8C5). All three subunits showed a decreased expression, indicating an acidification process at the early development stage (Supplemental Table 2). The identified CASP and Coatomer subunit delta (RET2p; COPI) were highly abundant at 6 DAP, pointing to an active retrograde transport as CASP is involved in tethering to Golgi cisternae of a subpopulation of COPI vesicles presumed to be involved in retrograde transport to the ER (Malsam et al., 2005;Latijnhouwers et al., 2007). Using STRING, a functional association between dynamins and cytoskeleton-related proteins was visualized ( Figure 3B), both necessary for plant endocytic processes (Samaj et al., 2004;Samaj et al., 2005). Additionally, STRING revealed a functional correlation between several ATPases that were abundant at 6 and 10 DAP ( Figure 3B). The functional association between Golgin candidate 5, RABH1b and further GTPases shown by STRING points to an active protein sorting in the trans-Golgi or at the TGN (Latijnhouwers et al., 2007). Taken together, the proteins that are highly abundant at 6 DAP point to the necessity of endocytosis, acidification, and sorting during the development stage I.
To investigate whether endocytosis is most active in this stage, we first performed time-lapse FM4-64 live cell imaging to track the endocytic pathway on sections prepared from 6 and 12 DAP. The endocytic tracer FM4-64 could be observed after 10 min at vesicles at the plasma membrane and intracellularly at membranes at 6 DAP, indicating endocytosis ( Figure 4A). However, only the plasma membrane was labelled at 12 DAP ( Figure 4A), indicating less endocytic activity at this stage. In addition to proteins involved in endocytosis, cytoskeletonrelated proteins were highly abundant at 6 and 10 DAP (Figure 3). Interestingly, immunofluorescence studies of actin and tubulin with the antibodies anti-actin and antitubulin-α showed strong signals within PBs at 6 DAP, becoming weaker at 12 and ≥ 20 DAP, respectively ( Figure 4B-D). Interestingly, a faint actin signal was observed at the plasma membrane at ≥ 20 DAP, whereas the signal of tubulin at the plasma membrane was strong at 6 DAP but was reduced at 12 and ≥ 20 DAP ( Figure  To determine the time-dependent subcellular distribution of ATPase, we used fluorescence microscopy with anti-V-ATPase subunit epsilon antibody on sections at 6 and ≥ 20 DAP ( Figure  4G). Whereas at 6 DAP a punctate structure was observed at the plasma membrane and a strong signal within PBs, a weaker labelling could be observed at the periphery of starch granules and within PBs at 12 DAP (Supplemental Figure 4B). At ≥ 20 DAP, an additional strong signal appeared in aleurone, whereas the signal was weak within PBs and at the periphery of starch granules in the starchy endosperm ( Figure 4G). No signal could be detected in the negative controls for immunofluorescence for 6, 10, 12 and ≥ 20 DAP (Supplemental Figure  5B). Overall, our proteomics and in situ microscopic results point to a high abundance of proteins involved in endocytosis, cytoskeleton regulation and acidification of PBs at early barley grain development stage.
Using STRING, a functional association between the ESCRT TOL3, clathrin, and an auxilinrelated protein appeared ( Figure 5B). TOLs have putative clathrin-binding motifs (Korbei et al., 2013) and auxilin-related protein 1 functions as clathrin-uncoating factors (Adamowski et al., 2018), and both were described to be involved in endocytosis. Interestingly, proteins involved in sorting, transport, and degradation have been already identified in the development stage I.

Compartment-specific proteins and trafficking regulators participating in storage protein targeting, transport, and deposition are accumulating at development stage III
In total 40 proteins were associated with development stage III, most of them highly abundant at ≥ 20 DAP. Interestingly, 11 proteins involved in the secretory pathway (e.g., COPII/COPI), and 9 proteins that function in the protein sorting (e.g., VSR1, retromer, and ESCRT) were identified ( Figure 6A). Additionally, proteins involved in degradation (autophagy-related 8c, vacuolar processing enzyme1) were highly abundant at ≥ 20 DAP. Among other proteins, one SNARE was identified as well as 11 trafficking regulators associated with the ATPase, GTPase or the GTP-binding group. Strikingly, endomembrane-associated proteins characterizing the stage III presented the highest interconnectivity within the STRING database compared to stages I and II ( Figure 6B).
As expected, important ER markers such as the disulfide isomerase protein group (HvPDIL1-1 and ERO1) were found in this group, being in line with the accumulation of SSPs, described within Supplemental Figure 2. The STRING network indicates that the group of ER markers is linked with proteins associated with the secretory pathway and their associated factors such as a RAB regulator GDI1/2 (A0A287FZ46). Secretory proteins are synthesized on cytoplasmic ribosomes and finally translocated across the ER membrane through a channel formed primarily by the Sec61protein. There, chaperones support the proteins reaching their correct native conformations, which subsequently exit the ER and move to the Golgi apparatus. Figure  6B shows the central core formed by the coat protein complex (COPII) together with COPI. COPII is responsible for the ER-Golgi transport and is composed of Sar1, Sec23, Sec24, Sec13, and Sec31, while COPI regulates Golgi-ER transport and is represented here by coatomer subunits (A0A287KUM9 and F2E4V3) as well as other regulators (A0A287T0X1 and A0A287HI31). Interestingly, this core set of proteins are functional associated with diverse proteins including ESCRT related SKD1 protein (A0A287K2S5), autophagy-related protein 8c (M0YZY8), Sorting nexins, as well as regulatory factors related proteins (GTPases, GTP binding proteins).
However, VSR1, which participates in vacuolar sorting of 12S globulins and 2S albumins in Arabidopsis seeds (Zouhar et al., 2010), was strongly localized in the aleurone layer instead of the starchy endosperm at ≥ 20 DAP (Supplemental Figure 6A), indicating predominantly functional activity in the aleurone layer during development or during germination. This observation is supported by the fact that programmed cell death (PCD) of the starchy endosperm starts in development stage II around 15 DAP (Dominguez and Cejudo, 2014). Here, our electrolyte leakage (EL) assay, which indicates cell membrane integrity damage caused by PCD (Jiang et al., 2017), showed a strong increase between 12 and ≥ 20 DAP (Supplemental Figure 6B), supporting that PCD happens in the starchy endosperm at development stage II and III and subsequently indicates tissue-specific functions of proteins.
Some of the identified functional groups clustered within each development stage of barley grain development, suggesting a temporal regulation of the identified mechanisms. For example, vacuolar processing enzymes (VPEs) are known to be involved in PCD (Hara-Nishimura et al., 1998;Hatsugai et al., 2015;Radchuk et al., 2018) as well as in processing seed storage proteins in seeds (Shimada et al., 1994;Shimada et al., 2003;Wang et al., 2009). Similarly, ESCRT-related proteins contained members of each development stage.

ESCRT-III HvSNF7 associates to MVBs at development stage I and both localize to and within PBs at development stage II and III
Proteomic analysis identified eight proteins related to ESCRT-0, ESCRT-III and SKD1 complex. Interestingly, they showed different expression patterns and subsequently belonged to different clusters ( Figure 7A). Barley homologs of TOL1, TOL2, TOL8, and VPS20.1 showed a high abundance at development stage I, but a continuous decrease over the development of barley grain ( Figure 7A). TOL3, SNF7.1, and SNF7.2 exhibit an expression peak during development stage II, while VPS4 continuously accumulated during the early grain development up to development stage III.
To gain insight into to expression behavior of identified ESCRT members, we analyzed the transcript levels of TOLs, SNF7s, VPS20.1, and VPS4 during barley grain development by RT-qPCR (Supplemental Figure 7). RNA was isolated from whole grains harvested at 6, 10, 12 and ≥ 20 DAP, and the previously characterized most stable genes were used to normalize the ESCRT transcripts, as described by Shabrangy et al. (2018). A high correlation between transcript and protein abundances for all identified ESCRT members was observed, except for VPS4. Even though the VPS4 transcript follows the same trend as SNF7 transcripts and proteins, 20 VPS4 protein increased, suggesting a delay of the response or a fine-tuning of VPS4 translation (Supplemental Figure 7). These results indicate that the expression of ESCRT is temporally regulated during barley grain development.
ESCRT originally refers to a protein-protein interaction network in yeast and metazoan cells that coordinates sorting of ubiquitinated membrane proteins into intraluminal vesicles (ILVs) of the MVB (Katzmann et al., 2001;Babst et al., 2002b;Babst et al., 2002a). So far, no observations of MVBs have been performed in barley endosperm tissues. Thus, transmission electron microscopy (TEM) analyses of 12 DAP grains were performed to study the localization of MVBs. Interestingly, an MVB was identified in proximity to a fused PB ( Figure 7B).
In order to study in vivo the subcellular localization of MVBs, we generated the transgenic barley line YFP-2xFYVE that functions as a fluorescent PtdIns3P-specific biosensor that can act as an endosomal marker (Samaj et al., 2004;Vermeer et al., 2006). Confocal live cell imaging revealed faint, punctate structures that were associated with PBs at 10 DAP ( Figure 7C). Treatment with the phosphatidylinositol-3 (PI-3) kinase inhibitor wortmannin (WM), which targets MVBs and leads to swelling of these compartments (Tse et al., 2004;daSilva et al., 2005), indicated that the YFP-2xFYVE structures were sensitive to WM at 10 DAP ( Figure 7C).
Specifically, ESCRT-III is known to be necessary for membrane remodeling that drives the biogenesis of MVBs (Hurley and Emr, 2006;Hurley and Hanson, 2010). Recent electron tomography studies in A. thaliana revealed that intraluminal vesicles form as large networks of interconnected or concatenated vesicles. AtSNF7 was detected in the intervesicle bridges, suggesting that ESCRT-III proteins remain trapped inside the vesicle cluster in MVBs and are finally delivered together with the cargo into the vacuole (Yoshida et al., 2013;Buono et al., 2017;Otegui, 2018). We asked whether HvSNF7 is localized at MVBs in barley endosperm. Therefore, the transgenic line SNF7.1-mEosFP driven by the hordein promotor p6U was generated to study the localization in vivo. Confocal live cell imaging revealed faint, small punctate signals that were associated with PBs at 12 DAP ( Figure 7D). To determine the localization of HvSNF7.1 in more detail, we treated the p6U::SNF7.1-mEosFP grains with WM. Live cell imaging of 12 DAP old sections indicated swelling of SNF7.1-mEosFP positively labelled structures, indicating that SNF7.1 is localized at MVBs ( Figure 7D).
We next asked if the localization of HvSNF7 is temporally regulated. To answer this question, we first studied the localization of HvSNF7 in vivo at the early development stage using the transgenic line p6U::SNF7.1-mEosFP. Confocal live cell imaging of the p6U::SNF7.1-mEosFP transgenic line revealed vesicular structures and few agglomerations around PBs at 6 and 10 DAP ( Figure 8A). AtSNF7.1 is known to form homodimers and thus can possibly lead to agglomerations (Richardson et al., 2011). Indeed, BiFC and Y2H analyses revealed homodimerization of HvSNF7.1 (Supplemental Figures 8A, B). As live cell imaging in developing endosperm is limited to mid-and late development stages (Ibl et al., 2014), we used an anti-SNF7 antibody to analyze the localization by immunofluorescence microscopy.
Immunofluorescence analyses of sections prepared from all development stages confirmed observations of small vesicular structures and few agglomerations at 6 DAP (Supplemental Figure 7C) and strong signals within PBs at 12 and ≥ 20 DAP ( Figure 8B). The punctured structures that could be observed inside the PBs possibly appeared by fusing of several smaller PBs. (Figure 8B). Additionally, a weak signal could be observed at the periphery of starch granules ( Figure 8B) at 12 and ≥ 20 DAP. Interestingly, faint signals of small vesicular structures could be observed at single PBs, indicating that the localization of SNF7 between 10 and 12 DAP is fluent. These findings indicate that HvSNF7 is associated with single PBs and is localized inside fused PBs.

Proteomics and in situ microscopic analyses enable the mapping of the endomembrane system of developing barley endosperm
Many proteomic analyses in barley grain allow us to identify and understand the molecular composition of developing barley grain (Finnie et al., 2002;Finnie et al., 2011;Mock et al., 2018). However, less is known about proteins regulating the endomembrane system in developing barley grains. Only one of the proteins identified in our study, Membrane steroidbinding protein 1 (F2CS48), was recently characterized in barley . Recent studies suggested that the endomembrane system and SSP trafficking are co-spatiotemporally regulated in developing barley endosperm between 8 and 12 DAP Recent studies have shown that the endomembrane system and SSP trafficking are spatio-temporally regulated in developing barley endosperm between 8 and 12 DAP (Ibl et al., 2014;Roustan et al., 2018). This time-dependent regulation was confirmed by PCA analysis of our proteomics data, which revealed stage-specificity of protein expression during grain development.
A detailed analysis of the 95 proteins associated with the endomembrane system associated them with three development stages: I, II and III. Altogether with previously published work, correlation of proteomics data with in situ microscopic analysis allowed us to provide the first temporal map of endomembrane-related proteins involved in stage-specific regulation of different endomembrane processes (Figure 9): in development stage I, prolamin and glutelin RNAs are localized to two subdomains of the cortical endoplasmic reticulum (ER), and targeted to the Golgi or to PBs by RNA-binding proteins and the cytoskeleton (Hamada et al., 2003a;Hamada et al., 2003b;Crofts et al., 2004;Crofts et al., 2010;Tian et al., 2019). Additionally, development stage I displays endocytic activities involving plasma membrane rearrangement. Such processes have been shown to be associated with specific cytoskeleton dynamics: besides the necessity of the plant cytoskeleton in vesicle trafficking and organelle movement (Thomas and Staiger, 2014), actin is required for the auxin-dependent convolution and deconvolution of the vacuole (Scheuring et al., 2016).
Specifically, SNAREs, actin and its associated motor protein myosin shape the vacuole by actindependent constrictions. While the protein accumulation starts at 6 DAP with observable PBs at 8 DAP, the size of PSVs decreases between 8 and 10 DAP. As we could observe actin as well as tubulin associated with PBs, we conclude that the cytoskeleton proteins present at early stages of barley endosperm development regulate the PBs formation/trafficking and the size of PSVs. Concurrent with the endocytosis activity, microscopic analyses indicate an increase of the acidity within PBs. During development stage II, processes associated with the sorting system seem to be highly active. Recently, MVBs were discussed to be taken up into the vacuole by autophagy (Abirached-Darmency et al., 2012). Additionally, studies in A. thaliani root cells have shown that central vacuoles were derived from MVB to small vacuole transition and subsequent fusions of small vacuoles (Cui et al., 2019).
Thus, we propose that MVBs loaded with HvSNF7 possibly contribute to PSV rearrangement events, resulting in large PSVs containing PBs associated with MVBs and HvSNF7, as it is known that PBs are taken up by PSVs at 10 DAP (Ibl et al., 2014). Finally, during development stage III, PBs, MVBs, and HvSNF7 were found in proximity to the protein matrix at the periphery of starch granules, while the presence of the cytoskeleton was reduced.
Here, we point out the necessity to correlate proteomics data with microscopic analysis for reasons of spatio-temporal specificity. Although the bulk of the barley grain is mainly occupied by the starchy endosperm, we cannot exclude that identified proteins can reflect different spatial activity. Thus, proteomics from dissected sections would be useful to identify tissuespecific proteins. In order to refine our analysis to a spatio-temporal level, we compared our dataset with previously published LMD-based proteomics analysis (Roustan et al., 2018). However, only four out of our 95 identified endomembrane related proteins could be detected (F2DY31, Actin-depolymerization factor 4; A0A287NWD9, HvPDIL1-1; A0A287RLW1, HvPDIL2-2; and F2CQ27, GTP-binding nuclear protein). This indicates a detection range limitation of proteomic analyses of dissected samples prepared for laser microdissection. Consequently, a combined analysis of proteomics together with microscopy appears as the most appropriate strategy.
For example, VPE4 was described to be most expressed in the pericarp of barley between 8 and 10 DAP (Radchuk et al., 2011) and to be necessary for programmed cell death execution in the developing pericarp (Radchuk et al., 2018), thereby being responsible for the grain size, starch and lipid content. As our proteomics data identified VPE4 to be most abundant at 6 and 10 DAP and subsequently to be grouped into development stage I, we assume the main function of VPE4 is in the pericarp. In contrast, VPE2b, c, and d, which are most abundant between 10 and 12 DAP and were grouped to development stage II, have been described to be involved in nucellar PCD (Radchuk et al., 2011). Additionally, besides being expressed and putatively involved in sorting processes during grain development, VSR1 was mostly expressed at ≥ 20 DAP, but mainly localized in the aleurone layer, concluding that VSR1 could be putatively involved in recycling processes during germination.
It is worth mentioning that our data provide comprehensive coverage of the endomembraneassociated proteins involved in the rearrangement of the endomembrane system and protein trafficking. However, we cannot entirely exclude additional proteins involved in these processes or organelle-specific proteins involved in other processes. For example, even though the importance of the Golgi in protein trafficking was shown previously in wheat in the transition between stage I and II (Tosi et al., 2009;Arcalis et al., 2010;Tosi et al., 2011), only two Golgi-associated proteins could be found (Golgin 5 and CASP), both grouped to development stage I.
In order to define if Golgi-associated proteins are underrepresented in our dataset, we compared our data with previously identified proteins in A. thaliana (Dunkley et al., 2006). Interestingly, we could explicitly identify more than 10 Golgi-resident proteins such as cell wall synthesis-associated proteins (e.g., UDP-glucuronic acid decarboxylase, A0A287H8Z0), which are parts of the glycosylation processes (Supplemental Table 2). It is possible that these identified proteins are active in aleurone, as previous data characterized the barley aleurone N-glycoproteome, in which numerous N-glycosylation sites were identified that play key roles in protein processing and secretion (Barba-Espin et al., 2014).

How do ESCRTs and MVBs contribute to PB formation?
Although MVBs were reported to be responsible for targeting proteins to the storage vacuole in maize aleurone cells (Tian et al., 2007;Reyes et al., 2011), their existence and localization during cereal endosperm development has so far not been investigated. Here, our proteomics data revealed VPS29, SNX1, and SNX2b, proteins from the retromer complex, mediating the recycling pathway at the TGN and at the MVB (reviewed by (Paez Valencia et al., 2016). Additionally, our proteomics data identified eight ESCRT proteins (four from ESCRT-0, three from ESCRT-III and one from the SKD1 complex) quantified over all three development stages. MVBs have been suggested to arise by TGN maturation (Scheuring et al., 2011;Singh et al., 2014) with the support of Rab GTPases (Cui et al., 2016). It is worth mentioning that several Rab GTPases were detected over all three different development stages, possibly supporting MVB maturation.
Additionally, ESCRT are known to be necessary to drive the formation of ILVs in MVBs (Katzmann et al., 2001;Babst et al., 2002b;Babst et al., 2002a). Only few studies concerning ESCRT proteins are described in cereals: in maize, supernumerary aleurone layer1 (Sal1), that encodes the maize homolog of VPS46/CHMP1 (ESCRT-III associated), was found to restrict aleurone cell identity to the outer cell layer of endosperm (Shen et al., 2003). SAL1 maintains the proper plasma membrane concentration of DEFECTIVE KERNEL1 (DEK1) and CRINKLY4 (CR4), both involved in aleurone cell fate specification, by internalization and degradation of SAL1 positive endosomes (Tian et al., 2007). The rice AAA ATPase LRD6-6, which is homologous to the AAA ATPase VPS4/SKD1, was identified as an interactor with OsSNF7b/c (Os06g40620/Os12g02830) and OsVPS2.2 (Os03g43860), supporting its putative MVB-mediated vesicular trafficking function (Zhu et al., 2016). OsLRD6-6 is described to be localized at MVBs, to be required for MVBs-mediated vesicular trafficking and to inhibit the biosynthesis of antimicrobial compounds for the immune response in rice (Zhu et al., 2016).
Recently, the overexpressed ESCRT-III-associated component HvVPS60 was shown to be involved in protein targeting in developing barley endosperm (Hilscher et al., 2016). Here, seven of the identified ESCRT proteins were most expressed at development stage I and II, indicating a possible involvement in MVB body formation. The high abundances of the ESCRT-0 proteins, HvTOL1, HVTOL2 and HvTOL8 in the development stage I indicate early steps of cargo endocytic events to the vacuole: originally, nine Tom1 (target of Myb1) proteins with a domain structure similar to the VHS domain of ESCRT-0 were identified and it was speculated that these proteins are responsible to load the ESCRT machinery (Winter and Hauser, 2006). Tom1 proteins were further characterized as members of the A. thaliana TOL family (TOM1-LIKE), which are able to bind ubiquitin directly and participate in the endocytic trafficking of plasma membrane proteins, such as the auxin efflux facilitator PIN2 (Korbei et al., 2013 Our proteomics, RT-qPCR, microscopic and biochemical analyses showed a highly similar protein expression and transcription behaviour of both HvSNF7.1/2, which is reasonable as interaction studies showed heterodimerization (Richardson et al., 2011). Snf7/VPS32 induce membrane curvature at MVBs by assembling into long spiral filaments and are discussed to be involved in corralling the ESCRT cargo at the vesicle bud (Henne et al., 2012;Shen et al., 2014;Chiaruttini et al., 2015). Given the transgenic lines 35S:YFP-2xFYVE and p6U::SNF7.1-mEosFP, which showed punctate structure and is most properly labelling MVBs, we were able to detect MVBs around PBs at early and within fused PBs at later barley endosperm development stages.
Taken together, our findings suggest a putative role of MVBs and ESCRT proteins for targeting proteins to PBs. As proteins involved in protein sorting by MVBs were detected as most abundant at each specific development stage, we conclude that protein sorting by MVBs is a constant factor during endosperm development, thereby constantly delivering different cargos to PBs.

Cargo-driven protein trafficking versus mechanically-stressed protein dynamics in barley endosperm
Grain development is marked by desiccation of the seed and the onset of dormancy. In barley, the starchy endosperm cells do not survive this desiccation and undergo PCD during grain development (Young et al., 1997), starting around 16 DAP (development stage II) (Dominguez and Cejudo, 2014). Different hallmarks such as DNA fragmentation, vacuolar tonoplast disruption and activity of caspases starting and executing apoptosis were described for starchy endosperm PCD (reviewed in (Dominguez and Cejudo, 2014;Tran et al., 2014). Besides VPEs, caspase-6-like proteolytic activity with VEIDase activity has been described to be active in autophagosomes and is involved in developmental programmed cell death in barley (Boren et al., 2006). Our proteomics results revealed two autophagy-related proteins (Atg3, Atg8c) grouped into development stage I and III, respectively. VPE1, which was described to be most expressed at late development stages in barley (Radchuk et al., 2011), grouped to development stage III. Studies in rice and in tobacco suggested that VPE1 is necessary for the maturation of rice glutelins and for PCD, respectively (Wang et al., 2009;Hatsugai et al., 2015). Additionally, crosstalk between MVB, ESCRT and autophagy was described, as the A. thaliana DUB, AMSH1 (Associated Molecule with the Sh3 Domain Of STAM1), and the ESCRT-III subunit VPS2.1 are important for autophagic degradation and autophagy-mediated physiological processes (Katsiarimpa et al., 2013).
Two classes of PCDs are known: vacuolar cell death and necrosis (reviewed in (van Doorn et al., 2011). Both have different morphological features such as formation of actin cables, rupture of tonoplast, accumulation of autophagosomes and small lytic vacuoles for vacuolar cell death and early rupture of the plasma membrane and shrinking of the protoplast for necrosis. Interestingly, lytic vacuoles have only been identified in the aleurone layer of barley grains (Swanson et al., 1998;Swanson et al., 2003). Given recent published data, where it was shown that the tonoplast degenerates in the starchy endosperm during development (Ibl et al., 2014) and our proteomics and microscopic data, where autophagy-related proteins, VPEs, acidification of PBs and less cytoskeleton related proteins are detected at development stage III, we identified putative key players for PCD in developing starchy endosperm already at development stage II. However, SSPs accumulate strongly at late development stages, raising the question of how proteins can be transcribed and translated during PCD. Is it possible that in the development stage II protein dynamics is driven by mechanical stress?
To conclude, our proteomic approach combined with in situ microscopy, which focused on endomembrane associated proteins isolated from four different stages of barley grain development, provide only a snapshot of the endomembrane system dynamics. Future studies will be needed to verify the temporal protein-protein interactions identified by the STRING analysis. Furthermore, it would be of interest to study the endomembrane system remodeling under abiotic stress conditions that are likely to affect the SSPs synthesis and/or the production of recombinant proteins. Finally, experimental proof of the involvement of cytoskeleton-related proteins, MVBs and ESCRT in sorting proteins to the PBs, ideally obtained by investigating mutant barley lines impaired in ESCRT function using proteomics and in situ microscopy, has yet to be provided. Nevertheless, our identified proteins that are associated with the endomembrane system are useful targets for genetic engineering to modulate SSPs accumulation and/or to improve the production of recombinant proteins.