A cell fractionation and quantitative proteomics pipeline to enable functional analyses of cotton fiber development

SUMMARY

Cotton fibers are aerial trichoblasts that employ a highly polarized diffuse growth mechanism to emerge from the developing ovule epidermis. After executing a complicated morphogenetic program, the cells reach lengths over 2 cm and serve as the foundation of a multi‐billion‐dollar textile industry. Important traits such as fiber diameter, length, and strength are defined by the growth patterns and cell wall properties of individual cells. At present, the ability to engineer fiber traits is limited by our lack of understanding regarding the primary controls governing the rate, duration, and patterns of cell growth. To gain insights into the compartmentalized functions of proteins in cotton fiber cells, we developed a label‐free liquid chromatography mass spectrometry method for systems‐level analyses of fiber proteome. Purified fibers from a single locule were used to fractionate the fiber proteome into apoplast (APO_T), membrane‐associated (p200), and crude cytosolic (s200) fractions. Subsequently, proteins were identified, and their localizations and potential functions were analyzed using combinations of size exclusion chromatography, statistical and bioinformatic analyses. This method had good coverage of the p200 and APO_T fractions, the latter of which was dominated by proteins associated with particulate membrane‐enclosed compartments. The apoplastic proteome was diverse, the proteins were not degraded, and some displayed distinct multimerization states compared to their cytosolic pool. This quantitative proteomic pipeline can be used to improve coverage and functional analyses of the cotton fiber proteome as a function of developmental time or differing genotypes.

Significance Statement

A cell fractionation and quantitative proteomics pipeline has been developed to analyze cotton fiber proteins abundance and localization.

INTRODUCTION

Gossypium hirsutum is the most widely grown cotton, ranking among the world's most economically important crop species, and provides the largest source of renewable textiles (over $6 billion in raw product, FAOSTAT, http://www.fao.org). This global textile economy is based solely on the growth and morphogenesis of individual fiber cells that emerge from the developing seed coat. At about 1 day before flower opening and anthesis, conserved transcriptional control pathways drive subsets of epidermal precursors into the trichoblast fate (Szymanski et al., 2000; Wang, Wang, et al., 2020). Subsequently, the individual cells execute a complicated and poorly understood morphogenetic program (Delmer, 1999; Haigler et al., 2012; Kim & Triplett, 2001; Qin & Zhu, 2011) that includes an initial tapering phase that reduces fiber diameter (Applequist et al., 2001; Graham & Haigler, 2021; Yanagisawa et al., 2022). Daily phenotyping of fiber traits and cell wall molecular features reveal a progressively slowing fiber elongation rate during the early stages of development (Wilson et al., 2024). Well after fiber elongation slows, fibers execute a multi‐phased transition to the synthesis of a cellulose‐rich secondary cell wall that eventually consumes much of the cell cytoplasm prior to cell death, desiccation, and the opening of the mature boll. Domestication and breeding have generated modern cultivars with narrow fibers, extended developmental phases of cell elongation, and twisted morphologies that enable dried fibers to be spun into a usable product. Because the genetic diversity in the cottons is relatively low, future opportunities reside in the combined use of genetic engineering and breeding to improve fiber traits and yield.

Even when dealing with individual cells this is a complicated engineering challenge. Fiber development includes complex developmental interactions among central metabolism and turgor regulation (Ruan et al., 2001; Tuttle et al., 2015). The biomechanics of how turgor pressure interacts with the material properties of the growing trichoblast cell wall is not completely understood. Aerial trichoblasts, like most plant cells (Gu & Rasmussen, 2022), employ a diffuse growth strategy that couples persistent synthesis with assembly of a cellulose‐rich wall throughout the expanding cell surface (O'Kelley, 1952; Ryser, 1977; Yanagisawa et al., 2022). This enables the cell to not only maintain the wall strength needed to avoid cell rupture by elevated tensile stress (Fujita et al., 2013) but also maintain anisotropic wall material properties that enable polarized elongation with minimal radial (Baskin, 2005; Graham & Haigler, 2021; Yanagisawa et al., 2015, 2022). A microtubule‐patterned cellulose synthesis system operates throughout fiber development to promote persistent axial elongation and cell diameter control at the cell apex (Seagull, 1986; Wilson et al., 2024; Yanagisawa et al., 2022).

The microtubule‐cellulose control module is just one part of a much broader network of cellular systems that dictate the rates and patterns of cell expansion. As just one example, matrix polysaccharides like pectins, homogalacturonans, and rhamnogalacturonans are major components of the growing cell wall (Avci et al., 2013; Swaminathan et al., 2024), and their material properties can have strong effects on the rates and patterns of growth in leaf trichoblasts (Yanagisawa et al., 2015) and cotton fibers (Wilson et al., 2024). Pectin biosynthesis, transport, and modification involve dozens of genes that must be tightly regulated during growth, so that the thickness and material properties of the wall are maintained during a growth phase that proceeds for weeks (Meinert & Delmer, 1977; Schubert et al., 1973). There is a similar level of complexity to mediate the transition to a very different physiology of the cell as it rearranges its metabolism, endomembrane systems, and glycosyl transferase repertoire to form the secondary wall (Hoffmann et al., 2021; Zhong et al., 2019).

There are emerging opportunities to discover key molecules and systems‐level interactions that govern fiber traits. Forward genetic screens are complicated by genetic redundancy in tetraploid cottons; however, G. hirsutum fiber morphology mutants have enabled discovery of cytoskeleton‐related genes that affect fiber length and twist (Wan et al., 2016; Zang et al., 2021; Zhang et al., 2021). Because fiber cells develop in a synchronous manner and tens of thousands of fiber cells emerge from a single developing seed coat (Bowman et al., 2001), it is possible to conduct transcriptomic and metabolomic analyses using either purified fibers or developing ovules of different genotypes to predict pathways and genes that underlie fiber quality (Qin et al., 2019; Rapp et al., 2010; Tuttle et al., 2015; Wang et al., 2015; You et al., 2023).

Cotton fiber proteomic analyses are especially valuable because they can provide direct measurements of homoeolog utilization, protein abundance, and post‐translational modifications in polyploid species (Hu et al., 2014; Soltis et al., 2016; Wang, Cheng, et al., 2020). Gel‐free workflows and the increased sensitivity of modern spectrometers have enabled the quantification of thousands, rather than dozens, of proteins in a typical experiment (Hu et al., 2013, 2015; Lee & Szymanski, 2021; Yang et al., 2008, 2023). Proteins with an altered abundance in response to domestication (Bao et al., 2011; Hu et al., 2013; Qin et al., 2017), hormone or inhibitor responses (Wang, Liu, et al., 2020; Yang et al., 2023), and developmental stages (Jiang et al., 2022; Mujahid et al., 2016; Zhou et al., 2019) have been identified. There are also growing opportunities to use quantitative proteomics and correlation profiling as a systems‐level phenotyping tool that analyzes multimerization state, localization, and protein complex composition (Lee & Szymanski, 2021; McBride et al., 2019).

This paper describes a robust single locule proteomic analysis pipeline that can be adapted to analyze developmental progressions and differences among genotypes. The workflow combines cell fractionation with label‐free shotgun proteomics to quantify relative protein abundance and generate crude subcellular localization estimates for thousands of proteins. Because of the central importance of cell wall remodeling (Avci et al., 2013; Pettolino et al., 2022; Yanagisawa et al., 2022) and developmentally regulated apoplastic sugar metabolism (Ruan et al., 2001) during fiber development, we included a gentle locule submersion step to target loosely bound extracellular proteins. Similar strategies to analyze the apoplast in other plant systems detect both soluble secreted proteins and those associated with lipid‐enclosed extracellular vesicles. We detected diverse classes of proteins in the apoplastic fraction that could not be explained by contamination by broken cells. Soluble apoplastic proteins were not proteolytic remnants of cytosolic proteins, and many displayed unique multimerization states compared to a cytosol‐enriched pool. Most apoplastic proteins were associated with lipid‐enclosed vesicles. The apoplastic vesicle‐associated proteome closely resembled that of crude microsomes; however, there were hundreds of proteins and protein complexes that were significantly enriched in the apoplastic fraction. We provide an example database in which well‐annotated protein data are searchable as a function of subcellular location and relative abundance.

RESULTS

Cell fractionation and quantitative proteomics of purified fibers

We sought to create a method that would enable crude estimates of subcellular localization and quantitative analysis of abundance in differing locales. Developing cotton fibers dissected from individual locules from a developing boll (Figure 1a). Almost all fibers on the isolated locules at 9 day post anthesis (DPA) were intact (Figure 1b) and high magnification scanning electron microscopy (SEM) revealed a highly textured cell surface dotted with smooth punctae that could reflect extracellular vesicle‐like particles embedded in the wall matrix (Figure 1b(4,5)). To increase protein coverage and analyze protein localization in a crude manner, a total apoplast subcellular fraction (APO_T) of loosely bound proteins was obtained after careful dissection and dunking of a single locule, then crude microsomal proteins in the high‐speed pellet (p200), and total soluble (s200) fraction enriched in soluble cytosolic proteins were prepared. Due to fragility of the developing fibers, we generated the APO_T fraction of loosely bound apoplastic proteins by gently transferring the dissected locule to a microsome isolation buffer (MIB) solution with gentle rocking for 10 min. When the locules were analyzed by confocal microscopy after staining with propidium iodide to detect broken cells, the samples before (water) or after the 10‐min incubation procedure were similar with little evidence for broken cells (Figure 1c), and the fraction was not dominated by broken cells. After the APO_T, fibers were manually dissected from the seeds, homogenized, filtered, low‐speed spun, and ultracentrifuged to generate a total soluble protein (s200) and a crude microsomal fraction (p200). This centrifugation protocol does not efficiently sediment large protein complexes like Rubisco (Aryal et al., 2014), and the washed crude microsome fraction has been shown to be highly enriched with true membrane‐associated proteins (McBride et al., 2017). Therefore, we are not clearing s200 of large protein complexes and p200 consists primarily of peripheral, integral, and luminal proteins.

Figure 1.

Discovery of apoplast‐enriched cotton fiber proteins via cell fractionation and quantitative mass spectrometry.

(a) Protein isolation method from each locule at 9 day post anthesis (DPA) boll for shotgun proteomics.

(b) Scanning electron microscopy analysis of developing cotton fibers at 15 DPA. (1) Scanning electron micrograph demonstrating intact fibers. Bar, 200 μm. (2–5) Zoom‐in views of fiber surface. Bar, 50 μm (2), 10 μm (3), 5 μm (4), and 2 μm (5). Boxed regions with pink dotted lines were zoomed in.

(c) Propidium iodide‐stained images of the intact locules before (Water) and after (microsome isolation buffer [MIB]) 10 min‐dipping in MIB via 20× confocal microscope. Bar, 50 μm.

(d) A focus on proteins with a differential accumulation in the APO_T (apoplast). Spectral count and precursor peptide ion intensity‐based quantification to predict proteins with an increased abundance in the apoplast.

The single locule protein yields of APO_T and p200 were sufficient for shotgun proteomics experiments (Figure 2a). To promote consistent locus ID assignments between annotated G. hirsutum genome annotations, a locus ID conversion table is provided in Data S1. Triplicate samples were processed for label‐free shotgun proteomics using previously established methods (Cox et al., 2014; Rudolph & Cox, 2019; Tyanova, Temu, & Cox, 2016; Tyanova, Temu, Sinitcyn, et al., 2016), and the raw LC–MS/MS search results (Data S1) and raw LC/MS files are available at PRIDE (accession code: PXD051721). To facilitate consistent protein identifications and quantitative comparisons across all samples, the nine LC/MS raw files from the cell fractions were searched simultaneously using MaxQuant and LFQ normalization to standardize protein intensities across samples (Cox et al., 2014; Tyanova, Temu, & Cox, 2016). However, s200 yield was low (Data S1a) as was protein coverage (Figure 2a).

Figure 2.

Proteome coverage, reproducibility, and compositional differences among the three cell fractions.

(a) Number of proteins reproducibly identified by MS/MS counts in the APO_T (1), p200 (2), and s200 (3) cellular fractions. Numbers in bold indicate reproducible proteins identified in two out of three replicates.

(b) Hierarchical clustering and heatmap showing relative protein expression (log₂‐transformed LFQ protein intensities) of the differentially expressed proteins in the cellular fractions. Columns correspond to the three replicates of the fractions. Proteins that were not present in the datasets were rendered with black background color.

(c) Identification of apoplast‐specific (APO_specific) proteins based on MS/MS count data across the replicates and cell fraction.

(d) Protein intensity‐based identification of apoplast‐enriched (APO_enriched) proteins compared to p200 and s200 fractions. Expression levels of the quantified APO_T proteins (the circle with blue background color) were compared to either p200 or s200 using two‐sample t‐tests at 5% false discovery rate (FDR) and twofold changes.

(e, f) Abundance analyses of dually localized APO_T proteins in p200 (e) and s200 (f). Log₂‐transformed LFQ protein intensities.

(g, h) Gene Ontology (GO) enrichment analyses of the three cellular fractions. The agriGO v2.0 Singular Enrichment Analysis tool (http://systemsbiology.cau.edu.cn/agriGOv2/index.php) was implicated at 5% FDR. The heat maps show molecular function (g) and biological process (h) categories of the over‐represented GO terms in the APO_specific and APO_enriched protein list. Significance levels of the over‐represented GO terms in APO_T, p200, and s200 protein lists, and the common protein list between APO_T and p200 were shown in the heat maps. GO terms that were not over‐represented in each dataset were boxed with white background.

Three thousand twenty‐six proteins were classified as present based on at least one unique peptide detected in two of the three replicates for a single cell fraction. The protein identifications and quantification data are provided in the Data S2. The number of proteins identified within each cell fraction was reproducible (Figure 2a). The APO_T and p200 had similar protein compositions, and s200 was most different due in large part to lower coverage (Figure 2b–d; Data S2). The s200 sample contained only 79 enzymes, and 45 of them were scattered throughout central metabolism and oxidative stress pathways (Data S2d,e), clearly improved methods are needed for better coverage of this compartment. The APO_T and p200 fractions contained thousands of proteins including enzymes, receptor kinases, transporters, cytoskeletal proteins, kinases, and transcription factors. In about 40% of the cases, the peptide data were sufficient to resolve homoeologs, while about 46% of ambiguously assigned to both homoeologs (Data S2a). Also, the quantitative data for each protein is provided both for spectral counts and LFQ‐normalized precursor ion intensities (Data S2a). The data can be sorted to view proteins based on their ranked estimated abundance within the cell fraction. The data can also be searched by the Arabidopsis best‐hit locus ID or by orthologous group number based on Phytozome V12 (Goodstein et al., 2011). As an example, cellulose synthases (CESAs), Korrigan/glycosyl hydrolase 9 (GH9), cellulose synthase interactor (CSI), companion of cellulose synthase (CC), and COBRA proteins are directly involved in patterned cellulose synthesis (Polko & Kieber, 2019). In our dataset, we detected cotton orthologs for 8 of these known players (Data S2f) and large families of tubulins and of microtubule‐associated proteins (MAPs) that affect the dynamics and organization of the cortical microtubule array (Buschmann et al., 2010). There are 56 members of different orthologous groups that relate to microtubules and may affect fiber morphogenesis in Data S2(f). We expect this method to generate broadly useful data on medium‐ to high‐abundance proteins in the cell that carry out functions influencing fiber traits.

There was considerable overlap in the molecular functions and biological processes that were predicted to be enriched in p200 and APO_T (Data S3a,b,d). The apoplast proteome in this study also had substantial overlap in composition of apoplastic proteomes reported previously (Data S4). When the cotton protein IDs were converted to their closest Arabidopsis or rice orthologs, the cotton APO_T contained 54% of the proteins detected in an Arabidopsis fraction enriched in extracellular vesicles (Rutter & Innes, 2017). There was more than 40% overlap with most of the apoplast proteome surveys (Basu et al., 2006; Ge et al., 2011; Kaffarnik et al., 2009; Kwon et al., 2005; Tran & Plaxton, 2008; Trentin et al., 2015), and the lowest overlap was ~27% (Boudart et al., 2005; Cheng et al., 2009). None of the above published datasets contained a population of apoplastic proteins with narrowly defined predicted functions or localizations, and the same is true with our cotton data here (Data S4a).

The APO_T and p200 proteomes were strongly correlated in terms of the abundances of the dual localized proteins (Figure 2e). If APO_T proteins that were dually localized in the s200 fraction were due solely to contamination from cell breakage, one would expect a high correlation of protein abundance in the two fractions as observed with p200. This was not observed, even though the protein populations were medium to high‐abundance proteins in both cell fractions (Figure 2f). These data indicate that the protein route(s) to the apoplast (which are not known) is(are) not linked to a particular organelle or the APO_T samples are highly contaminated with broken cells that selectively contribute microsome‐associated proteins.

To better understand the possible contribution of broken cell contamination, we first tested for the effect of isolation buffers on protein yield. The APO_T sample contained proteins that were distributed between particulate (APO_p200) and soluble (APO_s200) fractions, with yields of ~64 and ~16 μg/boll, respectively. Similar protein yields were obtained with a wide range of buffers including deionized water (Data S5). Particles in the APO_p200 fraction were membrane‐enclosed based on fluorescence microscopy (see below). As an additional test, a completely different APO wash method was developed to avoid locule removal. When the lower half of the boll capsule wall was removed and the locules were dunked in deionized water (Figure S1a), abundant fluorescent particles were persistently released from the sample at the time scale of hours (Figure S1b–f) and seemed unlikely to arise solely from surficial organelles fragments generated by the relatively few broken cells from the capsule wall. However, in the same assay the isolated endocarp walls that remain after locule dissection were a potent source of FM4‐64 positive apoplastic vesicles. Few to no fluorescent particles were detected when the intact inner capsule wall was similarly exposed to buffers (see methods). It is therefore possible that some of the APO_T protein from dunked locules arise from surgically damaged cells from the capsule wall. As an independent test for the existence of bona fide extracellular vesicles, we turned to the in vitro cotton ovule and fiber culture system that avoided any dissection (Figure S2). Fresh media was introduced to the cultured fibers during the elongation phase at 15 DPA and collected at 22 DPA. Ultracentrifugation of this culture fluid consistently contained thousands of FM4‐64‐positive particles that were visually indistinguishable from those obtained from dissected locules. This supports the assertion that apoplastic vesicles do not arise solely from broken fibers or the dissected capsule wall. However, it remains formally possible that these apoplastic vesicles arose from cells that lysed during the 7 days of incubation with fresh media.

In another independent test for contamination, we measured the concentrations of glucose, fructose, and sucrose from triplicate APO_s200 and s200 samples analyzed at equal cell proportions. Surprisingly, fructose and glucose were significantly reduced in APO_s200 compared to s200, and sucrose was below the detection threshold only in the APO_s200 samples (Data S6). These data clearly indicate a chemically distinct composition for the apoplast wash samples that cannot be explained easily by contamination and suggest the presence of active invertase in the extracellular space. The precise origin and delivery route of these FM4‐64 positive particles remains to be determined. However, for the purposes of this paper, we will refer to the APO_p200 fraction as extracellular or apoplastic vesicles and only speculate about their origin or functional importance.

We next looked for strong candidates for apoplastic localization in the APO_T fraction by testing for subsets with an increased abundance. We defined apoplast‐specific (APO_specific) proteins as those that had spectral counts in two or three replicates in the APO_T fraction and zero counts in any of the replicates from the s200 and p200 fractions (Figure 2c). Forty‐four proteins satisfied these criteria (Data S7a), and only a small fraction of them are extracellular proteins based on the SUBAcon localization prediction database (Heazlewood et al., 2007). To further test for proteins with an elevated abundance in APO_T, relative protein abundances in the three cell fractions were statistically analyzed using LFQ‐normalized precursor ion intensities (Cox et al., 2014). In pairwise t‐tests between APO_T and p200 or s200 proteins at a false discovery rate of 5%, 55 proteins were specifically enriched in APO_T (Data S7b). These APO‐enriched (APO_enriched) proteins are unlikely to be contaminating high‐abundance proteins from broken cells because their ranked relative abundances were elevated compared to that in the s200 or p200 fractions (Data S7b). Terms of hydrolase/peptidase and a wide variety of transmembrane transporters were enriched GO terms in the APO_specific and APO_enriched proteins, compared to APO_T, p200 and s200 (Figure 2g,h; Data S3e). Surprisingly, RuBisCO large and small subunits are APO_specific and an enzyme involved in starch synthesis, Gorai.004G128700, is APO_specific. This did not correspond to a clear‐cut enrichment with plastid proteins in the APO_T fraction compared to p200, because 124 and 286 plastid proteins were detected in each cell fraction, respectively (Data S2b,c). The SUBAcon localization patterns in the data were analyzed further after the APO_T fraction was centrifuged to separately analyze soluble and particulate subfractions of the apoplast.

Analysis of protein size and protein multimerization in the apoplastic and cytosolic fractions

It is possible that the apoplast is a proteolytic compartment and we are only detecting degraded or partially degraded proteins. To address this question, apparent mass (M _app) of soluble apoplastic proteins (APO_s200) was determined using a label‐free quantification of SEC column fractions (Figure 3a; Data S8 and S9). We have developed robust methods for large‐scale analyses of protein multimerization under non‐denaturing conditions (Aryal et al., 2014; Havugimana et al., 2012; Kristensen et al., 2012; Lee et al., 2021; Lee & Szymanski, 2021; McBride et al., 2017, 2019; McWhite et al., 2020; Wan et al., 2015). Reliable Gaussian‐fitted peaks are generated from protein intensity values from multiple column fractions with >90% of the proteins having identical M _app measurements between replicates (Aryal et al., 2014; Lee et al., 2021; Lee & Szymanski, 2021; McBride et al., 2017, 2019). Of the 205 APO proteins detected in the SEC separations here, 142 could be fitted to a Gaussian profile, and all 46 detected in both replicates had an identical peak location and M _app value and were flagged as “replicated” (Figure 3b,c; Data S9). To leverage the high reliability of the data, those with a Gaussian peak on one replicate were retained and flagged as “single replicate”. Proteins in the APO_s200 fraction were not degraded. Using R _app, the ratio of M _app over the calculated monomeric protein mass (M _mono), as a metric protein multimerization (Liu et al., 2008), 69 proteins were predicted to be monomeric, and 61 had an R _app greater than 1.6 and were predicted to assemble into some sort of complexes (Figure 3d; Data S9a). Only 12 proteins, including a 20.3 kDa polygalacturonase fragment had an R _app <0.6 and potentially were not full length.

Figure 3.

Intactness and multimerization states of dual localized apoplastic proteins.

(a) The CFMS pipeline combined with size exclusion chromatography (SEC) is used to analyze the intactness and potential multimerization states of soluble apoplastic (APO_s200) proteins.

(b) Coverage of proteins identified in chromatography fractions from two replicate separations.

(c) High confidence peaks present based on two‐fraction shift between the two replicates.

(d) APO_s200 proteins are intact and are frequently predicted to be in a multimeric state. R _app is used as a diagnostic for multimerization.

(e) Comparison of the apparent mass of 104 apoplast‐localized proteins (APO_s200) with those that had previously reported M _app measurements from the S200 fraction (Lee & Szymanski, 2021). A and D suffixes reflect homoeolog identification.

One hundred and four APO_s200‐SEC proteins also had reported M _app in a previously published SEC profiling experiment of cytosol‐enriched proteins isolated from cotton fibers at the same developmental stage (Lee & Szymanski, 2021). Most of the measured M _app values for APO_s200‐SEC matched the cytosol‐enriched pool and fell along the diagonal (Figure 3e). However, there were several instances in which the multimerization state of a protein differed between the APO_s200‐SEC and cytosol‐enriched fractions. For example, a lipid transfer protein (Gohir.A10G150100 and Gohir.D10G116300) and a triose phosphate isomerase (Gohir.A01G163600 and Gohir.D01G155700) had higher assembly states in the APO_s200‐SEC fraction. The converse was true for the translation factor EF‐1B (Gohir.D08G211100) and aldolase (Gohir.A04G065600 and Gohir.D04G065600). Given that most proteins in the s200 and APO_s200 fractions have a similar M _app, we find no evidence of a systematic effect of protein isolation or subcellular compartment on the measured multimerization states. We predict that multimerization differences between APO_s200 and s200 reflect functionally distinct forms.

Analyses of the soluble particulate and proteomes in the apoplastic fraction

To further characterize the apoplastic proteome, we scaled up the isolation to four locules and subjected APO_T to ultracentrifugation to generate particulate (APO_p200) and soluble (APO_s200) fractions (Figure 4a). There was a visible pellet in APO_p200 and its protein was ~63 μg/boll, four times higher compared to APO_s200. Triplicate APO_p200 and APO_s200 samples were analyzed using LC/MS, identifying an additional 2335 apoplast‐localized proteins that were not identified previously (Data S10 and S11). There was compositional overlap between the two fractions, but APO_p200 had a much more diverse proteome, having 1312 APO_{p200‐specific} proteins based on the spectral count criteria used above (Figure 4b,c). It is the APO_p200 fraction that is driving the similarity between APO_T and p200. The APO_s200 fraction was dominated by 499 dual‐localized proteins. These dual‐localized proteins are likely to be peripherally associated with microsomes and only 50 were predicted to be integral membrane proteins based on bioinformatic screens for membrane‐spanning segments using DeepTMHMM (Hallgren et al., 2022). In the analysis, 267/499 were predicted to be globular and cytosolic (Data 11a). The LFQ intensity‐based comparisons of dual‐localized proteins identified 35 and 3 upregulated proteins in the APO_p200 and APO_s200, respectively (Figure 4d; Data S11c). Another two LTPs and a secreted glycosyl hydrolase were identified as APO_{s200‐enriched}, and the 35 APO_{p200‐enriched} proteins included a diverse set of enzymes, transcription factors, and vesicle coat proteins (Data S11c).

Figure 4.

Separation and identification of soluble (APO_s200) and vesicle‐associated (APO_p200) fractions in the apoplast (APO_T).

(a) The pipeline used to identify APO_p200 and APO_s200 proteins.

(b) Number of proteins reproducibly identified by MS/MS counts in the APO_p200 and APO_s200 fractions. Fraction‐specific proteins were identified based on MS/MS count data across the replicates and fractions.

(c) Hierarchical clustering and heatmap showing relative protein expression (log₂‐transformed LFQ protein intensities) of the proteins detected in both apoplast fractions. Columns correspond to the three replicates of both fractions. Missing values were imputed by normal distribution (width = 0.3, shift = 1.8) using Perseus version 2.0.7.0.

(d) Volcano plot depicting two‐sample t‐test result between all quantified proteins present in (c). The lines indicate 1% false discovery rate (FDR) and eightfold changes.

(e) Accounting of SUBAcon localization predictions for the different cell fractions including only proteins that have a predicted SUBAcon localization to a single subcellular compartment (proportion).

To gain more insights into the soluble proteins in our dataset we compared s200 with apoplast fractions. Among 392 proteins quantified in APO_s200, 95 were also quantified in s200. The abundance distributions of these dually localized proteins were not strongly correlated, and their ranked abundances were very different in the two fractions, arguing against broken cell contamination (Data S11e). The most highly enriched GO terms in the APO_s200 fraction included catalytic activity, hydrolase activity, and peptidase activity. The s200 fraction was highly enriched in triose‐phosphate isomerase, which was not present in the APO_s200 fraction (Data S3c and S12e). Transmembrane topology prediction by DeepTMHMM predicted a surprisingly large number of proteins with signal sequences and transmembrane domains in the s200 fraction, and this has also been observed in Arabidopsis and rice s200 fractions (Aryal et al., 2014; Lee et al., 2021). Only the APO_{s200‐specific} subset of proteins was enriched with members with a predicted signal sequence (Data S11e), and this group was also highly enriched in apoplast‐localized proteins according to SUBAcon (Figure 4e). The APO_{s200‐enriched} class contained two LTPs with signal sequences and a glycosyl hydrolase 32 protein with a predicted membrane‐spanning segment (Data S11c). These data suggest that many APO_{s200‐specific} and APO_{s200‐enriched} proteins reach their final destination through the known secretory pathway.

Having data on the APO_s200 and APO_p200 fractions allowed us to compare the predicted organelle localization among all cell fractions generated in this study (Figure 4e; Data S11a). We limited the analysis to the subset of proteins in each cell fraction that had a unique SUBAcon localization prediction to a specific organelle and calculated the proportion of proteins residing in each cellular locale. As expected, no clear differences were apparent among APO_T, s200, and p200. Also as expected, the APO_{s200‐specific} protein population was highly enriched in predicted extracellular proteins, including a collection of LTPs, cell wall proteins, glycosyl hydrolases, peroxidases, and proteases (Data S11a). The proteins classified as “enriched” or “specific” in APO_T relative to s200 and p200 also had an elevated representation of extracellular proteins compared to other cell fraction datasets. The vacuole was the only compartment that had an increased representation (0.1 to 0.2 of proportion) in the cell fractions enriched for apoplastic proteins, including the APO_specific, APO_enriched, and APO_{p200‐specific} fractions (Figure 4e). Potential reasons for this modest enrichment of vacuolar proteins are discussed below.

DISCUSSION

This paper describes a simple cell fractionation and quantitative proteomics workflow that provides useful information on the abundance, localization, and multimerization states of the cotton fiber proteome. The single locule analysis yielded low coverage of the crude cytosolic (s200) fraction. More input material and/or improved precipitation protocols will be needed for better access to this cell fraction. The crude microsome (p200) and apoplastic (APO_T) fractions included thousands of fiber proteins. These measured protein abundances spanned five orders of magnitude and provide a glimpse into the systems of proteins that enable the cells to reproducibly generate long, thin, spinnable fibers that feed the textile industry. The microsomal fraction contains known proteins that relate to cellulose synthesis and the dynamic organization of the microtubule array and therefore provides a way to analyze how these, and to be discovered proteins, operate as part of a cytoskeleton‐cellulose synthesis‐cell shape control module that is central to fiber morphogenesis. All of these data are available in well‐organized supplemental tables that are searchable based on both Gorai/Gohir locus IDs, Phytozome‐generated ortholog group numbers, and best‐hit Arabidopsis ortholog IDs.

The crude microsome (p200) fraction had high protein yields and good coverage in the single locule assay. The most abundant p200 proteins point to specific cellular activities and key homoeologs (or homoeologous pairs when the proteomic data are more ambiguous) that are likely to play a central role in fiber development. For example, the actin and microtubule cytoskeletons are required to pattern fiber growth (Bao et al., 2011; Graham & Haigler, 2021; Han et al., 2013; Seagull, 1986; Thyssen et al., 2017; Tiwari & Wilkins, 1995; Wang et al., 2009; Wang, Wang, et al., 2010; Yu et al., 2019; Zang et al., 2021). TUA2/Gorai.009G224800 (Gohir.A05G214700/Gohir.D05G218000) and TUB8/Gorai.003G126300 (Gohir.A03G047400/Gohir.D03G119800) are among the top 15 expressed proteins in p200 and are predicted to assemble into an abundant heterodimer in microtubules at 9 DPA. Additional orthologs predicted to be involved in microtubule cortical array organization and cellulose synthesis are summarized in Data S2c,f. The top 15 included GhACTIN7/Gorai.003G069800 (Gohir.A11G228966/Gohir.D11G231100) and GhACTIN11/Gorai.012G070300 (Gohir.A05G349400/Gohir.D04G062900) orthologs. The GhACTIN11 ortholog predicted to correspond to the dominant Ligon lintless‐1 (Li1) mutant of GhACT_LI1 (Thyssen et al., 2017) was the fifth most abundant actin isoform in p200. Flux of reduced carbon into cell wall polysaccharides is central to cell wall biogenesis and fiber development. The p200 top 15 included UDP‐glucose pyrophosphorylase/Gorai.007G188400 (Gohir.A11G169600/Gohir.D11G175800) and Sucrose synthase/Gorai.009G038000 (Gohir.A05G036000/Gohir.D05G037500), two key enzymes implicated in this pathway (Ahmed et al., 2018; Amor et al., 1995; Coleman et al., 2007; Ruan et al., 2003). The top 15 also included enzymes in methionine biosynthesis, homocysteine S‐methyltransferase/Gorai.004G251500 (Gohir.A08G219500/Gohir.D08G236300), and two related to S‐adenosyl‐methionine‐based one‐carbon metabolism, adenosylhomocysteinase/Gorai.006G223900 (Gohir.A09G206600/Gohir.D09G200200) and serine hydroxymethyltransferase/Gorai.011G153900 (Gohir.A10G127300/Gohir.D10G138600). Methylation reactions during homogalacturonan biosynthesis (Du et al., 2020) and central carbon metabolism (Hanson & Roje, 2001) may be major pathways for carbon utilization.

We focused much of our attention on characterizing the apoplastic wash (APO_T) fraction because of its importance in cell wall biology and its unexpected protein diversity and abundance. The plant apoplast proteome has been analyzed in numerous previous publications (Data S4). This compartment frequently contains both soluble proteins and a particulate fraction that can include both membrane‐enclosed vesicles and large protein/RNA complexes that can sediment during ultracentrifugation (Borniego & Innes, 2023; Zand Karimi et al., 2022). We used ultracentrifugation conditions that do not efficiently sediment large protein complexes like Rubisco (Aryal et al., 2014). Ultracentrifugation of the apoplast fluid and confocal microscopy of the particulate fraction demonstrated that apoplast fraction contains both a soluble pool and a particulate fraction contains membrane‐enclosed vesicles (Figures 1 and 4; Data S11). Across kingdoms, the extracellular proteome is cloaked in mystery and notoriously difficult to analyze due to variability in methodology and the poorly understood contributions of broken or dead cells (Consortium et al., 2017; Rutter et al., 2017; Rutter & Innes, 2020).

For decades, numerous localization and cell fractionation papers have reported the unexplained existence of known cytosolic proteins like calmodulins or 14‐3‐3/GRF proteins in the extracellular space (Ling & Assmann, 1992; Wen et al., 2007). Our apoplastic datasets are also quantitatively enriched in calmodulin, calmodulin‐like, and 14‐3‐3/GRF proteins (Data S7 and S11). Serine/threonine‐protein kinases are known to phosphorylate RNA‐binding proteins of the splicing machinery (Rodriguez Gallo et al., 2022). The kinase ortholog GhSRPK3A/Gorai.006G102900 (Gohir.A09G086500/Gohir.D09G086460) was the highest count protein in 2/3 APO_s200 replicates (Data S11b) even though it had no predicted signal sequences or TM. The presence of diverse protein types in the apoplast cannot be explained by contamination from broken cells. For dual localized proteins, the APO_T abundance is not similar to that in the s200 fraction (Figure 2f). The APO_s200 fraction was also highly enriched in known extracellular proteins that harbor conventional signal sequences to enable secretion (Figure 4e; Data S11). These included numerous glycosyl hydrolases, e.g. GhGH32A/Gorai.008G216800 (Gohir.A12G199100/Gohir.D12G201800) and xyloglucosyl transferase/Gorai.007G095000 (Gohir.A11G084800/Gohir.D11G088900). Even more convincingly, several soluble apoplastic proteins displayed unique multimerization states compared to their crude cytosolic pool (Figure 3e). For example, GhLTP5A/Gorai.011G128200 (Gohir.A10G150100/Gohir.D10G116300) assembled into a large complex but was predicted to be monomeric in the s200. GhLTP5A is one of several high‐abundance apoplastic LTP/PR/Bet v 1 proteins, e.g. GhLTP1A/Gorai.011G128100 (Gohir.A10G150100/Gohir.D10G116300), GhLTP1B/Gorai.005G038900 (Gohir.A02G027300/Gohir.D02G035200), GhLTP1C/Gorai.011G128300 (Gohir.A10G150100/Gohir.D10G116400), GhBetV1A/Gorai.012G129000 (Gohir.A04G103900/Gohir.D04G143600), and GhBetV1B/Gorai.011G197500 (Gohir.A10G170000/Gohir.D10G176900), that may be involved in extracellular non‐vesicular transport of a variety of hydrophobic molecules to the extracellular space (DeBono et al., 2009; Morris et al., 2021; Sterk et al., 1991) and cuticle formation (Kim et al., 2012, 2023; Liu et al., 2014; Thompson et al., 2017; Yatsu et al., 1983). The presence of abundant LTPs in all cell fractions implies this class of protein has a generally important role in fiber cells.

The origin and presence of easily extracted membrane‐enclosed vesicles in the APO_T fraction are even more difficult to explain. Unconventional routes for protein transport to the apoplast have been described in the plant literature (Robinson et al., 2016; Wang, Ding, et al., 2010), and membrane‐enclosed organelles in the apoplast have been observed in electron micrographs of cryo‐preserved samples (Akita et al., 2017; De Bellis et al., 2022). However, genetic experiments in yeast (Oliveira et al., 2010) and Arabidopsis (De Bellis et al., 2022) indicate a general importance of anterograde trafficking, but no single trafficking pathway has been found to be required for extracellular vesicle production. Like all other previous studies, we cannot rule out a contribution of broken cells to our apoplast protein samples, but if the identities and abundances of proteins in the APO_T and APO_p200 proteomes reflect a true extracellular pool, the data point to a rather non‐specific route of vesicle/organelle fragment delivery to the apoplast. The slight enrichment of vacuole localized proteins (Figure 4e) might reflect the enhanced ability of the vacuole to fuse with or donate membranes to the extracellular space (Hatsugai et al., 2009). The size distribution of the vesicles is also impossible to reconcile with passive diffusion of vesicles through the wall. Most vesicles diameters range from ~100 to 350 nm, and this exceeds the porosity/void space estimates of the cell wall by more than an order of magnitude (Read & Bacic, 1996; Titel et al., 1997; Zheng et al., 2017). One possible explanation is that the apoplastic vesicles are generated during nano‐scale cell wall rupture as bundles of densely packed and strongly adherent cells (Singh et al., 2009) grow at different rates as they elongate and bend to fill the volume of the developing boll capsule. This mechanism would explain the non‐specific nature of membrane/organelle export as any organelle that is near the site of rupture would spew into the extracellular space. A localized cell wall rupture could also provide an escape route for the vesicles into an unrestricted apoplastic space.

This subcellular proteomics identified pools of apoplast proteins that are not degraded and display complex multimerization behaviors. Also, the apoplast proteome resembles that of crude microsomes but has enriched protein markers. Those identified APO proteins are present in many anabolic pathways, including RuBisCO shunt and pathways, such as gluconeogenesis, glutamine biosynthesis, and pentose phosphate pathway, that involve NADH and NADPH concentrations (Figures S3 and S4). Interestingly, malate dehydrogenase (1.1.1.40 and 1.1.1.37), isocitrate dehydrogenase (1.1.1.42 and 1.1.1.41/1.1.1.286), glucose‐6‐phosphate dehydrogenase (1.1.1.49), and phosphogluconate dehydrogenase (1.1.1.44) were present in APO_T and APO_p200 fractions, suggesting specific functions within the apoplast. These enzymes provide the reducing power necessary for biosynthetic pathways, such as fatty acid and cell wall biosynthesis, which are essential for fiber elongation. It is plausible that apoplastic vesicles are biochemically active; however, further work is needed to characterize the pH, redox status, and substrate concentrations in different apoplast compartments.

Improvements and future challenges

There is great potential to use this proteomics‐based pipeline to analyze fiber development over time, across genotypes, or as a function of environmental conditions. The full potential to use the proteomic data to predict regulatory mechanisms and protein function will require integration with gene expression, metabolite, and glycome dynamics (Grover et al., 2024; Swaminathan et al., 2024; Tuttle et al., 2015) as they are measured across the same developmental timeline. The proteomic method described here provides a new way to gain improved coverage and crude estimates of subcellular localization by analyzing coupling of quantitative proteomics with cell fractionation. The single locule method here was not sufficient to analyze the crude cytosolic (s200) fraction. Either more input material or improved protein precipitation methods will be needed to get acceptable coverage of this important subcellular compartment. The data here suggest that fibers have a complex extracellular proteome that includes more than just proteins that are secreted through standard anterograde trafficking. Based on the intactness and unique multimerization of apoplastic proteins, these are not simply degraded proteins. Further research is needed to characterize the chemistry and potential compartmentalized metabolism of the apoplast and how it might relate to cell wall remodeling and coordinating growth within and between cells.

MATERIALS AND METHODS

Cotton cultivars and growth conditions

Cotton plants (G. hirsutum cv. TM1) were cultivated in a Conviron^® E15 growth chamber (Conviron, Pembina, ND, USA) at the College of Agriculture Plant Growth Center at Purdue University. Seeds were sown in 3‐gallon pots with a soil mixture prepared as 4:2:2:1 soil:perlite:bark:chicken grit. The growth chamber was controlled to generate 50% relative humidity and remain with a day/night setting of 28/23°C and 16/8 h at a light intensity of 500 μmol m⁻² sec⁻¹ (fluorescent lamps of 28 Sylvania F72T12/CW/VHO 100 W and 4 Sylvania F24T12/CW/HO 35 W; incandescent bulbs of GE 60 W light 130 V A19). The 30‐min two‐step ramp‐up and 30‐min two‐step ramp‐down periods (15 min at 166 μmol photons + 15 min at 336 μmol photons) were programmed at the daytime beginning and the ending, respectively. Cotton flowers were marked at anthesis as 0 DPA and harvested at 9 DPA. The harvested cotton bolls were maintained on ice and dissected immediately to obtain intact ovules from each locule.

Collection of apoplast fluid fraction (APO_T )

The fibers in isolated locules were fragile and did not tolerate vacuum infiltration and high salt buffers that have been used to isolate extracellular fluids. Therefore, individual locules (~700 mg each) were dunked in 5 ml of a MIB [50 mm HEPES/KOH (pH 7.5), 250 mm sorbitol, 50 mm KOAc, 2 mm Mg(OAc)₂, 1 mm EDTA, 1 mm EGTA, 1 mm dithiothreitol (DTT), 2 mm PMSF and 1% (v/v) protein inhibitor cocktail (160 mg ml⁻¹ benzamidine‐HCl, 100 mg ml⁻¹ leupeptin, 12 mg ml⁻¹ phenanthroline, 0.1 mg ml⁻¹ aprotinin, and 0.1 mg ml⁻¹ pepstatin A)] and then incubated for 10 min under gentle shaking (Figure 1a). The resulting solution was recovered through two layers of cheesecloth and then the entire APO fluid (APO_T) cellular fraction was used for protein precipitation using a cold acetone precipitation method (sample to 100% acetone ratio was 1:4) as described previously (McBride et al., 2017). Three biological replicates were prepared.

Isolation of cytosolic (s200) and microsome (p200) fractions

From the material used for the above APO_T fraction collection, fibers were directly collected as described previously (Lee & Szymanski, 2021; McBride et al., 2017). Using forceps, fibers were isolated from seeds without any damage to the seeds under 1 ml of MIB solution (sample to MIB ratio was 1:4). The fresh fiber tissues were ground under the cold MIB using a Polytron homogenizer (Brinkman Instruments, New York, NY, USA) with chilled blade tip with 10 sec grinding, 1 min rest on ice, and another 10 sec grinding. The homogenate was filtered with four layers of cheesecloth pre‐soaked in cold MIB, and the cheesecloth was further squeezed to get a residual sample in the cheesecloth. Then the filtered homogenate was further centrifuged on an Allegra X‐30R centrifuge (Beckman Coulter Life Sciences, Indianapolis, IN, USA) at 1000  g for 10 min to remove debris. The supernatant was fractionated at 200 000  g for 20 min at 4°C using a Beckman Optima Ultracentrifuge with a TLA110 rotor (Beckman Coulter Life Sciences, Indianapolis, IN, USA) into a cytosolic (s200) fraction and pellet. The entire s200 fraction was used for protein precipitation using the acetone precipitation method as described above. To get a microsome fraction (p200), the resulting pellet was resuspended with MIB, incubated on ice for 10 min, and ultracentrifuged at the same condition as shown above. The supernatant was discarded, and this washing step was repeated. Three biological replicates were prepared for both cellular fractions.

Determination of protein concentration

For the APO_T and s200 cellular fractions, dried pellets from the acetone precipitation were dissolved and denatured in 100 μl of 8 m urea for 1 h at room temperature. For the p200 fraction, 200 μl of 8 m urea was directly added over the final pellet and incubated for 1 h at room temperature to denature proteins from membranes. Undissolved debris was removed by centrifugation at 12 000  g for 15 min using an Allegra X‐30R centrifuge (Beckman Coulter Life Sciences). Protein assay was performed using a BCA assay kit, according to the manufacturer's protocol (Thermo Fisher Scientific, Inc., Waltham, MA, USA). Protein yields for fresh fiber weight were 0.015, 0.25, and 0.012 mg g⁻¹ in APO_T, p200, and s200, respectively.

Separation of soluble (APO_s200 ) and particulate (APO_p200 ) subfractions of APO_T by ultracentrifugation

The APO_T fraction was obtained from four locules in the same boll as described in the APO_T isolation method. Apoplastic extracellular vesicles (APO_p200) were enriched from the APO_T using a Beckman Optima Ultracentrifuge (Beckman Coulter Life Sciences) at 200 000  g for 20 min at 4°C (Figure 4a). The supernatant (APO_s200) was used as a control in a quantitative proteomics analysis for apoplast vesicle and cargo protein identification. To deplete weakly associated soluble proteins, the pellet was washed twice with 10 ml of MIB as shown above. The APO_p200 proteins were solubilized with 150 μl of 8 m Urea and then ultracentrifuged to remove insoluble materials from the APO_p200. Triplicates were prepared.

Apoplast protein complexes for SEC and co‐fractionation mass spectrometry

For size exclusion chromatography (SEC), APO_T proteins, that were freshly obtained from four locules from one cotton boll, were pooled and then centrifuged to get APO_s200 as described in the above APO_p200 experiment. SEC was performed in the same conditions (Figure 3a) as described previously (Lee & Szymanski, 2021). The method is not highly sensitive to protein loading. Briefly, ~100 μg of APO_s200 proteins were resolved in a Superdex Increase 200 10/300 GL (GE Healthcare, Chicago, IL, USA) on an AKTA FPLC system (GE Healthcare). The mobile phase was (50 mm HEPES‐KOH, pH 7.5, 100 mm NaCl, 10 mm MgCl₂, 5% glycerol, and 1 mm DTT) and the flow rate was set at 0.65 ml min⁻¹. Fractions (APO_s200‐SEC) were collected at every 500 μl of elution volume. The column was calibrated using a gel filtration kit 1000 MWGF1000 (Sigma‐Aldrich, Inc., St. Louis, MO, USA), and the void fraction was determined using blue dextran. Twenty‐seven fractions were collected including the first two void fractions. A cold acetone method was used for precipitating proteins in each fraction. Two biological replicates were prepared.

LC–MS/MS sample preparation

For LC–MS/MS analysis, proteins were digested using trypsin as described previously (McBride et al., 2019). We used all protein samples from the APO_T, APO_p200, APO_s200, APO_s200‐SEC, and s200 preparations, and 50 μg of proteins in the p200 cell fractions. Denatured protein samples were reduced in 10 mm DTT for 45 min at 60°C, and then alkylated with 20 mm iodoacetamide for 45 min at room temperature under the dark. The urea concentration of the solution was brought down to below 1.5 m by additional ammonium bicarbonate before trypsin (Sigma‐Aldrich) digestion. The digestion was proceeded at an enzyme‐to‐protein ratio of 1 to 50 at 37°C. After overnight digestion, trifluoroacetic acid was added to end digestion. The digested peptides were purified using C18 Micro Spin Columns 74‐4601 (Harvard Apparatus, Holistan, MA, USA). Peptide concentration was measured by a BCA assay kit (Thermo Fisher Scientific, Inc.). All the peptide samples were adjusted to equal volume. The most concentrated sample had a peptide concentration of 1 μg μl⁻¹, and 1 μl of each sample was injected onto the LC–MS/MS system. For the APO_p200 and APO_s200 samples, 1 μg of each sample was injected.

LC–MS/MS data acquisition

LC–MS/MS analysis was carried out as described previously (Barabas et al., 2019). Peptides samples were analyzed by a reverse phase LC‐ESI‐MS/MS system using a Dionex UltiMate 3000 RSLCnano System coupled with an Orbitrap Fusion Lumos Tribrid Mass Spectrometer (Thermo Fisher Scientific, Inc.). Peptides were loaded onto a trap column (300 μm ID × 5 mm) packed with 5 μm 100 Å PepMap C18 medium and then separated on a reverse phase column (75 μm ID × 50 cm) packed with 2 μm 100 Å PepMap C18 medium (Thermo Fisher Scientific, Inc.). Peptides were resolved at a flow rate of 200 nl min⁻¹ over a 160‐min LC gradient. All the data were acquired in the positive ion mode in the Orbitrap mass analyzer using a higher energy collisional dissociation (HCD) fragmentation scheme. The MS scan range was from 375 to 1500 m/z at a resolution of 120 000. The automatic gain control target was set as standard, maximum injection time (50 msec), dynamic exclusion 60 sec, and intensity threshold 5.0e3. MS data were acquired in data‐dependent mode with a cycle time of 3 sec scan⁻¹. MS/MS data were collected at a resolution of 7500.

Peptide identification and quantification

Andromeda search engine on MaxQuant (version 1.6.14.0) was used for relative protein abundance quantification and protein identification (Cox et al., 2014; Tyanova, Temu, & Cox, 2016). The search was conducted as described (Lee & Szymanski, 2021). Briefly, the search was conducted with all three cellular fractions obtained from the biological triplicates at a single search, but the three different parameter groups for the three cellular fractions were defined to specify different parameters for each cell fraction. The search parameters were as follows: the match between runs function was set with a maximum matching time window of 0.7 min as default; proteins identified by a single unique peptide were selected; the cotton reference was provided by Dr. Corrinne Grover (Grover et al., 2024); Label‐free quantification was selected within each parameter group; all other parameters were set as default. For the APO_p200 and APO_s200 search, the same settings were applied. For the APO_s200‐SEC search, we used the same settings except for the Label‐free quantification method.

Measurement of sugar levels in s200 and APO_s200

Triplicate s200 and APO_s200 samples were made from triplicate bolls using the same methods described for cell fractionation and centrifugation. Samples were analyzed on a Thermo Dionex HPLC system with pulsed amperometric detection as previously described (Gangola et al., 2014). Samples were analyzed using Thermo CarboPak‐PA1 column (Thermo Fisher Scientific, Inc.) with a simple increase of sodium acetate from 0 to 100 mm for 40–50 min of the linear gradient, beginning at 100 mm NaOH at 0 min. Standard curves were constructed using purified sugars (Sigma‐Aldrich).

Data filtering for protein identification in each cellular fraction (APO_T , p200, s200)

Two layers of filtering strategies were applied for each cellular fraction dataset to identify cellular fraction‐specific proteins and proteins enriched in the APO_T fraction. First, those proteins identified with MS/MS count(s) ≥1 in at least two out of three replicates were chosen as identified proteins in each cellular fraction (Count‐based data filtering; Data S2). Here apoplast‐specific (APO_specific) proteins were determined when any proteins in the APO_T dataset had no MS/MS count in both p200 and s200 fractions (Data S7a). Secondly, quantified proteins with valid LFQ values in at least two out of three replicates in each cellular fraction were selected from the above count‐based dataset (Data S7b). Also, as the benchmark for abundance of proteins in different fractions, the percentile rank of each protein was estimated based on protein abundances in each dataset (Data S7). To identify apoplast enriched (APO_enriched) proteins, LFQ intensities were log₂‐transformed, and then missing values were replaced using imputation function as default on Perseus (version 2.0.6.0) (Rudolph & Cox, 2019; Tyanova, Temu, Sinitcyn, et al., 2016). Two‐sample tests between APO_T and p200 or s200 were carried out at 5% false discovery rate (FDR). The APO_enriched proteins were defined as apoplastic proteins that were statistically significant at the tests at 5% FDR and had twofold higher protein abundance than one in both p200 and s200 datasets (Data S7b).

To identify APO_p200 proteins, this two‐step filtering strategy was also applied. Briefly, the APO_p200 proteins should be identified at least two out of three replicates by MS/MS count(s), but not shown in the control. Those proteins present in both the APO_p200 and APO_s200 datasets should be statistically significant in a Two‐sample t‐test at 1% FDR with eightfold change in abundance. The list of proteins in the APO_p200 was reported in Data S11.

Determination of protein oligomerization states using APO_s200‐SEC

We applied the optimized Gaussian fitting algorithm to convert raw profile data into fitted peaks (McBride et al., 2017). The fitted peak locations were used to determine the apparent mass (M _app) values of proteins using the SEC calibration curve created above. If a protein elution profile was not fitted to a Gaussian curve, the global maximum (G _max) was replaced with its peak location. We selected peaks that were present within two‐fraction shifts between duplicates and Gaussian‐fitted peaks that were present in either replicate for further analysis. The protein multimerization state (R _app) was defined as the ratio of the M _app of a protein to the theoretical monomer mass (M _mono) of the protein (Aryal et al., 2014; Lee & Szymanski, 2021; Liu et al., 2008). The R _app of ≥1.6 thresholds indicates that a protein is in a complex. For the R _app comparison between APO_s200‐SEC proteins and cytosolic proteins, the previously published cotton fiber dataset was downloaded from table S2 in Lee and Szymanski (2021), and then, the R _app values were searched using cotton gene IDs.

SEM of developing cotton fibers

Locules were collected from a dissected boll at 15 DPA. After dissection, the distal end of the locule was cut with a razor blade rotated roughly 30° off of the proximodistal axis to generate a roughly uniform sample a few millimeters thick while preserving the undisturbed locule surface for imaging. The sample was immediately submerged in liquid nitrogen after attachment to the stub. Samples were imaged on a Thermo Scientific Apreo 2S (Thermo Fisher Scientific, Inc.,) equipped with a Quorum PP3010T Cryo‐SEM Preparation System (Quorum Technologies, Ltd., Laughton, East Sussex, UK). Sublimation was performed in 5‐min increments, and additional sublimation steps were used as needed to remove surface ice.

Fluorescence microscopy of isolated extracellular vesicles (APO_p200 )

The APOp200 samples were stained with 1 μm FM4‐64 (Thermo Fisher Scientific, Inc.) for 2 h at 4°C, mounted on a slide, and visualized using a spinning disk confocal microscope system with a 100x Plan‐APO 1.46 NA oil‐immersion objective. Images were collected using a spinning disk CSU‐X1 confocal head (Yokogawa Electric, Musashino, Tokyo, Japan) and a Prime 95B camera (Photometrix, Tuscon, AZ, USA) mounted on a Zeiss Observer.Z1 inverted microscope (Zeiss, Jena, Germany) controlled using a Slidebook software (Intelligent Imaging Innovation, Denver, CO, USA). FM4‐64 was excited with 488 nm light and emitted light was detected using a 617/73 bandpass filter. SEM imaging for cotton fibers was done using a Quorum PP3010T cryo system mounted on a Thermo Fisher Apreo 2S scanning electron microscope. Dissected locules were shaved on the opposite side from the viewed surface area to enable them to be mounted on the cryo holder using a mixture of OCT and aqueous colloidal carbon. Freezing was carried out in nitrogen slush before transferring under vacuum to the preparation chamber attached to the microscope. Sublimation was carried out at −90°C for 5 min followed by sputter coating with platinum. Images were recorded using an accelerating voltage of 5KV and a beam current of 0.1 nA.

Apoplast wash following capsule wall dissection and locule dunking

As an alternative method to isolate an APO wash sample, outer capsule wall of a 15 DPA boll was carefully dissected without the underlying fiber cells (Figure S1). The dissected boll was rinsed in a beaker of deionized water at room temperature. Then the stem was used to suspend the boll in 25 ml of water so that it did not touch the beaker, the exposed locules were submerged, and the cut edge of the capsule wall was above the waterline. Three 10‐min rinses were done, and then a final 1‐h rinse. Twenty‐five millilitres of each rinse was collected and centrifuged in an ultracentrifuge for 25 min at 200 000  g . The pellets were then each resuspended in 80 μl ddH₂O. These resuspended particles were stored on ice. Ten microliters of aliquot were diluted twofold and stained with 1 μm FM4‐64 for 2 h at 4°C. Then, 1 μl of the stained sample was transferred to a slide and mounted with a 25 mm × 25 mm cover glass. To test the possibility that the particles originated from the intact inner capsule wall of the boll, the concave domain of a capsule wall was flooded with 100 μl of buffer or water for 1 h, taking care to not expose the solution to the cut edge of the capsule wall.

In vitro cotton ovule culture

Harvested ovaries at 1 DPA were surface sterilized with 80% ethanol, and three or four ovules were removed from the ovary and placed into wells of 6‐well tissue culture dishes containing 4 ml of modified BT culture media per well (Beasley & Ting, 1973). During ovule cultivation at 30°C, the modified BT culture fluid was changed every 5 to 8 days. At 22 DPA, 7 days after the last media change, 3 ml of culture fluid was carefully withdrawn from each well without touching the fiber clusters. The fluid was centrifuged at 1000  g for 10 min at 4°C using an Allegra X‐30R centrifuge (Beckman Coulter Life Sciences). The resulting supernatant was centrifuged at 200 000  g for 30 min at 4°C using a Beckman Optima Ultracentrifuge with TLA110 rotor (Beckman Coulter Life Sciences). The pellet was resuspended in 30 μl of cold ddH₂O and stored at −80°C until further use.

Five microliters of aliquots from samples with visible pellets were diluted twofold and stained with 1 μm FM4‐64 at 4°C as described above. Then 1 μl (~1.4% of the resuspended pellet) was placed on a slide, covered with a cover slip (25 mm × 25 mm), and imaged with a spinning disk confocal microscope. Excitation was with 488 nm light and fluorescence was detected using a 617/73 bandpass filter.

Comparison of cotton APO_T proteins with prior proteomics studies

We created a combined list from 13 published apoplast/extracellular vesicle proteomics studies to validate the results and obtain a clearer picture of the progress made in the secretome field to date. To generate a combined list, 12 external datasets from Arabidopsis and 1 dataset from rice secretome studies were used and duplicates from each dataset were eliminated (Data S4). All of the Arabidopsis locus IDs were converted to cotton locus IDs by performing the Phytozome‐based ortholog mapping as described previously (Lee & Szymanski, 2021). To identify overlaps for each protein and to find cotton‐specific secretory proteins, the list of 1643 APO_T proteins was compared with the combined list generated from external datasets (Data S4b). The percent overlaps were calculated by dividing the number of Arabidopsis IDs commonly found in the APO_T dataset and external dataset by the total number of Arabidopsis IDs identified in each external dataset.

Gene ontology enrichment analysis

Gene Ontology (GO) enrichment analysis was performed using the Singular Enrichment Analysis (http://systemsbiology.cau.edu.cn/agriGOv2/index.php) on agriGO v2.0 (Tian et al., 2017). The enrichment was calculated using Fisher's exact test against Cotton D locus ID v2.1 (Phytozome v11.0) background. The minimum number of mapping entries was 5, and Yekutieli (FDR under dependency) was used as a multi‐test adjustment method. Significantly enriched GO terms in biological process (BP), GO molecular function (MF), and GO cellular component (CC) categories were reported at 5% FDR.

Statistic tests and data analyses

Statistical analysis was performed using R version 4.2.0 (R Core Team, 2018) on RStudio 2022.07.1 (RStudio Team, 2018). Gaussian fitting (https://github.com/dlchenstat/Gaussian‐fitting) was applied using MATLAB_R2022a. Microsoft Excel on Office 365 for Mac was used to organize and display the analyzed data.

AUTHOR CONTRIBUTIONS

Conceptualization, DBS; Methodology, YL and DBS; Formal Analysis, YL and DBS; Investigation, YL, ELM, HR, and DBS; Data Curation, YL and HR; Writing – original draft, YL and DBS; Writing – review and editing, YL, ELM, HR, and DBS; Visualization, YL; Supervision, DBS; Funding acquisition, DBS.

CONFLICT OF INTEREST

The authors declare no competing interests.

Supporting information

Data S1. Raw abundances of peptides and proteins identified in the APO_T, p200, and s200 of cotton fibers at 9 DPA.

Data S2. Proteins identified and quantified in the APO_T, p200, and s200 cellular fractions.

Data S3. Gene ontology enrichment analysis of APO_T, p200, and s200.

Data S4. Overlap between APO_T proteome in this study and proteomes of the extracellular space in the previous 13 proteomics studies.

Data S5. APO_p200 and APO_s200 protein yields in different buffers.

Data S6. Concentrations of glucose, fructose, and sucrose in APO_s200 and s200.

Data S7. APO_specific and APO_enriched proteins.

Data S8. Raw profiles of peptides and proteins identified in the APO_s200 cellular fraction of the 9 DPA cotton fiber by CFMS (APO_{s200‐ SEC}).

Data S9. Apoplastic protein complex profiling analysis by CFMS (APO_s200‐SEC).

Data S10. Raw abundances of peptides and proteins identified in the extracellular vesicle (APO_p200) samples.

Data S11. Proteins specifically identified in APO_p200 and APO_s200.

Data S12. Gene ontology enrichment analysis of APO_p200 and APO_s200.

Figure S1. FM4‐64 positive vesicles in the apoplast wash following capsule wall dissection and locule dunking.

Figure S2. FM4‐64 positive vesicles in the culture fluid of cotton ovules at 22 DPA.

Figure S3. Functional vesicles for catabolic and anabolic pathways in apoplast?

Figure S4. Additional pathways that concentrate NADPH concentrations.

DATA AVAILABILITY STATEMENT

All LC/MS .raw files are annotated and made available at Pride (accession code: PXD051721, PXD051728, and PXD051739).