ZmCTLP1 is required for the maintenance of lipid homeostasis and the basal endosperm transfer layer in maize kernels

Summary Maize kernel weight is influenced by the unloading of nutrients from the maternal placenta and their passage through the transfer tissue of the basal endosperm transfer layer (BETL) and the basal intermediate zone (BIZ) to the upper part of the endosperm. Here, we show that Small kernel 10 (Smk10) encodes a choline transporter‐like protein 1 (ZmCTLP1) that facilitates choline uptake and is located in the trans‐Golgi network (TGN). Its loss of function results in reduced choline content, leading to smaller kernels with a lower starch content. Mutation of ZmCTLP1 disrupts membrane lipid homeostasis and the normal development of wall in‐growths. Expression levels of Mn1 and ZmSWEET4c, two kernel filling‐related genes, are downregulated in the smk10, which is likely to be one of the major causes of incompletely differentiated transfer cells. Mutation of ZmCTLP1 also reduces the number of plasmodesmata (PD) in transfer cells, indicating that the smk10 mutant is impaired in PD formation. Intriguingly, we also observed premature cell death in the BETL and BIZ of the smk10 mutant. Together, our results suggest that ZmCTLP1‐mediated choline transport affects kernel development, highlighting its important role in lipid homeostasis, wall in‐growth formation and PD development in transfer cells.

Adjacent to the PC are the BETL and BIZ, two elongated cell types located at the bottom of the endosperm (Shannon, 1972;Becraft & Gutierrez-Marcos, 2012;Leroux et al., 2014;Larkins, 2019). The BETL, a unique endosperm cell layer with wall ingrowths that increase the surface area of the plasma membrane and cell wall, plays a major part in the transfer of nutrient solutes from the maternal placenta to the developing endosperm (Davis et al., 1990;Wang et al., 1994). Several genes have been reported to influence nutrient uptake and allocation in this region, thereby regulating kernel development. Mn1 (Miniature1) is located in the BETL and the PED and catalyses the cleavage of sucrose into hexoses to generate a physiological gradient of photosynthate that passes symplastically and apoplastically into the PC, then apoplastically through the BETL to the endosperm (Miller & Chourey, 1992). Loss of Mn1 function results in smaller kernel size (Miller & Chourey, 1992;Cheng et al., 1996;Cheng & Chourey, 1999;Kang et al., 2009;Sosso et al., 2015), whereas Mn1 overexpression improves kernel filling and markedly increases kernel size and weight (Wang et al., 2008;Li et al., 2013). Its rice homologue GIF1 (Grain Incomplete Filling 1) is also recognised as a positive regulator of seed filling (Wang et al., 2008).
A second gene, ZmSWEET4c (Sugars Will Eventually be Exported Transporter 4c), encodes a sugar transporter reported to transport hexoses produced by Mn1 to the endosperm (Sosso et al., 2015). Its soybean homologue GmSWEET10 regulates oil content and seed size by transporting sucrose and hexose from the seed coat to the developing embryo (Wang et al., 2020). The development of fully differentiated BETL with wall in-growths requires the normal functions of Mn1 and ZmSWEET4c (Kang et al., 2009;Sosso et al., 2015). Wall in-growth formation enlarges the surface area of the plasma membrane and therefore requires abundant lipid precursors such as phosphatidic acid, triglyceride, acetyl-CoA and choline (van Greevenbroek et al., 1995;Michel et al., 2006;Correa et al., 2020).
Choline is a quaternary amine that plays a fundamental role in cell membrane phospholipid metabolism and the synthesis of the neurotransmitters acetylcholine and betaine (Michel et al., 2006). Disorders of choline uptake, transport and metabolism affect cell proliferation and differentiation in animals, thereby contributing to neurodegenerative disorders such as Alzheimer's disease and Parkinson's disease (Michel et al., 2006;Ueland, 2011). Choline transport is mediated by three different systems: high-affinity choline transporters (CHT1s), intermediate-affinity choline transporter-like (CTL) proteins, and low-affinity polyspecific organic cation transporters (OCTs) (Michel et al., 2006). CTL1 is widely expressed in animal tissues, and its encoded protein specifically regulates choline supply for phospholipid and sphingolipid synthesis (Machov a et al., 2009;Wang et al., 2017). In plants, the first choline transporter AtCTL1 was isolated in Arabidopsis thaliana, and it has been shown to have choline transport activity (Dettmer et al., 2014). AtCTL1 also regulates auxin distribution by promoting the trafficking of auxin efflux transporters during plant growth and development . AtCTL1 is essential for the formation and development of PD and plays vital roles in the formation of sieve tubes; mutation of AtCTL1 alters ion homeostasis in rosette leaves from one 5-wk-old plant, ultimately causing a severely stunted growth phenotype (Dettmer et al., 2014;Gao et al., 2017;Kraner et al., 2017;Wang et al., 2017). Nonetheless, information on the choline transporter in plants is very limited, and its roles in kernel development and kernel biology remain largely unknown.
In this study, we identified the maize CTL1 choline transporter homologue ZmCTLP1 through map-based cloning of smk10. Mutation of ZmCTLP1 disrupts kernel development, dramatically decreasing kernel size and weight. In smk10, the nutrient transfer cells of the BETL and BIZ exhibit typical features of dysfunction, including irregular cell shape, decreased PD number, premature cell death and impaired lipid homeostasis. RNA-seq and immunoblot analysis revealed that the mutant shows altered expression of many genes involved in nutrient metabolism and transport. Collectively, these findings demonstrate that ZmCTLP1 is required for kernel development and participates in lipid homeostasis, wall in-growth formation and PD development in transfer tissues.

Plant materials
The ethyl methanesulfonate (EMS)-generated smk10 mutant was obtained from mutant populations produced in our laboratory on the B73 genetic background. The smk10 mutant was crossed with Mo17 and then self-pollinated to generate F2 and F3 populations. The F3 population was used for fine mapping of smk10. The wild-type and the smk10 were used for cytological and biochemical analyses. Leaves and roots were collected from at least three 10-d-old B73 seedlings. Leaf, root, internode, node, silk, cob and husk tissues were collected at the R1 stage 70 d after sowing. Immature B73 kernels were harvested at 2, 4,6,8,10,12,14,16 and 20 d after pollination (DAP). Maize plants were grown under natural conditions at the Shang-Zhuang experimental field of China Agricultural University in Beijing.

Light microscopy, fluorescence microscopy and transmission electron microscopy
For light microscopy, 4-12 DAP wild-type and smk10 kernels were harvested. Paraffin sections were prepared using the methods described by Li et al. (2018). The sections were stained with fuchsin or eosin and observed using an Olympus SZ51 microscope. Detection of PD after aniline blue staining (0.1% aniline blue in double-distilled water and 1 M glycine, pH 9.5, at a volume ratio of 2 : 3) was performed using a ZEISS LSM710 confocal microscope (Germany) with scan speed set at 'Best signal'. The excitation wavelength for PD signal was 440 nm, and the emission wavelengths were 450-511 nm.
For transmission electron microscopy observation, 12 DAP immature kernels from the wild-type and smk10 were cut along the horizontal axis for resin section preparation. Each sample was fixed with glutaraldehyde and osmic acid and then embedded with low viscosity resin (Electron Microscopy Sciences, cat. #14300). Sections were stained with lead citrate and observed using a Hitachi H7600 transmission electron microscope (Japan).
Determination of soluble sugar, starch, protein, choline and lipid contents Soluble sugars, starch and protein Mature kernels (15 g) were pulverised and extracted with water. The samples were dispersed with an ultrasonic dispersion instrument for 30 min and then measured using a Chromeleon chromatography management system (Dionex, Sunnyvale, CA, USA). Starch and protein contents were measured as previously described (Clegg, 1956;Wang et al., 2011).
Choline and lipids Samples were prepared and measured using the methods described by Kraner et al. (2017). In brief, c. 30 mg of maize kernel tissue were frozen in liquid N 2 and ground to a fine powder in a cold mortar. Metabolites were extracted using a MeOH : CHCl 3 : H 2 O (6 : 2 : 2) extraction buffer. MS analysis for the quantification of choline was performed using an Ultimate 3000 UHPLC system coupled to a Q-Exactive MS instrument (Thermo Scientific, Bremen, Germany). Lipid contents were determined according to a previously described protocol (Shui et al., 2010;Gao et al., 2017). In brief, the sample (100 mg) was combined with 900 ll of chloroform : methanol (1 : 2) extraction buffer and acid-washed glass beads (Sigma), vortexed for 10 min, and incubated overnight at 4°C with shaking at 1100 rpm. Next, 600 ll of chloroform : H 2 O (1 : 1) was added, and the sample was vortexed for 30 s. The mixtures were centrifuged at 7500 g for 5 min, and the lower organic phase was collected and dried with a Savant SpeedVac vacuum system (Thermo Fisher Scientific, Milford, MA, USA). The samples were stored at À80°C before mass spectrometric analysis.

Map-based cloning of smk10
To clone smk10, an F2 population was produced by crossing smk10 with Mo17 and then self-pollinating their F1 progeny. Thirty-five individual miniature kernels were pooled, and 65 genetic markers were used for genotyping. The causal locus was mapped to a 2.3-Mb genomic region on chromosome 2. To fine map the candidate gene, 3264 F3 individuals were yielded from the cross of smk10 and Mo17. Five genetic markers within the 2.3-Mb genomic region further narrowed the smk10-containing region to 300 kb between markers M096 and M127. The candidate genes within this region were sequenced.

Maize transformation
To obtain ZmCTLP1 CRISPR lines, a 20-bp sgRNA-specific target site in the first exon of ZmCTLP1 was cloned into the backbone of the pCAMBIA3301 vector and transformed into maize by Agrobacterium-mediated transformation as previously described (Zhu et al., 2016). Transgenic plants were identified by PCR, and the targeted fragments were sequenced to detect variable sites. Two independent Cas9-edited knockout lines (cas-1 and cas-2) were obtained.

Expression in Xenopus oocytes and choline uptake assays
Isolation and RNA injection of Xenopus oocytes Coding sequences of ZmCTLP1 and AtCTL1 were codon optimised and synthesised. These cDNAs were cloned into the pT7TS expression vector. After linearisation of pT7TS plasmids with BamHI, RNA was transcribed in vitro using an mRNA synthesis kit (mMESSAGE mMACHINE T7 kit, Ambion). Xenopus laevis oocytes were isolated in 25 ml ND96 (Ca 2+ free) solution containing 43 mg collagenase and 12.5 mg trypsin inhibitor for 1.5 h and were then recovered in ND96 for 24 h. Oocytes were injected with cRNA (25 ng in 25 nl) after recovery and were incubated at 18°C in MBS solution (88 mM NaCl, 1 mM KCl, 2.4 mM NaHCO 3 , 0.71 mM CaCl 2 , 0.82 mM MgSO 4 , and 15 mM HEPES, pH 7.5) with gentamycin (0.1 mg ml -1 ) and streptomycin (0.1 mg ml -1 ). The incubation buffer was changed every 24 h. Oocytes injected with water were used as the negative control.
[ 14 C]-labelled choline uptake by Xenopus oocytes The assay was performed as described previously (Kraner et al., 2017) with some modifications. At 72 h after injection, the oocytes were transferred into 500 ll Na-Ringer (115 mM NCl, 2 mM KCl, 1 mM MgCl 2 , 1.8 mM CaC1 2 , and 10 mM MES-Tris (pH 5.5) or 10 mM HEPES-NaOH (pH 7.4)) with 0.1 mg ml -1 gentamycin and 5 µM choline ([methyl-14 C]-choline chloride 55 mCi mmol -1 ; ARC). After incubation at 20°C for 2 h, oocytes were transferred to ice-cold Barth's medium, washed four times, and solubilised with 100 ml 1% (w/v) SDS. Three oocytes were used as one sample, and the accumulated radioactivity in the oocytes was measured. The experiments were repeated independently using three different batches of oocytes with similar results.

Subcellular localisation
The full-length open reading frame (ORF) of ZmCTLP1 was amplified by PCR from maize B73 and cloned into pCAMBIA1300-35S-green fluorescent protein (GFP) using restriction enzymes and ligases. The fused pCAMBIA1300-35S-ZmCTLP1-GFP construct and RFP-SYP61 (a TGN marker) were transiently transformed into maize mesophyll protoplasts as previously described (Yoo et al., 2007). The GFP signal was observed and merged with the RFP signal using a ZEISS LSM710 confocal microscope (Germany) with scan speed set at 'Best signal'. The excitation wave lengths for GFP and RFP signals were 488 and 561 nm, and the emission wave lengths were 500-586 and 598-656 nm, respectively.
qPCR and RNA-seq RNA was extracted from the leaves and roots of 10-d-old seedlings, immature kernels (2-20 DAP at 2-d intervals), and leaf, root, internode, node, silk, cob and husk tissue were harvested at the R1 stage (70 d after sowing) using the TRIzol reagent according to the manufacturer's instructions (Invitrogen). cDNA was synthesised from the extracted RNA using PrimeScript RT Master Mix (Takara, Dalian, China), and qPCR was performed using the TB Green Premix Ex Taq kit (Takara) following the manufacturer's protocols. Gene expression levels were calculated using the 2 ÀDDCt relative quantification method with Actin as the endogenous control gene. For RNA-seq analysis, total RNA was extracted from two biological replicates of immature 6 and 12 DAP wildtype and smk10 kernels using TRIzol reagent (Invitrogen). RNAseq libraries were constructed according to the protocol of the VAHTS mRNA-seq v2 Library Prep Kit (Vazyme, Nanjing, China) and sequenced on the Illumina HiSeq 2500 platform to generate 150-bp paired-end reads. The reads were mapped to the B73 reference genome using HISAT2 2.0.4 (Kim et al., 2015). The unique bam files were acquired using CUFFLINKS (v.2.2.0) (Ghosh & Chan, 2016), and FPKM values were calculated to identify differentially expressed genes (DEGs) (P < 0.05, log 2 fold-change > 1.0). GO enrichment analysis of the DEGs was performed using the agriGO singular enrichment analysis tool (http://bioinfo.cau.edu.cn/agriGO), and the DEGs were visualised using MeV (https://sourceforge.net/projects/mev-tm4/).

Antibodies
For production of monoclonal antibody against Mn1, a specific cDNA fragment (129-450 sequence site, representing 43-150 amino acids) of Mn1 was cloned into pET-28a expression vector

Research
New Phytologist (Amersham Biosciences, Piscataway, NJ, USA) and transformed into BL21 cells (TransGen Biotech, Beijing, China). Mn1-His fusion protein was purified using Ni Sepharose TM 6 Fast Flow (GE Healthcare, Uppsala, Sweden), and production of antibodies in mice was performed according to standard protocol of Abmart Inc. of China. Anti-a-tubulin (anti-a-tubulin mouse monoclonal antibody, BE0031; Easybio) was purchased from EASYBIO Inc. of China.

Immunoblot analysis
Total proteins extracted from immature wild-type and smk10 kernels at 4, 8 and 12 DAP were used for western blot using the methods described by Wang et al. (2018). The nitrocellulose membranes with the protein samples attached were incubated with anti-Mn1 and anti-a-tubulin in TBST (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.05% Tween-20) with 5% milk for 2.5 h at 25°C and then washed three times using TBST for 10 min. The nitrocellulose membrane and attached protein were then incubated with the secondary antibody (anti-rabbit IgG, horseradish peroxidase (HRP)-linked antibody; Proteintech, Rosemont, IL, USA). HRP was detected using the Immobilon Crescendo Western HRP substrate kit (Millipore), and the signals were visualised with a C300 imaging system (Azure Biosystems, USA).

In situ hybridisation
Paraffin sections were prepared as described above, and genespecific ZmSWEET4c and ZmCTLP1 probes with locked nucleic acid modification was generated (Supporting Information  Table S2, see later). In situ hybridisation was performed according to previously described protocols (Ding et al., 2015).

TUNEL assay
Sections from wild-type and smk10 mutant kernels collected at 8 and 12 DAP were dewaxed using the HistoChoice Clearing Agent (Sigma) and hydrated through an ethanol series (95%, 90%, 80%, 70%). The sections were treated with proteinase K in phosphate buffer (pH 7.4), and the terminal deoxynucleotidyl transferase (TdT)-mediated dNTP nick end labelling (TUNEL) assay was performed using the DeadEnd Fluorometric TUNEL Kit (Promega) according to the manufacturer's protocol. Cell nuclei were stained using propidium iodide (PI) (Beyotime Biotechnology, Shanghai, China). The green fluorescence (the TUNEL signal) and the red fluorescence of PI were observed using a ZEISS LSM710 confocal microscope (Germany) with the parameters as described in the section of 'Subcellular localisation'.

Phylogenetic tree
ZmCTLP1 and its orthologous protein sequences were downloaded using MaizeGDB (https://www.maizegdb.org/) and NCBI (https://www.ncbi.nlm.nih.gov/) databases. The obtained orthologous full-length proteins grouped together and their sequences were aligned to compare equivalent residues using the program CLUSTALW. (https://www.megasoftware.net/). Neighbourjoining phylogenetic trees were constructed from full-length protein sequences, and the numbers at the nodes represent the percentage support from 1000 bootstrap replicates.

Phenotypic and genetic characterisation of smk10
We screened a mutant library generated by EMS mutagenesis on the B73 background, aiming to identify novel regulators of kernel development based on the phenotype of small kernel size. We screened one mutant, designated smk10 and performed three continuous backcrosses with B73 to clean the smk10 background. We confirmed that smk10 harboured a recessive mutation by genetic analysis, as the F2 ears segregated at a 3 : 1 ratio (v 2 = 0.16-0.92 < v 2 0.05 = 3.84) of wild-type vs smk10 kernels. The average hundred-kernel weight of smk10 was significantly lower than that of the wild-type (Fig. S1a). On the B73 genetic background, smk10 exhibited reduced size of both endosperm and embryo (Figs 1a-c, S1b,c). The smk10 mutant had a higher embryo : endosperm ratio than that of wild-type (Fig. S1d), and its kernel development was slower (Fig. S1e). Soluble sugar and storage protein contents were much higher in mature smk10 kernels, whereas total starch content was much lower (Fig. S2), suggesting that the kernel development process was attenuated in smk10. The growth of seedling leaves and roots was slower in smk10 than that of the wild-type, which could mainly be due to the smaller endosperm and embryo. However, there were no obvious differences between the wild-type and smk10 in important agronomic traits such as kernel row number, ear length, plant height, and tassel branch number at the mature stage (Figs 1d,e, S3-S5).
Paraffin sections of immature wild-type and smk10 kernels collected 4-12 DAP were prepared to observe morphological features during kernel development (Fig. 1f,g). The mutant kernel phenotype was distinguishable at 6 DAP, when the endosperm and pericarp showed an irregular periphery. Phenotypic differences in the embryo could be discerned at 8 DAP: the wild-type embryo showed a visible scutellum, whereas the smk10 embryo was still in the proembryo stage. Clear phenotypic differences were observed at 12 DAP: smk10 exhibited an empty pericarp and a small, irregular endosperm and embryo (Figs 1f,g,S6a,b). Starch granules were smaller in smk10 than in the wild-type at 12 DAP and in mature kernels ( Fig. S6c-h).
These phenotypic changes motivated us to investigate the morphological differences in different endosperm compartments. At 12 DAP, the wild-type conducting zone (CZ) and central starchy endosperm (CSE) cells were fairly uniform in size and shape, whereas the smk10 CZ and CSE cells were smaller and nonuniform with highly variable sizes (Figs 2a,b, S7a-d). The wild-type AL cells were similar to those of smk10 (Fig. 2c,d). In the wildtype, the BETL displayed a characteristic slightly elongated shape with labyrinth-like wall in-growths (Figs 2e, S7e). By contrast, the mutant exhibited misshapen BETL cells with few thickenings or wall in-growths (Figs 2f, S7f). The wild-type had extremely elongated BIZ cells with wall in-growths that decreased gradually from the second to the fourth cell layers, but the elongation and wall in-growths were less pronounced in smk10, and some of its cells were rounded like those of the starchy endosperm ( Fig. 2eg). These results suggested that the smk10 endosperm develops dysfunctional nutrient transfer cells, especially the BETL and BIZ cells, and this may block nutrient transport from the maternal tissue to endosperm.

Map-based cloning of smk10
To identify the causal gene of smk10, an F2 mapping population of 70 individuals was obtained from a cross between heterozygous smk10 and Mo17. The smk10 mutant locus was mapped to a 2.3-Mb physical interval on chromosome 2 that was flanked by the two InDel markers Chr2-M018 and Chr2-M248. To narrow down the interval, a high-resolution map was obtained from 3264 F3 individuals, and smk10 was fine mapped to a 300-kb region that contained 10 candidate genes following by sequencing ( Fig. 3a; Tables S1, S2). RNA-seq analysis indicated that there was a single nucleotide transition (G?A) in the 5 0 splicing site of the fourth intron of the candidate Zm00001d001803, encoding a putative choline transporter with 10 transmembrane helices ( Fig. S8a,b). Consistent with this nucleotide transition in the intron splicing site, PCR analysis with a gene-specific primer pair (F, R) based on the Zm00001d001803 sequence revealed at least four different transcripts in smk10 kernels, visualised by agarose gel electrophoresis (Fig. S8c). A primer set (DJ-1) was designed to amplify the full-length cDNA. Subsequent cloning and sequencing of the products indicated that five alternative spliced variants with premature termination codons (PTCs) generated in upstream regions of the intact mRNA were found (Fig. S8d). Interestingly, the five alternative splicing transcripts were at least partially escaped from nonsense-mediated mRNA decay (NMD), a surveillance pathway that degrades mRNAs containing PTCs. Although the cause of the escape remains unclear, several studies had also reported examples of NMD escapes (Ruiz-Echevarria et al., 1998;Nyik o et al., 2009;Lindeboom et al., 2019). CRISPR-Cas9 knockouts were created to confirm that Zm00001d001803 was responsible for the small kernel phenotype. A guide RNA was designed to target the first exon, and two knockout events with frameshifts in the first exon were identified (Figs 3b, S9, S10a). The kernels of knockout plants exhibited a serious defective kernel phenotype with empty pericarps at the crown kernels, very similar to smk10. An allelism test was performed by crossing smk10 with heterozygous cas-1 (cas-1/+), heterozygous cas-2 (cas-2/+) and cas-1; neither cas-1 nor cas-2 complemented smk10 ( Fig. S10b; Table S3). In addition, cas-1 developed defective basal endosperm cells that resembled those of smk10 (Fig. S10c). To investigate whether there is maternal effect of ZmCTLP1 or not, we performed phenotypic analysis by using kernels of reciprocal-crossed ears derived from wild-type (WT) (+/+) and cas-1 (À/À). No significant differences were found for 100-kernel weight, suggesting that there was no visible maternal effect of ZmCTLP1 (Fig. S11). All these results confirmed that a null mutation of Zm00001d001803 is the molecular basis for the smk10 phenotype.
Smk10 encodes the choline transporter ZmCTLP1 that facilitates choline transport Smk10 and its homologous proteins were used to construct a neighbour-joining phylogenetic tree to investigate their evolutionary relationships (Fig. S12). The sequences were divided into three clades: dicots (clade I), monocots (clade II), and humanmouse (clade III). Smk10 was grouped with the monocots, and AtCTL1 was grouped with the dicots. A phylogenetic tree containing Smk10 and its paralogues indicated that Smk10 and Zm00001d052748 were highly similar to AtCTL1 (Fig. 3c), which is a CTL protein located at the TGN (Fig. S13) (Dettmer et al., 2014). However, the expression of Zm00001d052748 (named ZmCTLP2) was much lower than that of ZmCTLP1 (Fig. S14a). Therefore, Smk10 was named ZmCTLP1. ZmCTLP1, like other CTL proteins, is a transmembrane protein

Research
New Phytologist with 10 highly conserved transmembrane helices and a large extracellular loop at the N terminus, as predicted by TOPO2 (http://www.sacs.ucsf.edu/cgi-bin/open-topo2.py). In maize, five genes containing a conserved plasma-membrane choline transporter domain were identified (Fig. S15). They contain four conserved amino acids (V 422 G 500 R 516 N 560 ), and these sites may be functionally important. The other three maize genes with lower expression levels were less similar (<26%) to AtCTL1 (Fig. S14a).
ZmCTLP1 is closest in homology to animal and Arabidopsis thaliana CTL1 proteins (Fig. S12), which have been shown to have choline transport activity (O'Regan et al., 2000;O'Regan & Meunier, 2003;Kommareddi et al., 2010;Dettmer et al., 2014). To examine the choline transport activity of ZmCTLP1, we expressed both ZmCTLP1 and AtCTL1 in Xenopus laevis oocytes and measured their choline uptake capacity at two pH levels (pH 5.5 and pH 7.5). The results showed that the choline uptake capacity of ZmCTLP1 was similar to that of AtCTL1 (Fig. 3d).

Subcellular localisation and expression patterns of ZmCTLP1
In previous studies, AtCTL1 was localised to the TGN (Dettmer et al., 2014;Kraner et al., 2017). To examine the subcellular localisation of ZmCTLP1, the full-length ORF of ZmCTLP1 from the wild-type was fused to the N terminus of GFP driven by the CaMV 35S promoter. The construct was co-transformed into maize protoplasts with SYP61-RFP (a TGN marker; Poulsen et al., 2014;Rosquete et al., 2018). Transient expression of ZmCTLP1-GFP showed that it co-localised with SYP61 (Fig. 3e), suggesting that ZmCTLP1 was located at the TGN in maize. ZmCTLP1 was constitutively expressed in various tissues, with higher expression levels in kernels and lower expression levels in other examined tissues (Fig. 3f). During kernel development, the expression of ZmCTLP1 was higher at 2-6 DAP and slowly decreased after 8 DAP (Figs 3f, S14a). At the protein level, the ZmCTLP1 content was higher in endosperms at 10-12 DAP (Fig. S14b). To accurately delineate the expression patterns of ZmCTLP1 within the kernel, we performed RNA in situ hybridisation on 12 DAP kernels. ZmCTLP1 expression was higher in the BETL, BIZ, aleurone layer (Al), PC, and embryo (EMB), but lower in the CSE and in the endosperm adjacent to scutellum (EAS) (Fig. S16).

ZmCTLP1 is involved in lipid homeostasis during kernel development
ZmCTLP1 functions in choline transport, and we therefore measured choline content in 12 DAP kernels using high-performance liquid chromatography (HPLC). The smk10 mutant had significantly lower choline content than the wild-type (Fig. 4a), implying that ZmCTLP1 was required for normal choline homeostasis. Choline serves as a precursor for phosphatidylcholines and sphingolipids, which are the main components of the cell membrane , and we therefore investigated the contents of 24 classes of membrane lipids using HPLC-mass spectrometry (HPLC-MS). Content of total lipids was higher in the smk10 mutant (Fig. 4b), and free fatty acid (FFA) levels were also greater
Furthermore, we also examined the global lipid profiles of wild-type and smk10 kernels by measuring 423 molecular species from 24 classes ( Fig. 4d; Table S4). Over 30% of the individual lipids were markedly higher in the smk10 mutant, and 6.9% were markedly lower. All FFA and most phospholipid molecular species showed significant alteration in smk10. Although the total contents of lyso-PC (LPC), phosphatidylethanolamines (PE), and lyso-PE (LPE) were similar between smk10 and the wild-type, more than half of the individual molecular species were significantly altered in smk10 ( Fig. 4d; Table S4). Phosphatidylcholines are the major phospholipid components of the plasma membrane. There was no significant difference in total phosphatidylcholine content between the genotypes, but 29% of the individual phosphatidylcholine species differed markedly between the wild-type and smk10 (Fig. S17). Interestingly, although the absolute lipid contents were higher in smk10, the relative proportions of some lipid classes were significantly lower. The phospholipids including LPC and phosphatidylinositols (PI), glycolipids including monogalactosyl diacylglycerols (MGDG) and digalactosyl diacylglycerols (DGDG) etc., were markedly lower in the smk10 than in the wild-type (Fig. 4e). Over 16% of the individual lipids were markedly lower in the smk10, and 8.5% were markedly higher (Fig. 4f). As the endosperm cell size in smk10 was smaller than that of wild-type (Fig. S7a,b), the increased absolute contents of lipids in smk10 might due to the large number of cells per unit weight of kernel. In fact, the total lipid content and most of individual lipids per kernel were significantly decreased in smk10 than that of the wild-type (Fig. S18). Overall, our results indicated that lipid homeostasis was severely disrupted in smk10.

ZmCTLP1 is crucial for PD formation in endosperm transfer cells
Because the transfer cells in smk10 were misshapen, we further investigated whether the symplastic pathway of nutrient transport among transfer cells was affected. To this end, we performed aniline blue staining of paraffin sections followed by confocal microscopy to identify the location of callose deposition around the neck regions of the PD. Signals at the BIZ cell walls were significantly diminished in smk10 compared with the wild-type at 12 DAP ( Fig. 5a-d). The number of PD in the BIZ region at 12 DAP was dramatically lower in smk10, but there were no obvious differences in PD morphology between the wild-type and smk10 (Fig. 5e-m). These observations are consistent with previous reports in Arabidopsis thaliana (Kraner et al., 2017) and show that ZmCTLP1 is required for PD formation in endosperm transfer cells.

Loss of ZmCTLP1 function alters the expression of key kernel development genes
To better understand the impacts of smk10 on global gene expression, RNA-seq analysis was performed on 6 and 12 DAP smk10 and wild-type kernels. In total, we identified 578 DEGs (97 upregulated and 481 downregulated in smk10) in 6 DAP kernels and 3477 DEGs (1322 upregulated and 2155 downregulated) in 12 DAP kernels (P < 0.05, log 2 fold-change > 1.0) (Table S5). Here, 10 DEGs were selected for qRT-PCR validation of the RNA-seq results, and both methods showed similar expression differences between the wild-type and smk10 (Fig. S19). Gene Ontology (GO) analysis of DEGs in 6 DAP kernels revealed enrichment of the GO terms 'response to abiotic stimulus', 'nutrient reservoir activity', 'response to lipid', 'cell wall macromolecule metabolic process' and 'killing of cells of other organism' (Fig. 6a). The DEGs at 12 DAP were enriched in GO terms related to nutrient metabolism and transport such as 'nutrient reservoir activity', 'carbohydrate metabolic process', 'transmembrane transporter activity, and 'cell-cell junction (PD)' (Fig. 6b). To further explore GO terms enriched in DEGs from different endosperm compartments, we analysed available data from a laser-capture microdissection experiment (Zhan et al., 2015). Enriched GO terms included 'cell killing' in the BETL at 6 and 12 DAP, 'hydrolase activity, acting on glycosyl bonds' in the PE, and 'ion homeostasis' and 'ion transport' in the PED at 12 DAP ( Fig. S20; Table S6). Eight BETL-specific genes, including Mn1 and ZmSWEET4c, were significantly downregulated in smk10 (Fig. 6c). Tissuespecific transporters for ions, amino acids and sugars located in the PED, PC, and BETL were also differentially expressed in smk10 at 12 DAP (Fig. 6d). In fact, c. 11.5% (90/784) of all transporters in smk10 were differentially expressed at 12 DAP; these included 27 ion transporters, 13 sugar transporters, nine amino acid transporters, and two lipid transporters (Table S7).
Mn1 and ZmSWEET4c are essential for the development of fully differentiated transfer cells. To measure the abundance of Mn1 protein, total proteins were isolated from immature wildtype and smk10 kernels at 4, 8 and 12 DAP and analysed by western blotting with anti-Mn1 mouse monoclonal antibody. Mn1 protein accumulation was significantly lower in smk10 than in the wild-type (Fig. 6e). To accurately determine the expression level of ZmSWEET4c in smk10 and the wild-type, we performed

Research
New Phytologist RNA in situ hybridisation with 12 DAP kernels. Consistent with the RNA-seq results, ZmSWEET4c expression was again significantly lower in smk10 than in the wild-type (Fig. 6f-h). These results indicated that ZmCTLP1 influenced the expression of key kernel development genes.

Loss of ZmCTLP1 causes premature cell death in transfer tissues
Disruption of choline and lipid homeostasis can trigger cell death in plants and animals (Li et al., 1998(Li et al., , 2020Hirasawa et al., 2004;Lin et al., 2008), and we were therefore motivated to investigate cell death in the BETL and BIZ of smk10. We performed a TUNEL assay to examine this cell death signal in situ at 8 and 12 DAP. Intense positive TUNEL signals that co-localised with cell nuclei were detected at the BETL and BIZ positions in smk10, whereas no TUNEL signals were detected in the wild-type (Fig. 7). We also observed that both the PC and EAS exhibited similar cell death in the wild-type and the mutant, as reported by Kladnik et al. (2004) and Doll et al. (2020) (Figs 7, S21). The transfer cells are extremely active and do not undergo programmed cell death before maturity in normal maize kernels (Gao et al., 1998). Therefore, these findings indicated that smk10 exhibited premature BETL and BIZ cell death, which presumably attenuates nutrient transport from the maternal placenta to the filial tissues.

Discussion
Choline and its metabolism have been studied extensively in animals and humans. Despite its important physiological functions, little information is known about the role of choline in plants (Michel et al., 2006;Dettmer et al., 2014;Iwao et al., 2016). In Arabidopsis thaliana, choline deficiency impairs phloem development and conductivity by reducing the number of sieve pores and altering the pore structure in sieve areas (Dettmer et al., 2014). In this study, the smk10 mutant contained less choline than the wild-type and exhibited an irregular and small endosperm, a small embryo and significantly lower kernel weight. BETL and BIZ cells in the basal endosperm region of the mutant displayed irregular shapes and had few wall thickenings and wall in-growths. We therefore concluded that loss of ZmCTLP1 function disrupted choline and lipid homeostasis of transfer cells, thereby reducing nutrient transport from the maternal placenta to the developing endosperm.
ZmCTLP1 is required for the normal expression of Mn1 and other nutrient transporters Membrane lipid composition is crucial for maintaining the correct structure and function of membrane proteins (Sanders & Mittendorf, 2011). Choline, an essential membrane lipid component, is predominantly used for the synthesis of phosphatidylcholine and sphingomyelin (Michel et al., 2006;Wang et al., 2017). HPC1 (High Phosphatidylcholine 1) encodes a phospholipase A1 enzyme; when introgressed from teosinte Mexicana, it induces early maize flowering by modulating phosphatidylcholine levels (Rodr ıguez-Zapata et al., 2021). Here, we showed that ZmCTLP1 is required for kernel development and that it appears to regulate choline transport to maintain the homeostasis of membrane lipids. Nutrient transfer tissues such as the BETL and BIZ have convoluted wall in-growths that increase the surface area of the plasma membrane, thereby enhancing nutrient transport efficiency (Kang et al., 2009). The formation of cell wall architecture is a membranerelated process that relies on the supply of lipids, proteins and other components Rosquete et al., 2018). Here, loss of ZmCTLP1 function significantly disrupted choline and lipid homeostasis, and this appeared to block the formation of wall in-growths. Mn1 is synthesised on the endoplasmic reticulum and delivered to the wall in-growths by the TGN (Kang et al., 2009). Mn1 and wall in-growths are interdependent, reinforcing each other (Kang et al., 2009). Our results suggested that ZmCTLP1mediated regulation of membrane lipid homeostasis is required for the normal accumulation of Mn1 (Figs 6e, 8). Lipid-dependent protein sorting plays a central role in cargo sorting and trafficking to the cell surface (Surma et al., 2012). Disruption of membrane lipid homeostasis in the smk10 mutant appeared to strongly influence membrane transport capacity, as 11.5% of all transporter genes were differentially expressed in the mutant, including ion transporters, amino acid transporters and sugar transporters (Table S7). In situ hybridisation suggested that ZmSWEET4c-mediated hexose import may have been reduced in smk10 (Fig. 6f-h). It has been proposed that ZmSWEET4cmediated sugar accumulation in the BETL induces hexose import by increasing membrane area of the basal cells in developing endosperm, therefore the capacity to hold transporters (including ZmSWEET4c) exists as a result of increased ZmSWEET4c activity. In our study, loss of function of ZmCTLP1 disrupted lipid homeostasis and led to incomplete wall in-growth of BETL, which could decrease the membrane area in transfer cells, resulting in reduced accumulations of Mn1 and ZmSWEET4c in BETL. Thereby, the overall activities of hydrolysing sucrose by Mn1 and importing hexose into BETL by ZmSWEET4c were decreased, which might be one of the major causes for small kernel phenotype in smk10.
ZmCTLP1 is essential for PD formation and development PD are membrane-lined channels between neighbouring cells that provide a direct route for the symplastic movement of proteins, RNAs, sugars and other small molecules Sager & Lee, 2018). During the development of primary PD, materials derived from ER tubules (desmotubules) are enclosed by cytoplasmic strands, and TGN secretory vesicles provide wall materials and membrane lipids (Ehlers & Kollmann, 2001). Sphingolipids that are enriched in the TGN perform essential functions to regulate PD permeability (Albright et al., 1996). Our results showed that sphingolipid levels differed significantly between smk10 and the wild-type, and it is plausible that disordered membrane lipid homeostasis in smk10 may have caused the defects in PD formation. There were fewer PDs in the smk10 mutant than in the wild-type, indicating that the mutant was impaired in PD formation and development. The morphological abnormalities of the BETL cells in smk10 were partly attributable to reduced numbers of PD, which would have disrupted cell-tocell movement of nutrients and other signals (Fig. 8). PD numbers are also reduced in the shoot apical meristems and developing leaves of the atctl1 mutant, causing a severely stunted growth phenotype. Soluble sugar and starch contents are significantly increased in the atctl1 mutant, suggesting that nutrient homeostasis is disrupted in the leaves (Kraner et al., 2017). Our results showed that starch, sugar and protein contents were significantly altered in smk10 kernels, and this may have resulted from defects in cell-cell symplastic transport mediated by PD. Zm00001d001803 (ZmCTLP1); Zm00001d052748 (ZmCTLP2); Zm00001d007070; Zm00001d016972; Zm00001d033497. ZmCTLP1 orthologues cited in this article can be found in the NCBI's Nucleotide database under the following accession numbers: AT3G15380 (AtCTL1); AT1G25500; AT3G03700; AT3G04440; AT4G38640; AT5G13760; AT5G17830 in Arabidopsis thaliana, and AJ420812; AJ245621 in human, and AJ420809; AAH31535 in mouse. The generated raw reads have been uploaded to NCBI's Sequence Read Archive (SRA) database and are available under the accession number PRJNA720421.

Supporting Information
Additional Supporting Information may be found online in the Supporting Information section at the end of the article.                and the abundance of ZmCTLP1 protein (Walley et al., 2016) in various tissues.

Fig. S21
Cell death detected by the TUNEL assay in 12 DAP wild-type (a-c) and smk10 (d-f) kernels.

Table S1
The candidate genes predicted by maizeGDB in the 300-kb smk10 interval.

Table S2
Primers used in this study.

Table S5
Expressed genes in 6 and 12 DAP smk10 kernels compared with wild-type.

Table S6
Enriched GO terms analysis of compartments-specific genes in smk10 mutant compared with wild-type.