Cellulose synthase-like D1 controls organ size in maize

Background Plant architecture is a critical factor that affects planting density and, consequently, grain yield in maize. The genes or loci that determine organ size are the key regulators of plant architecture. Thus, understanding the genetic and molecular mechanisms of organ size will inform the use of a molecular manipulation approach to improve maize plant architecture and grain yield. Results A total of 18 unique quantitative trait loci (QTLs) were identified for 11 agronomic traits in the F2 and F2:3 segregating populations derived from a cross between a double haploid line with a small plant architecture (MT03-1) and an inbred line with a large plant architecture (LEE-12). Subsequently, we showed that one QTL, qLW10, for multiple agronomic traits that relate to plant organ size reflects allelic variation in ZmCSLD1, which encodes a cellulose synthase-like D protein. ZmCSLD1 was localized to the trans-Golgi and was highly expressed in the rapidly growing regions. The loss of ZmCSLD1 function decreased cell division, which resulted in smaller organs with fewer cell numbers and, in turn, pleiotropic effects on multiple agronomic traits. In addition, intragenic complementation was investigated for two Zmcsld1 alleles with nonsynonymous SNPs in different functional domains, and the mechanism of this complementation was determined to be through homodimeric interactions. Conclusions Through positional cloning by using two populations and allelism tests, qLW10 for organ size was resolved to be a cellulose synthase-like D family gene, ZmCSLD1. ZmCSLD1 has pleiotropic effects on multiple agronomic traits that alter plant organ size by changing the process of cell division. These findings provide new insight into the regulatory mechanism that underlies plant organ development. Electronic supplementary material The online version of this article (10.1186/s12870-018-1453-8) contains supplementary material, which is available to authorized users.


Background
Maize (Zea mays ssp. mays) is the highest yielding and most widely grown crop in the world. Increasing grain yield has always been a primary goal of maize breeding. Plant architecture is a result of many trait interactions during plant development and growth, and is a critical factor that affects plant density and, consequently, grain yield. In general, maize plant architecture includes plant height (PH), leaf number (LN), leaf angle (LA), leaf area, and tassel traits. An ideal plant architecture includes a short stature to prevent lodging, an erect LA and a moderate leaf area to maximize light interception, and small tassels to partition more energy and nutrients for ear and seed development [1][2][3][4]. Most of the traits that contribute to plant architecture, such as PH, leaf length (LL), leaf width (LW), and tassel length (TL) can collectively be referred to as organ size, which is determined by cell number and cell size and results from the two successive processes of cell proliferation and cell expansion. Thus, understanding the genetic and molecular mechanism of organ size will inform the development of a molecular manipulation approach to improve maize plant architecture and grain yield.
Plant organ size is a product of the interaction of genotype and environmental influences [5]. The genetic architecture of organ size has been well studied in maize by using different genetic populations [6][7][8][9][10][11]. For example, a nested association mapping (NAM) population, which contained~5,000 recombinant inbred lines and was developed by crossing 25 diverse lines to B73, was used to identify the quantitative trait loci (QTLs) of multiple traits that relate to organ size [7,9,10]. By using joint linkage analysis, a total of 89, 92, 37, 36, and 34 QTLs were detected for PH, ear height (EH), TL, LL and LW, respectively. Plant architecture traits were also studied in a natural population that contained 513 diverse maize lines by a genome-wide association study, and 185 SNPs were significantly associated with PH, EH, LL, LW and TL [11]. Recently, approximately 314 QTLs were identified to control the phenotypic variation in organ size-related traits (PH, EH, LL and LW) in 10 maize populations [8]. These QTL studies indicate that the formation of organ size is complex and characterized by small effects for most QTLs, large effects for several QTLs, and a certain number of pleiotropic QTLs. Studies of the model plant Arabidopsis have revealed that the formation of plant organ size are modulated by several molecular mechanisms, including transcriptional regulation [12][13][14][15], hormone signalling [15][16][17][18], ubiquitin-mediated proteolysis [19,20], and cell wall biosynthesis [21]. However, only several genes for organ size have been isolated and characterized in maize, such as BR2 [22,23] and ZmGA3ox2 or D1 [24,25] for PH; DIL1, GA20-OXIDASE1, GRF1 and GRF10 for LL and PH [26][27][28][29]; ZmGE2 [30] and BIGE1 [31] for embryo size; and ZmBRI1, CNR1, CNR2 and ZmPLA1 for overall organ size [32][33][34]. Similar to Arabidopsis, organ size in maize is controlled by multiple genes in various regulatory pathways. For the genes known to regulate organ size, the molecular mechanisms of most genes regulating organ size relate to hormone signalling [23,25,34]. As expected, all the genes control organ size by changing cell division and cell expansion to alter cell number or cell size, respectively. For example, overexpressing ZmPLA1 increased organ size by prolonging cell division, and knocking down ZmBRI1 reduced organ size by decreasing cell division and cell elongation [32,34]. Besides, the regulation ways to organ growth were also different in the same tissue. Both BR2 and ZmGA3ox2 control PH, although, BR2 plays a role in cell division, while ZmGA3ox2 plays a role in cell expansion [23,25]. Similarly, LL was regulated by GRFs, with overexpression of GRF1 R and GRF10 increasing LL by cell division and expansion, respectively [28,29]. In addition, not like the role of ZmGA3ox2 in stem, gibberellin greatly enlarged the cell division zone, and increased LL in GA20-OXIDASE1 overexpression plants [27,35]. These results had provided summary knowledge about the genes that control organ size, although the regulatory pathways in which these genes operate are still poorly understood in maize.
In this study, we developed a double haploid line with a small plant architecture, MT03-1, which was crossed with an inbred line with a large plant architecture, LEE-12, to determine the genetic architecture of the maize agronomic traits that relate to organ size. A major QTL for organ size, qLW10, was identified in the F 2 and F 2:3 segregating populations derived from a cross between MT03-1 and LEE-12. This locus was further cloned to be ZmCSLD1, which encodes a cellulose synthase-like D1 protein that affects plant growth through cell division or expansion. The molecular mechanism of ZmCSLD1 that affects plant organ size was extensively addressed through genetic analysis, subcellular localization, an expression profile and the analysis of cell number and cell size.

Results
A large-effect QTL cluster controls multiple agronomic traits MT03-1 is a double haploid line with a relatively small plant architecture (Additional file 1: Table S1). To investigate the genetic basis of these traits, QTL mapping was performed for the 11 measured traits by using a genetic map of 1833.66 cM (Additional file 1: Tables S1;  Additional file 2: Tables S2), and a total of 19 and 25 QTLs were detected in the F 2 and F 2:3 populations, respectively ( Fig. 1a; Additional file 3: Table S3). These QTLs were distributed in 18 genomic regions across 8 of the 10 chromosomes, except for chromosomes 6 and 9, and each QTL explained 3.27 to 75.05% of the phenotypic variation. Out of 18 unique QTLs, 6 and 2 were associated with at least two traits in the F 2 and F 2:3 populations, respectively, which indicates that they had pleiotropic effects or were closely linked loci.
Among these QTLs, one QTL cluster with a large effect on chromosome 10, which was flanked by SYN17100 and PZE110041758, was identified for nine traits in both the F 2 and F 2:3 populations. On average, this locus explained 27.35% of the phenotypic variation in all nine traits and differed from 75.05% for LW to 9.18% for EH. In addition, all the alleles at this locus associated with decreased size came from MT03-1. This result was highly consistent with the phenotypic correlation among traits (Fig. 1b). All nine traits with the QTL identified at this locus had a moderate to high correlation with one another (r = 0.44-0.94, F 2:3 population), whereas the two remaining traits without the identified QTL at this locus, namely, the LA and 100-kernel weight (KW100), were weakly correlated with the nine traits (r = -0.2-0.34, F 2:3 population). This result suggests that the nine traits may share a common pleiotropic genetic variant at this locus.
ZmCSLD1 is the causal gene for qLW10 To clone the causal gene for the QTL cluster, we selected LW as the trait for positional cloning because the greatest difference was observed for LW between the parents, LEE-12 and MT03-1 ( Fig. 2a and b; Additional file 1: Table S1), and this locus had the largest effect on LW ( Fig. 2c; Additional file 3: Table S3). This locus was then designated qLW10. The position of qLW10 was refined to a 3.3-Mb segment flanked by markers IDP9050 and MTL02 by using 3,292 BC 2 F 1 and 669 BC 3 F 1 plants (Fig. 2d). To confirm the narrowed interval, a heterozygous inbred family BYK-HIF, which segregated in LW (Additional file 4: Figure S1A-C) was also used to fine map qLW10. Nine recombinant types were identified in 720 F 2 plants selfed from BYK-HIF, and the progeny test of homozygous segregants narrowed the interval to a 3.8-Mb region flanked by the markers IDP9050 and IDP7754 (Additional file 4: Figure S1D). The nucleotide sequence analysis based on B73 reference sequences predicted 33 and 36 genes in the 3.3-and 3.8-Mb intervals, respectively. Simultaneously, two pairs of near-isogenic lines (NILs) were developed. qLW10 MTL and qlw10 MTL carried 3.3-Mb fragments from LEE-12 and MT03-1, respectively. qLW10 BYK and qlw10 BYK harboured different alleles at the target locus, with only a 0.4% difference in the genome background. Compared with qLW10 MTL and qLW10 BYK , qlw10 MTL and qlw10 BYK had smaller plant architectures (Additional file 5: Figure S2; Additional file 6: Figure S3).
Among these genes in the qLW10 interval, GRMZM2G015886 (ZmCSLD1) has been previously reported to narrow LW and reduce growth after a loss of function [36], which is very similar to the effect of qLW10. Therefore, ZmCSLD1 was proposed as the best candidate gene for qLW10. To confirm this hypothesis, we first sequenced the full-length ZmCSLD1 in two pairs of NILs. Compared with the qLW10 MTL allele, the qlw10 MTL allele contained a 467-bp insertion and a 63-bp deletion in the third exon, resulting in a frameshift that prevented the production of the mature protein because of the introduction of a premature stop codon (Fig. 2e). For the qLW10 BYK and qlw10 BYK lines, there was a G to A transition in the third exon, causing a conserved amino acid with a major change from glycine to aspartic acid ( Fig. 2e; Additional file 4: Figure S1E; Additional file 7: Figure S4).
To confirm that ZmCSLD1 is the causal gene for qLW10, two additional Zmcsld1 mutants, Zmcsld1 MO17 and Zmcsld1 W22 , were developed (Fig. 2e). A genomic sequence analysis revealed that Zmcsld1 MO17 had a T-to-C substitution (tryptophan-to-arginine) in the third exon and a conserved amino mutation in the last transmembrane domain of ZmCSLD1 (Additional file 7: Figure S4), and Zmcsld1 W22 had a Mu-transposon insertion in the second exon. As expected, the LW in each mutant was significantly decreased compared with their corresponding wild-type lines, similar to qlw10 MTL and qlw10 BYK (Fig. 2f ). The allelic effect was the largest for Zmcsld1 W22 , followed by qlw10 MTL , qlw10 BYK and Zmcsld1 Mo17 . To further evaluate the allelic effect, Zmcsld1 W22 was crossed with qlw10 MTL , qlw10 BYK , Zmcsld1 MO17 and their wild types, as designated in Additional file 8: Table S4. Compared with the F 1 hybrids with a normal allele, all three F 1 hybrids with both mutant alleles had  Table S2 narrower leaves (Fig. 2g), which was further validated in the six F 2 populations produced by the four mutant combinations (Additional file 9: Figure S5A and C). These results indicate that ZmCSLD1 is the causal gene for qLW10.

Characterization of ZmCSLD1
ZmCSLD1 is a member of the conserved family of CSLD proteins in maize that are important for cell growth and development [36]. ZmCSLD1 contains a "D,D,D,QXXRW" motif with a RING-type zinc finger-like domain and two transmembrane domains at the N-terminus and six transmembrane domains in the C-terminal region (Additional file 7: Figure S4). The phylogenetic tree of ZmCSLD1 orthologues across 32 angiosperms indicate that ZmCSLD1 is highly conserved in different classes such as asterids, core eudicots, rosids and poales (Additional file 10: Figure S6).
To examine the subcellular localization of ZmCSLD1, we transiently expressed pSuper:ZmCSLD1-GFP in maize c LOD profile of qLW10, which was identified in the F 2 and F 2:3 populations. d qLW10 was mapped to a 3.3-Mb genomic DNA region between the markers IDP9050 and MTL02 by using 3,292 BC 2 F 1 and 669 BC 3 F 1 plants. The numbers (Num) and LW are shown for the recombinant plants (I-III) and the non-recombinant plants (IV). The black and grey bars represent the chromosomal segments for the homozygous MT03-1 and heterozygous alleles, respectively. The data are shown as the mean ± SD. Scale bar = 2 Mb. e Four allelic variations in the candidate gene ZmCSLD1; the insertions (triangles) and point mutations are shown. Scale bar = 1 kb. f-g LW in the lines with four pairs of allelic samples (f) and three pairs of F 1 hybrids (g). The data are shown as the mean ± SD (n = 30). **P < 0.01 (Student's t-test) protoplast cells under the blank control of pSuper:GFP and co-transformed it with mCherry-HDEL-RFP (endoplasmic reticulum (ER) marker), Man1-mCherry-RFP (cis-Golgi marker) and RFP-SYP61 (trans-Golgi marker) as the controls for subcellular localization (Fig. 3a-p). A punctate pattern of GFP signals was observed in the protoplasts when ZmCSLD1-GFP was expressed (Fig. 3c, f and n), and this pattern was different from the ER marker location pattern (Fig. 3e). The partial overlap of ZmCSLD1-GFP with the cis-Golgi marker and complete merging with the trans-Golgi marker indicate that ZmCSLD1 localizes to the trans-Golgi ( Fig. 3k and e); this finding is consistent with the well-known subcellular localization of its orthologues in rice (OsCSLD4) and Arabidopsis (AtCSLD5) [21,37].
In previous studies, xylan and homogalacturonan synthase activity were reduced in Atcsld5 [21], and the content of xylan, cellulose and homogalacturonan were  [37]. These results suggest that Zmcsld1 might participate in the cell wall polysaccharides biosynthesis due to their protein identity. To test the association between ZmCSLD1 and cell wall compositions, the cell wall monosaccharide composition and cellulose content of the ear leaves in ZmCSLD1 W22 and Zmcsld1 W22 were measured at 7 days after flowering. Compared with ZmCSLD1 W22 , Zmcsld1 W22 showed a significant increase of mannose and glucose content, but a decrease tendency in xylose, galactose and cellulose content (Additional file 11: Table S5). These results confirmed that ZmCSLD1 played an important role in the cell wall polysaccharide biosynthesis.

ZmCSLD1 has pleiotropic and heterogeneous effects
To validate that ZmCSLD1 is also the causal gene that affects the other eight traits, 16 traits were examined in qLW10 MTL and qlw10 MTL , qLW10 BYK and qlw10 BYK , ZmCSLD1 W22 and Zmcsld1 W22 and ZmCSLD1 Mo17 and Zmcsld1 Mo17 (Additional file 12: Figure S7). Twelve traits showed large differences between qLW10 MTL and qlw10 MTL , and the qlw10 MTL allele conferred overall decreases in these traits of 8.99% (veinlet number, VN) to 33.47% (LW). The same QTL at the ZmCSLD1 locus was identified for each of the nine common traits in the LEE-12 × MT03-1 F 2 population, which further indicates the pleiotropic effects of ZmCSLD1. Furthermore, 10 of the traits, including narrow leaves, dwarf plants, and thin stems, showed consistent patterns in all the lines with the loss-of-function alleles, whereas some traits showed significant differences between qLW10 BYK and qlw10 BYK and between ZmCSLD1 W22 and Zmcsld1 W22 but not between qLW10 MTL and qlw10 MTL and ZmCSLD1 Mo17 and Zmcsld1 Mo17 . The tassel branch number (TBN), for example, which is determined by the branching ability of the shoot apical meristem, varied in the different genomic backgrounds. Even for the traits with striking effects of the loss-of-function alleles, the effects varied among the different alleles, e.g., the effects of LW ranging from 17 to 42% (Additional file 12: Figure S7). This phenomenon may be due to the heterogeneous effects of different alleles because the causal sites occurred in different protein domains (Fig. 4a). Both qlw10 MTL and Zmcsld1 W22 contained large insertions before the "D,D,D,QXXRW" motif and transmembrane domains that were predicted to lead to a complete loss of the ZmCSLD1 function, whereas the missense mutations in the "D,D,D,QXXRW" motif of qlw10 BYK (G839D) and in the last transmembrane domain of Zmcsld1 Mo17 (W1184R) had weak effects on the traits.
To further confirm the pleiotropic and heterogeneous effects of ZmCSLD1, the 16 traits were also investigated in four F 1 hybrids, including the three pairs described above that were crossed with Zmcsld1 W22 and two other combinations, namely, Zmcsld1 Mo17 × qlw10 BYK and ZmCSLD1 Mo17 × qLW10 BYK (Additional file 8: Table S4). As expected, the performance of all the measured traits in all three F 1 hybrid crosses between the strongest allele and three other alleles supported the pleiotropic and heterogeneous effects of ZmCSLD1. Unexpectedly, the defective phenotype for most traits in the Zmcsld1 Mo17 × qlw10 BYK F 1 hybrid was completely complemented (Fig. 4b), similar to the phenomenon of intragenic complementation in the cellulose synthase gene in Arabidopsis AtCESA3 [38]. A further comparison of the LW among genotypes in six F 2 populations indicated that the effect of the qlw10 MTL allele was comparable to the effect of the Zmcsld1 W22 allele because both alleles produced an abnormal ZmCSLD1 protein, whereas the qlw10 BYK and Zmcsld1 Mo17 alleles had weaker effects (Additional file 9: Figure S5). In addition, the leaves of the individuals with heterozygous (C/A) genotypes were significantly wider than the leaves of the individuals with both homozygous genotypes (C/C and A/A) in the Zmcsld1 Mo17 × qlw10 BYK F 2 population (Additional file 9: Figure S5A); this finding strongly supports the Zmcsld1 Mo17 × qlw10 BYK F 1 intragenic complementation.
To test whether ZmCSLD1 could form a homodimer at molecular level, we cloned two fragments, S 1-331 from 1 to 331 amino acids and S 400-984 from 400 to 984 amino acids of ZmCSLD1, into both pGBKT7-BD and pGADT7-AD vectors. The yeast two hybrid analysis indicated that segment S 1-331 could interact with itself, but segment S 400-984 couldn't (Fig. 4c). In addition, the mutation of qlw10 BYK occurs in the cytoplasmic domain of ZmCSLD1, which possesses glycosyltransferase activity, and the Zmcsld1 Mo17 mutation is located in the last transmembrane domain of ZmCSLD1. Taken together, we raised a model to explain the intra-genic complementation between qlw10 BYK and Zmcsld1 Mo17 (Fig. 4d). Both catalytic and transmembrane domains are necessary to the function of ZmCSLD1. Either mutation in these two domains in homologous lines will weaken the function of ZmCSLD1, while the function of ZmCSLD1 will be complemented in F 1 hybrids of two mutation alleles in different domains.
Decreased cell division rate contributes to a reduced organ size Cell division and expansion are the primary ways of plant growth. The relationship between ZmCSLD1 and cell division or expansion raises the question of whether ZmCSLD1 has pleiotropic effects on the size of multiple organs in maize. To investigate the relationship between organ size and ZmCSLD1, the LL and LW of all unfold leaf blades were measured to assess the leaf morphology of 40-day-old seedlings (Fig. 5a). An increasing difference between qLW10 MTL and qlw10 MTL was observed for LW after the second leaf stage and for LL after the third leaf stage, and the fourth LW and length of qLW10 MTL reached 1.36 and 1.21 times the LW and LL of qlw10 MTL , respectively. This result was highly consistent with the morphology of mature leaves ( Fig. 2a and b). To address the relationship between LW and cell division or cell expansion, the cell size was quantified and the cell number was calculated for the fourth leaf, which had the greatest variation in the leaf stages (Additional file 13: Figure S8). Compared with qLW10 MTL , cell width increased by 19.52% in qlw10 MTL , whereas cell length decreased by 4.44% (Fig. 5b); the cell number in both the lateral and longitudinal axes was notably reduced by 40.55% and 13.79%, respectively (Fig. 5c). Similarly, Fig. 4 ZmCSLD1 regulates plant architecture, and the effects of ZmCSLD1 vary among different alleles. a Schematic diagram of the domains and the location of causative alleles in Zmcsld1. The protein domains were predicted according to the descriptions of Zeng and Keegstra [92] and Hunter et al. [36]. b The pleiotropic effects of Zmcsld1 by comparing four Zmcsld1-combined F 1 hybrids with their ZmCSLD1-combined F 1 hybrids, respectively. The significance levels of the differences analysed by Student's t-test (n = 30) are indicated in different colours. Traits abbreviations are listed in Additional file 2: Table S2. c Yeast two-hybrid system showing the interactions between AD-S 1-331 and BD-S 1-331 . pGADT7-T/pGBKT7-53 was used as a positive control. d Schematic diagram of the intragenic complementation between qlw10 BYK and Zmcsld1 Mo17 all of three other alleles, qlw10 BYK , Zmcsld1 W22 , and Zmcsld1 Mo17 , showed significant decrease in LW and cell number in the lateral axes of the third leaf from 20-day-old seedlings (Additional file 14: Figure S9). These results imply that the narrowing of the qlw10 MTL leaf was mainly caused by a decrease in cell number due to a reduction in cell division.
In addition, based on the expression levels of ZmCSLD1 in 13 tissues at various developmental stages [39][40][41][42][43], ZmCSLD1 has been determined to be expressed in most maize tissues (Additional file 15: Figure S10). ZmCSLD1 expression was high in the ear primordia, tassel primordia, shoot apical meristem, third leaf base and ovule but was low in the root and embryo at 10 days after pollination (DAP), in the endosperm at 6 DAP, and in the seeds at 3 DAP, and no expression was detected in the third leaf tip, pre-emergence tassel, silk or anthers (Additional file 15: Figure S10A). In addition, ZmCSLD1 was highly expressed in the early stage of seed development, and dropped successively during the seed maturation process (Additional file 15: Figure S10B), which corresponds to a phase of mitotic cell proliferation during maize embryogenesis and endosperm development [40,44,45]. To further validate the association of ZmCSLD1 with cell proliferation, we sampled three root tissues and shoot tissue from 5-day-old seedlings, and four leaf tissues of the third leaf tissues from 10-day-old seedling in qLW10 MTL and qlw10 MTL (Fig. 6a and b). As expected, ZmCSLD1 were highly expressed in primary root tip, shoot base and the third leaf base tissues, and expressed at low levels in primary root middle, third leaf middle (proximal and distal) and tip (Fig. 6c). The expression pattern of ZmCSLD1 was highly consistent with that of three cell circle related genes CycB1;4, CycD3;1b and histone H2B which expressed during G2/M transition, G1/S transition and S phase in cell cycle, respectively [46,47] (Fig. 6d; Additional file 16: Figure S11A and B). Specially, the expression level and pattern of ZmCSLD1 was most similar to CycB1;4 (r = 0.91), suggesting that ZmCSLD1 may also be expressed in the G2/ M transition-phase during cell division. Additionally, ZmCSLD1 was also highly expressed in primary root base tissues which might be caused by the development of lateral roots. Collectively, ZmCSLD1 was highly expressed in the early stages during rapid growth in all tissues, whereas this expression was low in the mature stages ( Fig. 6c; Additional file 15: Figure S10B), which is consistent with the function of ZmCSLD1 that relates to cell division and development.

ZmCSLD1 participates in the cell wall formation during cell division
Organ size growth is a complex process that consists of cell division and expansion. The cell wall is an integral structure in plant cells and mainly comprises cellulose, hemicellulose and pectin. These components play essential roles in plant cell division and expansion, plant growth and, consequently, morphogenesis [48]. There are ten subfamilies in the cellulose synthase (CESA) superfamily according to protein sequence similarity, including nine cellulose synthase-like (CSL) families and one CESA family. These subfamilies contain a "D,D,D,QXXRW" motif thought to catalyse the synthesis of the β-glycosyl unit structure of cellulose and hemicellulose [49][50][51]. All CESA proteins localize to the plasma membrane to synthesize cellulose on the cell surface [52,53]. Most CSL proteins, such as AtCSLA9, AtCSLC4, BdCSLF6, HvCSLH1 and AtCSLD5, accumulate in the Golgi, where they are thought to produce hemicellulose and pectin polysaccharides for subsequent vesicular secretion to the extracellular space [21,[54][55][56]. Heterologous expression has demonstrated that the CSLA family is related to β-mannan or glucomannan polysaccharide synthesis. CSLC proteins function in assembling the β-1,4 glucan backbone of xyloglucan, and the grass-specific families of CSLF and CSLH encode the mixed linkage glucan synthases [55,[57][58][59]. The other CSL families, namely, CSLB, CSLD, CSLE, CSLG and CSLJ, also likely participate in synthesizing hemicellulose polysaccharides, although it is still unclear which products they synthesize.
Among the CSL families, the CSLD proteins display the greatest amino acid sequence similarity to the CESA Fig. 6 Expression patterns of ZmCSLD1 and CycB1;4 in root, shoot and leaf tissues in qLW10 MTL and qlw10 MTL . a-b Positions of primary roots and shoot from 5-day-old seedlings (a) and the third leaf tissue from 10-day-old seedlings (b), which are indicated by white frame, were sampled for expression analysis. Scale bar = 2 cm. c-d Relative expression of ZmCSLD1 (c) and CycB1;4 (d) in eight tissues corresponding to (a) and (b) in qLW10 MTL and qlw10 MTL were detected by qRT-PCR, data are shown as the mean ± SE (n = 3) family [50]; therefore, the CSLD proteins were predicted to have similar functions in synthesizing cell wall components. This prediction was confirmed by the finding that the CSLD proteins participate in the synthesis of cell wall polysaccharides, such as non-crystalline cellulose and xyloglucan [21,37,60,61]. Our finding that ZmCSLD1 is a trans-Golgi protein is consistent with the subcellular localization of its orthologues AtCSLD5 and OsCSLD4; this finding suggests that ZmCSLD1 is likely to participate in the synthesis of hemicelluloses and non-crystalline cellulose in the Golgi apparatus, or cellulose with the assembly of CSLD proteins in the Golgi followed by transport to the plasma membrane [21,37,60,[62][63][64]. In addition, the cell wall composition changed in Zmcsld1, especially a decrease tendency in the main components of cell wall backbone, xylose and cellulose content (Additional file 15: Table S5). These results, together with a previous study by Hunter et al. [36], suggest that ZmCSLD1 participates in the cell wall polysaccharide biosynthesis of cell plate during cell division. Whereas, the nature of the function of ZmCSLD1 as an enzyme, including its substrates and products, remains to be identified.

ZmCLSD1 affects cell division and organ size and has pleiotropic effects
Pleiotropic effects, which refer to one gene that can affect many traits, have been identified in many species [65]. These effects play an important role in development and evolution because of the complexity of the biochemical and developmental networks [66]. Pleiotropic effects often cause genetic correlations in bi-or multi-parent-derived populations [6,8]. In our study, nine traits with QTL identified at the qLW10 locus had a moderate to high correlation with one another (r = 0.44-0.94, F 2:3 population) (Fig. 1b), which differs from other studies [6][7][8][9][10][11]. For example, the correlation between LW and LL was 0.62 in our F 2:3 population but only 0.08 in an NAM population [10]. In addition, the NILs and mutants of ZmCSLD1 showed pleiotropic effects (Additional file 12: Figure S7), which also confirms that ZmCSLD1 underlies the pleiotropic effects of qLW10.
Pleiotropic genes often participate in transcription regulation, hormone signalling, ubiquitin-mediated proteolysis and cell wall biosynthesis pathways, which control organ size by regulating cell proliferation [5,67]. The size and growth of organs are determined by the characteristics of their cells, such as their width, length and number. Because ZmCSLD1 encodes an enzyme in cell wall biosynthesis, it is interesting to investigate whether the pleiotropic effects of ZmCSLD1 due to organ size variation are caused by cell division or expansion. In the current study, the reduction in cell number in both the lateral and longitudinal leaf axes was far more pronounced than the reduction in cell size (Fig. 5b and c; Additional file 14: Figure S9). Thus, the reduction in cell number appears to be the main cause of the narrow organ phenotype, which is highly consistent with the results of Hunter et al. [36]. Furthermore, ZmCSLD1 was highly expressed in immature tissues, such as in the root tip, immature leaf base, ear primordia, shoot apex and tassel primordia and young-embryo endosperm and seed ( Fig. 6c; Additional file 15: Figure S10A), but the transcript was nearly undetectable in mature tissues ( Fig. 6c; Additional file 15: Figure S10B). Similar expression patterns of its orthologues in Arabidopsis and rice were validated by a β-glucuronidase (GUS) expression from the AtCSLD5 and OsCSLD4 promoters, respectively [21,37]. The common characteristics of these tissues in which ZmCSLD1 is highly expressed is that these tissues are regions of rapidly dividing cells. In addition, OsCSLD4 is specifically expressed during the M phase of the cell cycle and the AtCSLD5 functions in the cell plate formation period, both of which are critical phases in mitosis [68,69]. The sequence conservation of ZmCSLD1 with OsCSLD4 and AtCSLD5 strongly supports the role of ZmCSLD1 in cell division.

Mechanism of intragenic complementation between different ZmCSLD1 alleles
The recovery of the narrow-leaf phenotype in the heterozygotes (Fig. 4b) indicates that the Zmcsld1 Mo17 and qlw10 BYK alleles of the ZmCLSD1 gene exhibit intragenic complementation. Intragenic complementation has been reported in plants, e.g., among the pattern formation gene GNOM [70], the abscisic acid biosynthesis gene ABA2 [71], the cytokinin receptor gene WOL/CRE1 [72], the brassinosteroid receptor gene BRI1 [73], and the cellulose biosynthesis genes AtCESA1 [74] and AtCESA3 [38]. The common characteristics of these genes is that they encode proteins with multiple functional domains or proteins that form homomeric or higher order homomultimeric complexes. For example, AtCESA1 and AtCESA3 are a part of the large cellulose synthase complex that contains approximately 36 CESA polypeptides in Arabidopsis and that is divided into six rosette subunits, each of which consists of three CESA isoforms [53,54]. The three CESA isoforms form cellulose synthase complexes through various heterodimeric and/ or homodimeric combinations to synthesize the cellulose microfibrils of the cell wall [75][76][77]. Therefore, these CESA genes can complement one another when the product of one allele functionally compensates for the product of a second allele in the same complex.
Similarly, the proteins of the CLSD family also likely form a heterodimeric or homodimeric enzyme complex that functions in generating non-crystalline cellulose or the β-1,4 glucan backbone of hemicellulose [78]. Notably, the mutations in qlw10 BYK and Zmcsld1 Mo17 result in different amino acid residues in the "D,D,D,QXXRW" motif (G839D) and the last transmembrane domain (W1184R), respectively (Fig. 4a). Thus, the mechanism of intragenic complementation in the F 1 hybrids between qlw10 BYK and Zmcsld1 Mo17 by the formation of the homodimers of the two mutant alleles that compensate for the defective cytosolic catalytic domain and the disabled transmembrane domain in the same complex, which consequently reduces or recovers the narrow-leaf phenotype in the heterozygotes (Fig. 4b-d). However, further studies are necessary to confirm how many ZmCLSD1 subunits are present in the complex and to determine whether other protein subunits are involved.

Conclusion
Our results show that qLW10 was resolved to be ZmCSLD1 through map-based cloning. ZmCSLD1 encodes an enzyme in cell wall biosynthesis and controls organ size by altering cell division, which results in pleotropic effects on multiple traits in maize. The subcellular localization of ZmCSLD1 in the trans-Golgi shows that ZmCSLD1 participates in cell wall polysaccharides formation during cell division. In addition, ZmCSLD1 may form a homomeric or higher order homomultimeric complex to catalyse polysaccharide biosynthesis because of intragenic complementation.

Plant materials and growth conditions
Two maize double haploid lines, MT03-1 with a small plant architecture and LEE-12 with a large plant architecture, were used to develop an F 2 population including 197 individuals. A heterozygous inbred family, BYK-HIF, was also used, which was derived from a By815 × K22 recombinant inbred line population. A pair of NILs, ZmCSLD1 MO17 and Zmcsld1 MO17 , were developed from a line in a Mo17 × X26-4 (Zea mays ssp. mexicana) BC 2 F 5 population by backcrossing once with Mo17 and selfing four times. A UniformMu line (UFMu-04904) was ordered from the Maize Genetics Cooperation Stock Center and backcrossed twice with W22 to reduce the background, which generated ZmCSLD1 W22 and Zmcsld1 W22 . The plants were grown in the fields for QTL mapping, fine mapping and allelism tests.

Construction of a genetic linkage map and QTL mapping
All 197 F 2 individuals and F 2:3 families in the LEE-12 × MT03-1 F 2 population, along with both parents, were grown in Beijing (40°N, 116°E), China, in 2012 and Shenyang (42°N, 123°E), China, in 2013, respectively. All F 2 individuals and their parents were genotyped by using the MaizeSNP3K subset (3072 SNPs) of the Illumina MaizeSNP50 BeadChip [79]. SNP genotyping was performed on the Illumina Gold-enGate SNP genotyping platform [80] at the National Maize Improvement Center of China, China Agricultural University. The quality of each SNP was checked manually, and poor-quality SNPs were excluded. Using 767 high-quality polymorphic SNPs, a genetic map of 1833.66 cM with an average interval of 2.42 cM between adjacent markers was constructed by R/qtl. QTL mapping for 11 measured traits (Additional file 1: Table S1; Additional file 2: Table S2) in both the F 2 and F 2:3 populations was performed by using composite interval mapping [81] that was implemented in Windows QTL Cartographer 2.5 [82]. The threshold logarithm of odds (LOD) value to declare the putative QTL was estimated by permutation tests with a minimum of 1,000 replicates at a significance level of p < 0.05 [83]. The confidence interval of the QTL position was determined by using the 1.5-LOD support interval method [84].

Positional cloning of qLW10
Fine mapping of qLW10 was based on 3,292 BC 2 F 1 and 669 BC 3 F 1 plants bred from the backcross between MT03-1 and LEE-12 with MT03-1 as the recurrent parent. Additionally, an F 6 family, BYK-HIF, in the By815 × K22 RIL population that had heterozygous alleles at the qLW10 locus and that varied in LW was also used to fine map qLW10 based on 720 F 2 plants descended from BYK-HIF. The markers used for positional cloning by using the two populations are listed in Additional file 17: Table S6.

Genotyping of various Zmcsld1 alleles
To determine the causal variants of the four Zmcsld1 alleles, full-length ZmCSLD1 was sequenced from two pairs of NILs, qLW10 MTL /qlw10 MTL and qlw10 BYK / qLW10 BYK , and two pairs of mutants and their wild type, Zmcsld1 MO17 /ZmCSLD1 MO17 and ZmCSLD1 W22 / Zmcsld1 W22 . The genotype of Zmcsld1 W22 was confirmed with the Mu-transposon primer TIR6 and gene-specific primer CSLD1-P3. A list of primers for this genotyping is given in Additional file 17: Table S6. Furthermore, four Zmcsld1 alleles and their wild-type counterparts were genotyped by using the MaizeSNP6K subset (5,259 SNPs) of the Illumina MaizeSNP50 BeadChip to check their genetic backgrounds.

Allelism tests
Ten F 1 hybrids were generated from eight lines with different ZmCSLD1 alleles to evaluate the allelic effects as described in Additional file 8: Table S4. All the hybrids were planted in a randomized complete block design with two replications in Sanya (18°N, 110°E), China, in 2015. Each hybrid was grown in a two-row plot with 5-m rows and 0.67 m between rows. The 16 traits were measured in at least 30 plants per hybrid as described in Additional file 2: Table S2. In addition,~120 individuals of six F 2 populations, including qlw10 MTL × qlw10 BYK , qlw10 MTL × Zmcsld1 W22 , qlw10 BYK × Zmcsld1 W22 , Zmcsld1 Mo17 × qlw10 MTL , Zmcsld1 Mo17 × qlw10 BYK and Zmcsld1 Mo17 × Zmcsld1 W22 , were planted in Sanya (18°N, 110°E) in 2015. These individuals were genotyped by SNP and Mu-transposon insertion markers (Additional file 17: Table S6) and were phenotyped for LW, LL, PH, EH (Additional file 2: Table S2).

Construction of the phylogenetic tree
The predicted protein sequences of ZmCSLD1 orthologues were searched by using BLAST against the UniProt database (http://www.uniprot.org/blast/) for sequences with alignment scores > 3,870, and one sequence in each species was kept for analysis. These protein sequences were aligned by using MUSCLE [85], and the aligned sequences were used to construct the phylogenetic tree by using the maximum likelihood method with 1,000 bootstrap replications in MEGA version 6.0 [86].

Subcellular localization
The coding sequence of ZmCSLD1 was amplified by using a cDNA template from 5-day-old B73 seedlings. The PCR product was amplified by using primer rCSLD1-03 (Additional file 17: Table S6) and cloned into the SUP1300 vector at the XbaI and SpeI sites to encode a ZmCSLD1-GFP fusion protein with the Super promoter. Maize protoplasts were obtained as described by Burdo et al. [87], except that B73 was used instead of B73 × Mo17 F 1 . Plasmids (15 μg) were transformed into protoplasts and incubated for 16 to 20 h in the dark at 25°C before monitoring their GFP expression by using confocal microscopy (Zeiss 710, Germany).

Measurement of leaf and cell size
To measure leaf size and cell number, the NILs, qLW10 MTL and qlw10 MTL , were grown in the field, and the completely extended fourth leaf from 40-day-old seedlings was sampled. Simultaneously, the LL and LW of the six unfolded leaves (L2-7) were measured. For the remaining qlw10 BYK /qLW10 BYK , and two additional mutants and their wide type, Zmcsld1 MO17 /ZmCSLD1 MO17 , and ZmCSLD1 W22 /Zmcsld1 W22 , were planted in greenhouse, and the leaf and cell size of the third leaf from 20-day-old seedlings was measured.The replication of the abaxial epidermal surface in the middle leaf was carried out as described by Moon et al. [88]. Briefly, colorless nail polish was applied to the leaf abaxial surface and allowed to dry completely, and then the dried nail polish replicas were peeled off by transparent adhesive tape and pasted to glass slides. Leaf epidermal cells were observed and photographed by using light microscopy (Leica DM2000 LED, Germany) with a Leica DFC450 camera. For each sample, cell sizes were measured using ImageJ 1.45 s (ImageJ, National Institutes of Health, USA) in a 0.585 mm by 0.785 mm field, and at least 50 normal cells except stomata guard cell, smaller and irregular cells were measured per sample.

RNA isolation and qRT-PCR
qLW10 MTL and qlw10 MTL were planted in greenhouse with 16-h-light/8-h-dark photoperiod for expression analysis. The primary root tip (5 mm, cell division region), middle (5 mm, cell elongation region), base (5 mm, root hair region), and shoot base (2 mm, including shoot apical meristem) from 5-day-old seedlings (Fig. 6a), and the third leaf base (1 cm), proximal middle (1 cm), distal middle (1 cm), and tip (1 cm) from 10-day-old seedlings ( Fig. 6b) were sampled for analyzing ZmCSLD1, CycB1;4, CycD3;1b and histone H2B expression. Each tissue had three independent biological replicates with ten plants per biological replicate. Total RNA was extracted using a TianGen plant RNA extraction kit (China). First-strand cDNA was synthesized using the PrimeScript 1st Strand cDNA Synthesis kit (TaKaRa, Japan). qRT-PCR was carried out in triplicate for each sample using the SYBRGreen I kit (TaKaRa) on a 7500 Real-Time PCR System (Applied Biosystems, USA). Maize TUBG was used for normalization between samples [89]. Quantification of relative expression was based on the comparative threshold cycle method