Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
Photosynthesis underpins the viability of most ecosystems, with C4 plants that exhibit ‘Kranz’ anatomy being the most efficient primary producers. Kranz anatomy is characterized by closely spaced veins that are encircled by two morphologically distinct photosynthetic cell types. Although Kranz anatomy evolved multiple times, the underlying genetic mechanisms remain largely elusive, with only the maize scarecrow gene so far implicated in Kranz patterning. To provide a broader insight into the regulation of Kranz differentiation, we performed a genome-wide comparative analysis of developmental trajectories in Kranz (foliar leaf blade) and non-Kranz (husk leaf sheath) leaves of the C4 plant maize. Using profile classification of gene expression in early leaf primordia, we identified cohorts of genes associated with procambium initiation and vascular patterning. In addition, we used supervised classification criteria inferred from anatomical and developmental analyses of five developmental stages to identify candidate regulators of cell-type specification. Our analysis supports the suggestion that Kranz anatomy is patterned, at least in part, by a SCARECROW/SHORTROOT regulatory network, and suggests likely components of that network. Furthermore, the data imply a role for additional pathways in the development of Kranz leaves.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The development of an anatomical and biochemical framework upon which photosynthesis operates is a pre-requisite for plant survival, with the specifics of this framework differing between plant groups. Most plants fix atmospheric carbon dioxide into a three-carbon compound and hence are referred to as C3 plants. In contrast, C4 plants such as maize (Zea mays) have evolved an alternative pathway in which the first product of photosynthesis is a four-carbon compound. The majority of C4 plants split the biochemical reactions of photosynthesis between two morphologically distinct cell types known as bundle sheath (BS) and mesophyll (M) (Langdale, 2011). BS cells and M cells are arranged in concentric wreaths around veins (V) to form ‘Kranz’ anatomy (Haberlandt, 1896), in which all veins are separated by two BS and two M cells in a repeating V–BS–M–M–BS–V pattern. C4 metabolism in leaves with Kranz anatomy is the most productive photosynthetic pathway on earth.
Concerns about future food supplies have led to the suggestion that C4 traits should be introduced into C3 plants such as rice (Oryza sativa) to increase crop yields (Hibberd et al., 2008; von Caemmerer et al., 2012). Clearly, this ambitious goal requires a fundamental understanding of how leaf development in C4 plants differs from that in C3 plants. Comparisons between leaf transcriptomes of closely related C3 and C4 species in the genera Cleome and Flaveria showed that at least 500 genes are differentially expressed in mature C3 and C4 leaves, and suggested that an upper limit of 3500 gene changes were associated with the evolutionary transition from C3 to C4 (Brautigam et al., 2011; Gowik et al., 2011). Other studies have used systems approaches to analyze developmental transitions along the maize leaf blade (Li et al., 2010; Majeran et al., 2010; Pick et al., 2011), and to identify similarities and differences between differentiated BS and M cells (Li et al., 2010; Chang et al., 2012). These latter studies benefit from comparisons in the context of a single species, and have provided an overview of transcriptome dynamics during the induction and maintenance of the C4 pathway in the leaf. However, in all cases, the vein spacing and cellular arrangement associated with Kranz anatomy were already established. A more recent study examined earlier developmental stages, but embryonic leaf primordia of various ages and displaying various degrees of Kranz differentiation were pooled for analysis (Liu et al., 2013). As such, no overview of changes in gene activity that occur during the development of Kranz anatomy is currently available.
In maize, the morphological trajectory associated with the development of Kranz anatomy has been well documented through histological and cell lineage analyses (Sharman, 1942; Esau, 1943; Langdale et al., 1989). Importantly, veins develop with associated BS cells, and the positioning of veins determines the number of M cells separating them. The timing of vein formation is such that the leaf mid-vein is initiated first, and then lateral veins begin to appear at plastochron 2 (P2; a plastochron is the time interval between initiation of primordia, and P1 is always the most recently initiated; Esau, 1943; Sharman, 1942). A previous transcriptome study of maize shoot development identified gene cohorts in P1 primordia and in the associated shoot meristem (Takacs et al., 2012). However, it is the development of intermediate veins during P5 that leads to the vein spacing pattern that is characteristic of Kranz anatomy (Sharman, 1942). Intermediate veins are initiated at the tip of P5 primordia, and, by the end of the plastochron, have extended from the leaf blade, through the developing ligule region, into the leaf sheath. However, anastomoses (fusions) at the ligule result in fewer intermediate veins in the sheath than in the blade. Kranz anatomy is thus a feature of the leaf blade but not of the leaf sheath. This difference is seen most dramatically in the husk leaves that surround the ear, where most if not all of the leaf is sheath tissue. Veins in husk leaf sheaths are surrounded by a wreath of BS cells, but then each vascular bundle is separated by up to 20 m cells rather than the two seen in Kranz anatomy (Langdale et al., 1988). In combination, these temporal and anatomical differences suggest that positive regulators of Kranz anatomy act between P2 and P5 in foliar leaf primordia, and are repressed or absent in equivalent husk leaf primordia. Conversely, negative regulators act in husk primordia but not in foliar primordia.
Although the histological manifestation of Kranz anatomy in maize is understood, very little is known about the genetic regulation of Kranz development. Importantly, a recent reverse genetics study reported that the maize scarecrow1 (Zmscr1) mutant exhibits perturbed leaf venation patterns and ectopic BS cells (Slewinski et al., 2012). On the basis of this phenotype, it was proposed that the SCARECROW (SCR)/SHORTROOT (SHR) pathway, which regulates radial patterning in the stem and roots of C3 plants, was recruited into the leaf during evolution of Kranz anatomy (Slewinski et al., 2012). However, no association between other genes in the SCR/SHR pathway and the development of Kranz anatomy has yet been demonstrated. To obtain a comprehensive overview of changes in gene activity during the patterning of Kranz anatomy, we have used Illumina sequencing to generate transcriptomes of developing foliar (Kranz) and husk (non-Kranz) primordia. These transcriptome profiles have been compared with later stages of leaf development where the initiation or maintenance of C4 metabolism is superimposed on the pre-established anatomical framework. Our results reveal cohorts of genes that are active during early leaf development. From within these cohorts, we have identified a group of transcription factors associated with the patterning of Kranz anatomy. The data support a role for the SCR/SHR pathway in Kranz development, but also reveal additional candidate regulators.
To characterize transitions in transcriptome signatures throughout maize leaf development, a range of biological samples representing various developmental trajectories and anatomical traits were harvested for RNA sequencing analysis. Figure 1(a) shows the position on the plant and the overall morphology of the samples used. Foliar and husk leaf samples were both harvested at five stages of development covering the range from P1 to P9. To define the anatomical status of all ten samples, transverse sections were examined by light microscopy (Figure 1b–k). In both foliar and husk samples, P1 and P2 primordia comprise cytoplasmically dense cells that lack vacuoles, and procambium associated with the mid-vein is barely visible (Figure 1b,g). By P4, however, the mid-vein and developing lateral veins may be distinguished, and the distance between them is already smaller in foliar primordia (Figure 1c) than in husk primordia (Figure 1h). At P5, foliar primordia have initiated intermediate veins and the bauplan of Kranz anatomy is apparent: each vascular centre is surrounded by a wreath of small BS cells, and is separated from the next centre by one or two intervening M cells (Figure 1d). In contrast, intermediate veins are not formed in husk primordia. Instead, cell division and expansion in M cells increases the space and cell number between veins from five at P4, to approximately ten at P5 (Figure 1i), and ultimately to 15 or more in mature husk leaves (Figure 1j,k). Vein spacing does not change after P5, and, as such, both foliar immature (FI; Figure 1e) and foliar expanded (FE) (Figure 1f) leaf blades exhibit the classical cellular arrangement associated with Kranz anatomy (V–BS–M–M–BS–V). A summary of the key anatomical differences between tissues is shown in Table 1.
Table 1. Key anatomical traits and transcriptome profiles of foliar and husk leaf samples
Number of M cells between veins
BS cell size
BS plastid size
Number of signature genes in dataset
Metabolic pathways significantly enriched in signature gene sets
Signature genes are those whose transcripts were detected at significantly higher levels in the named sample compared to the other nine samples (P ≤ 0.05).
Mid-vein Laterals Intermediates
Ribonucleotide synthesis DNA synthesis
Mid-vein Laterals Intermediates
Protein synthesis Lipid metabolism Tetrapyrrole biosynthesis
Distinct transcriptome signatures define different stages of foliar and husk leaf development
To determine whether specific transcriptome signatures define the five developmental stages represented in both foliar and husk samples, we first deduced the number of signature genes in each sample. Signature genes were defined as those whose relative mRNA abundance was significantly higher (P ≤0.05) in the named sample than in all other samples. mRNA abundance estimates, gene annotations and enrichment analyses for all of the signature genes in each sample are provided in Tables S1–S10. A comparative summary (Table 1) shows that the highest number of signature genes (1479) was found in the most photosynthetically active sample (FE), but only ten signature genes were identified in the foliar P3/4 (FP3/4) and husk P3/4 (HP3/4) samples. Generally, there is an increase in the number of signature genes in both foliar and husk tissue as leaves mature and become more metabolically complex. The one exception is seen between the foliar P1/2 (FP)/husk P1/2 (HP) and FP3/4/HP3/4 stages, where the number decreases from approximately 20 to 10. This decrease probably reflects the inclusion of meristematic tissue in the FP and HP samples, such that a suite of meristem-specific genes is represented in the FP and HP transcriptomes, but not in the FP3/4 and HP3/4 transcriptomes.
To provide a comparative overview of the developmental and metabolic differences between foliar and husk leaves, gene ontology (GO) terms, MaizeCyc pathway terms, Mapman terms and Pfam domains were assessed in the signature gene sets (Table 1, Figure 2, Figures S1 and S2, and Tables S1–S10). Although enriched metabolic pathways were not identified in any of the primordia samples, GO term enrichment indicated transcriptional activity in both FP and HP samples, and also in FP5 samples (Figure S1). The shared terms between FP and HP samples, and the lack of corresponding terms in FP3/4 and HP3/4 samples, suggest that the transcriptional activity represented in FP/HP tissues is associated with meristem function, as opposed to leaf formation. At P5, enriched terms indicative of transcriptional activity are seen in foliar but not husk leaf samples, suggesting that foliar leaf-specific differentiation events have been initiated. The difference between FP5 and HP5 tissues is further evidenced by the accumulation profile of transcription factors: transcripts of 22 genes encompassing ten transcription factor families are detected in foliar samples, but only 13 genes are represented in husk samples (Figure 2). In FI/husk inner (HI) and FE/husk exposed (HE) tissues, both shared and organ-specific enriched pathways are detected (Figure S2). Overall, global transcriptome profiles are consistent with a developmental trajectory in foliar leaves from proliferating non-photosynthetic leaf primordia (P1–P5), through expanding leaves that are initiating photosynthesis (FI), to actively photosynthetic FE leaves (Table 1). In contrast, husk transcriptome profiles are consistent with the development of an inner leaf that is expanding to surround the developing ear, and an outer leaf that is photosynthesizing at a low level but is also starting to senesce.
Profile classification analysis distinguishes foliar and husk primordia development
To identify genes associated with the earliest patterning processes that operate in maize leaf primordia, we assigned all genes that were detected in primordia samples to one of seven predetermined profiles (Figure 3). Gene lists in each of the seven profiles were generated for both foliar and husk leaves (Tables S11–S24). Each profile was designed to identify genes associated with specific aspects of leaf development. For example, descending profiles (D1, D2, D3) should contain genes required for meristem function and for very early events in leaf development such as mid-vein formation and adaxial/abaxial axis formation. In contrast, ascending profiles (A1, A2, A3) should contain genes required for the differentiation of lateral leaf veins, patterning of intermediate veins, specification of cell types and early plastid biogenesis. Importantly, genes with known function appeared in the expected profiles (summarized in Table 2). For example, knotted1-like homeobox (KNOX) genes that are required for both shoot and axillary meristem function (Jackson et al., 1994) are present in descending profiles, as are most of the genes required for adaxial–abaxial patterning in the leaf (Husbands et al., 2009). Genes required for patterning medio-lateral and proximo-distal leaf axes are also present in the expected profiles, with a notable difference in mRNA accumulation dynamics for the liguleless1 and 2 genes (Becraft et al., 1990; Walsh et al., 1997) in foliar versus husk leaves. This difference correlates with the presence and absence of ligules in the foliar and husk leaf samples, respectively. The golden2-like (glk) regulators of chloroplast development (Rossini et al., 2001) also exhibit the expected trajectories, in that g2 transcripts show equivalent accumulation dynamics in foliar and husk leaves, consistent with the loss-of-function mutant phenotype (Langdale and Kidner, 1994), whereas Zmglk1 transcripts are specific to foliar A2 profiles and thus to the development of active C4 metabolism. In combination, these observations validate the use of profile-based analysis to identify regulators of leaf development.
Table 2. mRNA abundance dynamics for regulatory genes known to be associated with meristem maintenance, axis patterning in the leaf and chloroplast biogenesis
Columns show gene accession number in maizesequence organism, gene name, foliar and husk profile identification (ID), mean transcript reads in reads per kilobase per million in FP, FP3/4, FP5, HP, HP3/4 and HP5 samples, and non-normalized trendlines of transcript abundance. The discrepancies between profile ID and trendline trajectories (e.g. rs2 is in FN but read data show a descending trajectory) are due to the fact that profiles are defined by significant differences between samples as opposed to general trends.
Lateral organ initiation and adaxial/abaxial patterning
Profile D1: meristematic function
D1 profiles contain transcripts that are detected at significantly higher levels in samples that contain both meristematic and leaf tissue than those that contain only leaf tissue (Tables S11 and S18). As expected, knotted1-like homeobox (KNOX) genes that are required for meristem maintenance (Jackson et al., 1994) are represented in both the FD1 and HD1 profiles, as are homologues of the rice DROOPING LEAF (DL) gene that is required for mid-vein formation (Yamaguchi et al., 2004), and the DROOPING LEAF repressor OsMADS32 (Sang et al., 2012). Two genes that are specific to the FD1 profile may play roles in meristem maintenance: one (GRMZM2G151223, www.maizesequence.org) is related to genes known to be involved in cytokinin signal transduction (Muller and Sheen, 2007), and has been annotated as maize histidine kinase 1 (Zmhk1) (Yonekura-Sakakibara et al., 2004); the other (GRMZM2G065496) is a member of the reproductive meristem (REM) sub-family of B3 domain transcription factors, named after the Brassica oleracea gene REM1 (Wang et al., 2012).
Profile D2: meristematic function and early leaf development
As with D1, D2 profiles contain transcripts that accumulate at higher levels in meristem-containing samples than leaf-only samples (Tables S12 and S19). However, whereas D1 profiles contain transcripts that are detected at significantly higher levels in P3/4 primordia than in P5 primordia, transcripts in D2 profiles are detected at similar levels in the two samples. D2 profile genes are therefore likely to play a role in both the meristem and early leaf primordia. Both FD2 and HD2 profiles contain genes that are known to be expressed within the meristem, for example nam2 (Zimmermann and Werr, 2005) and wox9B (Nardmann et al., 2007), but neither profile contains the KNOX genes that are required for meristem identity per se. Other shared genes in foliar and husk D2 profiles include homologues of Arabidopsis genes that are required for meristem specification (BEL-like/AtBLR; Rutjens et al., 2009) and axial patterning in the leaf (KAN-like; Kerstetter et al., 2001), plus genes involved in auxin signalling (arf4 and arf29; Xing et al., 2011). Although these examples demonstrate overlap between FD2 and HD2 profiles, over 60% of the transcription factors in D2 profiles are specific to either foliar or husk samples, with specificity manifested in two ways. In the first, specificity reflects the presence of distinct gene family members. For example, two homologues of OVATE genes (that are known to repress KNOX genes in Arabidopsis) (Wang et al., 2011) are present in both FD2 and HD2 profiles, but both profiles contain a unique gene family member. The second type of specificity reveals processes unique to foliar and husk samples. For example, only the FD2 profile contains homologues of Arabidopsis and rice genes involved in ethylene signalling (OsERF4 and AtEIN3-like; Guo and Ecker, 2004)] In contrast, the HD2 profile contains homologues of Arabidopsis genes involved in timing of the floral transition (LHY and CO; Putterill et al., 2004), plus maize genes that are known to regulate the floral phase change (mads1; Heuer et al., 2001). The GO term ‘trehalose biosynthetic process’ is also uniquely enriched in the HD2 profile, consistent with the role of ramosa3 and the trehalose metabolic pathway in branching of the maize inflorescence (Satoh-Nagasawa et al., 2006). The differences between FD2 and HD2 profiles thus reflect distinct developmental trajectories towards active vegetative growth and the transition to inflorescence development.
Profile D3: early leaf development
D3 profiles were designed to identify genes that regulate leaf developmental processes prior to P4, in that transcript levels may be equivalent in P and P3/4 samples but must be significantly lower in P5 samples (Tables S13 and S20). Shared genes include homologues of those expressed in early primordia (ANT-like; Mizukami and Fischer, 2000) and those required for axial patterning (zyb9; Juarez et al., 2004). The FD3 profile additionally contains homologues of genes that are expressed in dividing primordia [PAN-like (Chuang et al., 1999), ULT-like (Carles et al., 2005), ig1 (Evans, 2007), zfl1 (Bomblies et al., 2003), zmm16 (Munster et al., 2001)], genes required for axial patterning [kan5 (Zhang et al., 2009), dl2 (Yamaguchi et al., 2004; Ishikawa et al., 2009)], genes involved in epidermal patterning [SPCH-like, SCRM-like (Pires and Dolan, 2010), MIXTA-like (Dubos et al., 2010)], and genes responsive to auxin (iaa4, iaa16 and iaa26; Wang et al., 2010). In contrast to FD3, and consistent with the identity of the axillary meristem, the HD3 profile contains homologues of genes that play roles in light signalling (LAF1-like; Ballesteros et al., 2001), drought/abscisic acid/stress responses [SNAC-like, NAP-like (Zhu et al., 2012), MYB (Dubos et al., 2010), WRKY (Wei et al., 2012), TINY-like (Sun et al., 2008)], flowering time [U2AF35 (Wang and Brendel, 2006), CO-like (Putterill et al., 2004)], and axillary bud outgrowth [gt1 (Whipple et al., 2011), TB1-like (Martin-Trillo and Cubas, 2010), OsNAC2-like (Mao et al., 2007)]. This profile is consistent with integration of signals for the floral transition and with the fact that HP5 samples were harvested from axillary meristems that were clearly differentiated as inflorescences (Figure 1). Notably, over 80% of the transcription factors in the D3 profiles are specific to either FD3 or HD3, indicative of organ-specific differences in early leaf patterning mechanisms. Many of the identified transcription factors cannot be assigned a putative function by homology with characterized proteins (Tables S25), and hence are defined as factors that distinguish the formation of foliar and husk leaf primordia.
Profile A1: organ identity and metabolic function
A1 profiles comprise transcripts that accumulate at consecutively higher levels in P, P3/4 and P5 samples (Tables S14 and S21). These profiles were designed to identify genes that progressively establish organ identity and function. As with the descending profiles, substantially fewer genes are present in HA1 than FA1, and the only shared transcription factor is a BME1-like GATA protein (Reyes et al., 2004). More overlap is found between FA1 and HA2/3 profiles, and between HA1 and FA3 profiles. This pattern is consistent with shared developmental processes that exhibit variant timing in foliar versus husk primordia. Most of the genes encoding transcription factors that are shared between FA1 and HA2/3 profiles are homologues of genes involved in organ outgrowth/differentiation [ATBH16-like (Harris et al., 2011), AmDIV-like (Galego and Almeida, 2002), TSO-like (Hauser et al., 1998), TCP-like (Martin-Trillo and Cubas, 2010)], and photomorphogenesis [e.g. >OBP3-like (Ward et al., 2005), GATA2-like (Reyes et al., 2004)]. The number and range of genes in the FA versus the HA profile suggest greater complexity in photomorphogenesis and differentiation programs in foliar as opposed to husk primordia.
Profile A2/A3: patterning of leaf venation and cell-type differentiation
The A2 and A3 profiles were designed to identify genes required for patterning leaf venation and regulating cell-type differentiation in foliar leaves. The profiles contain transcripts that accumulate to higher levels in P3/4 and P5 samples than in P1/2 (A2 profiles; Tables S15 and S22) or to higher levels in P5 than the other two primordia samples (A3 profiles; Tables S16 and S23). As lateral veins are being extended during P4 and P5, whilst intermediate veins are being patterned, the presence of genes in the FA2/3 profiles that are related to those that regulate xylem differentiation (e.g. XYLEM NAC DOMAIN 1; Zhao et al., 2008) and phloem differentiation (e.g. ALTERED PHLOEM 1; Bonke et al., 2003) was expected (Tables S15, S16 and S25). Similarly, the presence of foliar-specific. AUXIN RESPONSE FACTOR (ARF) and AUX/IAA homologues in the FA3 profile (Figure 4a, Figures S3 and S4, Tables S16 and S25) is consistent with the known role of auxin in vascular differentiation (Mockaitis and Estelle, 2008). However, because very little is known about the patterning of leaf venation or the regulation of BS and M differentiation, it was expected that most of the genes in the FA2 profile would not have been previously characterized and/or annotated.
There are 16 genes of note that are specific to the FA2 profile (Figure 4a, and Tables S15 and S25). First, there are three bHLH genes that may regulate cellular differentiation given the role of other bHLH proteins in patterning processes (Pires and Dolan, 2010). One of the genes is related to a target of the auxin-dependent transcription factor MONOPTEROS, which regulates vascular development in Arabidopsis (Schlereth et al., 2010). Second, there is a bZIP gene (Gibalova et al., 2009), and three DoF zinc finger genes that share clades with the Arabidopsis DoF gene AFFECTING GERMINATION (DAG)-like (Gualberti et al., 2002; Gibalova et al., 2009) and the HIGH CAMBIAL ACTIVITY 2 (HCA2) gene (Guo et al., 2009) (Figure S5), which are expressed in vascular tissue. Notably, transcripts of the DAG-like genes were detected specifically in BS cells of expanded maize leaves (Li et al., 2010). Third, two GRAS family SHORTROOT (SHR)-like genes are present, homologues of which play a role in radial patterning around vasculature in the Arabidopsis root, hypocotyl and stem (Fukaki et al., 1998; Helariutta et al., 2000). Fourth and most notable are seven C2H2 zinc finger proteins. Two are related to the previously characterized Arabidopsis gene DEFECTIVELY ORGANIZED TRIBUTARIES 5 (DOT5) that regulates vascular patterning (Petricka et al., 2008), and three are related to SHOOT GRAVITROPISM 5 (SGR5) (Morita et al., 2006) and JACKDAW (JKD) (Welch et al., 2007) (Figure S6), which are known components of the SHR pathway in Arabidopsis. Recently, orthologues of four of these genes (the bZIP gene, two DAG-like DoF zinc finger genes and a SGR5-like C2H2 zinc finger gene) were shown to be up-regulated in Arabidopsis provascular cells compared to other leaf cell types, suggesting conserved roles in early vascular development (Gandotra et al., 2013). The remaining two C2H2 ZnF proteins are related to maize MYB-related protein 1-interacting proteins (Royo et al., 2009). Intriguingly, MYB-related protein 1-interacting proteins interact with MYB-related protein 1 through a C-terminal domain that is shared with SHR target proteins (Levesque et al., 2006; Royo et al., 2009; Cui et al., 2011). The phylogenetic relationships and expression profiles of all 16 of these genes suggest that they are likely regulators of vascular patterning and cellular differentiation in maize foliar leaves (Figure 4b).
Consistent with the fact that intermediate veins do not form in husk leaves and that cellular differentiation processes are less complex than in foliar leaves (Figure 1), the HA2/3 profiles are dominated by homologues of genes involved in the floral transition and in stress responses, as opposed to genes that are likely to be involved in leaf development (Tables S22, S23 and S25). The only striking feature is the presence of eight R2R3 MYBs in the HA3 profile, five of which are husk-specific (Figure 4a, and Tables S23 and S25). Two of the shared genes are MIXTA-like, and thus are probably involved in epidermal patterning (Perez-Rodriguez et al., 2005). The function of the remaining shared gene and of the five husk-specific genes is unclear, but four of the husk-specific genes share a clade with an Arabidopsis gene that is involved in regulating mucilage deposition and phenylpropanoid metabolism (Penfield et al., 2001; Newman et al., 2004), and the other shares a clade with genes that are involved in defence responses (Dubos et al., 2010) (Figure S7). This observation, plus the presence of SHINE (SHN)1-like homologues that regulate epidermal wax deposition in Arabidopsis (Aharoni et al., 2004), suggests considerable modification of the husk leaf epidermis between P4 and P5, possibly as a defence strategy to protect the developing inflorescence.
Supervised classification identifies genes associated with the development of Kranz anatomy
Supervised classification criteria inferred from anatomical and developmental analyses (Figure 1) were used to identify both positive and negative candidate regulators of Kranz anatomy. For positive regulators, it was assumed that genes would be expressed at significantly higher levels in foliar leaves than in husk leaves, and that Kranz patterning genes would be expressed prior to and/or during P5. For negative regulators, it was assumed that genes would be expressed at significantly higher levels in husk leaf primordia than in foliar leaf primordia, and that expression levels would be significantly higher in primordia than in expanded leaves with established anatomy. Table 3 summarizes the classification criteria applied, and shows that 283 putative positive and 142 putative negative regulators of Kranz anatomy were identified. The corresponding gene lists are supplied in Tables S26 and S27, and include accession numbers, mRNA abundance estimates, and enrichment analyses for GO terms, MaizeCyc pathways, MapMan terms and Pfam domains.
Table 3. Filtration steps to identify putative regulators of Kranz anatomy
All tests were performed at a P value cut-off of ≤0.05.
n = 35 770
n = 35 770
FP or FP3/4 or FP5 significantly > FE + HP + HI + HE
HP or HP3/4 or HP5 significantly > FE + HI + HE
Genes must pass test in two of three FP samples.
Genes must pass test in two of three HP samples
n = 918
n = 2567
FP3/4 or FP5 significantly > or not different to FP
FP not significantly > HP
FP3/4 not significantly < FP
FP3/4 not significantly > HP3/4
HP3/4 not significantly > FP3/4
FP5 not significantly > HP5
HP5 not significantly > FP5
FE or HI or HE not significantly > FP or
HP not significantly > FP
HP5 significantly < FP3/4
HP significantly < FP3/4
HP3/4 significantly > FP
FP5 or FP3/4 > 0
HP5 significantly > FP
HI not significantly > FP, FP3/4, FP5, FI or FE
n = 160
HE not significantly > FP, FP3/4, FP5, FI or FE
FE not significantly > FP, FP3/4, FP5 or FI
n = 334
If in leaf gradient dataset:
If in leaf gradient dataset:
Base not significantly < 1, 4 cm or tip
Tip not significantly > 4 cm, 1 cm or base
1 cm not significantly < 4 cm, tip
4 cm not significantly > 1 cm or base
4 cm not significantly < tip
1 cm not significantly > base
n = 283
n = 142
Amongst the 283 putative positive regulators of Kranz, GO term enrichment identified a cohort of genes associated with the cytoskeleton and with ribosomes (Figure 5a). As the classification criteria selected genes that were highly expressed in developing foliar primordia, this result is expected, as pre-P5 primordia consist of actively dividing cells. However, it is unlikely that such genes play a central role in regulating Kranz anatomy. The enrichment of genes annotated with the term ‘ice binding’ is also unlikely to reflect an association with the regulation of Kranz anatomy. The remaining enriched terms in the positive regulator dataset and all of those in the negative regulator dataset (Figure 5a,b) are of greater interest because they suggest roles in the regulation of transcription (e.g. nucleus, DNA binding).
Within the lists of positive and negative regulators of Kranz anatomy, 71 genes were identified that are predicted to encode transcription factors (Figure 5c). Although only some of these transcription factors are likely to play a role in Kranz patterning, a number of observations support the suggestion that regulatory roles will be confirmed from within either the positive or negative dataset. For putative negative regulators to be feasible candidates, orthologues must be expressed in developing rice leaves. This is the case for the majority of transcription factors in the negative regulator list, in that transcripts are detected in rice seedlings (Table S28). With respect to positive regulators, the presence of Zmscr1 within the list is supportive of a role for at least a subset of this cohort in Kranz development given that Zmscr1 is expressed during early vascular development in maize (Lim et al., 2005), and mutant alleles show some disruption to Kranz anatomy (Slewinski et al., 2012). Similar support is provided by the presence of DOT5 orthologues in the list, given the role of DOT5 in patterning leaf venation in Arabidopsis (Petricka et al., 2008), and by the presence of orthologues of a further five genes that are up-regulated in Arabidopsis provascular cells (Gandotra et al., 2013). Finally, of the positive regulators identified as transcription factors, approximately 81% of the genes (including Zmscr1 plus the SHR and DOT5 orthologues) have been classified as members of co-expression modules associated with increased Kranz differentiation in a pooled embryonic leaf dataset (compared with approximately 27% of transcription factors in the total embryonic leaf dataset: co-expression modules ≥C13; Liu et al., 2013). Key regulators of Kranz anatomy are thus likely to be present within the list of transcription factors shown in Figure 5(c).
Table S29 summarizes data for all detected genes, and shows maize ID, EntrezGene classification (http://www.ncbi.nlm.nih.gov/gene), reads per kilobase per million for replicates of all ten samples, presence in the signature gene set (if relevant), plus affiliation with specific foliar and husk profiles. The co-expression of gene cohorts in spatial and temporal patterns that are associated with the development of Kranz anatomy in maize supports the role of the SCR/SHR pathway in patterning Kranz anatomy, and identifies likely additional components of that pathway. The data further suggest that other pathways are also involved, particularly in the elaboration of venation patterns.
Plant material and growth conditions
Maize inbred line B73 was grown in soil in a greenhouse with a diurnal light regime of 16 h light (supplemented to 300 μm m−2 sec−1) and 8 h dark, a mean daytime temperature of 28°C and a mean night-time temperature of 20°C. Tissue was harvested from 2-, 4- or 8-week-old plants, and placed directly into liquid nitrogen. FP comprised the vegetative apical shoot meristem plus P1 and P2 leaf primordia, FP3/4 comprised P3 and P4 primordia harvested from the same apex and then pooled, FP5 comprised P5 primordia only, FI comprised expanding 5th leaves at P7, and FE comprised fully expanded 3rd leaves at P9. FP5 primordia, FI and FE leaves were all harvested above the ligule to ensure that only blade (Kranz) tissue was represented. HP samples were harvested from the same plants as the foliar samples, from the axils of P9 and P10 leaves. HP samples included the axillary inflorescence meristem plus P1 and P2 husk primordia (the prophyll was discarded). HP3/4 and HP5 samples were harvested from the upper, ear-forming nodes of 4-week-old plants, and inner (HI) and outermost exposed (HE) husk leaves were harvested from the ears of 8-week-old plants. Under our growth conditions, husk leaves did not form blades, and thus all samples comprised only sheath (non-Kranz) tissue.
Leaf samples were fixed overnight in FAA (4% formaldehyde, 5% acetic acid, 50% ethanol), and embedded in Paraplast Plus (Sigma-Aldrich, http://www.sigmaaldrich.com) as described previously (Langdale, 1994). Sections (8 μm) were stained with Safranin/Fast Green (Sigma-Aldrich) and viewed using a Leica DMRB microscope (Leica, http://www.leica-microsystems.com/).
RNA extraction and cDNA synthesis
Multiple individual isolates were pooled prior to RNA extractions. Each sample pool comprised tissues for n = 1200 (FP), 300 (FP3/4), 100 (FP5), 12 (FI), 12 (FE), 2400 (HP), 300 (FP3/4), 100 (FP5), 12 (HI) and 12 (HE) plants. Two independent biological replicates were constructed according to the pool structure above, and each replicate for each sample was subjected to RNA-seq. This pooled replicate strategy was adopted to circumvent the need for RNA amplification and to minimize biological noise from individual samples. RNA was extracted using a mirVana™ miRNA isolation kit (Applied Biosystems, http://www.invitrogen.com/). RNA integrity was analysed by formaldehyde gel electrophoresis, and samples were sent to the Beijing Genome Institute (http://www.genomics.cn/en/) for further quality control.
cDNA library preparation and sequencing were performed at the Beijing Genome Institute. Each RNA sample was treated in the same manner. Total RNA was treated with DNase I and then purified over an oligo(dT) column. Enriched mRNA was sheared and converted into cDNA using standard Illumina protocols (http://www.illumina.com/). cDNAs were subsequently ligated to Illumina adaptors and subjected to standard Illumina paired-end read library preparation. Paired end reads of 72 bp were generated for samples FP1, FI1, FE1, HP1, HI1 and HE1. All other samples were sequenced using 90 bp paired end reads.
Transcript quantification and differential gene expression analysis
Paired end reads were subject to quality-based trimming using the FASTX toolkit (Goecks et al., 2010), setting the PHRED quality threshold at 20 and discarding reads <21 nucleotides in length. Transcripts were quantified using RSEM (Li and Dewey, 2011) (using default parameters) and the predicted coding sequences from version 5b of the maize genome (http://maizesequence.org/). For differential gene expression analysis, expected transcript counts originating from the same gene locus were summed, and all possible pairwise comparisons between biologically replicated samples were performed using DESeq (Anders and Huber, 2010). Expression values were normalized by DESeq using the default method, and, in all cases, differentially expressed genes were identified as those genes with a Benjamini–Hochberg corrected P value ≤ 0.05 (Benjamini and Hochberg, 1995).
Validation of RNA-seq data by quantitative RT-PCR
Total RNA (1 μg) was retained from samples FP1, HP1 and FI1 for independent validation of the Illumina sequencing. RNA was treated with RQ1 DNAase (Promega, http://www.promega.com), and cDNA was synthesized using Superscript® II reverse transcriptase (Invitrogen, http://www.invitrogen.com/) according to the manufacturer's instructions. Primer pairs (Table S30) were designed to amplify regions of single exons (predicted from the RNA-seq data to be present at different levels in mature transcripts from all three tissue types) of between 70 and 200 bp for 97 genes in the dataset. PCR amplification was performed using GoTaq® Hot Start polymerase (Promega), and amplification was detected using 1/60 000 SYBR® Green II (Sigma-Aldrich) and the Mx3000P QPCR system (Agilent, http://www.home.agilent.com/). Using fivefold dilution standard curves, 20 primer pairs were found to have appropriate amplification efficiencies (90–110%) and R-sq values >98.5 for accurate quantitative analysis. For each sample, two PCR replicates were performed for each of two independent cDNA synthesis reactions. After manual checking of dissociation curves and Ct values, the mean value across the replicates was calculated. Fold changes for each gene in each sample were then calculated using the Pfaffl method (Pfaffl, 2001), and correlated against fold changes of the same genes in the same samples as measured by RNA-seq (Figure S8).
Enrichment analyses were performed for gene ontology (GO) terms (http://www.geneontology.org/), MaizeCyc pathways (http://maizecyc.maizegdb.org/), MapMan terms (http://www.mapman.gabipd.org/) and Pfam domains (http://pfam.sanger.ac.uk/). P values were obtained by approximating Wallenius' non-central hypergeometric distribution (Wallenius, 1963). GOseq (Young et al., 2010) was used to compensate for over-detection of differential expression for long and highly expressed transcripts. The resulting P values were subject to multiple hypothesis test correction to correct for type I family-wise error using the Benjamini–Hochberg method (Benjamini and Hochberg, 1995). Significantly enriched annotation terms were identified as those that showed a corrected P value of ≤ 0.05.
Profile classification analysis
Profiles were generated such that descending profiles (D1, D2, D3) contained only genes whose relative mRNA abundance decreased significantly from P1/2 to P5, while ascending profiles (A1, A2, A3) included genes whose relative mRNA abundance increased significantly from P1/2 to P5, and neutral profiles (N) had non-significantly different mRNA abundance estimates in all three (P1/2, P3/4 and P5) primordia samples (P ≤0.05). Genes were assigned to expression profiles using custom Perl scripts. The input data for these scripts were the Benjamini–Hochberg-corrected P values obtained from the pairwise DESeq-based significance tests (see above). The 383 automatically annotated transcription factors in the descending and ascending profiles were subject to manual classification by Reciprocal Best BLAST. In cases where orthology was not clear, additional phylogenetic analyses were performed to identify the most closely related homologues, and published data were used to infer the most likely gene function.
Classification criteria for selection of candidate Kranz regulators
Filters for the identification of Kranz anatomy regulators were based on the following biological criteria. For positive regulators, it was assumed that genes would be expressed at significantly higher levels in foliar leaves than in husk leaves, and that patterning genes would be expressed prior to and/or during P5. It was further assumed that if genes were expressed in the expanding leaf, transcript levels would be significantly higher in the immature basal regions of the leaf than in more distal regions. For negative regulators, it was assumed that genes would be expressed at significantly higher levels in husk leaf primordia than in foliar leaf primordia, and that expression levels would be significantly higher in primordia than in expanded leaves with established anatomy. It was further assumed that if genes were expressed in the expanding leaf, transcript levels would not be significantly higher in the distal regions of the leaf than in the immature basal regions. The expanding leaf gradient dataset was generated by processing the raw reads from Li et al. (2010) and re-mapping them to the genome in the same manner as all other samples.
Phylogenetic tree inference
Iterative hidden Markov model searches were performed as previously described (Kelly et al., 2011) using the genomes of Arabidopsis, rice, Brachypodium distachyon, Sorghum bicolour and maize, and initiating each iterative query using a maize gene from the candidate list. For each search, the identified protein sequences were aligned using MergeAlign-91 (Collingridge and Kelly, 2012), and a maximum-likelihood phylogenetic tree was inferred using FastTree (Price et al., 2010) using the JTT model of sequence evolution (Jones et al. 1992) and CAT rate heterogeneity (Lartillot and Philippe, 2004). In all cases, SH-like support values (Shimodaira and Hasegawa, 1999) are shown at branch nodes. Pfam domains exceeding a threshold expectation value of 1 × 10−5 were identified, and used to construct domain cartoons beside each sequence shown in each phylogenetic tree.
We are grateful to Julie Bull for technical support and John Baker for photography. We thank Mara Schuler, Sayuri Ando, Olga Sedelnikova and Tom Hughes for contributions to the C4 project, and all members of the lab at the Department of Plant Sciences, University of Oxford, Oxford, for helpful discussions. Andy Plackett and Laura Moody provided constructive comments on the manuscript. This work was funded by a grant to J.A.L. from the Bill and Melinda Gates Foundation Grant through the International Rice Research Institute, Los Banos, Philippines, and by the Oxford Martin School. S.K. was funded as a Biotechnology & Biological Sciences Research Council Systems Biology Fellow, and J.F. was funded by a Biotechnology & Biological Sciences Research Council Doctoral Training Account. We thank all of our colleagues in the C4 rice consortium for sharing ideas and data ahead of publication. P.W. prepared plant material and extracted RNA, S.K. designed the bioinformatics methodology and wrote the code, and J.F. validated the quality of sequencing data and performed the profile analyses. J.A.L. proposed the original experimental design and obtained funding. All authors designed the data filtration criteria, analysed datasets and wrote the paper.