The bulk of plant biomass is comprised of secondary cell walls (SCWs), consisting of a cross-linked matrix of cellulose, hemicellulose and lignin biopolymers. The latter form the basic scaffold of fibers and vessels found in angiosperm xylem. In addition to providing mechanical support, SCWs facilitate critical biological processes, such as water and nutrient transport, anther dehiscence, silique shattering, plant organ movement and response to pathogens (Caño-Delgado et al., 2003; Mitsuda et al., 2005; Fratzl et al., 2008; Mitsuda and Ohme-Takagi, 2008). Candidate genes involved in the biosynthesis of SCWs have been studied in both woody and herbaceous model species (e.g., Brown et al., 2005; Mellerowicz and Sundberg, 2008). These structural genes are under strict transcriptional control during xylogenesis (Hertzberg et al., 2001; Schrader et al., 2004), highlighting the central role of transcription factors in this regard (Du and Groover, 2010). Understanding the regulation of SCW deposition is important because of (1) the widespread use of lignocellulosic biomass in pulp, paper and cellulose-derived products, (2) the potential of second-generation biofuel feedstocks such as short-rotation hardwoods (e.g., Populus, Eucalyptus) (Rockwood et al., 2008; Carroll and Somerville, 2009; Hinchee et al., 2010), and (3) the role of cell wall material in nutrition and health (Fincher, 2009; Doblin et al., 2010; McCann and Rose, 2010). However, the challenges to studying transcriptional regulation in non-model organisms impede the improvement of lignocellulosic biomass for fiber, raw cellulose and biofuels.

Considerable progress has been made in understanding how TFs regulate SCW structural genes. To this end, various model organisms (Arabidopsis, Oryza, Populus) (e.g., Kubo et al., 2005; Grant et al., 2010; Zhong et al., 2011a) as well as Zinnia and Arabidopsis (trans)differentiation systems (Fukuda and Komamine, 1980; Oda et al., 2005) have been instrumental. In the last decade, studies in Arabidopsis in particular have revealed the existence of an extensive transcriptional network regulating SCW deposition in vessels, fibers, anther endothecium and structures (replum, endocarp, valve margin) within the silique (reviewed in Yamaguchi and Demura, 2010; Zhong et al., 2010a). Whilst a considerable diversity of TF families participate in SCW transcriptional regulation, the most prominent families of TFs involved in this network appear to be the NAC (NAM/ATAF/CUC) and R2R3-type MYB (MYELOBLASTOSIS) family proteins, both characterized by conserved N-terminal DNA-binding domains and diverse C-termini that participate in transcriptional regulation (Ooka et al., 2003; Dubos et al., 2010).

General structures of SCW transcriptional networks have been illustrated in a number of reviews, based on knowledge of Arabidopsis (Umezawa, 2009; Zhong and Ye, 2009; Caño-Delgado et al., 2010; Yamaguchi and Demura, 2010; Zhang et al., 2010; Zhong et al., 2010a; Wang and Dixon, 2011; Zhao and Dixon, 2011; Pimrote et al., 2012; Schuetz et al., 2013), and monocots (Handakumbura and Hazen, 2012). A few primary research articles also depict schematic representations of the Arabidopsis SCW network, incorporating data from Populus and limited knowledge of Eucalyptus and Pinus SCW transcriptional networks (Zhong et al., 2008a, 2011b; McCarthy et al., 2009). However, aside from Umezawa (2009) who focused on the cinnamate/monolignol pathway, these representations have not fully captured individual protein–DNA and protein–protein interactions reported in the literature. In addition, the regulatory dynamics of SCW transcriptional regulation are poorly understood compared to network structure (i.e., connectivity). Furthermore, the methodologies used to generate evidence lines for SCW network reconstruction have not been extensively reviewed. Here we comprehensively integrate and illustrate the complexity of known protein–DNA and protein–protein interactions in the Arabidopsis SCW transcriptional network. We discuss the roles of putative regulatory modules in the network, highlighting known and hypothetical balancing mechanisms that may influence network behavior. Finally, we provide a critical review of the methodologies currently used to infer SCW transcriptional networks and recommend approaches for increasing reliability in inferring SCW transcriptional network structure.

The deposition of SCWs and the initiation of programmed cell death (Bollhöner et al., 2012) together represent the culmination of developmental signals that cue vascular tissue specification and cell fate determination (Figure 1). This specification begins with establishing a population of meristematic cells known as the procambium via the combinatorial effect of hormones such as auxin, cytokinins and brassinosteroids (BRs). The procambium in turn gives rise to the primary vascular tissues (xylem, phloem) in the shoot vascular bundles and root vasculature (Turner et al., 2007; Caño-Delgado et al., 2010). In root and shoot tips, a pre-procambial state is established via PIN1-mediated polar auxin transport along files of parenchyma cells, effectively channeling auxin to what will become the procambium (Dettmer et al., 2009). In leaf veins, a preprocambial state is associated with expression of ATHB8, which is directly activated by the auxin response factor MP/ARF5 (reviewed in Zhang et al., 2010). In addition to procambium specification, auxin promotes cell division in the procambium in combination with cytokinins (reviewed in Caño-Delgado et al., 2010). The vascular cambium, from which all secondary xylem and phloem tissues arise during secondary growth, develops from the procambium and interfascicular parenchyma (Plomion et al., 2001; Baucher et al., 2007). As per the convention of Dettmer et al. (2009), we generally refer to procambiums and (secondary) vascular cambiums as vascular meristems, which are thought to be regulated in a similar, but not identical, fashion to shoot and root apical meristems (Sanchez et al., 2012; Milhinhos and Miguel, 2013) (Figure 1).

The establishment of xylem and phloem cell fate is influenced by hormones, TFs, miRNAs, mobile peptides and proteoglycans acting on nascent mother cells produced in the vascular meristems (Figure 1) (see Carlsbecker and Helariutta, 2005; Du and Groover, 2010; Zhang et al., 2010; Schuetz et al., 2013 for review). Auxin concentrations lower than those encountered at the vascular meristem promote xylem differentiation in the presence of cytokinin (Sorce et al., 2013). In the root, xylem differentiation is in contrast thought to be promoted by high auxin concentrations, brought about by cytokinin-mediated activation of a phosphorylation cascade in the procambium that results in polar auxin transport toward the protoxylem (reviewed in Aichinger et al., 2012). Five members of class III homeodomain leucine zipper (HD-ZIP III) TFs, including ATHB8, IFL1/REV, PHB, and PHV, are induced by auxin and generally promote xylem differentiation (Zhong et al., 1997; Baima et al., 2001; Ohashi-Ito and Fukuda, 2003; Ilegems et al., 2010; Schuetz et al., 2013). However, some HD-ZIP III genes, such as ATHB8 and ATHB15, appear to be antagonistic to REV in meristem formation, embryo patterning and interfascicular fiber development (Prigge et al., 2005). For example, ATHB15 seems to negatively affect xylem development, while miR166-mediated cleavage of ATHB15 transcript (see below) promotes xylem differentiation (Kim et al., 2005). Xylogen, a secreted proteoglycan, has also been implicated in xylem specification (Motose et al., 2004), while gibberellic acid (GA) promotes fiber elongation and general xylogenesis (Eriksson et al., 2000; Israelsson et al., 2003; Mauriat and Moritz, 2009). Brassinosteroids (BRs) have been associated with xylem differentiation in Arabidopsis, and in trans differentiating Zinnia cell cultures BRs are required for the expression of a homolog of ATHB8 (reviewed in Jung and Park, 2007). Ethylene is essential for in vitro tracheary element (TE) differentiation in cultured Zinnia cells (Pesquet and Tuominen, 2011). In planta, ethylene is thought to diffuse from its site of synthesis in maturing TEs through to the cambium (Pesquet and Tuominen, 2011), where it promotes cell division (Love et al., 2009).

On the opposite side of the cambium, phloem differentiation occurs under the influence of APL, a MYB-related TF (Bonke et al., 2003; Ilegems et al., 2010), whilst KAN1/KAN2/KAN3/KAN4 TFs indirectly promote phloem differentiation by repressing (pro)cambium maintenance and restricting class III HD-ZIP TF expression through repression of polar auxin transport (Emery et al., 2003; Izhaki and Bowman, 2007; Schuetz et al., 2013). Phloem-expressed miR165/166, which are upregulated by SHR and SCR in roots, post-transcriptionally inhibit HD-ZIP III genes (Tang et al., 2003; Mallory et al., 2004; McHale and Koning, 2004; Zhong and Ye, 2004, 2007; Williams et al., 2005; Carlsbecker et al., 2010). Ectopic xylem formation is inhibited by a dodecapeptide ligand TDIF/CLE41/CLE44, which is produced in the phloem and diffuses to the xylem side of the vascular meristem (Ito et al., 2006). The peptide also co-ordinates the orientation of cell divisions in the cambium via the perception of a peptide concentration gradient by the LRR receptor-like kinase PXY in procambial cell membranes and induction of WOX4 (Etchells and Turner, 2010; Hirakawa et al., 2010). Xylem differentiation may be further suppressed in the phloem in part by XIP1, which is related to PXY (Bryan et al., 2011) (Figure 1).

Once xylem mother cell fate has been established and cell elongation has ceased in immature xylem, SCW deposition occurs. This is activated by the TFs VND6 and VND7 in the case of xylem vessels, and SND1 and NST1 in fibers. These “master regulators” initiate a SCW transcriptional network, successively activating at least two tiers of intermediate TFs which, in addition to the master regulators, activate the structural genes for SCW biosynthesis (Figure 1, gray blocks). In the remainder of this review we focus on the SCW transcriptional network and the tools available to study its structure and function.

A simplified representation of the SCW regulatory network is shown in Figure 1, which depicts the putative positions of associated TFs and their direct or indirect targets. We have also reconstructed a SCW-regulating protein–DNA and protein–protein interaction network from the Arabidopsis literature using BioTapestry (Longabaugh et al., 2005), showing cell type contexts where known (Figure 2). Aside from the indicated exceptions, we represent only direct protein–DNA interactions, as elucidated using yeast one-hybrid, electrophoretic mobility shift assay, chromatin immunoprecipitation, or post-translationally induced protoplast transactivation (see section Methodologies for the Study of SCW Transcriptional Regulation). Such interactions are referred to as direct regulation in this review. Finally, we provide as a supplementary file (Additional file 1) a more detailed network capturing the vast majority of demonstrated direct and indirect protein–DNA interactions and all known protein–protein interactions. This resource can be interactively visualized and manipulated with the freeware program Cytoscape (Shannon et al., 2003), and is accompanied by a list of the literature supporting each of the 435 captured interactions (Data sheet 1). To the best of our knowledge, this is the most exhaustive network representation compiled to date. The Cytoscape representation has several uses. First, it assists the generation of hypotheses related to biological function of poorly characterized proteins based on their interactions with known proteins. Second, additional attributes such as expression data may be integrated into the network to better understand network function and behavior. This is further enhanced by the fact that the network layout can easily be converted into built-in or customized views, and new interactions added as they are reported in the literature. In future, researchers may be able to use the network to provide priori structural information for the building of probabilistic causal networks that integrate diverse types of data, as performed in yeast by Zhu et al. (2012). Third, the network serves as a reliable basis for template-based construction of SCW transcriptional networks in sequenced non-model organisms (Babu et al., 2009).

At least three main tiers of TFs can be identified in the network that ultimately regulate a suite of structural genes involved in cellulose, hemicellulose and lignin biosynthesis, signal transduction, the cytoskeleton, programmed cell death and proteins with unknown functions (Figures 1, 2). We designated TF tiers from the bottom upwards, relative to a reliable reference point, i.e., the structural genes. A similar convention has been adopted before (Jothi et al., 2009). First-tier TFs are only known to directly regulate structural genes, second-tier TFs directly regulate first-tier TFs in additional to structural genes, and so forth. We stress that this assignment is not rigid and that TFs may be re-assigned, where possible, to a different tier as additional data arises. Furthermore, extensive feedback may occur between tiers.

SCW transcriptional networks in different cell types that synthesize SCWs are initiated by distinct, functionally redundant pairs of NAC proteins, which have been broadly referred to as secondary wall NACs (SWNs) (Zhong et al., 2010c) (Figure 2; third tier). Specifically, SCW deposition in xylary and interfascicular fibers (Mitsuda et al., 2007; Zhong et al., 2007b, 2008a) as well as silique valve endocarps and valve margins (Mitsuda and Ohme-Takagi, 2008) is redundantly regulated by NAC SECONDARY WALL THICKENING PROMOTING FACTOR1 (NST1) and SECONDARY WALL ASSOCIATED NAC DOMAIN PROTEIN1 (SND1). SND1 has also been referred to as NST3 and ANAC012 (Ko et al., 2007; Mitsuda et al., 2007; Mitsuda and Ohme-Takagi, 2008); to avoid confusion, we refer to this protein as SND1. In meta- and protoxylem vessels, a SCW deposition is regulated by VASCULAR RELATED NAC DOMAIN6 (VND6) and VND7, respectively (Kubo et al., 2005; Yamaguchi et al., 2008, 2010a; Zhong et al., 2008a). NST1 and NST2 are SCW master regulators in the endothecium of anthers (Mitsuda et al., 2005). To date, comparatively little data are available for the regulatory functions of NST2. MYB26 activates NST1 and NST2 in the endothecium through an as yet unknown mechanism (Yang et al., 2007), suggesting the existence of a fourth tier (Figure 2).

While the SCW master regulators in fibers, vessels, siliques and anther endothecia differ from one another, current data suggest that they regulate a common core transcriptional network (Figure 2). VND6/VND7 and NST2 regulatory functions largely overlap with those of SND1/NST1, but a number of targets are unique to VND6, VND7, or SND1 (Figure 2). Notably, vessel differentiation is distinguished from fiber development by strong VND6/VND7-mediated activation of genes involved in programmed cell death (PCD); in contrast, PCD gene activation by SND1/NST1 is weak (Ohashi-Ito et al., 2010; Zhong et al., 2010c) (Figure 2). A second notable difference is the fact that VND6/VND7 participate in a positive feedback loop with ASYMMETRIC LEAVES2/LATERAL ORGAN BOUNDARIES DOMAIN TFs ASL19 and ASL20 (Soyano et al., 2008), and VND7 additionally interacts with the transcriptional repressor protein VNI2 (Yamaguchi et al., 2010b) (see section Network Dynamics) which has not been identified in other cell types. VND7 also interacts with VND1, VND2, and VND3 (Additional file 1) which do not have clearly defined functions, whereas VND6 interacts primarily with itself and probably binds as a homodimer in vivo (Yamaguchi et al., 2008).

Third-tier SWNs directly regulate common second-tier MYB domain TFs MYB46, MYB83, and MYB103, NAC domain TFs XND1 and SND3, and ASYMMETRIC LEAVES2/LATERAL ORGAN BOUNDARIES DOMAIN TF ASL11 (Zhong et al., 2010c; Yamaguchi et al., 2011) (Figure 2). MYB46 and MYB83, which are functionally redundant, appear to form a common regulatory hub in the second-tier that directly regulate first-tier TFs MYB6, MYB43, MYB52, MYB54, MYB58, and MYB63, the functionally redundant trio MYB4/MYB7/MYB32 (see section Network Dynamics), a C3H-type zinc finger gene C3H14 and homeobox TF KNAT7 (Ko et al., 2009; McCarthy et al., 2009; Nakano et al., 2010; Zhong and Ye, 2012) (Figure 2). KNAT7 is directly activated by all the SWNs (Zhong et al., 2008a). In turn, KNAT7 represses cellulose, hemicelluloses and lignin biosynthetic genes directly or indirectly (Li et al., 2012a) (Figure 1). KNAT7-mediated repression is dependent on protein–protein interactions with MYB75, a weak transcriptional activator which has no known targets or direct regulators (Bhargava et al., 2010, 2013) (Figure 2). MYB20, associated with the regulation of lignin biosynthetic genes, is likely a first-tier candidate since it is an indirect SND1 target but downregulated in the myb103 mutant (Zhong et al., 2008a; Öhman et al., 2012) (Figure 1). A number of novel bZIP, homeodomain, BEL1-like and zinc finger TFs that have not been linked to SCW regulation were also listed as MYB46/MYB83 direct targets (Zhong and Ye, 2012) (Additional file 1, Data sheet 1). Dominant repression of MYB52 and MYB54 result in reduced fiber SCW deposition (Zhong et al., 2008a). Enhanced drought tolerance in MYB52 overexpression lines (Park et al., 2011) suggests a pleiotropic role for this gene in both fiber development and abiotic stress response.

The first-tier TFs regulate various SCW biosynthetic genes although some members of the second-tier (MYB46, MYB61, and MYB83) and third tier (SND1, VND6, and VND7) also directly activate structural genes. BP, ATHB18, a C2H2-type zinc finger protein At3g46080, MYB20, MYB69, MYB79, MYB85, and the functionally redundant pair MYB58/MYB63 are known only to directly or indirectly regulate lignin biosynthetic genes (Mele et al., 2003; Zhou et al., 2009; Mitsuda et al., 2010), whereas BES1 is the only TF currently shown to bind to cellulose synthase (CesA) genes in both primary and SCWs (Xie et al., 2011) (Figures 1, 2). BP is a KNOX gene family member that maintains shoot apical meristems (Sanchez et al., 2012) and strongly represses lignification in inflorescence stems (Mele et al., 2003). MYB85 appears to specifically regulate the lignin pathway (Zhong et al., 2008a) and appears to be regulated by MYB46/MYB83 (Figure 1, Additional file 1). All other TFs regulate structural genes involved in the biosynthesis of more than one SCW biopolymer. SND2 has an unclear position in the network: it is known to be indirectly activated by SND1 (Zhong et al., 2008a), it is downregulated in the myb103 mutant (Öhman et al., 2012), and appears to regulate genes related to signaling, hemicellulose and lignin polymerization in addition to the secondary wall CesA genes (Hussey et al., 2011; Öhman et al., 2012) (Figure 1, Additional file 1). Therefore, we have tentatively placed it in tier 1.

A number of techniques have been employed to study SCW-regulating TFs. We provide a summary of the advantages and challenges of common approaches used in the literature for TF functional annotation and SCW transcriptional network inference (Table 2). We have roughly arranged these techniques in increasing resolution of the regulatory information obtained in each; that is, increasing understanding of the in vivo direct gene targets of a given TF and its bound cis-element. Here, we discuss in greater detail the widespread use of reverse genetics and protoplast transfection approaches in model organisms in SCW regulation studies, and review approaches better suited to non-model organisms.

Classical reverse genetics approaches employing overexpression and knock-out mutagenesis have been central to the functional annotation of SCW TFs in Arabidopsis and Populus (e.g., Zhong et al., 2007b; McCarthy et al., 2009; Grant et al., 2010). Direct or indirect targets of a TF subjected to knock-out or overexpression may be inferred under the premise that the transcriptional regulation of those targets is altered, leading to their differential expression relative to the wild type. However, we would like to highlight some problems associated with overexpression that have emerged in studies of SCW regulation, namely the level and site of overexpression.

SND1, now accepted as a master transcriptional activator of Arabidopsis SCW biosynthesis in fibers (Mitsuda et al., 2007; Zhong et al., 2007b, 2008a), was reported to suppress fiber SCW deposition when excessively overexpressed (Zhong et al., 2006). SND2, an indirect target of SND1, exhibited increased fiber SCW thickness when overexpressed and a mirrored reduction in SCW deposition in dominant repression lines (Zhong et al., 2008a). However, when our laboratory analyzed independent Arabidopsis lines overexpressing SND2 (Hussey et al., 2011), we observed a decrease in fiber SCW deposition which we attributed to SND2 transcript levels far-exceeding those reported in the previous study. Such phenomena could be explained by transcriptional squelching, defined as the repressive effect of a transcriptional activator beyond a certain threshold of abundance, due to the sequestration of interacting co-regulators or general transcription factors (Cahill et al., 1994; Orphanides et al., 2006). Alternatively a “dosage balance” mechanism (Birchler et al., 2005) holds that, for multi-subunit TF complexes, a relative increase in the abundance of one particular subunit does not lead to an increase in the yield of the assembled complex, but rather a stoichiometric reduction in the abundance of complete complexes and an increase in the abundance of non-functional sub-complexes (Birchler and Veitia, 2007, 2010). Together, these inconsistencies in functional studies of cell wall-related TFs suggest that overexpression differences may introduce indirect or even conflicting phenotypes.

Ectopic expression can also modify TF function. Regulating primary cell wall (PCW) deposition in the Arabidopsis root cap, three partially redundant TFs closely related to clade IIb NACs NST1, SND1, VND6, and VND7 have been described, namely SOMBRERO (SMB), BEARSKIN1 (BRN1), and BRN2 (Bennett et al., 2010). When constitutively driven by the 35S CaMV promoter, they are sufficient for ectopic deposition of lignified SCWs in several tissues, a phenotype resembling that of NST1, VND6, and VND7 overexpression (Bennett et al., 2010). Since SCWs are not found in the root cap where the TFs normally function, ectopic expression resulted in a modification of the gene targets that SMB and BRN1/2 naturally regulate, perhaps due to differences in co-regulators or other regulatory factors between tissues. This mechanism may also explain results reported by Bomal et al. (2008), where ectopic overexpression of the xylem-associated pine gene PtMYB8 in spruce caused misregulation of flavonoid-associated transcripts, which have preferential expression in tissues that correspond to regions of low expression of native PtMYB8 in pine.

The examples cited above of confounding effects due to the level and site of a candidate TF's expression provide compelling grounds to substantiate with additional evidence some of the conclusions arising from overexpression approaches. Such concerns have been echoed in a review of gain-of-function mutagenesis (Kondou et al., 2010). To avoid these problems, loss-of-function mutagenesis and non-transgenic approaches such as ChIP-seq may be more reliable. Although conventional mutagenesis is frequently unsuitable for SCW-regulating TFs due to the high degree of functional redundancy between homologs, the use of chimeric repressor silencing technology (CRES-T; Hiratsu et al., 2003) has circumvented this problem, at least for transcriptional activators. In CRES-T, dominant loss-of-function transgenic plants overexpress a candidate TF fused to a hexapeptide dominant repression domain (Hiratsu et al., 2004; Mitsuda and Ohme-Takagi, 2008; Zhong et al., 2008a). The hexapeptide repressor opposes the transcriptional activation function at loci bound by the TF, in addition to the complementation that functional homologs may exert at those loci. Ectopic noise arising from overexpression may also be reduced by the use of tissue-specific promoters or laser capture-microdissection to capture only those cell types where the TF candidate is naturally expressed. Similarly, inducible expression may limit knock-on effects of long-term overexpression.

Promoter transactivation by an induced candidate TF in Arabidopsis mesophyll protoplasts (Wehner et al., 2011) has proved particularly useful in the identification of Arabidopsis gene targets, and may be used to complement approaches such as ChIP-seq which does not strictly indicate active target regulation (Table 2). The assay typically involves co-transfection with different plasmids, one harbouring a constitutively expressed candidate TF gene (the “effector”), a promoter::GUS reporter vector and a luciferase expression vector to allow for normalization of transfection efficiency. Inferring direct targets using this system is complicated by the potential ability of a candidate TF to induce transcription of an intermediate TF in the host cell that is responsible for activating a target gene. This has been addressed by translational fusion of the candidate TF to the regulatory region of the human estrogen receptor (Zuo et al., 2000). The chimeric protein is post-translationally induced by β-estradiol, allowing for an inhibitor of protein synthesis to be added to the system prior to induction to block the translation of intermediate TFs. The activated chimeric TF is then able to regulate transcription of target genes using the existing transcriptional machinery of the cell, and direct targets can be inferred with promoter::GUS RT-qPCR analysis (Zhong et al., 2008a; Zhou et al., 2009) or microarray analysis of the host cell transcriptome (Zhong et al., 2010c). For this reason, only those protoplast transactivation experiments that used post-translational induction were considered as evidence for protein–DNA interactions represented in Figure 2, to avoid the possibility that putative targets may have been indirectly regulated.

Arabidopsis mesophyll protoplasts have also been used to assess in vivo promoter transactivation by Populus and Eucalyptus TFs using GUS reporters fused to candidate promoters from these species (Zhong et al., 2011b). A genome-wide analysis of endogenous promoter transactivation of TFs from non-model species would require protoplasts derived from the same, or a related, species. Although the promoter::GUS approach suffers from much lower throughput, it mitigates the effects that the chromatin structure of the host protoplast may exert on the regulation of endogenous target genes. DNaseI hypersensitivity sites, an indicator of open chromatin marking most active TF binding sites, vary considerably across cell types (Thurman et al., 2012). Therefore, protoplasts should ideally be sourced from the same tissue in which a candidate TF is expressed. The recent report of isolation and transfection of Populus secondary xylem protoplasts (Li et al., 2012b) sets the stage for genome-wide analysis of poplar genes transactivated by poplar TFs involved in wood development.

Several approaches exist for in vitro, in vivo, and in planta analysis of TFs from non-model organisms that are not yet easily transformable (Table 2). Systems genetics allows for gene regulatory networks to be reconstructed by co-expression analysis across large numbers of segregating progeny (Ayroles et al., 2009). Microarray or RNA-seq expression data are obtained for tissues of interest from a structured segregating population. The addition of genetic markers allows for the identification of expression quantitative trait loci (eQTLs) (Jansen and Nap, 2001), which can be differentiated into those acting in cis or trans. Trans-eQTLs likely represent polymorphisms in transcriptional regulators, and due to their ability to affect expression of many genes, trans-eQTL “hotspots” may be mapped that contain significantly more eQTLs than the genome average. The combination of eQTLs and co-variation in transcript levels allows the prediction of causal relationships (Zhu et al., 2007) and candidate regulators (Drost et al., 2010) and is the basis on which regulatory networks can be constructed a posteriori, or hypothetical a priori networks tested (Kliebenstein, 2009). One considerable limitation of current systems genetics studies is the difficulty of studying the segregation of transcript abundance in specific cell types of organs. However, recent advances in obtaining high-throughput cell type-specific transcriptome data may make this challenge more feasible (Chitwood and Sinha, 2013).

Yeast one-hybrid (Y1H) has been widely used to identify TFs that interact directly with SCW-related promoters (Lin et al., 2010; Kim et al., 2012b). These assays can be performed either through direct cloning of candidate TF coding sequences and systematically testing interactions with different potential target promoters (Mitsuda et al., 2010), or via screening of cDNA expression libraries which have the advantage of discovering novel interacting proteins (Lopato et al., 2006). Recent advancements in Y1H screening, including smart pooling and robotics, have increased the generally low throughput of this technique (reviewed in Reece-Hoyes and Walhout, 2012). However, Y1H interactions occasionally fail independent validation assays. For example, although SPL8 was isolated from 20 of 72 yeast colonies showing a positive interaction with the CCoAOMT1 promoter, it failed to activate the promoter in particle-bombarded Arabidopsis leaves (Mitsuda et al., 2010). A significant disadvantage of Y1H is that protein–DNA interactions which require cofactors or bind as complexes will not be identified (Table 2).

A poorly researched area in SCW transcriptional regulation is the symplastic movement of regulatory proteins between cells. It is well known that some TFs may move from cell to cell through plasmodesmata (Burch-Smith et al., 2011; Wu and Gallagher, 2012). Aside from the SHR example in section Spatial Specificity of Fiber and Vessel Development (reviewed by Kurata et al., 2005), in plants this has been found predominantly in meristems and involves mainly the KNOX (e.g., KNOTTED1) and MADS-box TF families (Zambryski and Crawford, 2000). However, at least one MYB-like protein is known to be non-cell-autonomous (Wada et al., 2002), and it is possible that some SCW-associated TFs act non-cell autonomously. This necessarily implies that the use of in situ hybridization, which has been widely used to study SCW-associated TF transcript abundance (e.g., Zhong et al., 2010b), may not accurately reflect a candidate TF's biological function. For species that can be stably transformed, TF movement can be tracked by fluorescent protein fusion experiments. For example, Kim et al. (2003) expressed GFP~KN1 fusion proteins (where ~ denotes a linker sequence) in mesophyll or epidermal cells using tissue-specific promoters, and compared the movement of GFP~KN1 between the mesophyll and the epidermis with that of free GFP and GFP fused to a viral movement protein. Microinjection of fluorescently labeled recombinant TFs into the cytoplasm of cells of interest can also be performed, but this approach is technically cumbersome and limited to larger cells (Lucas et al., 1995; Wang et al., 2007b). For non-model organisms, immunolocalization methods using an antibody against a TF of interest can be used to detect its presence in planta. Whilst low-abundance TFs may be difficult to detect using immunohistochemical methods, both alkaline phosphatase staining and immunogold labeling have been used to detect TF proteins at cellular and subcellular levels (Rodriguez-Uribe and O'Connell, 2006).

Chromatin immunoprecipitation combined with high-throughput sequencing (ChIP-seq) offers many advantages that are particularly suited to non-model organisms where genomic information is available (Table 2). It has been shown that even fragmented genome assemblies are acceptable for ChIP-seq read mapping (Buisine and Sachs, 2009), evading the need for genome assemblies on par with model plants. However, ChIP-seq in plants currently suffers from a lack of protocols for isolation of sufficient amounts of chromatin from individual tissues, and each tissue and species may require customized modifications to chromatin fixation, nuclei isolation and chromatin shearing (Haring et al., 2007). To our knowledge no ChIP procedures have yet been applied to developing xylem from woody stems, but a report of successful mapping of the ARBORKNOX1 TF in poplar vascular cambium using ChIP-seq (Andrew Groover, personal communication) sets the stage for its implementation in xylem. A range of improvements have been made to the basic ChIP-seq principle (reviewed in Furey, 2012), amongst them the ability to amplify sufficient amounts of ChIP DNA for Illumina sequencing from limited cell numbers (Adli and Bernstein, 2011). The latter is a particularly exciting advancement as it may allow for ChIP to be applied for the first time to plant tissues where chromatin yield is poor or where a TF's expression is low.

While these and other technologies advance—especially those involving second-generation sequencing—systems approaches to study SCW transcriptional regulation are still lacking. Systems biology attempts to integrate various high-throughput datasets into a holistic biological model, or achieve meaningful dynamic modeling of extensive biological data. Despite considerable progress in plant systems biology (Yuan et al., 2008) and the existence of several genome- and transcriptome-wide datasets relating to Arabidopsis SCW biosynthesis, few attempts have yet been made to integrate such data with other -omics platforms. The integration of metabolomic data into transcriptional networks, for example, can link modifications of TFs and their interactions to phenotypic outputs. Two model Arabidopsis studies, one using multiple knockout mutants of lignin biosynthetic pathway enzymes (Vanholme et al., 2012) and another analyzing five TF overexpression lines involved in glucosinolate biosynthesis (Malitsky et al., 2008), have integrated transcriptomic and metabolomic data to reveal novel aspects of metabolic pathway flux and regulation. In future, however, such analyses will have to be extended to cell-specific gene expression and interactions, especially in the field of transcriptional regulation. Overlaying cell type-specific expression profiles with Y1H and Y2H interaction data has been successfully achieved in the Arabidopsis root using enzymatic cell wall maceration and fluorescence-activated cell sorting of target cell protoplasts expressing a GFP marker (Brady et al., 2011). Another approach developed in Arabidopsis involves the purification of tagged nuclei from specific cells for transcriptome and ChIP-seq analysis (Deal and Henikoff, 2010). These and other innovations will undoubtedly contribute to a systems-level understanding of SCW regulation in the near future.

In this review we aimed to provide a comprehensive summary of what is currently known about Arabidopsis SCW transcriptional regulation, highlighting current gaps in our understanding of the transcriptional network. We have also emphasized that an understanding of protein–protein interactions, spatial specificity and network dynamics (modules and hubs, regulatory motifs, and temporal regulation) is severely underdeveloped compared to what is known about the network's connectivity. The immediate goal of future research is to comprehensively identify the physical interactions (protein–DNA and protein–protein) involved in SCW transcriptional regulation. This includes the identification of not only interacting partners of known TFs, but also their cell-type context that might influence the functions of TFs in different ways. This goal will allow us to identify TFs and transcriptional modules that regulate genes involved in the biosynthesis of specific SCW biopolymers. This, together with systems approaches, will also reveal to what degree regulation of different genes and metabolic pathways is independent. Currently, only the lignin pathway seems to be specifically targeted by TFs such as MYB58, MYB63, and MYB85, and it may not be possible to uncouple the transcriptional regulation of cellulose and hemicellulose biopolymers. However, two recent studies have used components of the SCW transcriptional network to engineer plants with favorable biofuel properties by restoring vessel wall integrity in xylan (Petersen et al., 2012) and lignin mutants (Yang et al., 2013) or reinforcing polysaccharide deposition in fiber SCWs (Yang et al., 2013).

The ability to predict the regulatory outcome of perturbations in transcriptional networks through network modeling is invaluable to the field of biotechnology. A detailed knowledge of the strength of interaction for each edge connecting two nodes and a mathematical understanding of how the network responds to perturbations in expression, as well as genetic and environmental modulation, has not yet been attained. Systems biology experiments in Arabidopsis, for which knock-out lines are readily available to quantify network dynamics in response to genetic perturbations, will contribute extensively in this regard. For non-model organisms, Y1H and ChIP-seq are expected to be two key techniques used to identify protein–DNA interactions in the near future. However, systems genetics, which facilitates network reconstruction, modeling and quantification from perturbations caused by natural genetic variation, is gaining momentum in agronomically important species (Ingvarsson and Street, 2010; Mizrachi et al., 2012). Identification of trait QTLs and eQTLs additionally allow for the assessment of phenotypic impact of expression variation in TFs, the strength of association of TFs with regulons of co-expressed genes, and the ability to apply molecular breeding strategies to populations.

An understanding of the integration of intercellular signals, miRNAs, chromatin changes and temporal dynamics in transcription during xylem development remains a future challenge, marred by a limited understanding of regulatory mechanisms. For example, the occurrence of alternative splicing as a form of SND1 regulation in Populus (Li et al., 2012b) underscores an overlooked regulatory mechanism in SCW deposition, and there exists the possibility that certain RNA-binding proteins may participate in alternative splicing during xylogenesis. There is currently no data on cell type-specific chromatin modifications, DNA methylation or chromatin states during various aspects of fiber and vessel development that may influence availability of TF binding sites. We have no knowledge of the degree to which the SCW-associated TFs downstream of the HD-ZIP III TFs (Figure 1) are post-transcriptionally regulated by miRNAs, or of the transcriptional changes associated with the transitions between S1, S2, and S3 layer deposition in SCWs. Finally, the findings that fibers in close proximity to vessels show a vessel-like lignin composition (Gorzsás et al., 2011) and that lignification of tracheary elements may occur post-mortem due to monolignol transport from live cells (Pesquet et al., 2013) highlights the need to better understand the role of cell non-autonomous regulation of xylogenesis. Clearly there are plenty of opportunities for further study in this exciting field.