This is mainly the wall of reaction wood fibres that exhibits specific biochemical, ultra-structural, and mechanical features. In the primary wall of xylem cells, cellulose microfibrils are randomly oriented, whereas, in the secondary wall, they are highly oriented, being fixed by the hemicellulose gels to form a rigid, twisted honeycomb structure [4]. Lignification occurs firstly at the cell corner and middle lamella, then extends to the secondary wall. Therefore, the cell wall can be assimilated to a composite made of a matrix of lignins and hemicelluloses, reinforced by long bundles of cellulose microfibrils [5]. Tension wood anatomy is derived from this general pattern and varies depending on hardwood species. Using image analysis technology, Jourez et al. [6] gathered a large number of data on the anatomy of tension wood and opposite wood on poplar cuttings induced to form tension wood upon a gravitational stimulus. Both fibres and vessels were significantly longer in tension wood, which may result from a prolonged differentiation of tension wood cells in comparison to normal xylem cells. These authors also observed a significant reduction in the number and size of vessels and a corresponding increase in the proportion of fibres that exhibit very thick cell walls. Indeed, for a lot of hardwood species, the major anatomical characteristic of tension wood is the occurrence of xylem fibres with a thick extra layer at the inner side of the secondary wall, which is not very cohesive with the other cell wall layers (Fig. 1) [2]. Because of its translucent and gelatinous appearance, this extra-layer has been named G-layer. G-fibres (tension wood fibres containing a G-layer) have on average a smaller radial diameter than normal fibres in tension wood [6]. It must be pointed out that this G-layer is absent from the tension wood of a number of hardwood species: for example, among 346 Japanese hardwood species surveyed, G-layers was not found in 137 (about 40%) of them [7]. However, in species without G-layer, the S2 layer tends to develop features related to those of the G-layer that are described thereafter [8].

Tension wood is generally described as containing a high proportion of cellulose and being less lignified than normal wood [9–11]. In fact, whereas the vessel cell wall appeared generally well lignified, this is mostly the fibre cell wall that seems partly unlignified.

From analyses performed on ten hardwood species, Timell [12] determined an average chemical composition of tension wood and concluded that the lower overall lignin content generally measured in tension wood results from the incorporation of more cellulose in the G-fibres. Indeed, when expressed on the basis of equal cellulose content, tension wood, in comparison with normal wood contained less glucomannan, two to five times as much galactan [13,14], an equal or a slightly lesser amount of xylan, and the same quantity of lignin. It appears that the parts of the cell wall in tension wood that are deposited prior to the development of the G-layer, usually the S1 and S2 layers contain at least as much lignin and xylan as in normal wood. Moreover, as S1 and S2 layers are generally thinner in tension wood fibres than in normal fibres, Timell [12] indicated that the concentration of lignin must be considerably higher in these layers than in the corresponding layers of a normal fibre. Timell concluded that a lack of lignification should not be regarded as characteristic of tension wood as a whole but only of the G-layer.

Indeed, the G-layer is described as almost entirely composed of cellulose, with highly crystalline microfibrils oriented nearly parallel to the longitudinal axis [2]. Histological staining reactions, such as safranin-light green or phloroglucinol–HCl, seem to indicate that the G-layer is usually non-lignified or lignified to a very limited degree, whereas the wall layers external to the G-layer are strongly lignified.

Norberg and Meier [15] succeeded in the isolation and characterisation of pure G-layers by means of an ultrasonic treatment. The qualitative sugar analysis of a hydrolysate of G-layers strongly indicates a purely cellulosic nature of the G-layer, but these authors did not perform any phenolic analysis. Using UV microscopy and staining techniques, Scurfield and Wardrop [16] observed in several species that the G-layer could sometimes be partly lignified, especially in transitional regions between normal and tension wood. This has been also reported by Araki et al. [17] in Robinia pseudoacacia and Populus euramericana using transmission electron microscopy and KMnO₄ staining. Using UV microspectrophotometry, Yoshida et al. [18] detected lignin in small amount in the G-layers of tension wood fibres of black locust. Lignin has also been immunocytochemically detected in the G-layer of poplar tension wood [19]. Therefore, in tension wood, lignification seems to be reduced only in the G-layer.

Early studies performed on different hardwood species have consistently identified a correlation between Klason lignin content and lignin composition: high guaiacyl lignin correlate with high lignin contents and conversely low lignin contents correspond to high syringyl lignin. This relation seems to apply in tension wood: in 1971, Sarkanen and Hergert [20] already reported that tension wood in birch and madronna exhibits both a lower lignin content and more S units than normal wood. Likewise, Bland and Scurfield [21] using autoradiography demonstrated a lesser extent of methylation in opposite wood compared to tension wood.

Although a few reports (e.g., [22]) are contradictory, most of the studies agree on an increased S/G ratio in tension wood. Yoshida et al. [23] determined in situ lignin content using UV microspectrophotometry in yellow poplar, a tree species that does not form G-fibres, but whose secondary wall becomes gelatinous layer-like in the area with high tensile strength. They correlated an increased tensile strength, indicative of tension wood area, with a decreased microfibril angle (or MFA, that is the angle of cellulose microfibrils with the fibre axis) and both a decreased lignin content and an increased S/G ratio. Similar results were obtained in black locust [18], poplar [11], and eucalypts [24,25].

Recent reports indicate that the low amount of lignin present in the G-layer is also enriched in S units. Using antibodies specifically directed against non-condensed lignin, Joseleau et al. observed by immunogold electron microscopy the presence of guaiacyl-syringyl lignin epitopes within the G-layer of tension wood fibres [26]. Moreover, using another antibody specific for S units (Fig. 2), they estimated that the abundance of S units into the G-layer was about 40% of that in S2. Using microspectrophotometry coupled with the Maüle reaction (Yoshinaga, unpublished results) was also able to detect such a small amount of lignin rich in S units in the G-layer of poplar tension wood fibres.

It has been proposed that the increased S/G ratio generally observed in tension wood results from an enzymatic regulation. Scurfield [27] already reported an increase in peroxidase activity in the G-layer of living reaction wood cells. Aoyama et al. [25] observed an increased activity of wall-bound peroxidases that catalyse preferentially syringyl polymerisation during tension wood formation in Eucalyptus viminalis (a species that differentiates G-layers). A similar preference of cell-wall-bound peroxidases for S unit has also been observed in poplar tension wood [28].

Finally, Akiyama et al. [29] reported a slight but significant increased proportion of the erythro form in β-O-4 structures (the predominant linkage types in lignin) in the tension wood of yellow poplar, which strongly correlated with an increased content in methoxyl group of the lignin.

All these data suggest that lignification in G-layer is specifically regulated in favour of syringyl units. It remains to determine what could be the effects on the cell wall mechanical properties of the syringyl enrichment in G-layer lignin and if this effect results from a higher proportion of the erythreo form of β-O-4 structures in tension wood lignin.

Although the lignin polymer is often considered as a random network, its deposition seems to be spatially and temporally regulated [5]. Among other factors, it has been proposed that both the chemical nature of the carbohydrate matrix and the orientation of cellulose microfibrils may influence lignin deposition [30].

During the assembly of the secondary cell wall, the polysaccharide matrix is always deposited prior to lignification: this suggests a coordinated regulation of lignin and cellulose deposition. Siegel [31] already pointed out the importance of the nature of the polysaccharide matrix in which lignin is polymerised: he proposed that the association of lignin precursors with the polysaccharide surfaces might result in some organisation of these precursors prior to polymerisation. Houtman and Atalla [32] suggested that lignin, although hydrophobic, could adsorb on polysaccharide surface such as cellulose microfibrils. They proposed that polysaccharides, both cellulose and hemicelluloses, are templates for the assembly of lignin and therefore polysaccharide composition would control lignin structure. In their view, different sugar moieties will interact with lignin precursors in different ways or to different degrees. Such interactions may provide the basis for selective binding of the different lignin precursors in a manner that can impart spatial organisation to the lignin.

In their model of lignification in restricted spaces, Roussel and Lim [33] found that the geometry of the hemicellulose matrix is an important factor. Besombes and Mazeau (unpublished results) investigated the interactions between cellulose and lignin at the atomistic scale using molecular modelling. They identified hydrophobic interactions between lignin aromatic rings and apolar CH groups of cellulose. Importantly, these interactions lead to a preferential parallel orientation of lignin aromatic groups relative to cellulose surface, which appeared linked to the molecular mass and the structure of lignin. Notably, these authors found that the presence of branching and some particular inter-unit linkages were a hindrance to this parallel orientation.

From these general studies on lignin–polysaccharide interactions, it is tempting to speculate on potential links between the very peculiar polysaccharide composition of the G-layer highly crystalline cellulose, low MFA, increased galactan and decreased xylan, and the lignin modifications generally observed in this layer.

Thanks to the development of high-throughput sequencing technologies, it has been possible to generate a high number of expressed sequence tags (EST), reflecting the population of genes expressed in different types of wood, which lead to the discovery of genes specifically involved in wood formation. With the recent sequencing of its genome (http://genome.jgi-psf.org/poplar0/poplar0.home.html), poplar is now the model species for genomic studies focused on tree specific features such as wood formation. Today, more than 140 000 EST from Populus have been sequenced from a wide variety of tissues and are available in public databases [40,41]. Sterky et al. [42] were the first to report the sequencing of 5692 non-redundant EST from wood-forming tissues in poplar species. In a similar study focused on tension wood formation, Déjardin et al. [43] have sequenced more than 10 000 EST corresponding to genes expressed in woody tissues from trees induced to differentiate tension wood. The analyses of these EST gave a picture of the genes expressed in the differentiating xylem both in tension and opposite wood. In tension wood, different genes coding for arabinogalactan proteins (AGP) were strongly over-represented as well as genes from cellulose biosynthesis. AGP are highly glycosylated proteins whose function remains unclear. They are strong candidates to be mediators of cell–cell interactions and regulators of cell growth [44]. Recently, an Arabidopsis AGP mutant has been shown to exhibit perturbed cell walls and an increased sensitivity to salt stress that causes the swelling of the cells in root elongation zone [45].

Among lignin biosynthetic genes, a CCoAOMT gene was significantly under-represented in the tension wood, suggesting a lower level of expression in tension wood xylem. Interestingly, Chen et al. [46] have demonstrated in poplar that two CCoAOMT promoters (corresponding to the two CCoAOMT genes identified in poplar) were responsive to mechanical stress. Indeed, upon bending or leaning of the stem, their expression was induced in all cell types whereas without stress, their expression seems to be restricted to vessels and contact rays. Unfortunately, this study has been performed only nine days after the application of the mechanical stress and the formation of tension wood was not yet anatomically visible [46].

Only a few studies have been carried out at the protein level on tension wood formation. Baba et al. [47] identified five different proteins attached to the cell wall and the plasma membrane that were specifically present in tension wood tissues, two weeks after induction. Likewise, Plomion et al. [48] identified in E. gunii 12 different proteins significantly associated with growth strains. All these proteins still need to be identified by microsequencing or mass spectrophotometry.

As exemplified by Hertzberg et al. [49], DNA microarray allows for determining the pattern of expression of a high number of genes during wood formation. Preliminary experiments on global gene expression in tension wood performed in INRA–Orléans (Leplé et al., unpublished results), confirmed the over-expression of several AGP genes and of genes coding for enzymes from the cellulose synthase complex already detected after EST sequencing [43]. Global gene expression has also revealed that a number of other genes are regulated during tension wood formation such as, as already shown [50], a gene coding for amino-cyclo-propane (ACC) oxidase that is strongly over-expressed in tension wood. Important differences between the expression of lignin biosynthetic genes were also recorded (Table 1). Two phenylalanine ammonia-lyase (PAL), a 4-coumarate-CoA ligase (4CL), a caffeic acid O-methyltransferase (COMT), a caffeoyl-CoA O-methyltransferase (CCoAOMT), a ferulate-5-hydroxylase, a cinnamyl alcohol dehydrogenase (CAD) and a peroxidase appeared highly expressed in wood, whereas cinnamoyl-CoA reductase (CCR), two trans-cinnamate 4-monooxygenase (C4H) and a laccase exhibited only low levels of expression. More interestingly, among the highly expressed genes, only the two PAL genes and a 4CL gene were strongly (2 to 4 times) down regulated in tension wood. The expression of the genes coding for the enzymes involved in the synthesis of syringyl units (F5H, COMT and CAD) does not appear significantly affected in tension wood. Due to their low level of expression, it was not possible to conclude on the pattern of expression for CCR, C4H and laccase genes. Although, these results still need to be confirmed, they suggest that both the reduced lignin content and the increased in syringyl units observed in the G-layer may result from a change in the expression of genes coding for enzymes of the lignin biosynthetic pathway. Indeed, transgenic plants down regulated for either PAL or 4CL (both strongly down-regulated in tension wood) have been shown to exhibit lower lignin content with a preferred reduced content in G units [51,52]. PAL and 4CL genes have also been shown to be up regulated in Pinus taeda cell cultures upon phenylalanine addition [53]. This suggests that lignin reduction in tension wood may result from a shortage in lignin precursors due to a decrease in phenylalanine availability.

The availability of accurate microscopy techniques based on immunochemistry will make it possible to define precisely in which cells of the wood and even where in these cells the proteins encoded by genes regulated in tension wood are localised. Moreover, the assessment of transgenic trees for their ability to differentiate tension wood will certainly allow for unravelling the function of unknown proteins expressed in tension wood.

Transgenic trees (mostly poplar trees) targeted for nearly all the different lignin biosynthetic genes have already been produced. These trees exhibit a wide array of modifications at the level of the lignin polymer (see [54] for a recent review). The evaluation of these transgenic trees for potential alterations in their ability to form tension wood may shade some light on potential causal links between lignin characteristics and cell-wall mechanical behaviour. For example, it has been demonstrated in 4CL down-regulated poplars that a repression of lignin biosynthesis promotes cellulose accumulation [52]. It remains to determine whether this strong interrelation occurs at a metabolic level through carbon reallocation or at a developmental level through a delayed xylem differentiation and prolonged cellulose deposition.

CAD down-regulated transgenic poplar trees are characterised by a strong red coloration in the differentiating xylem [55] (Fig. 3). Interestingly, it appeared that this red coloration is absent from tension wood formed in bent stems (Photo 3). Although the nature of the compound responsible for this coloration is still not elucidated, it is likely incorporated in the lignin polymer. If this is the case, one may wonder why the red coloration seems absent from tension wood. Upon gravity induction, several other transgenic poplar trees with much reduced levels of CAD activity [56] seemed altered in their ability to form gelatinous fibres (Pilate et al., paper in preparation). This suggests a causal relationship between lignification and G-layer differentiation. Such transgenic trees certainly constitute a precious material to assess the potential function of lignin in tree physiology and in particular in the generation of growth stress. Using integrative approach, associating molecular biology, biochemistry, anatomy and biomechanics, we may expect that the function of lignification in tension wood formation will be in the near future better understood through the evaluation of the ability of these different transgenic trees to form tension wood.

Previous biochemical observations have been confirmed by genomic studies: this is the case for lignin and cellulose biosynthesis during tension wood formation. Genomic studies have also allowed the identification of genes and proteins potentially involved in tension wood formation. This is the case for AGP and also for several other genes of unknown function that were highly expressed in tension wood. In the future, we can anticipate that our knowledge on different aspects of tension wood formation will increase: this is already the case for the role of plant-growth regulators. Indeed, redistribution of growth regulators is thought to be involved in the transduction of the stimulus responsible for the induction of reaction wood formation [57]. For a long time, auxin (indole acetic acid) was the preferred candidate. However, quantification of endogenous auxin levels in poplar stems induced to differentiate tension wood leads to the conclusion that auxin distribution is most likely not the primary factor [58]. Ethylene appears now as a strong candidate. Indeed, Andersson-Gunnerås et al. [50] have recently demonstrated that ethylene production occurs asymmetrically within poplar stem upon induction by a gravitational stimulation and that a gene coding for ACC oxidase is strongly regulated in tension wood. It is worth reminding here the existence of a link between ethylene and lignification as evidenced in the phenotype of an Arabidopsis mutant [59], where the mutation of a chitinase-like gene causes an ectopic expression of CCoAOMT, an ectopic deposition of lignin and an overproduction of ethylene. Moreover, the mutation affected some aspects of plant development including an exaggerated hook curvature. It remains to determine if such a relation between lignification and ethylene also exists in the stem of a tree subjected to a mechanical stress. The endogenous substrate of chitinase-like genes in plant is still a matter of debate: interestingly, chitinase has been shown to be necessary for somatic embryo development [60], which is also controlled by AGP [61]. The possibility that certain AGP contain N-acetyl-glucosamine residues that may be processed by chitinase to produce a cellular signal certainly deserves further examination. Further studies are certainly needed to assess any potential relations during tension wood formation between lignification, ethylene production, chitinase activity, and AGP.