A revised architecture of primary cell walls based on biomechanical changes induced by substrate-specific endoglucanases.

Xyloglucan is widely believed to function as a tether between cellulose microfibrils in the primary cell wall, limiting cell enlargement by restricting the ability of microfibrils to separate laterally. To test the biomechanical predictions of this "tethered network" model, we assessed the ability of cucumber (Cucumis sativus) hypocotyl walls to undergo creep (long-term, irreversible extension) in response to three family-12 endo-β-1,4-glucanases that can specifically hydrolyze xyloglucan, cellulose, or both. Xyloglucan-specific endoglucanase (XEG from Aspergillus aculeatus) failed to induce cell wall creep, whereas an endoglucanase that hydrolyzes both xyloglucan and cellulose (Cel12A from Hypocrea jecorina) induced a high creep rate. A cellulose-specific endoglucanase (CEG from Aspergillus niger) did not cause cell wall creep, either by itself or in combination with XEG. Tests with additional enzymes, including a family-5 endoglucanase, confirmed the conclusion that to cause creep, endoglucanases must cut both xyloglucan and cellulose. Similar results were obtained with measurements of elastic and plastic compliance. Both XEG and Cel12A hydrolyzed xyloglucan in intact walls, but Cel12A could hydrolyze a minor xyloglucan compartment recalcitrant to XEG digestion. Xyloglucan involvement in these enzyme responses was confirmed by experiments with Arabidopsis (Arabidopsis thaliana) hypocotyls, where Cel12A induced creep in wild-type but not in xyloglucan-deficient (xxt1/xxt2) walls. Our results are incompatible with the common depiction of xyloglucan as a load-bearing tether spanning the 20- to 40-nm spacing between cellulose microfibrils, but they do implicate a minor xyloglucan component in wall mechanics. The structurally important xyloglucan may be located in limited regions of tight contact between microfibrils.

Since the 1970s, xyloglucan (XyG) has been a central figure in theories relating primary cell wall architecture to plant cell enlargement (Keegstra et al., 1973;Labavitch and Ray, 1974;Hayashi, 1989). As the major physical restraint to cell expansion, the primary cell wall plays a pivotal role in cell growth, morphogenesis, and tissue mechanics (Cosgrove, 2005;Schopfer, 2006;Hamant and Traas, 2010). Understanding of cell wall architecture is a cornerstone for conceiving the possible molecular mechanisms by which plant cells modify their walls to control growth as well as influence cell mechanics, water relations, and other properties. Although the biochemical structures of cell wall polymers are well established (Harris and Stone, 2008;Caffall and Mohnen, 2009;Albersheim et al., 2011;Burton et al., 2010;Scheller and Ulvskov, 2010), less certain are their conformations in the wall and their interactions to make a functional network with high tensile strength, resilience, and ability to expand in surface area. Most current depictions of primary cell walls show XyG in an extended conformation, coating the surface of well-separated cellulose microfibrils and spanning the 20-to 40-nm distance between adjacent microfibrils (Somerville et al., 2004;Cosgrove, 2005;Albersheim et al., 2011;Doblin et al., 2010;Fig. 1). Pauly et al. (1999) operationally identified three XyG domains: a xyloglucanase-accessible domain that includes the hypothetical tethers, loops, and free strands between microfibrils; a second domain tightly bound to the surface of cellulose microfibrils (defined as KOH soluble but inaccessible to xyloglucanase); and a third XyG domain only released upon complete wall digestion and thought to be trapped inside or between cellulose microfibrils. This "tethered network model" is based on numerous, but indirect, results, including the extractability and binding behavior of XyG as well as microscopy indicating that XyG is present in the spaces between microfibrils (Hayashi and Maclachlan, 1984;Hayashi et al., 1987;McCann et al., 1990;Pauly et al., 1999). The concept has been illustrated by a variety of cartoons that vary in complexity and detail and typically show pectic polysaccharides as an independent, coextensive phase embedding the cellulose-XyG network.
As noted by Scheller and Ulvskov (2010), "cell wall models that are created to illustrate how wall polymers are organized in higher-order structures influence the thinking about biological functions of the hemicelluloses." Thus, any molecular model of the primary cell wall leads to predictions and expectations about how it will respond to mechanical forces, what kinds of enzymes and biochemical processes might be involved in its expansive growth, and how new polymers could be integrated into its structure. The tethered network model inevitably emphasizes the hypothetical XyG tethers between microfibrils as key determinants of cell wall mechanics and as points for control of cell wall expansion. Although this model is widely accepted, the biomechanical evidence in support of XyG tethers is indirect and open to alternative interpretations. Moreover, some experimental results seem at odds with its predictions (e.g. the failure of XyG endohydrolase/ transglucosylase to induce significant cell wall creep comparable to expansin's action [McQueen-Mason et al., 1993;Saladié et al., 2006] and the near-normal phenotype of a XyG-deficient Arabidopsis [Arabidopsis thaliana] mutant [Cavalier et al., 2008;Park and Cosgrove, 2012]). Moreover, NMR studies have not detected the extensive XyG-cellulose interactions proposed in the model (Bootten et al., 2004;Dick-Pérez et al., 2011), and neutral pectins have been shown to bind to cellulose in vitro, potentially functioning as tethers in parallel with XyG (Zykwinska et al., 2005). Thus, there are good grounds for reexamining the role of XyG in cell wall mechanics.
In this study, we tested the biomechanical predictions of the tethered network model. For our approach, we digested primary walls with endoglucanases that differ in their substrate specificities and measured the mechanical consequences with assays of cell wall creep and mechanical compliance. This is related to the approach originally used to identify expansins (McQueen-Mason et al., 1992) and provides a direct assessment of the structural role of specific wall polymers. We reasoned that if cellulose microfibrils were connected primarily by XyG tethers, as depicted in Figure 1, digestion of the enzyme-accessible XyG between microfibrils should provoke wall extension and mechanical weakening. The results proved contrary to these predictions, instead pointing to a minor, relatively inaccessible XyG-cellulose structure with importance in cell wall mechanics and with consequences for models of cell wall architecture and mechanisms of cell wall loosening.

Distinctive Actions of Family-12 Endoglucanases
For initial experiments, we made use of three endob-1,4-glucanases from glycosyl hydrolase family 12 (www.cazy.org). The advantage of this approach is that the enzymes have similar structures and enzymatic mechanisms (Sandgren et al., 2005), so differences in wall action are more likely to be due to substrate specificity than to differences in the size or shape of the protein. The enzymes are small onedomain proteins (25 kD, approximately 4 nm in diameter), lack a separate carbohydrate-binding domain, and differ in substrate specificities. Their size is below the porosity limit of primary cell walls (Carpita et al., 1979;Baron-Epel et al., 1988) and less than the spacing between cellulose microfibrils, so they should have access to polysaccharides in the hydrated matrix between microfibrils (e.g. the hypothesized XyG tethers). One enzyme was XEG, the XyG-specific endoglucanase, or "xyloglucanase," from Aspergillus aculeatus used by Pauly et al. (1999) to define the enzyme-accessible tethers, loops, and free strands of XyG between cellulose microfibrils. The second enzyme, Cel12A from Hypocrea jecorina, digests both XyG and cellulose (Yuan et al., 2001;Karlsson et al., 2002). The third enzyme, CEG from Aspergillus niger, is a cellulose-specific endoglucanase, or "cellulase," according to the supplier (www.megazyme.org). We first tested the substrate specificity of these enzymes with XyG and cellulose substrates, confirming that XEG indeed hydrolyzed XyGs but not cellulose, CEG hydrolyzed cellulose but not XyG, and Cel12A possessed both activities (Fig. 2). We also tested the three enzymes for hydrolytic activities against the other major wall polysaccharides (other than glucans), but the activities were negligible (Supplemental Fig. S1).
To test their ability to induce cell wall creep, we added the enzymes individually to the solution surrounding cell wall specimens clamped in a constantforce extensometer (Cosgrove, 1989;Yuan et al., 2001). The wall specimens were prepared from growing cucumber (Cucumis sativus) hypocotyls briefly heated to inactivate endogenous wall-loosening activities and lightly abraded to promote rapid enzyme penetration through the cuticle. Based on the tethered network model, we expected XEG and Cel12A to induce creep, assuming that they could access the XyG tethers between microfibrils. At 0.1 mg mL 21 , Cel12A indeed Figure 1. Cartoon of the tethered network model of the primary cell wall, in which xyloglucans (thin strands) bind to the surface of cellulose microfibrils (thick rods), forming a load-bearing network (redrawn after Albersheim et al. [2011] and Pauly et al. [1999]). The semicircle (top left) represents the approximate size of a family-12 endoglucanase used in this study, assuming microfibrils with 20-nm spacing. The microfibrils and the xyloglucan would be longer than what is represented here.
[See online article for color version of this figure.] induced a high creep rate (Fig. 3A), confirming a previous report (Yuan et al., 2001). However, XEG was totally ineffective, even at 1,000 times greater concentration (Fig. 3B). This was unexpected. Likewise, CEG was ineffective (as expected).
As XEG and Cel12A were able to digest XyG in the cucumber walls (see below), the lack of wall extension by XEG indicated that hydrolysis of enzyme-accessible XyG tethers was not sufficient to loosen the wall, whereas the positive result with Cel12A, combined with the negative result with CEG, suggested that digestion of both XyG and an unbranched b-1,4-Dglucan (cellulose) was required for endoglucanaseinduced creep. We tested this idea by simultaneous treatment with XEG and CEG (Fig. 3C); at 10 mg mL 21 , the combined enzymes had no effect on creep, and at 100 mg mL 21 , the combination induced a very small creep rate, which was negligible compared with that induced by 1,000 times less Cel12A. Thus, simultaneous digestion of XyG and cellulose by separate enzymes was unable to mimic the creep effect of Cel12A. In other experiments, we found that simultaneous treatment with CEG reduced the effectiveness of Cel12A (Fig. 3D), suggesting the CEG nonproductively occupied a site that Cel12A could digest and thereby induce wall creep.
These enigmatic results suggested various possible explanations: (1) Cel12A might digest some other structural (nonglucan) component of the wall. We found no evidence to support this possibility from assays with a range of cell wall polysaccharides (Supplemental Fig. S1).
(2) XEG access to interfibrillar XyG might be blocked by pectins (Marcus et al., 2008;Zhao et al., 2008). This seemed unlikely in view of the strong effect of Cel12A, but we tested this idea directly by pretreating the wall with pectate lyase and the calcium chelator cyclohexane diamine tetraacetic acid (CDTA) to reduce the postulated blockage by pectin. These pretreatments transiently weakened the wall, but XEG still did not cause creep (Fig. 4), so we reject this explanation. (3) The hypothetical XyG tethers might assume an unusual conformation or packing that XEG cannot digest, but it is digestible by Cel12A. (4) The hypothetical connection between microfibrils might be  These curves are averages of six to 19 responses. Arrows indicate when enzyme was added; the spike at this point is a mechanical artifact due to buffer exchange. No enzyme was added in the control. A, Wall creep induced by Cel12A (0.1 mg mL 21 ; 13 # n # 15). B, Wall creep was not induced by XEG or CEG (each 100 mg mL 21 ; 13 # n # 19). C, Wall creep was not induced by XEG + CEG mixtures (8 # n # 11). D, Wall extension induced by Cel12A (0.1 mg mL 21 ) alone and by Cel12A (0.1 mg mL 21 ) + CEG (10 mg mL 21 ; n = 6). Error bars show 95% confidence intervals. * P , 0.05 in t test comparison of means at selected time points.
Probing Cell Wall Architecture with Endoglucanases an amalgam of XyG and cellulose, whose efficient digestion requires one enzyme with both activities (i.e. separate enzymes would be ineffective due to the need to alternatively and repeatedly hydrolyze the two intertwined polymers).
To test possibilities 3 and 4, we assayed two additional endoglucanases with xyloglucanase and cellulase activities: BlXG12, a family-12 enzyme from Bacillus licheniformis, and PpXG5, a family-5 enzyme from Paenibacillus pabuli (Gloster et al., 2007). Our tests showed that the enzymes possessed both xyloglucanase and cellulase activities (Supplemental Fig. S2) and that both enzymes induced cell wall creep (Fig. 5), although with different lag times that depend on the concentration and effectiveness of the enzymes (Supplemental Fig. S3; Yuan et al., 2001). We also tested AnXEG, a family-12 enzyme from Aspergillus nidulans that hydrolyzes XyG but not cellulose (Bauer et al., 2006). Like the XEG from A. aculeatus, AnXEG was ineffective in creep assays (Fig. 5). These results show that the creep activity of Cel12A is not unique to that enzyme and support hypothesis 4, that endoglucanaseinduced creep requires an enzyme able to digest both XyG and cellulose. They also undercut hypothesis 3, because it is unlikely that an unusual conformational variant of XyG would be substrate for both Cel12A and the nonhomologous family-5 enzyme (PpXG5), which have different structures and substrate-binding grooves (Sandgren et al., 2005;Gloster et al., 2007), yet would not be digested by two family-12 xyloglucanases homologous to Cel12A.
Additional results confirming the involvement of XyG in Cel12A-induced creep came from parallel experiments with walls from an Arabidopsis mutant (xxt1/xxt2) lacking XyG (Cavalier et al., 2008). The walls of this mutant should lack the hypothetical XyGcellulose structure that is digested by Cel12A to induce creep. Therefore, we tested the wall-creep activity of Cel12A using hypocotyl walls from wild-type and xxt1/xxt2 Arabidopsis. With wild-type walls, Cel12A induced creep very similar to its action on cucumber walls, whereas no creep response was detected with walls from the xxt1/xxt2 mutant (Fig. 6A). Additionally, XEG did not cause cell wall creep in the wild-type Arabidopsis walls, similar to the results with cucumber walls (Fig. 6B). These results show that XyG is necessary for endoglucanase-induced creep, and they provide additional support for the concept of a XyGcellulose structure digested by Cel12A but recalcitrant to XEG.

Endoglucanase Effects on Cell Wall Compliance and Artificial Cellulosic Composites
We extended the foregoing results with additional experiments to assess the effects of family-12 endoglucanases on the mechanical compliance of cucumber hypocotyl walls. For these measurements, the walls were predigested with enzyme for 1 h and then extended in two cycles (Cosgrove, 1989). The slope (Dlength/Dforce) of the second extension yields a measure of the elastic (reversible) compliance, whereas the plastic compliance is calculated from the difference between the first and second extensions. As shown in Figure 7, Cel12A at 0.1 mg mL 21 increased the plastic compliance by 70% and the elastic compliance by 30% compared with the control, but no significant changes were found after digestion with XEG or CEG, even at . Wall-creep responses to XEG are not sensitized by pectindisrupting pretreatments. A, Walls predigested with pectate lyase (initial 30 min) followed by XEG (mean of 10 curves). B, Walls preincubated in 50 mM CDTA (pH 6.5) for the first 30 min followed by XEG (average of 12 responses). The control included CDTA pretreatment but not XEG. Figure 5. Wall-creep responses to family-5 and family-12 endoglucanases (means of eight to 10 responses). PpXG5 and BlXG12 have both cellulase and xyloglucanase activities, whereas AnXEG (family 12) has xyloglucanase but not cellulase activity. The solid arrow indicates when the enzymes (30 mg mL 21 ) were added. Error bars show 95% confidence intervals (for AnXEG, the bar is smaller than the line width). * P , 0.05 by Student's two-tailed t test for difference before and after enzyme treatment.
1,000 times higher concentrations. These results parallel the effects seen in creep assays and indicate that the particular XyG-cellulose structure digested by Cel12A is a determinant of cell wall mechanics as well as enzymeinduced creep, whereas the XEG-accessible XyG has a negligible impact on wall mechanics, at least as measured by creep and stress/strain methods.
To gain further insight into the wall-loosening action of these enzymes, we turned to cellulosic networks ("pellicles") produced by cultures of Gluconacetobacter xylinus (formerly Acetobacter xylinus). Pellicles can be produced in the presence of XyG or other matrix polysaccharides, which get incorporated to form composites that have been considered analogs of plant cell walls (Chanliaud et al., 2002(Chanliaud et al., , 2004. We first tested the effect of the three family-12 enzymes on pure cellulosic pellicles (no polysaccharide additives) in creep assays (Fig. 8A). Both CEG and Cel12A induced creep, followed by breakage after approximately 1 h (3%-4% extension), whereas XEG was ineffective. These results are consistent with cellulose friction and entanglements determining the rheology of such networks (Chanliaud et al., 2002).
With cellulosic composites produced in the presence of 0.5% tamarind (Tamarindus indica) XyG, Cel12A still induced a high creep rate, but the response to CEG was greatly diminished (Fig. 8B). XEG also induced a substantial creep rate in the XyG composite, unlike its effect on plant cell walls. These results indicate that direct cellulose entanglements are less limiting for creep for this XyG-containing material compared with the pure cellulosic composite, most likely as a result of the large-scale ordering of microfibrils in the XyGcontaining composite (Chanliaud et al., 2002;Whitney et al., 2006). The XEG result is also consistent with XyG cross-linking of the bacterial composite but stands in stark contrast to the result obtained with plant cell walls. We conclude that XyG in the bacterial composite does not closely mimic its structural role in native plant cell walls.
Evidence for a Special XyG Component Accessible to Cel12A To verify that XyG in the cucumber wall was accessible to XEG and Cel12A, as well as to look for structural differences in the corresponding XyGs, we treated whole walls with XEG or Cel12A and measured the released xyloglucan oligosaccharides (XGOs) through high-performance anion-exchange chromatographypulsed amperometric detection (HPAEC-PAD) analysis. At an enzyme loading of 10 mg mL 21 , Cel12A released about 40% more XGOs than XEG did, but the profiles of XGOs were similar (Fig. 9A). The quantitative difference between the two enzymes was not sufficient to account for their different effects on cell wall mechanics. The main XGOs liberated by the two enzymes were XXXG and XXFG (using the shorthand for XGOs proposed by Fry et al. [1993]). Smaller amounts of XXG, XXGG, XLFG, and other XGOs were also released. This XGO pattern is similar to that reported for primary cell walls of other dicots (Pauly et al., 1999;Hsieh and Harris, 2009). The similarity in XGO patterns indicates the XyGs targeted by the two enzymes were structurally similar.
To examine this last point in further detail, we sequentially digested and extracted the cucumber walls with XEG, Cel12A, 4 M NaOH, and finally a cellulase that completely solubilized the residue (endoglucanase I from Megazyme). These sequential fractions corresponded to 50%, 8%, 33%, and 9% of the total XyG in the wall (Supplemental Fig. S4A); only minor differences in the XGO profiles were seen in these fractions (Supplemental Fig. S4, B and C), indicating only subtle differences in XyG structure in these various compartments. We also detected cello-oligosaccharides in the Cel12A digest but not in the XEG digest (Supplemental Fig.   Figure 6. Creep responses of etiolated Arabidopsis wild-type (ecotype Columbia) and XyG-deficient mutant (xxt1/xxt2) hypocotyl walls to Cel12A and XEG treatment (average values of seven to eight responses). A, Creep responses to Cel12A. The arrow indicates when Cel12A (1 mg mL 21 ) was added. Error bars show 95% confidence intervals. * P , 0.05 for t test comparison of the two treatments at the time indicated. B, Creep responses to XEG. The arrow indicates when XEG (10 mg mL 21 ) was added. Figure 7. Alteration of wall compliance by family-12 enzyme treatment. Error bars show 95% confidence intervals (10 # n # 12). * P , 0.05 by Student's two-tailed t test for difference from buffer control.
Probing Cell Wall Architecture with Endoglucanases S4D), consistent with the idea the Cel12A-induced creep involves cellulose digestion. Figure 9B shows the release of XGOs as a function of XEG and Cel12A concentration. At 0.1 mg mL 21 , Cel12A released much more XGO than did XEG. More significantly, the XEG curve reached a plateau at 100 mg mL 21 , whereas Cel12A continued to release XGOs, exceeding the amount of XEG-released XGO by approximately 20%. These XGO-release curves support the conclusion that the wall contains a minor XyG compartment that is digestible by Cel12A but not by XEG. This compartment may be identical to the special XyG-cellulose structure detected above in the biomechanics assays. Consistent with this idea, a plot of creep rate as a function of XGOs released by the two enzymes shows that creep rate increased monotonically as XGOs were released by Cel12A, but the XEG curve plateaued (Fig. 9C). To test this idea in yet another way, walls were exhaustively digested with one of these enzymes and washed, and then XGOs released by the other enzyme were assayed. When XEG predigestion was followed by Cel12A, much higher levels of XGOs were released than when the order of the enzyme treatments was reversed (Fig. 9D). These additional results support the idea that Cel12A can digest a XyG compartment that is recalcitrant to XEG.
Finally, we evaluated the total XyG content of cucumber walls by measuring isoprimeverose [xylosyla-(1,6)-glucose] after exhaustive digestion with Driselase (Cavalier et al., 2008). Approximately 45% of the XyG was removed from the wall by exhaustive XEG digestion, whereas 55% was removed by Cel12A (Supplemental Fig. S5). Thus, the physical effects of Cel12A are associated with digestion of a XyG compartment that constitutes a minor fraction of the total XyG (maximally 10% [i.e. the difference between 55% and 45%]). In fact, the biomechanical changes induced by Cel12A began at much lower levels of XyG digestion. For instance, at 0.1 mg mL 21 , Cel12A digestion increased wall compliance (Fig. 7) and creep rate (Fig. 3) within 1 h. Only 1.5% of the XyG was solubilized by such a treatment (Supplemental Fig. S6), and only onefifth of this XyG is likely to be load bearing and Cel12A specific (compare with Supplemental Fig. S4A). Thus, the large biomechanical effects caused by Cel12A may be traced to the digestion of a remarkably small fraction (perhaps 0.3%) of the total wall XyG.

DISCUSSION
We used a set of homologous endoglucanases with varying substrate specificities to test the predictions of the tethered network model (Fig. 1), reasoning that digestion of xyloglucanase-accessible XyGs between microfibrils should result in cell wall weakening, which we assayed by creep and stress/strain measurements. The results are at odds with this prediction and thus argue against a major mechanical role for XyG tethers spanning the space between adjacent microfibrils. Instead, our results point to a minor, relatively inaccessible XyG compartment that may be intertwined or otherwise complexed with cellulose, forming a load-bearing connection between microfibrils. These results have implications for cell wall architecture and put constraints on the potential for enzymatic mechanisms of cell wall loosening.

Microfibrils Connected Not by XyG Tethers but a XyG-Cellulose Amalgam
According to current views (Pauly et al., 1999;Albersheim et al., 2011), XyG tethers spanning the 20-to 40-nm distance between adjacent microfibrils (Fig. 1) should be part of the XEG-susceptible domain. We confirmed that XEG indeed digests nearly half of the wall XyG, but without a concomitant change in either the creep behavior or stress/strain responses of the wall. Thus, if XyG tethers exist, they are evidently not significant determinants of the mechanics of these walls, at least by these measures. We note that tetherlike strands have been observed in some primary walls Error bars show 95% confidence intervals. * P , 0.05 by Student's two-tailed t test for before and after enzyme treatment. For Ce112A and CEG treatments, half the samples broke at approximately 90 to 120 min, while the other half continued to creep at diminished rates. B, Creep of cellulose-XyG composite in response to Cel12A (3 mg mL 21 ), XEG (100 mg mL 21 ), and CEG (100 mg mL 21 ) in 20 mM sodium acetate (pH 5.0) with 7.5 g of force. Error bars show 95% confidence intervals. * P , 0.05 by Student's two-tailed t test for before and after enzyme treatment. (McCann et al., 1990;Itoh and Ogawa, 1997) but not in others (Fujino et al., 2000). Moreover, even when they are observed, their contribution to cell wall mechanics is difficult to assess from such images. Our XEG results do not support the tethered network concept, at least for hypocotyl walls from cucumber and Arabidopsis.
The lack of a XEG effect contrasts with the pronounced action of the closely related endoglucanase Cel12A, where we found evidence that a quantitatively minor XyG-cellulose complex contributes to cell wall mechanics. We base this conclusion on the induction of cell wall creep by three endoglucanases that hydrolyze both XyG and cellulose (Cel12A, BlXG12, and PpXG5), whereas no creep effect was seen when walls were treated with a cocktail containing both XEG and CEG, endoglucanases specific for XyG and cellulose, respectively. The involvement of XyG in these responses was further established by use of the XyGdeficient (xxt1/xxt2) mutant of Arabidopsis: Cel12A induced creep in the wild-type walls but not in the XyG-deficient walls. These and other results reported here lead us to postulate that endoglucanase-induced wall loosening requires the digestion of a XyG-cellulose complex that links microfibrils together in a distinctive way.
The exact location of the postulated XyG-cellulose complex is uncertain, but the fact that it is not digested by the combination of XEG and CEG suggests that it is located in a relatively inaccessible site and that its hydrolysis requires repeated alternating attacks on XyG and cellulose chains. Otherwise, one would expect that separate attack by the two enzymes would digest the complex. One possible scheme that is consistent with our results is depicted in Figure 10A, where the important XyG is depicted as a one-molecule-thick adhesive between adjacent microfibrils. In this scheme, microfibril adhesion is mediated by lateral, noncovalent bonding by a single XyG layer between two microfibril surfaces. With its ability to digest both cellulose and XyG, Cel12A could digest this "sequestered" XyG, perhaps after an appreciable lag, and the microfibrils would then be freed to move in response to wall stress. Microfibril movement after release would presumably Figure 9. Dionex HPLC analysis of XGOs of cucumber hypocotyls released by Cel12A and XEG. A, HPLC profiles of XGOs released by Cel12A and XEG (10 mg mL 21 ). B, Total XGOs released as a function of Cel12A and XEG concentration. Error bars show 95% confidence intervals (n = 2). C, Plot of peak wall extension rate against total XGOs released. Error bars show 95% confidence intervals (HPLC, n = 2; wall extension assay, 10 # n # 19). D, Amount of XGOs released by XEG or Cel12A following exhaustive predigestion with the other enzyme. Values are given for release by the second enzyme treatment. Error bars show 95% confidence intervals (n = 2). * P , 0.05 by Student's two-tailed t test for difference between means. Probing Cell Wall Architecture with Endoglucanases be limited by viscous drag and the entanglement of other matrix polymers surrounding or bound to the microfibril. In this scheme, microfibril bundling, mediated by XyG, would play a key role in wall biomechanics and the responses to Cel12A digestion. The fact that Cel12A increases cell wall compliance (Fig. 7) suggests that it irreversibly reduces the number of such microfibrilmicrofibril contacts, which we propose are important determinants of primary cell wall mechanics.
A variant scheme is depicted in Figure 10B, where the important XyG at microfibril junctions is pictured as entangled with disordered surface glucans of adjacent microfibrils. The result is a local region of microfibril adhesion that could only be undone by an enzyme with both xyloglucanase and cellulase activities. A variation of this scheme might entail deeper penetration of the XyG into the cellulose microfibrils, resulting in a disordered region that might be prone to adhere to similarly disordered regions of adjacent microfibrils.
This new architecture involves XyG, but not as a gap-spanning tether stretched taut between adjacent microfibrils, as commonly depicted. Instead, a minor fraction of the XyG acts as a microfibril adhesive in the limited places where microfibrils come in sufficiently close contact and a XyG is suitably located to become sandwiched between the two cellulose surfaces, perhaps entangled with surface glucans of the microfibrils. In this scheme, turgor-generated wall stress is not transmitted between microfibrils by the XyG backbone, as envisioned in the tethered network model, but involves numerous lateral interactions proportional to the contact area between cellulose and XyG. Such an architecture might be stronger than the tethered network of Figure 1, where XyG chains are liable to detach from the cellulose surface, one glucan at a time.
This alternative architecture requires regions of close contact (or, more correctly, XyG-mediated contact) between microfibrils. This differs from most current models, where a coating of XyG and other matrix polymers is thought to prevent microfibril contacts (Albersheim et al., 2011). However, there are reasons to suspect that microfibril contacts likely exist. First, scrutiny of cell wall images often reveals potential points of microfibril contact and bundling where these close contacts may occur (Thimm et al., 2000;Ding and Himmel, 2006). Second, NMR studies have concluded that only a small proportion (less than 10%) of the cellulose surface is coated with XyG (Bootten et al., 2004;Dick-Pérez et al., 2011). Third, cell walls (not extracted to remove hemicelluloses) are readily labeled with affinity reagents that bind to cellulose surfaces (e.g. reagents such as the direct cotton dye Pontamine Fast Scarlet 4B [Anderson et al., 2010] and carbohydrate-binding modules that bind crystalline cellulose [Knox, 2008;Lacayo et al., 2010]). Such labeling may indicate unoccupied cellulose surfaces. Fourth, the primary cell wall of the xxt1/xxt2 mutant, which lacks XyG, maintains most of its functionality (Cavalier et al., 2008;Park and Cosgrove, 2012), which would be unlikely if an essential role of XyG were to coat cellulose surfaces to prevent microfibril coalescence.

Implications for Plant Cell Wall Loosening
These results also have various implications for mechanisms of plant cell wall loosening. First, XyG endotransglucosylase is a plant enzyme that cuts and ligates XyG (Fry et al., 1992;Nishitani and Tominaga, 1992), and in view of Figure 1, it would be expected to loosen the cell wall, leading to cell growth. However, previous tests by creep and stress/strain assays have not confirmed this prediction (McQueen-Mason et al., 1993;Saladié et al., 2006), which is understandable if the enzyme-accessible XyG is not load bearing. By restructuring XyGs in the XEG-accessible domain, the action of this enzyme might result in subtle cell wall changes (Kaku et al., 2004;Van Sandt et al., 2007), but if it cannot access the structurally important XyGcellulose compartment identified here, its biomechanical actions are likely to be indirect and limited.
Second, we note that Cel12A is not a plant enzyme and that plants lack family-12 glycosyl hydrolases (www.cazy.org). Plant genomes encode glycosyl hydrolases from 34 other enzyme families, two of which (families 9 and 16) may have endo-b-1,4-glucanase activity (Lopez-Casado et al., 2008), but it is unclear whether any of them exert the biomechanical effects of Cel12A. The plant growth effects observed with ectopic expression of plant cellulases may be mediated by a similar mechanism or by some other route (Park et al., 2003;Shani et al., 2004;Hartati et al., 2008). Auxin treatments result in slow increases in wall plastic and elastic compliances (Cleland, 1984), so the cell wall changes caused by Cel12A might be similar to those that occur during long-term auxin-induced growth. Whether this is related to the auxin-induced metabolism of XyG (Labavitch and Ray, 1974;Nishitani and Masuda, 1983;Talbott and Ray, 1992) or to other structural changes in the wall remains to be determined.
Finally, we note that Arabidopsis xxt1/xxt2 walls are deficient in creep response both to Cel12A (Fig. 6) and to a-expansin (Park and Cosgrove, 2012), suggesting a common target of these two proteins. If this is true, then the XyG-cellulose structure identified here may participate in growth responses mediated by expansins and perhaps other wall-loosening agents.
Our revised model of the cellulose-XyG network resolves serious contradictions of the tethered network model posed by results in this and other studies and suggests that XyG-mediated control of wall extension may be restricted to relatively inaccessible contact surfaces between microfibrils of the primary cell wall. However, this is not the only possible model, and we note some of its limitations. It focuses on the structural and biomechanical architecture of the cellulose-XyG network and does not speak to contributions by other matrix polymers (pectins and arabinoxylans in partic-ular), which may be important in primary cell walls of different composition, at other developmental states, or for other physical aspects of primary cell wall function. The mechanical tests employed here made use of external uniaxial forces, which do not mimic turgor-generated forces in all respects, and microfibril movements may differ in some aspects. Numerous details of the revised model are not defined, such as the origin, distribution, density, structure, and modification of the biomechanical "hot spots." We offer the revision as a working hypothesis to direct further testing and refinement of our understanding of primary cell walls and how they expand.

Expression and Purification of Cel12A and AnXEG
Cel12A from Hypocrea jecorina was expressed in Escherichia coli Tuner (DE3) and purified as described by Yuan et al. (2001). AnXEG from Aspergillus nidulans was expressed in Pichia pastoris, and purification was performed as described (Cregg et al., 2000;Bauer et al., 2006). The M r and quality of purified proteins were determined by SDS-PAGE and Coomassie blue staining. CEG from Aspergillus niger was purchased from Megazyme (E-CELAN). XEG (from Aspergillus aculeatus), PpXG5 (from Paenibacillus pabuli), and BlXG12 (from Bacillus licheniformis) were obtained from Novozymes. Protein concentrations were quantified by the method of Bradford (1976) or based on supplier data.

Wall Extension (Creep) Assays
Cucumber (Cucumis sativus) and Arabidopsis (Arabidopsis thaliana) hypocotyls from 4-d-old etiolated seedlings were collected on ice water, and 1-cm segments were excised from the region immediately below the apical hook and frozen at 280°C as described (Cosgrove, 1989;McQueen-Mason et al., 1992). Cuticles were abraded by rubbing the frozen segments with a carborundum slurry (300-mesh size) to facilitate buffer and protein penetration, and the abraded segments were washed extensively with distilled, deionized water. Cucumber walls were pressed under a weight for 5 min to remove excess water, but no weight was applied to Arabidopsis walls because they were very fragile. Pressed specimens were heat inactivated in boiling water for 15 s (cucumber) or 12 s (Arabidopsis). The wall specimens were clamped (5 mm between clamps) under a constant force of 20 g (cucumber) and 0.5 g (Arabidopsis). The clamped wall segments were incubated with 20 mM sodium acetate (pH 5.0) for 30 min, the incubation buffer was replaced with the same buffer containing enzyme, and wall extension was recorded digitally every 30 s for 2 h (Cosgrove, 1989;Durachko and Cosgrove, 2009). Samples with excessive initial creep rates or showing signs of premature breakage were excluded from analysis (approximately one out of 10 samples). Creep-rate time courses were averaged, and appropriate time points were analyzed for statistical differences. Typical variability in creep responses is illustrated in Supplemental Figure S7, which shows the individual data sets averaged in Figure 3. Wall extension assays, unless otherwise mentioned, were performed with cucumber hypocotyl walls.

Stress/Strain Assays
Heat-inactivated cucumber walls prepared as above were treated with enzyme in 20 mM sodium acetate (pH 5.0) at 27°C for 1 h. The walls were clamped in a custom-made extensometer with 5 mm between clamps and extended at a rate of 3 mm min 21 until a force of 20 g was reached; plastic and elastic compliances were calculated from the slopes of the extension/force curves, as described (Cosgrove, 1989;Yuan et al., 2001).

Bacterial Cellulose Composite Preparation and Extension Assay
Gluconacetoacter xylinus (ATGC 53524) was cultured in Hestrin-Schramm medium (Hestrin and Schramm, 1954) containing 2% Glc with or without 0.5% tamarind (Tamarindus indica) XyG (Megazyme) at 30°C under static conditions. After 3 d of fermentation, bacterial pellicles (0.23 mm thick) were harvested and washed thoroughly with distilled, deionized water. Cellulose and cellulose/XyG materials were cut into strips (0.55 mm 3 0.23 mm in cross-section and 10 mm in length), clamped in a constant force extensometer (5 mm between clamps), and extended with 15 and 7.5 g of force, respectively, with 20 mM sodium acetate (pH 5.0) buffer. Enzymes were added after 30 min.

XGO Analysis
To assess XGOs released by Cel12A and XEG, frozen and thawed cucumber hypocotyl segments (apical 10 mm from eight etiolated hypocotyls; 5 mg dry weight total) were abraded with a carborundum slurry, pressed under a weight for 5 min, heat inactivated in boiling water for 15 s, and incubated with Cel12A or XEG in 100 mM ammonium acetate (pH 5.0) with 0.02% NaN 3 (to inhibit microbial growth) at 27°C for 24 h or for other times as indicated. Supernatant was collected and boiled for 5 min to inactivate the enzymes. For exhaustive XyG digestion, 5 mg (dry weight) of cucumber walls prepared as described above was predigested with one enzyme (XEG or Cel12A; each 100 mg mL 21 ) in 100 mM ammonium acetate (pH 5.0) with 0.02% NaN 3 at 27°C for 48 h with shaking, washed with distilled, deionized water, boiled in distilled, deionized water for 5 min to inactive the enzymes, and digested with the other enzyme (each 10 mg mL 21 ) in 100 mM ammonium acetate (pH 5.0) with 0.02% NaN 3 at 27°C for 24 h with shaking. The supernatant was collected and dried under vacuum, and the XGOs were dissolved in 1 mL of 80% ethanol. For sequential XyG digestion, 5 mg (dry weight) of cucumber hypocotyls was prepared as described, heat inactivated, and finely chopped with a razor blade. Chopped walls were incubated in 1 mL of 100 mM ammonium acetate (pH 5.0) with XEG (100 mg mL 21 ) containing 0.02% NaN 3 at 27°C for 24 h with shaking. The supernatant was collected and dried under vacuum, and XGOs generated by XEG were dissolved in 1 mL of 80% ethanol. The XEG-digested walls were boiled in water for 15 s and incubated in 1 mL of 100 mM ammonium acetate (pH 5.0) with Cel12A (100 mg mL 21 ) containing 0.02% NaN 3 at 27°C for 24 h with shaking. The supernatant was collected and dried under vacuum, and XGOs (released by Cel12A) were dissolved in 80% ethanol. Cel12A-digested walls were repeatedly digested with Cel12A to confirm complete XyG digestion by Cel12A as described. XyG of Cel12Adigested walls was extracted sequentially three times with 4 M NaOH containing 20 mM NaBH 4 for 6, 6, and 12 h at room temperature with shaking. The supernatant containing XyG was neutralized with acetic acid, dialyzed with distilled, deionized water, and dried under vacuum. The XyG solubilized by NaOH extraction was digested with XEG (100 mg mL 21 ) in 1 mL of 100 mM ammonium acetate (pH 5.0) containing 0.02% NaN 3 at 27°C for 24 h with shaking. Residues of walls extracted with 4 M NaOH were thoroughly washed, neutralized with distilled, deionized water, and incubated with 100 mg mL 21 cellulase (endoglucanase I) from Trichoderma longibrachiatum (Megazyme; E-CELTR) in 1 mL of 100 mM ammonium acetate (pH 5.0) containing 0.02% NaN 3 at 37°C for 24 h with shaking. The supernatant was dried under vacuum, and the XGOs (released by cellulase) were dissolved in 80% ethanol. For HPAEC-PAD analysis, XGO-containing samples were centrifuged at 10,000 rpm for 5 min; the supernatants were completely dried, redissolved in distilled, deionized water, and filtered through a 0.22-mm centrifuge filter (Amicon). XGOs were eluted from a Dionex CarboPac PA-200 column at a flow rate of 0.5 mL min 21 using a linear gradient over 35 min starting with 0.1 M NaOH to a 0.1 M sodium acetate and 0.1 M NaOH (1:1, v/v) mixture (Park and Cosgrove, 2012). XGOs were detected with a pulsed amperometric detector.

Isoprimeverose Analysis of Cucumber Hypocotyls by Driselase Digestion
Heat-inactivated cucumber hypocotyls (5 mg dry weight) were exhaustively digested with 100 mg of XEG or 100 mg of Cel12A with 1 mL of 100 mM ammonium acetate (pH 5.0) containing 0.02% NaN 3 at 27°C for 48 h with shaking. Nondigested cucumber walls were used for a control. Walls were incubated with 500 mL of 0.5% Driselase (Sigma) in pyridine:acetic acid:water (1:1:98 [pH 4.7]; containing 0.5% chlorobutanol and 0.02% sodium azide) at 37°C for 72 h with shaking as described (Gardner et al., 2002). The Driselase digest was added to 5 volumes of ethanol, incubated at 220°C for 16 h, and centrifuged. The supernatant was dried and resuspended in 500 mL of distilled, deionized water. For HPAEC-PAD analysis, the sample was filtered through a 0.22-mm filter, separated with a Dionex Carbo PA-20 column at 0.5 mL min 21 from 0 to 20 min under 125 mM NaOH isocratic conditions, and detected with a pulsed amperometric detector.

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S3. Concentration dependence of creep induced by family-5 and family-12 endoglucanases.
Supplemental Figure S7. Individual creep curves shown in Fig. 3, A and D.