A collection of genetically engineered Populus trees reveals wood biomass traits that predict glucose yield from enzymatic hydrolysis

Abstract

Wood represents a promising source of sugars to produce bio-based renewables, including biofuels. However, breaking down lignocellulose requires costly pretreatments because lignocellulose is recalcitrant to enzymatic saccharification. Increasing saccharification potential would greatly contribute to make wood a competitive alternative to petroleum, but this requires improving wood properties. To identify wood biomass traits associated with saccharification, we analyzed a total of 65 traits related to wood chemistry, anatomy and structure, biomass production and saccharification in 40 genetically engineered Populus tree lines. These lines exhibited broad variation in quantitative traits, allowing for multivariate analyses and mathematical modeling. Modeling revealed that seven wood biomass traits associated in a predictive manner with saccharification of glucose after pretreatment. Four of these seven traits were also negatively associated with biomass production, suggesting a trade-off between saccharification potential and total biomass, which has previously been observed to offset the overall sugar yield from whole trees. We therefore estimated the "total-wood glucose yield" (TWG) from whole trees and found 22 biomass traits predictive of TWG after pretreatment. Both saccharification and TWG were associated with low abundant, often overlooked matrix polysaccharides such as arabinose and rhamnose which possibly represent new markers for improved Populus feedstocks.

Figure 1

The BioImprove Populus collection provides a wide variation in major traits. (a,b) Growth-related traits: stem height (a) and stem diameter (b). (c,d) Biomass recalcitrance-related traits: proportion of lignin within the detected pyrolysate from biomass (c) and ratio of S- to G-units within the lignin polymer (d). (e,f) Saccharification-related traits: glucose release after a 72 h enzymatic hydrolysis without (e) or after (f) pretreatment. Histograms represent the average value for transgenic lines (color) and wild type (black). Error bars represent standard deviation. * and ^ indicate statistically significant differences from wild type (p < 0.05 and p < 0.1 respectively) following a post-ANOVA Fisher’s test (n = 3–5).

Figure 2

The BioImprove lines display a range of total-wood glucose yield (TWG). (a) Formula for estimation of a tree’s total-wood glucose yield after pretreatment and 72 h enzymatic hydrolysis, assuming conical shape, negligible bark contribution to diameter and homogeneous wood density. (b) TWG of the BioImprove Populus lines. Each histogram represents the average value for a transgenic Populus line (color) or wild type (black). Error bars represent standard deviation. * and ^ indicate statistically significant differences from wild-type (p < 0.05 and p < 0.1 respectively) following a post-ANOVA Fisher’s test (n = 3–5).

Figure 3

Certain traits contribute more than others to predicting TWG. (a) OPLS scatter plot showing the separation of the Populus lines (dots) horizontally along the predictive component for total-wood glucose yield (TWG). Vertical separation indicates variation not correlated with TWG. The lines were coloured by TWG. (b) Plots showing the variable importance for the projection (VIP) value for each trait for the predictive part of the model (up) and for the orthogonal part of the model (down). VIP values over 1 indicate important traits. (c) Contribution of each trait to the OPLS model. Apart from saccharification traits, traits with a VIP value over 1 for the predictive part of the model were emphasized by black text and arrows. Traits marked by (*) and annotated in grey are important (VIP value over 1) for both the predictive and the orthogonal part of the model. Q² scores over 0.5 indicate significant predictivity of a model.

Figure 4

TWG can be predicted by a specific subset of traits in a composite model. Scatter plot showing for each Populus line (dots) the observed total-wood glucose yield (TWG, x-axis) versus the predicted TWG (y-axis). Q² scores over 0.5 indicate significant predictivity of a model.