The experimental field is located in the Hauts-de-France region of Northern France (49°53 N, 3°00 E) at the INRA experimental unit in Estrées-Mons.

The trial was planted by hand in spring 2007 at a rhizome planting density of 2 plants per m². The plants were watered immediately after planting to ensure good root contact with the soil. No irrigation was applied during the following years of cultivation. No fertilizer was applied. The weeds were controlled each year by hand and machine hoeing.

Twenty-one Miscanthus genotypes were planted in 2007 in a randomized complete block experimental design with three blocks (for details, see Arnoult et al., 2015a). Among these 21 genotypes, the following five genotypes were studied: two genotypes were identified as M. sinensis (ROT and SIL), and three genotypes were identified as M. × giganteus (FLO, GIB, and H8). Among the three M. × giganteus genotypes, H8 was considered an M. × giganteus genotype as it was a hybrid between M. sacchariflorus and M. sinensis. Using Amplified Fragment Length Polymorphism (AFLP) markers, the genotype named “FLO” belonged to the M. × giganteus species (Rambaud, personal communication). These five genotypes were chosen among the 21 previously studied genotypes (Arnoult et al., 2015a) due to (i) their relatively high biomass production per hectare and (ii) their contrasted biomass composition, particularly for cellulose and lignin contents.

These five genotypes were harvested from the experimental field at the end of the winter of years 4, 5, and 7 (2011, 2012, and 2014, respectively). A surface of 16 m² was cut 5 cm above the ground using a reed harvester.

Two sets of samples were collected for each of the 5 genotypes as follows: (i) a sample of 70 kg of biomass harvested in the whole 3 blocks in year 7 was used for the pilot-scale assay, and (ii) a sample of ~500 g of biomass randomly chosen from the biomass harvested on each of the 3 blocks of the trial in years 4, 5, and 7 was used for the high-throughput assays and cell wall components quantification. For the pilot-scale assay, (i) we harvested 7th year miscanthus in 2014, which corresponds to the first year when the pilot was operational, and (ii) we needed to pool the biomass from three blocks of the trial to provide enough biomass for the pilot assays (70 kg).

Using data from the high-throughput assays, we calculated the biomass composition from the mean of three technical replicates, while the saccharification results were obtained from the mean of four technical replicates for each genotype. All sample CVs were <10%. Samples with CVs >10% were removed.

RStudio (version 3.3.2) was also used to perform an analysis of variance (ANOVA) to assess the effect of the harvesting year and genotype on the biomass production, cell wall composition, and saccharification yields. We also used the Tukey–Kramer test to achieve multiple comparison tests.

For these analyses, two linear models were built;

Investigation of the relationships between miscanthus traits and released glucose after enzymatic hydrolysis of untreated and hot water-treated samples was conducted by focusing on the effects of genotype and harvesting year.

Correlation between enzymatic hydrolysis efficiency and cellulose, hemicellulosic carbohydrates, and lignin contents was significantly affected by the genotype. Indeed, the results detailed in Table 2 indicate that saccharification of the untreated sample was slightly affected by extractives content (r = 0.52) and Biomass yield (r = −0.47) in FLO-GIB group while the group H8-ROT-SIL is mainly negatively impacted by lignin content (r = −0.55), Lignin/Cellulose (r = −0.59), and LCI (r = −0.52) ratios. Glucose release from hot water-treated samples belonging to FLO and GIB genotypes was strongly impacted by hemicellulosic carbohydrates content (r = 0.96) and Hemicellulosic Carbohydrates/Cellulose ratio (r = 0.95). This group is also negatively impacted by extractives (r = −0.53), cellulose (r = −0.52), and lignin (r = −0.78) contents. However, glucose released from pretreated genotypes H8, ROT, and SIL was less impacted by the cell wall composition and biomass production (Table 2).

Correlation analysis, based on harvesting year, has also highlighted the differential impact of biomass traits on enzymatic hydrolysis of untreated and pretreated miscanthus genotypes. Those samples harvested at year 4 and hot water-treated were strongly affected by cellulose (r = −0.88) and hemicellulosic carbohydrates (r = 0.82) contents, followed by slight effect of lignin content (r = −0.74). In contrast, untreated samples were not impacted by cell wall composition. At year 5, a strong effect of lignin was highlighted (r = −0.93) followed by cellulose (r = −0.89) and hemicellulosic carbohydrates (r = 0.89), which similarly affected digestion of pretreated miscanthus. Furthermore, compared to the year 4 samples, the three major cell wall polymers, especially cellulose (r = −0.74), displayed higher effect on digestion of 5th-year untreated miscanthus. Finally, strong correlations were established between hemicellulosic carbohydrates (r = 0.81), lignin (r = −0.83), LCI (r = −0.85), and enzymatic hydrolysis of untreated miscanthus at year 7. Glucose release from pretreated samples was also strongly impacted by hemicellulosic carbohydrates (r = 0.94), lignin (r = −0.91), and Hemicellulosic carbohydrates/Cellulose (r = 0.94), Lignin/Cellulose (r = −0.86), and LCI (r = −0.90) ratios. Cellulose conversion after hot water treatment was also positively affected by cell wall extractives on year 4 (r = 0.74), year 5 (r = 0.84), and year 7 (r = 0.73) while untreated genotypes were only slightly affected in year 5.

The enzymatic conversion yields of cellulose of steam exploded 7th-year miscanthus at the pilot-scale indicated that ROT genotype is the most digestible (68%), in contrast to H8 and SIL genotypes that express the same cellulose conversion yields (63%), and the less hydrolyzed ones, which are the FLO and GIB genotypes that reach only 52% (Figure 5A).

The comparison of these cellulose conversion yields, to those obtained using our miniaturized assay, revealed a very strong and positive correlation (Figure 5B). Indeed, compared to FLO and GIB, a coefficient of correlation r = 0.98 was obtained, confirming the better digestibility of the group formed by H8, ROT, and SIL genotypes.

Exploring biomass production and composition of the five miscanthus genotypes revealed a large genetic diversity. Based on variance analysis, this biomass quality diversity was found for the following three major cell wall polymers: cellulose, hemicellulosic carbohydrates, and lignin. The cell wall composition reported here is in agreement with previously reported M. × giganteus and M. sinensis data using Van Soest method (Arnoult et al., 2015a). This method allows sequential fractionation of the main cell wall polymers. However, non-cellulosic polysaccharides might be underestimated as neutral detergent solution can solubilize some pectins. In addition the fraction designated as hemicellulosic carbohydrates would thus mostly correspond to heteroxylans with few amounts of xyloglucan, mannan, and pectins (Le Ngoc Huyen et al., 2010; Costa et al., 2016). Furthermore, lignin concentrations as ADL residue were substantially lower than values obtained for miscanthus using Klason method (Le Ngoc Huyen et al., 2010), as previously reported for other grass species (Hatfield et al., 1994; Bertrand et al., 2006). The acid detergent used in the Van Soest method can solubilize acid soluble lignin (Hatfield et al., 1994), giving a lignin-like fraction with a higher degree of recalcitrance as shown in forage digestibility studies (Van Soest, 1963).

Regarding the relationships between biomass production and composition, we highlighted that biomass production was positively correlated to cellulose and lignin, but negatively correlated to hemicellulosic carbohydrates, in agreement with the literature (Xu et al., 2012; Arnoult et al., 2015a). This suggests that miscanthus growth may be accompanied by higher contents in cellulose and lignin while other carbohydrates proportions decrease.

Comparison of released glucose from untreated and hot water-treated genotypes has also revealed that in both conditions, harvesting year and genotypes differentially affect cellulose digestion. Indeed, after pretreatment, harvesting year effect was limited to year 5 (Figure 4), while genotype impact became stronger. In contrast to pretreated genotypes, relationships analysis mainly indicated that cellulose conversion of untreated miscanthus was less impacted by cell wall traits quantified in this present study, suggesting an incomplete and inefficient enzymatic hydrolysis. In fact, biomass production negatively impacted cellulose conversion of hot water-treated samples, but the strength of this relationship was significantly impacted by the two genotype groups and three harvesting years. Similar findings have been recently reported by Domon et al. (2013) while exploring the impact of cold acclimation on miscanthus cell wall composition.

Variations in cellulose conversion yields, according to genotypes and harvesting years, is related to the cell wall and tissue architectures between the less recalcitrant ROT, H8, and SIL genotypes and the more recalcitrant FLO and GIB ones. Indeed, based on our results, the highest cellulose conversion yields displayed by ROT, H8, and SIL can be related to the highest Hemicellulosic carbohydrates/Cellulose ratio and lowest Lignin/Cellulose and LCI ratios, while more recalcitrant FLO and GIB displayed the lowest Hemicellulosic carbohydrates/Cellulose ratio and highest Lignin/Cellulose and LCI ratios. This lignocellulose index (LCI) is reported to be high when biomass is less digestible (Moorhead et al., 2013). Based on correlation analysis, extractives do not seem to explain observed differences on cellulose conversion between the three harvesting years. Hot water pretreatment applied to these genotypes may result in dissolving of hemicelluloses and slight increasing lignin content in the pretreated biomass as reported by Li H.-Q. et al. (2013).

To better understand results highlighted in this study, other cell wall traits must be addressed such as monosaccharide composition of hemicellulose fraction, glycome profile, and cellulose crystallinity index. Indeed, wet chemistry can allow us determining xylan chain substitution by arabinose and other components. For instance, a higher substitution degree of xylan chain by arabinose was reported to positively affect miscanthus saccharification and microbial decomposition (Li F. et al., 2013; Amin et al., 2014). Acetylation of xylan chains has also been reported to create steric hindrance for binding cellulolytic enzymes (Selig et al., 2009; Gille and Pauly, 2012). Phenolic acids, p-coumaric and ferulic acid are also known to impact saccharification potential (Belmokhtar et al., 2013). Moreover, the role of minor cell wall components such as xyloglucans, rhamnogalacturonans, and homogalacturonans have been shown to be differentially distributed among different harvests, organs, and genotypes (Costa et al., 2016).

Variations in leaf/stem ratio can also explain differences in cellulose conversion between more and less recalcitrant genotypes as leaves have been reported to be more reactive to cellulases conversion as compared to stems (Le Ngoc Huyen et al., 2010; Costa et al., 2016). From the study of Arnoult et al. (2015a), ROT, H8, and SIL, displaying the highest cellulose conversion yields, show higher leaf/stem ratios (varied from 0.11 to 0.33 for the harvest in February of years 4 and 5) than FLO and GIB, displaying the lowest cellulose conversion yields (leaf stem ratios varied from 0.01 to 0.08 for the harvest in February of years 4 and 5). Indeed, ROT, H8, and SIL genotypes displayed more leaves at the February harvest than FLO and GIB, for which, the major part of the leaves fell to the ground before the harvest. As leaves contain less lignin than stems (Arnoult et al., 2015a), the higher quantity of leaves at harvest can be explained, in that ROT, H8, and SIL genotypes were less recalcitrant and then gave higher cellulose conversion yields. Indeed, the released glucose after enzymatic hydrolysis of hot water-treated miscanthus genotypes was positively correlated to the leaf/stem ratio (r-values of 0.57 and 0.79, respectively for the years 4 and 5). The correlation was also positive for the untreated miscanthus genotypes, even if this correlation was lower than for hot water-treated genotypes (r-values of 0.03 and 0.66, respectively for the years 4 and 5). The differences in cell wall and tissue architectures, between more and less recalcitrant genotypes, confirmed how hemicelluloses' interaction with cellulose and reduced lignin content was important for achieving higher cellulose conversion yields.

Cellulose conversion inhibition by high lignin and cellulose contents has been reported to be due to unproductive fixation of cellulolytic enzymes on lignin, while hemicelluloses are reported to deposit into cell walls via crosslink to cellulose by hydrogen bonds; thus, reducing cellulose crystallization (Xu et al., 2012; De Souza et al., 2015; van der Weijde et al., 2016). Indeed, reduction in cellulose crystallinity may result in efficient cellulase access to cellulose substrate (Yoshida et al., 2008). Lignin is also assumed to interact with hemicelluloses rather than with cellulose, which can reduce interactions between hemicelluloses and cellulose, thus increasing cellulose crystallinity.

To conclude, the genetic diversity on biomass production and cell wall composition can be used in breeding programs to improve the saccharification potential of genotypes for the objective of bioethanol production.

We investigated, for the first time at the pilot scale, the performance of five miscanthus genotypes contrasted for their biomass composition belonging to the species M. × giganteus (FLO, GIB, and H8) and M. sinensis (ROT and SIL). Steam explosion and enzymatic hydrolysis were carried out at the pilot scale in close agreement to previously reported data of other feedstocks (Schell et al., 2003; Thomsen et al., 2006; Rocha et al., 2012). Our results highlighted significant differences between the explored genotypes exhibiting distinct recalcitrance. Compared to the results shown by Schell et al. on corn stover, these results seem to be much lower, probably due to the intrinsic properties of miscanthus (Schell et al., 2003).

Furthermore, we performed hot water pretreatment and enzymatic hydrolysis with saturating amounts of cellulolytic enzymes to screen more samples for their digestibility based on the method recently described by Selig et al. (2010). A high-throughput screening method was initially developed at NREL (Golden, CO, USA) where the pretreatment step used a strong pilot-scale boiler to produce steam heated to 180°C. Here, we adapted this pretreatment step to a laboratory scale using a small steam generator provided from CellKraft® (Sweden). This resulted in a little more time for reaching target temperature but was able to maintain it during 40 min. Results accuracy and reliability have been assessed using 5 normalization controls for inter-plate and inter-assay comparisons. It consists on running a standard biomass in 4 wells in addition to a biomass control well not containing cellulolytic enzymes. During the current study, CV corresponding for these normalization controls was lesser than 1.5% (Supplementary Image 2). The advantage of our high-throughput method therefore, relies in its laboratory scale and reliability.

We highlighted a strong correlation between the pilot and miniaturized assay for cellulose conversion, displaying different patterns between the digestible H8, ROT, and SIL group and the more recalcitrant FLO and GIB group. However, cellulose conversion yields were relatively high according to the miscanthus genotype. This difference between the pilot and miniaturized assays can be explained by the different experimental conditions between the pilot and miniaturized assay, such as pretreatment technology and enzyme loading. Indeed, the better digestibility observed at the pilot scale for the FLO and GIB genotypes may be explained by the presence of sulfuric acid and fiber explosion during the pretreatment carried out at the pilot-scale; while the hot water treatment used in the miniaturized assay seemed less efficient in these recalcitrant biomasses. In contrast, the more digestible genotypes ROT, H8, and SIL responded better to the hot water treatment used in the miniaturized assay. Furthermore, this original approach allowed us to highlight the same trend in the cellulose conversion potential of the five genotypes regardless of the harvesting year. Moreover, the inter-year comparison revealed distinct digestibility, which confirms the importance of our approach based on using sub-optimal pretreatments, while giving high cellulose conversion yields.

Therefore, this high-throughput miniaturized and automated method accurately mimics the performance at the pilot scale and will be reliable for assessing the performance of miscanthus tested in small field plot trials.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.