Kramer CS was provided by the National Renewable Energy Laboratory (NREL; Golden, CO, USA). The biomass had been pre-milled and passed through a 10-mm screen at NREL. AFEX pretreatment for CS (AFEX-CS) was carried out at 130 ± 10°C, 60% moisture loading, 1:1 NH₃ to biomass loading (w/w) for 15 min total residence time (Chundawat et al., 2010). The pretreated substrate particle size was reduced using a centrifugal mill with an 80-µm sieve attachment (Chundawat et al., 2008). The glucan, xylan, and arabinan composition of AFEX-CS, determined using standard NREL biomass compositional analysis method (Sluiter et al., 2010), was 33.5 ± 1.1, 23.4 ± 0.3, and 3.5 ± 0.2%, respectively.

The following standard CE preparations were provided by Genencor-Danisco Inc. (Genencor Division, Rochester, NY, USA): Spezyme CP (CE 1), Multifect Xylanase (CE 3), Multifect Pectinase (CE 4), and Accellerase 1000 (CE 6). Depol 670L (CE 5) was provided by Biocatalysts Inc. (Cardiff, UK). Novozyme 188 (CE 2) was procured from Sigma-Aldrich (St Louis, MO, USA). The following eight research-grade enzymes (RE) preparations were a kind gift from Novozyme (Davis, CA, USA): NS22002 (RE 1), NS50010 (RE 2), NS50012 (RE 3), NS50013 (RE 4), NS50014 (RE 5), NS50029 (RE 6), NS50030 (RE 7), and NS50053 (RE 8). Protein concentrations were determined using the Kjeldahl nitrogen analysis method (in triplicates by Dairy One Cooperative Inc., Ithaca, NY, USA). The non-protein equivalents (e.g., urea and ammonium salts estimated by AOAC Method 941.04) was subtracted from the Kjeldahl total crude protein equivalents (estimated by AOAC Method 2001.11) to determine true protein content (6.25×% Nitrogen) and are originally reported elsewhere (Chundawat et al., 2011d).

Detailed protocols for microplate-based enzymatic hydrolysis and monosaccharide analysis (i.e., glucose, xylose) are provided elsewhere (Chundawat et al., 2008, 2011b; Gao et al., 2010a). Briefly, the hydrolysis experiments were performed in 2.2-ml deep-well microplates at 0.2% (w/w) total glucan loading along with 50 mM citrate buffer (pH 5.0) in a total volume of 500 µl per well. The microplates were maintained at 50°C in a shaking incubator at 200 rpm for defined time periods. All experiments were carried out in quadruplicate and average values and SDs are reported here.

Minitab (Version 15.0, Minitab Inc., State College, PA, USA) and SPSS (IBM SPSS Statistics, Chicago, IL, USA) based statistical software were used to create suitable experimental designs and analyze the data, respectively. Our regression analysis suggested that quadratic and quaternary interaction terms based model regressions did not accurately predict the variance (low R² and high σ values) in the data as compared to special cubic interaction terms based regression models. The σ (sigma or S) and R² values were also closely analyzed for comparing models; all σ-values suggested that the first SD of residuals was relatively small (<5%). The σ-values for xylan conversion were slightly higher than those for glucan conversion. High R² and R² (adjusted) values (>80%) indicate that the models account for majority of the variation in the dataset and were not a result of over fitting. High R² (predicted) (>80%) values indicate that the models had significant predictive capabilities.

All 14 procured enzymes cocktails from commercial vendors were tested for their respective activity on AFEX-CS at 4 different protein loadings (60, 30, 15, and 7.5 mg/g glucan) at two time points (6 and 24 h; Figure 1). The enzyme source, protein concentration determination, and relative specific activity on various model substrates are provided in Supplementary Material (see Appendix SA-I-4, SA-I-5, and SA-I-6 in Supplementary Material). Enzyme activity assays for all 14 CE on several model substrates (Avicel, carboxymethylcellulose, oat spelled xylan, p-nitrophenyl xyloside, p-nitrophenyl arabinoside, p-nitrophenyl acetate, and p-nitrophenyl glycoside) has been already reported elsewhere (Chundawat et al., 2011d) and reproduced here for the sake of comparison.

The individual enzyme blends mixtures gave widely varying activities on AFEX-CS (i.e., glucose and xylose yields) that seemed to correlate with the absence of certain glycosyl hydrolase families identified in a recent proteomics study (Chundawat et al., 2011d). Based on the results of this proteomics study, the composition for all original blends or mixtures prepared using these blends can be crudely estimated (see Supplementary Material for details on individual glycosyl hydrolase or GH family abundance). Six of the original fourteen enzyme blends (CE 1, CE 5, CE 6, RE 1, RE 4, and RE 8) gave greater than 50% glucan conversion within the first 24 h for the highest protein loading tested (60 mg/g glucan). Protein composition data revealed that these enzyme blends had greater than 25% cellobiohydrolase enzyme content (total protein basis), thus justifying the high cellulase activity seen on AFEX-CS. Other enzyme blends that gave lower glucan conversion were either deficient in exo/endocellulases or β-glucosidases. RE 2, which is abundant in β-glucosidases but deficient in exo-cellulases, gave close to 5–10% glucan conversion suggesting that several gluco-oligomers released from the cell wall after AFEX pretreatment are relatively easily digested to glucose by β-glucosidases. Recent analysis of oligomers released from the cell wall after AFEX pretreatment have revealed the presence of a range of low- to high-molecular weight oligosaccharides with degree of polymerization ranging from 2 to 20 (Chundawat et al., 2010; Vismeh et al., 2013).

All enzyme blends except CE 5 and CE 6 which gave high glucan conversions were also found to result in greater than 50% xylan conversion within 24 h for the highest protein loading employed. Protein composition data revealed that the high xylose yielding enzyme blends (CE 1, CE 3, CE 4, RE 1, RE 4, and RE 8) were abundant in endo- and exo-xylanases (e.g., β-xylosidase). In addition, CE 3 and CE 4 were also found to give high xylose yields even at lower protein loadings unlike the cellulase-abundant blends. CE 3 was abundant in endoxylanases, while CE 4 was abundant in other accessory enzymes (e.g., α-arabinofuranosidases) as well. CE 5 and RE 6 gave the lowest xylose yields as these blends were deficient in both endoxylanases and β-xylosidases. Some cellulase blends (like CE 6) gave lower xylose yields possibly due to deficiency in accessory enzymes (e.g., α-arabinofuranosidases, α-glucuronidase) necessary for efficient hemicellulose deconstruction. Interestingly, α-arabinofuranosidases (GH 54 in particular) and α-glucuronidase were the only two major accessory enzymes that were deficient or absent in CE 6 versus CE 1. This observation could explain the differences in xylose yields seen for both enzyme blends and has been verified using purified enzyme-based assays on AFEX-CS (Gao et al., 2011b). CE 2 is a common CE blend that is used extensively to supplement cellulase cocktails for cellobiase activity. However, CE 2 is also abundant in amylases and other non-cellulolytic proteins (which are greater than 70% in total protein abundance) with minor amounts of β-glucosidase. Based on these results, it seems that it would be more beneficial to use CE 4 to supplement accessory enzymes (inclusive of β-glucosidase along with several other accessory glycosyl hydrolases) for pretreated biomass rather than using CE 2.

Pearson correlation coefficients (r) were determined for each CE type and the slopes of glucan/xylan conversions across varying protein loadings (slope was determined as net % conversion per milligram total protein loaded; data not shown). This analysis was carried out to determine if there were any low-abundance enzyme components within each blend that were critical to hydrolysis yields. There are a number of significant correlations, including those between RE 1 and xylan slope (r = 0.425, p < 0.05), between RE 8 and glucan slope (r = 0.424, p < 0.05), and CE 1 and both xylan (r = 0.549, p < 0.05) and glucan slopes (r = 0.390, p < 0.05). Pearson correlations were also tested between groups of different enzyme families based on predicted proteomics composition and the slopes of glucan/xylan conversions across varying protein loadings. The enzyme compositions within each CE/RE mixture were grouped as follows (see Appendix SA-I in Supplementary Material for details on enzyme/protein family name acronyms): (a) total cellobiohydrolases (CBH), (b) total endoglucanases (EG), (c) total β-glucosidases (BG), (d) total CBH + EG, (e) total CBH + EG + BG, (f) total endoxylanases (EX), (g) total EX and β-xylosidase (BX), (h) total EX + BX + EG, (i) total arabinofuranosidases (ABF), (j) total ABF + BX, and (k) total ABF + EX + BX and α-glucuronidases (αGLUC). There are a number of significant correlations including those between total CBH and both glucan (r = 0.914, p < 0.01) and xylan (r = 0.551, p < 0.01) slopes, between total EG and both glucan (r = 0.888, p < 0.01) and xylan (r = 0.485, p < 0.01) slopes, between total CBH + EG and both glucan (r = 0.926, p < 0.05) and xylan (r = 0.545, p < 0.05) slopes, and between total CBH + EG + BG and both glucan (r = 0.919, p < 0.05) and xylan (r = 0.578, p < 0.05) slopes. These results validate previous findings that glucan/xylan yields are highly correlated and enzymes that facilitate glucan hydrolysis (CBH, EG, and BG) also facilitate increased digestibility of xylans by hemicellulases or multifunctional glycosyl hydrolases like EG I (Gao et al., 2010a; Chundawat et al., 2011b).

In addition, we found that the glucan-to-glucose conversion was strongly correlated (R² = 0.92) to the total CBH + EG + BG content (Figure 2). However, the correlation between xylan-to-xylose conversion and total ABF + EX + BX + αGLUC content was relatively weaker (R² = 0.11). This suggests that increased relative loading of core cellulases (CBH, EG, and BG) gives concomitant increase in glucan-to-glucose conversion. On the contrary, increased supplementation of major hemicellulases (EX, ABF, BX, or αGLUC) alone does not necessarily increase xylan-to-xylose conversion. This is likely due to missing synergistic, complementary families of accessory enzymes that are necessary to completely hydrolyze hemicellulose. Therefore, to achieve higher specific activity on alkaline pretreated biomass like AFEX-CS, it is necessary to complement the core cellulase cocktail with additional hemicellulases/accessory enzymes. A similar strategy using purified synergistic glycosyl hydrolases has resulted in maximizing AFEX-CS digestibility in recent years by our group (Gao et al., 2010a,b, 2011b) and others (Banerjee et al., 2010a,b). Based on the proteomics compositional analysis data and activities of the individual enzyme blends on AFEX-CS, six-core cellulases (CE 1, CE 5, CE 6, RE 1, RE 4, and RE 8) were selected for further studies.

From the previous results and proteomics study, it was clear that none of the individual commercial blends had all the enzymes necessary for simultaneously maximizing both glucan and xylan conversions (Chundawat et al., 2011d). The selected core cellulases (CE 1, CE 5, CE 6, RE 1, RE 4, and RE 8) contain significant amounts of exo- and endocellulases but were found to lack several crucial accessory enzymes that are necessary to completely digest both glucan/xylan. To address this issue and identify a single best cellulase blend for further screening efforts, we compared the relative activity of all six core cellulases supplemented with equal amounts of the seven predominantly accessory enzyme blends (CE 2, CE 3, CE 4, RE 3, RE 5, RE 6, and RE 7) in an equi-mass ratio.

Our previous proteomics study has shown that the hemicellulases and accessory enzymes in our collection represent nearly all major families of glycosyl hydrolases necessary for hydrolysis of lignocellulose to monomeric sugars (Chundawat et al., 2011d). Therefore, addition of all accessory enzymes in excess would allow us to make a reasonably fair comparison of relative efficacy of each core cellulase cocktails that otherwise contain highly variable but limited amounts of hemicellulolytic enzymes. The total enzyme loading was fixed at 30 mg/g glucan loading with the % core cellulase loading relative to the total accessory enzyme loaded varied from 0 to 100% (0, 25, 50, 75, and 100% core cellulase). The results from these assays are shown in Figure 3. CE 6 and RE 8 stood out as the best performing single cellulase blends with respect to glucan conversion. However, with respect to xylan conversion, RE 8 was the best performing core cellulase blend. Addition of only the accessory enzymes (no cellulase blends added) gave low glucan (25%) but high xylan (80%) conversion. Supplementing 25% of the total protein loading with the accessory enzymes resulted in all six-core cellulases giving much higher glucan and xylan conversions. RE 4 benefited the most upon supplementation of the accessory enzyme blend. However, RE 8 that was found to be the best performing cellulase blend (based on both glucan and xylan conversions), when supplemented with various ratios of accessory enzymes (6 and 24 h). Our previous work has shown that xylo-oligosaccharides are extracted and deposited on outer cell wall surfaces after AFEX pretreatment, which makes this polysaccharide fraction more readily accessible by hemicellulases (Chundawat et al., 2011c). However, insoluble glucan polymers like cellulose are still embedded within the cell wall that limits its accessibility to enzymes. This can partly explain the slower rate of glucan hydrolysis relative to xylan hydrolysis (6 versus 24 h). Similar differences in relative rate of hydrolysis of cellulose versus hemicellulose have also been reported using purified, single-component enzymes acting on AFEX-treated CS in the past as well (Gao et al., 2010a, 2011b).

Based on the previous results, RE 8 was identified for further screening studies. The core cellulase mixture RE 8 was supplemented with varying combinations of a minimal set of the accessory enzymes to determine which combination gave the highest glucan and xylan conversions (see Appendix SA-I-1 in Supplementary Material). Seven accessory enzymes were added in unary, binary, ternary, quaternary, quinary, senary, and septenary combinations (equi-mass loadings at 15 mg/g glucan).

The 128 enzyme mixture groups were rank ordered by glucan and xylan conversion and the top 20% performing mixtures for both sets were closely examined. Of the total number of mixtures that gave the highest glucan conversion, 100% had RE 7, 76% had CE 2, and 71% had CE 4. The next highest was RE 3, which was present in only 52% of all top-performing mixtures, followed by CE 3, RE 5, and RE 6, which were in 48, 38, and 33% of the best mixtures, respectively. Of the samples with the highest xylan conversion, 100% had CE 4 and 94% had RE 7. The next highest was CE 2, which was present in only 56% of samples, followed by CE 3, RE 6, RE 3, and RE 5, which were in 44, 38, 31, and 31% of the samples, respectively. Through this analysis, CE 4 and RE 7 were seen to be the dominant accessory enzymes present in mixtures that gave the highest glucan and xylan conversions.

From the combinatorial screening experiments, CE 4, RE 7, and CE 3 gave the highest activity in combination with RE 8. The optimum mixture composition for these four enzymes to maximize glucan, xylan, or glucan + xylan conversion was determined using a design of experiments approach for three different protein loadings (see Appendix SA-I-2 and SA-I-3 in Supplementary Material). Interestingly, the optimal mixture composition that maximized both glucan and xylan conversion was strongly dependent on the total protein loading employed. At lower total protein loadings (7.5 versus 30 mg/g glucan), a greater fraction of RE 8 (core cellulase) cocktail was required. These results validate recent findings that have also suggested that the relative ratio of cellulases is more important at lower total protein loadings (Gao et al., 2010a).

The optimal enzyme combinations of RE 8, CE 4, RE 7, and CE 3 identified earlier were tested against non-optimal, original individual enzymes (Figures 4–6). The hydrolysis was performed for three enzyme loadings (7.5, 15, and 30 mg/g glucan loading). We found that irrespective of the enzyme loading employed, the optimal enzyme blends always gave 10–25% higher (absolute conversion basis) glucan or xylan conversions to monosaccharides compared to the original, individual blends. An interesting point to note was that the difference in glucan conversion for the optimal versus non-optimal blends increased at lower total protein loadings. This result is consistent with studies indicating that residual lignin within pretreated biomass can bind cellulases non-productively (Gao et al., 2011a, 2014; Rahikainen et al., 2011). Therefore, with an increased fraction of non-productively bound enzymes at lower protein loadings, the need for optimal enzyme mixtures becomes even more apparent. Optimizing CE blends on AFEX-CS resulted in up to fourfold or greater reduction in total protein loading necessary to achieve equivalent glucan-to-glucose and xylan-to-xylose yields. Similarly, analysis of the glucan hydrolysis reaction half times (Table 1) suggested that there was up to threefold improvement in specific activity for the optimal versus non-optimal enzyme mixtures.

Proteomics analysis has been carried out previously to estimate the absolute abundance of key cellulolytic, hemicellulolytic, and accessory enzymes present in each of the 14 CE cocktails (Chundawat et al., 2011d). Protein abundance within each cocktail was determined based on the total observed peptide spectral counts (see Appendix SA-I-6 in Supplementary Material). Carbohydrate-active enzymes (CAZy) have been classified into various CAZy families based on sequence and structural similarity (Henrissat et al., 1991). Catalytic domains for glycosyl hydrolases (GH) from the same family have similar three-dimensional fold and substrate specificity. Despite some limitations associated with this proteomics-based compositional analysis approach (e.g., poor correlation between GH family and specific activity, presence of multifunctional enzymes), the protein composition of each CE blend based on various GH families can be used to determine the minimal set of enzymes necessary for efficient hydrolysis of pretreated biomass in a semi-quantitative manner. In order to further characterize the relationship between enzyme composition and hydrolytic performance, we designed a series of high-throughput assays using more than 750 unique enzyme combinations of the original 14 enzyme set (see Appendix SA-II in Supplementary Material for details). The diverse enzyme mixtures set were based on a simplex-centroid experimental design approach (Scheffé, 1963). In addition to creating a diverse collection of unique enzyme mixtures, this approach also allows calculation of a multidimensional polynomial-based mixture regression equation, with each regression term representing the relative effect of single enzymes, binary enzyme synergies (the product of the concentration of two enzyme cocktails and a regression coefficient), ternary enzyme synergies, and so forth. Depending on the order of the regression equation, the magnitude of the regression coefficient indicates the relative importance of any particular enzyme cocktail or enzyme synergies (see Eq. 1 below).

Y=∑1≤i≤qβixi+∑1≤i<j≤qβijxixj+∑1≤i<j<k≤qβijkxixjxk+⋯, where Y is the response variable (e.g., % glucan-to-glucose conversion); q is the number of enzymes; x_i is the proportion of the ith enzyme in the mixture; and β_ij…m is the coefficient of synergy for m enzymes. The magnitude of each β_ij…m indicates the strength of a particular enzyme or synergistic interaction.

Three simplex-centroid experiments were designed (Appendix SA-II-1, SA-II-6 and SA-II-9 in Supplementary Material), as reported elsewhere (Chundawat and Dale, 2009), and described here briefly: (1) 8 CE-RE Experiment: adding 8 mg total protein/g glucan of eight accessory enzymes—CE2, CE3, CE4, RE1, RE3, RE5, RE6, and RE7—to 8.5 mg of CE6/g glucan of AFEX-CS; (2) 10 CE-RE Experiment: adding 16.5 mg protein/g glucan of ten enzymes—CE1, CE3, CE4, CE5, CE6, RE1, RE2, RE4, RE7, and RE8; and (3) 5 CE-RE Experiment: adding 16.5 mg protein/g glucan of one major cellulase (either one of CE1, CE6, or RE4), CE4, CE5, RE1, and RE7. The assays were conducted for 24 h and end products were analyzed to estimate glucan/xylan conversions to their respective monosaccharides. The total sugar yield spanned from 20 to 540 g/kg biomass, while nearly all individual CE blends gave between 330 and 380 g/kg biomass (Figure 7). Third-degree mixed stepwise (α-to-enter/α-to-leave: 0.15) regressions were performed on the results (% glucan conversion, % xylan conversion, total glucose and xylose released as gram sugar per kilogram dry weight biomass) of each experimental dataset to generate models (see Appendix SA-II-2, SA-II-6 in Supplementary Material) predicting response as function of enzyme mixture composition, with the stepwise regression algorithm preserving statistically significant interaction effects (Box et al., 1978).

The validity of the regression models obtained from stepwise special cubic regressions of all mixture experiments was confirmed by analyzing residual histograms, normal distribution curves of residuals, residuals versus fits, and residuals versus order plots (data not shown). In summary, the estimated regression coefficients suggested the relative importance of individual enzyme cocktails or their synergistic mixtures to enhancing glucan and/or xylan conversion to monomeric sugars. These results further validated selection of the RE8, CE3, CE4, and RE7 for detailed optimization carried out previously in this study.

CE3, CE4, and RE7 are abundant in endo/exo-xylanases, arabinofuranosidases (GH 43, 51, 54, and 62) and α-glucuronidases that are either missing or are present in very low abundance in RE8. However, to obtain a more quantitative estimate of the relative abundance of the best performing enzyme mixtures (from the set of >750 mixtures tested) versus the original single cellulolytic blends (namely, CE1, CE6, and RE8) the protein composition of each enzyme mixture cocktail tested was determined using the known composition of the original 14 enzymes (based on the 56 individual components listed Appendix SA-I in Supplementary Material). All cellulolytic, hemicellulolytic, other accessory and pectin degrading enzymes were included in this list. All non-hydrolytic proteins (e.g., proteases) were clubbed together in the “OTHERS” category. The total number of tested mixtures (755 unique combinations) of the 14 original enzymes (total protein loading in all cases = 16.5 mg enzyme/g glucan), resulted in a 755 × 56 large matrix input parameter dataset.

Bivariate correlations were carried out for each of the 56 input variables with respect to the three responses (% glucan/xylan conversion, total sugar yield) to determine Pearson correlation coefficients (Appendix SA-II-7 in Supplementary Material) and their test of significance (p < 0.05). One of the major concerns with bivariate correlations between response and input variables is the assumption of linear dependence. However, the correlation between hydrolytic performance and amount of enzyme loaded is most likely to be non-linear, further confounded by the synergistic action of enzymes. Nevertheless, the bivariate Pearson correlation coefficients bring several interesting facts to light. First, there seems to be a very significant “opportunity cost” associated with individual enzymes that maximize glucan versus xylan conversion. That is, individual enzymes that maximize only xylan conversion have a negative Pearson coefficient (upon increasing the amount added) as compared to enzymes that maximize both glucan and xylan conversion.

Therefore, it would be also useful to examine the Pearson coefficients for total sugar yields that would be maximized by enhancing both glucan and xylan conversions. Individual enzymes/proteins that give a linear relationship with respect to total sugar yield, based on a positive and significant (p < 0.01) Pearson coefficient, include acetyl xylan esterases (AXE 1 and AXE 2), glucuronoyl esterase (CIP 2), swollenin (SWO), β-glucosidase (BG 3), β-xylosidase (BX 3), endoglucanase (EG 1, EG 2, EG 3, and EG 7), β-mannanase (BM), β-galactosidase (BGAL 2), cellobiohydrolase (CBH 2), amylase (AMY), endo-chitinase (CHI 1), and polygalacturonases (PGL). However, due to significant multicollinearity between certain protein components in the original 14 enzymes (Appendix SA-II-8 in Supplementary Material), the truly significant and necessary enzymes are partly confounded in this analysis.

Nevertheless, it was informative to plot the average protein composition of the top-performing single commercial cellulase cocktails (CE1, CE6, and RE8) versus the top 1% performing enzyme mixtures identified from within the 755 mixtures (based on total sugar yields). Figure 7 shows that the enzyme families with a significant difference (based on two-tailed Student’s t-test, p < 0.01) in composition between the two data sets were acetyl xylan esterases (AXE 1), glucuronoyl esterase (CIP 2), endoglucanase (EG 1), cellobiohydrolases (CBH 1 and CBH 2), pectinases (ARH, PME, PL, AG, ABGAL, AA, PGL), mannosidases (AMANO 1–2 and BMANO), and GH family 16/17. These results highlight the relative importance of hemicellulases and possibly other accessory enzymes that are enriched in the top 1% of the mixtures compared to the original single CE blend that gave improved activity on AFEX-CS.

Two of the primary limitations of optimizing crude enzyme blends using microplate-based assays are as follows: (1) substrate particle size issue: in order to pipette biomass slurries into the microplate, the biomass has to be milled into a fine powder (<100 μm) that is not necessarily representative of the substrate that would be used in a commercial biorefinery. We have found that finely milled biomass qualitatively requires a similar enzyme cocktail to maximize glucan-to-glucose and xylan-to-xylose conversions compared to unground material (Figure 8). Nevertheless, there were minor differences found in the relative performance of the various enzyme mixtures (Table 2) tested on finely milled versus unground biomass. These differences were more prominent on xylan versus glucan hydrolysis. (2) Multicollinearity issue: though the regression models used here do not have direct biological significance, the model interaction terms are indicative of the relative degree of synergy between 2 and more enzyme cocktails or purified enzyme components. For simple individual enzyme mixtures (composed of 3–10 components) such models can more clearly provide a quantitative degree of synergy factor between individual enzyme components as also shown by our group and others (Banerjee et al., 2010a; Gao et al., 2010a). One of the major issues encountered, when attempting to delineate the relative contribution of individual enzyme families on hydrolysis yields for more complex enzyme cocktails (>50 components) explored in this study, was the high multicollinearity between relative enzyme abundance. Multicollinearity is a statistical phenomenon in which two or more variables in a multidimensional model have similar effects. This will influence the extrapolation of the results obtained from the model (as also discussed in the previous section). Multicollinearity is indicated by high variance inflation factor (VIF) values (see Appendix SA-II-2, SA-II-6 in Supplementary Material). VIF values greater than 5 were observed for a number of the coefficients in the 10 enzyme and 8 enzyme experiments, indicating the presence of high degree of multi-collinearity. Multicollinearity in the model may have been due to the fact that several commercial and research enzymes target similar substrates, but with differing effectiveness. To improve the predictive abilities of the model, another set of experiments was designed by using a more diverse set of the five most effective enzymes determined from the previous two experiments. The enzymes were selected to minimize the overlaps in their substrate specificity. This issue can be addressed by studying high specificity, purified enzymes to allow development of more precise models, and potentially more active enzyme blends hitherto not being able to take advantage of multifunctional enzymes, which are relevant to reducing the amount and type of enzymes needed to deconstruct lignocellulosic biomass.