Resources, Conservation and Recycling 122 (2017) 307–318

Contents lists available at ScienceDirect

Resources, Conservation and Recycling

journal homepage: www.elsevier.com/locate/resconrec

Full length article

Multivariate data analysis applied in alkali-based pretreatment of

corn stover
Bin Li b , Li Ding a , Huanfei Xu a,b,∗ , Xindong Mu b , Haisong Wang b
a
College of Chemical Engineering, Qingdao University of Science and Technology, Qingdao, Shandong, 266042, People’s Republic of China
b
CAS Key Laboratory of Bio-Based Materials, Qingdao Institute of Bioenergy and Bioprocess Technology, Chinese Academy of Sciences, Qingdao, Shandong,
266101, People’s Republic of China

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, variables of Pulp Refining Instrument (PFI) refining assisted alkaline pretreatment and
Received 25 January 2016 hydrolysis saccharification of corn stover were analyzed. The process parameters were characterized
Received in revised form by multivariate data analysis methods including Principle Component Analysis and Partial Least Square
16 December 2016
(PCA and PLS) to investigate the specific relationships of primary variables in alkali-based pretreatment
Accepted 16 December 2016
of biomass. In this paper, pretreatment system was multivariate and the variables were inter-related to
each other. Total alkaline charge and removal rate of lignin had greatest impact on the pretreatment and
Keywords:
enzymatic hydrolysis.
Multivariate data analysis
Lignocellulosic biomass © 2016 Published by Elsevier B.V.
Alkali-based pretreatment
Enzymatic saccharification

1. Introduction Thus, the pretreatment was considered as the key process to break
the natural barriers of lignocellulosic biomass (Segal et al., 1959;
In view of the shortage of fossil fuels and the worsening environ- Luo and Zhu, 2011). High efficient pretreatment could reduce the
mental problems all over the world, more researchers have focused subsequent operating cost of the bioenergy conversion. There were
on developing of sustainable bio-energy alternatives (Bichraoui- many different biomass pretreatment categories (Oh et al., 2005),
Draper et al., 2015; Ayamga et al., 2015). Bio-energy was one of such as physical pretreatment, biological pretreatment, chemical
the sustainable energy alternatives (Ding et al., 2012; Yang et al., pretreatment, and physicochemical pretreatment (Liu et al., 2013).
2013; Von Schenck et al., 2013; Sun and Cheng, 2002). During the Therein, the physicochemical pretreatment was considered as one
bio-energy conversion process, enzymatic saccharification is one of the high efficiency pretreatment to improve the enzymatic sac-
of the key steps to produce fermentable sugars, which can be fur- charification (Agbor et al., 2011). Enzymatic saccharification was
ther converted to bio-fuels (such as bio-ethanol) (Yu et al., 2013). normally produced by enzymatic hydrolysis process, in which
The lignocellulosic biomass was considered as the most promising polysaccharide could be decomposed into fermentable sugar by
feedstock of the second-generation bio-ethanol (Ljungqvist et al., enzyme (e.g. cellulose, glucosidase).
2005). Main components of lignocellulosic biomass were polysac- Many different process variables were in the physicochemical
charide (cellulose and hemicellulose) and lignin (Zhu et al., 2009; pretreatment and enzymatic hydrolysis (Miura et al., 2012). More
Khoo et al., 2015). As polysaccharide, cellulose and hemicellulose or less, these variables were associated with each other. In consid-
could be degraded to monosaccharide (fermentable sugars), which eration of the relationship between the different variables (Li et al.,
could be converted into bio-ethanol (Himmel et al., 2007; Luo et al., 2011), studies of the pretreatment conditions, mechanical treat-
2011). ment, composition of pretreated corn stover, enzymatic hydrolysis
However, the recalcitrance of lignocellulosic biomass, such as effect could have an important guiding significance.
distribution and structure of lignin, was the bottleneck for the When multiple variables were involved, the data could be ana-
lignocelluloses energy conversion (Van Dyk and Pletschke, 2012). lyzed by a comprehensive dimension reduction method, which
could summarize the main aspects or information of a dataset (Lin
et al., 2015). The multivariate data analysis (PCA and PLS) were
especially useful for reducing high-dimension multivariate systems
∗ Corresponding author at: No.53 Zhengzhou Road, Qingdao, Shandong, 266042,
(Godoy et al., 2014). On account of multiple linear regressions, the
People’s Republic of China.
orthogonal transformation could be used by PCA method to trans-
E-mail address: [email protected] (H. Xu).

http://dx.doi.org/10.1016/j.resconrec.2016.12.007
0921-3449/© 2016 Published by Elsevier B.V.
308 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318

form the possibly correlated variables base set into a new base set For 1 g dry pretreated corn stover, the dosage of ␤-glucosidase was
of linearly uncorrelated variables (principal components). By PCA 5 IU/g and the celulase was 20 FPU. Then, the supernatant liquid
analysis, the variables could keep independent, and the data noise after enzymatic hydrolysis was collected and placed in a sealed bag
could be removed to help find the key variables for subsequent at −10 ◦ C for following experiment.
analysis. As a statistical method, PLS could reveal the relation-
ship of principal components regression. PLS method could not just 2.5. Analysis methods
give the hyperplanes of minimum variance (response vs indepen-
dent variables). PCA and PLS could simplify the complex process or National Renewable Energy Laboratory (NREL) protocol (Sluiter
issues, and reveal the essence of the variables (Li et al., 2014). et al., 2008) was used to test the chemical composition of corn
In this paper, the lignocellulosic biomass corn stover was treated stover. A high performance liquid chromatography (HPLC)(Model
by physicochemical pretreatment (PFI refining assisted alkaline 1200, Agilent, USA) with a Bio-Rad Aminex HPX-87H column
pretreatment) and enzymatic hydrolysis process. Variables of the (300 × 7.8 mm) was used to determine the enzymatic hydrolyzate
pretreatment and enzymatic hydrolysis were analyzed by PCA and at 55 ◦ C with 0.005 M sulfuric acid.
PLS method to find the inter-relationships between each other in The evaluation equations were as follows:
order to guide the further research of lignocellulosic biomass con-
Rsolid (%) = (weight of pretreated sample(g)/
version.
untreated sample(g)) × 100% (1)
2. Materials and methods

2.1. Raw material and chemicals Rglucan/xylan (%) = (Rsolid × Dglucan/xylan of pretreated corn stover )/

Corn stover (lignocellulosic biomass raw material) was har- Dglucan/xylan of raw corn stover × 100% (2)
vested in the October in 2014, from Pingdu, China. Firstly, corn
stover was milled by a grinder. Then, the milled samples were
screended to collect the corn stover with the target particle size Rdelignification (%) = 1 − (Rsolid × Dlignin of pretreated corn stover )/

(0.18–0.85 mm). After that, the samples were stored in a sealed Dlignin of raw corn stover × 100% (3)
bag at temperature for following research. (The moisture content
and composition analysis were shown in supporting information
Table.1). E glucan (%) = (Aglucose in hydrolyzate × 0.9/
Sodium hydroxide was bought from Sinopharm Chemical
Reagent Co.Ltd. Cellulase (Celluclast 1.5L) and ␤-glucosidase Aglucan in pretreated corn stover ) × 100% (4)
(Novozyme 188) were supplied by the Sigma-Aldrich China Inc. All
enzymes and chemicals were used without further purification.
E xylan (%) = (Axylose in hydrolyzate × 0.88/
2.2. Alkaline pretreatment of corn stover Axylan in pretreated corn stover ) × 100% (5)

The alkaline pretreatment was conducted in a cooking reac-

tor (Mode PL1-00, Xianyang TEST Equipment Co., Ltd., Xianyang, Rsolid was the recovery rate of solid, Rglucan/xylan was the recovery
China). Four small cooking seal pots (1 L) were in the cooking reac- rate of glucan or xylan, Rdelignification was the removal rate of lignin,
tor. For every small cooking pot, 50 g oven dried raw biomass was Eglucan was the enzymatic hydrolysis percentage of glucan, Exylan
added with alkaline chemicals at certain ratio of liquid to solid and was the enzymatic hydrolysis percentage of xylan. A was the weight
certain temperature for certain holding time (The pretreatment of sugar (g). D was the component content (wt.%).
conditions could be seen in supporting information Table.2). After
alkaline pretreatment, the cooking seal pots were cooled down 2.6. PCA and PLS analysis
quickly. Then, the pretreated corn stover samples was treated to
be neutral by washing with water in a 300 mesh bag. Finally, the The data analysis for PCA model and PLS model was calculated
pretreated corn stover samples were stored at 4 ◦ C for further study. by the SIMCA-P 12.0 (Soft Independent Modelling of Class Analogy,
Umetrics).
2.3. Mechanical refining treatment
3. Result and discussion
The corn stover samples with alkali-based pretreatment were
experienced the mechanical refining treatment (PFI) by a refiner 3.1. Principal component analysis of the whole process
(model PL11-00, Xianyang TEST Equipment Co., Ltd., China). The
revolution numbers (corn stover samples treated by PFI between In this paper, the multivariate data analysis included PCA and
the fixed disc and rotating disc) of the PFI treatment in this work PLS methods. The PCA was known as analysis for principal com-
were 4000. The oven dried weight of corn stover samples was 10 g. ponent, which was designed to reduce the dimension in order to
The corn stover solid content and water content of the pretreated reduce many indicators into a few composite ones. To comprehen-
samples in PFI treatment were 10% and 90%, respectively. The speed sively and systematically analyze the problems, the relationship
of the PFI mechanical refiner was 1400 rpm (0.24 mm refining gap). between multiple different factors should be considered. Com-
monly, these factors or indicators in multivariate statistical analysis
2.4. Enzymatic hydrolysis were known as variable. Since each variable reflected some infor-
mation of the research issue in some extent, and each variable
In the enzymatic hydrolysis process, the alkali-based pretreated had a certain correlation with each other, the statistics informa-
biomass samples were conducted in a 25 mL bottle with sodium tion reflected by all the variables was partly overlapped. In the
citrate buffer and 0.02% sodium azide (0.05 M, pH 4.8), at 50 ◦ C for multivariate analysis study, too many variables would increase the
48 h. The process was produced in a water bath shaker (90 rpm). amount of calculation and the analysis complexity. As a result,
 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318 309

Table 1 model showed four main variables. The value of R square was 0.795;
Classification of variables.
the value of Q square was 0.644. The value of R square could rep-
category of variables Name ofvariable Value range of variable resent the proportion of the variable for all the original variables,
The parameters of Alkaline (%) 5–13% and the closer value of R square was to 1, the better analysis results
pretreatment S/L ratio 4–8 the PCA model could get. The value of Q square could represent the
conditions T(◦ C) 120–160 predictive ability of the PCA model, which was calculated by the
H time (min) 10–50 min cross-validation method. When the value of the square of Q was
PFI 0 with 1 without
greater than 0.5, the PCA model could get a good predictive ability.
The compositions of Glucan (%) 40.1–55.5
pretreated corn stover Xylan (%) 22.5–27.4 As a result, the original experimental data generated a relatively
Extractive (%) 3.41–9.1 credible PCA model.
Lignin (%) 2.3–12.4 As can be seen in the scatter plot figure (Fig. 1b), the relation-
Evaluation of Solid Yield (%) 44.6–68.2
ship between the variables was reflected by the first two main
pretreatment D-lignin (%) 22.1–77.9
R-glucan (%) 53.2–92.5 component of PCA model (P1 and P2), which explained about 60%
R-xylan (%) 52.0−80.7 information of dataset. The farther away the variable point was
The evaluation of E-glucan (%) 51.1–85.4 from the origin point of coordinates the greater impact could the
enzymatic effect E-xylan (%) 34.5–75.1 variables provide on the whole process. In the same quadrant,
Y-glucan (%) 54.3–85.1
the variables which were adjacent to each other were positively
Y-xylan (%) 36.7–58.6
Y-total sugar (%) 47.9–74.4 correlated; meanwhile, the variables in different quadrant were
negatively correlated. As shown in Fig. 1, the total sugar yield,
glucan yield, xylan yield, the hydrolysis efficiency of glucan, the
in the quantitative analysis, the PCA method would reduce the hydrolysis efficiency of xylan, the total amount of alkaline, the
number of variables, and at the same time get more valuable infor- removal of lignin were close to each other in the same quadrant,
mation. In the PCA method, many models were established to help which showed that there was a positive correlation between these
the researchers understand the relationship between various vari- variables. A clear negative correlation example was the lignin con-
ables. The models would reveal the similarities and dissimilarities tent and the delignification rate. The two variables were far away
between the data points and recognize the abnormal value in the from each other in different quadrant. Obviously, when the D-lignin
issue (Paranhos and Petter, 2013). value was higher, the lignin content in the pretreated corn stover
The PLS method was a mathematical optimization technol- was lower. A large amount of lignin was removed during the pre-
ogy, which could find the best matching function by minimizing treatment, as a result, after alkaline pretreatment; the amount of
the square error. The systematic PLS model could be established lignin ingredient of corn stover samples was relatively small.
between the independent variables and the dependent variable, Seen from the P1 perspective, during the alkaline pretreatment
which could connect the main component in the principal compo- process, the total amount of alkaline had a very important influence
nent analysis with the meaningful data points. The PLS model could on the delignification results. The actual experimental results could
reveal the relationship between the independent variables and the also show that the greater the total alkaline amount, the greater
dependent variables in the multivariate analysis. the removal rate of lignin would get. The lignin was removed thor-
In this paper, the corn stover was pretreated by sodium hydrox- oughly and the value of lignin content of the pretreated samples
ide based alkaline pretreatment. The 18 key process parameters would be low.
in the pretreatment and enzymatic saccharification process were Seen from the P2 perspective, during the alkaline pretreatment
analyzed by PCA and PLS methods. The 18 key parameters were process, the relationship of pretreatment temperature and the
as follows: the total alkaline amount calculated based on the corn holding time was positive correlation. To some extent, temperature
stover (Alkaline%), the ratio of solid to liquid (S/L ratio), the maxi- and time could make up for each other. The conditions of shorter
mum holding time of the pretreatment temperature (H time), with holding time at relatively higher temperature and longer holding
or without PFI treatment (PFI), the glucan content of pretreated time at relatively lower temperature could get similar pretreatment
corn stover (Glucan%), the xylan content of pretreated corn stover effect. Pretreatment temperature, holding time and alkaline charge
(Xylan%), the extractive content of corn stover (Extractive%), the had positively correlated with lignin removal rate, but had nega-
lignin content of corn sotver (Lignin%), the solid yield after pretreat- tive correlation with lignin removal rate, recovery rate of glucan
ment (Solid Yield%), the desorption of lignin after pretreatment and recovery rate of xylan. The main reason was that higher tem-
(D-lignin%), the recovery rate of glucan (R-glucan%), the recov- perature, longer holding time and higher total alkaline charge could
ery rate of xylan (R-xylan%), the enzymatic hydrolysis efficiency cause the degradation of carbohydrates, breaking and removing of
of glucan (E-glucan), the enzymatic hydrolysis efficiency of xylan the macromolecular lignin chains. Thus, these pretreatment con-
(E-xylan%), the final total yield of glucan (Y-glucan%), the final total ditions could result in the lower value of total sugar yield, glucan
yield of xylan (Y-xylan%), the final total sugar yield (Y-total sugar%). yield and xylan yield.
These 18 parameters were divided into four categories, as shown in In the scatter plot figure of P3 and P4 (Fig. 1c), about 19% infor-
Table 1. All the data in this pretreatment research were imported mation could be explained by the third components and the fourth
into the SIMCA-P software. components. Seen from the P3 and P4 perspective, holding time
After the original data was analyzed by SIMCA-P software, the and temperature had negative correlation. The yield of Xylan had
PCA models were established, as shown in Fig. 1. positive correlation with xylan content, but had negatively corre-
The first step of multivariate analysis was principal compo- lated with temperature. The total sugar yield, glucan yield, xylan
nent analysis, which could obtain an overall graph based on the yield, glucan content and the recovery of glucan were close to each
entire data set. As mentioned above, PCA method could reduce the other in the same quadrant, which showed that there was a positive
number of the original variables for main component. The first prin- correlation between these variables.
cipal component (P1) was in the direction of the biggest change of The cumulated R2 and Q2 values for each variable of PCA could
the data set, and the second principal component (P2) was in the be seen in Fig. 1d. Most of variables have high green bars, R2, and
orthogonal direction of P1. In the successive principal component, high blue bars, Q2. Here, R2 indicates how well the variation of a
the amount of the variable of each principal component was less variable is explained, while Q2 indicates how well a variable can
than the amount of the previous one. As shown in Fig. 1.a, the PCA be predicted. Thus, variables were well modeled in PCA. Based on
310 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318

Fig. 1. PCA analysis.

(a: Score scatter plot; b: Loading scatter plot for p1 and p2; c: Loading scatter plot for p3 and p4; d: overview graphics (for each variable) of the values of cumulated R2 and
Q2.)
Note: The overall graph of PCA analysis based on the entire dataset. a: the scores t1, t2, t3 and t4, were summarizing the varibales. The scores were orthogonal, and they were
independent of each other, completely. The first component was showed by t1. t1 could showed the largest variation data information of X followed by t2, t3, t4. b: p1(for
first component) and p2 (for second component) showed the weights combining the original variables. c: p3(for third component) and p4 (for fourth component) showed
the weights combining the original variables. d: The fraction of the variation explained by this PCA model was presented by R2. This model could predict some fraction of
variations and it could be valued by Q2.
 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318 311

the PCA analysis, 4 PLS models were built (discussed in the follow- In addition of total alkaline charge, the second, third, fourth
ing) and the coefficient values have been calculated to illustrate the importance indexs (VIP) were temperature of pretreatment, pre-
relationship between different parameters. The coefficient values treatment time (holding time at the highest temperature), and ratio
of different PLS model could be seen in Supporting Information Fig. of liquid to solid, respectively. As discussed above, PFI had no con-
1. tribution to variables in this model, and its VIP was zero.

3.2. PLS model 1 analysis for the whole process

3.3. The PLS model 2 analysis for the whole process
As shown in Fig. 2, 13 variables of the pretreatment parame-
ters were picked from the 18 variables for the PLS model 1 analysis, As shown in Fig. 3, in the PLS model 2 analysis, the pretreat-
which including 5 independent variables (X) and 8 dependent vari- ment condition parameters were set as X, the chemical component,
able (Y). The Alkaline, S/L ratio, T, H time and PFI was set as X, pretreatment effect, and the hydrolysis effect were set as Y. There
and the chemical composition and pretreatment effect parameters were 18 variables in the PLS model 2, including 5 variables for X,
(Glucan, Xylan, Extractive, Lignin, Solid Yield, D-lignin, R-glucan, and 13 variables for Y. Alkaline, S/L ratio, T, H time, PFI were set as
and R-xylan) were set as Y. The PLS model 1 showed that for the X, Glucan, Xylan, Extractive, Lignin, Solid Yield, D-lignin, R-glucan,
dependent variable X parameters, R2X (cum) was 0.800, the R2Y R-xylan, E-glucan, E-xylan, Y-glucan, Y-xylan, Y-total sugar were
(cum) was 0.655 for the independent variable Y. The square of Q set as Y. In this PLS model 2, the R2X(cum) was 0.401, R2Y(cum)
was 0.517. In other words, in the PLS model 1, the two main com- was 0.536, Q2 was 0.436.
ponent could explain the 65.5% information of the pretreatment Similar with PLS model 1, PLS model 2 showed the relatively
process based on the 13 variables. The predictive ability of the PLS similar results. In the diagram layout, the holding time and pre-
model 1 was 52%. The results showed that the PLS model 1 based treatment temperature were close to each other, and away from the
on the 13 variables had a relatively good assessing capability. other variables point in other quadrants. The main reason was that
As can be seen in the figure, the PFI variables were at the origin the holding time and pretreatment temperature had negative cor-
of the coordinate graphs, indicating that there was no relationship relation with the pretreated corn stover ingredients, pretreatment
between the PFI variable and other variables in this PLS model. The efficiency and enzymatic efficiency. To some extent, the longer
main reason was that PFI was a kind of mechanical refining treat- holding time and higher pretreatment temperature could bring
ment after the alkaline pretreatment of the corn stover, and there negative impact on the final enzymatic effect. E-glucan, E-xylan,
was no chemical reaction or chemical charge during the PFI treat- Y-glucan, Y-xylan, Y-total sugar, Xylan, Glucan had a strong posi-
ment process. Generally speaking, after PFI mechanical refining tive correlation between each other, so they were very close to each
treatment, the significant difference of corn stover samples reflexed other in the PLS model 2 graph. When the glucan and xylan content
on the size, shape, and surface appearance. In the chemical com- were higher in the pretreated corn stover samples, the enzymatic
position, such as lignin content, extractive content, glucan content, saccharification process could get better effect, and the hydroly-
xylan content, and in the pretreatment effect, such as recovery rate sis efficiency of glucan and xylan could be improved, which also
of glucan, xylan, and removal rate of lignin, the corn stover samples improved the final total sugar yield.
with and without PFI treatment provided the same values. As can be seen in the PLS model 2 figure, the variables such as
As seen in the figure, the positive correlation and negative cor- Solid yield, R-xylan, R-glucan, Lignin, Extractive were close to each
relation between the variables showed in the both vertical and other, which reflected the close relationship between these vari-
horizontal coordinate axis were the same with the correlations in ables. This result could be explained by the relationship between
the PCA model analysis. Holding time and the pretreatment tem- the component of pretreated corn stover samples and the solid
perature were closely related to each other. Alkaline charge and recovery rate. The solid recovery rate was the ratio of weight of
removal rate of lignin showed positive correlations. Solid yield, pretreated samples to the weight of raw corn stover samples. Obvi-
recovery rate of glucan, recovery rate of xylan, extractive negatively ously, the weight of pretreated samples was decided by the content
correlated with alkaline charge. In view of the distance between of each ingredient of the corn stover after alkaline pretreatment.
the variable and the origin point, the impact of L/S ratio on the As a result, when the content of lignin, extractive, glucan, xylan
whole pretreatment process was smaller than the other variables. was relatively high, the recovery rate of glucan and xylan could get
As could be seen in the VIP graph of X (the importance of inde- high value, and the total weight of pretreated samples would be
pendent variables), during the pretreatment process, the greatest high, which could led to the higher solid recovery rate of the whole
impact was from the variable of alkaline charge, which provided process.
more important influence than any other process variables. Compared with PLS model 1, one of the main differences was
With the increase of the total alkaline charge, the removal rate that PFI was not in the origin of PLS model 2. In this model, the
of lignin improved significantly. For example, when the liquid to change of PFI variable could influence the other variables. As can
solid ratio was 6:1, temperature was 130 ◦ C, the removal rate of be seen in the two main components of the graph, PFI had good pos-
lignin was 21.1%, 33.72%, 51.7%, 73.9% and 77.54% respectively, and itive correlation with E-glucan, E-xylan, Y-glucan, Y-xylan, Y-total
the solid recovery rate was 68.16%, 59.82%, 59.87%, 50.61%, 51.52%, sugar. In the VIP analysis graph, the PFI variable could also have
respectively, according to the total alkaline charge (5%, 7%, 9%, 11%, contribution to the change of other variables results. For exam-
13%). Obviously, with the increase of the total alkaline charge, the ple, when the S/L ratio was 6, temperature was 130 ◦ C, the holding
solid recovery rate reduced. The main reason was that the alkaline time was 30 min, the total alkaline charge was 11%, the Y-glucan,
could promote the degradation of carbohydrate while removed the Y-xylan, Y-total sugar without PFI treatment were 79.49%, 52.76%,
lignin. So there was a certain loss of glucan and xylan which led to 69.65% respectively, however, the Y-glucan, Y-xylan, Y-total sugar
the reduction of solid recovery rate. As can be seen in the experi- with PFI treatment were 83.6%, 58.6%, 74.4% respectively. Obvi-
mental results, when the alkaline charge was 5%, the recovery rate ously, after PFI pretreatment, the Y-glucan, Y-xylan and Y-total
of glucan was 88.45%, and when the alkaline charge increased to sugar were increased, and increased by 4.11%, 5.84% and 4.75%
11%, the recovery rate of glucan was decreased to 83.63%. Mean- respectively (all original pretreatment and enzymatic hydrolysis
while, the recovery rate of xylan was decreased from 80.70% to data could be seen in Supporting Information Table.2). This showed
65.43%. that the PFI treatment in the whole pretreatment of corn stover
312 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318

Fig. 2. PLS model 1 analysis (a: score scatter plot, b: loading scatter plot, c: Variable Importance for the Projection).
Note: In this model, Alkaline, S/L ratio, T, H time and PFI was set as X, and the chemical composition and pretreatment effect parameters (Glucan, Xylan, Extractive, Lignin,
Solid Yield, D-lignin, R-glucan, and R-xylan) were set as Y. a: the scores t1 and t2, were summarizing the varibales. The scores are orthogonal, completely independent of
each other. the score t1 (first component) explains the largest variation of the X space, followed by t2. b: w*c[1] (for first component) and w*c[2] (for second component)
showed the weights combining the original variables. c: the VIP (Variable Importance for the Projection) plot summarized the importance of the variables both to explain X
and to correlate to Y. VIP plots are sorted by order of importance.
 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318 313

Fig. 3. PLS model 2 analysis (a: score scatter plot, b: loading scatter plot, c: Variable Importance for the Projection).
(Alkaline, S/L ratio, T, H time, PFI were set as X, Glucan, Xylan, Extractive, Lignin, Solid Yield, D-lignin, R-glucan, R-xylan, E-glucan, E-xylan, Y-glucan, Y-xylan, Y-total sugar
were set as Y.)
Note: a: the scores t1 and t2, were summarizing the varibales. The scores are orthogonal, completely independent of each other. the score t1 (first component) explains the
largest variation of the X space, followed by t2. b: w*c[1] (for first component) and w*c[2] (for second component) showed the weights combining the original variables. c:
the VIP (Variable Importance for the Projection) plot summarized the importance of the variables both to explain X and to correlate to Y. VIP plots are sorted by order of
importance.
314 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318

Fig. 4. PLS model 3 analysis (a: score scatter plot, b: loading scatter plot, c: Variable Importance for the Projection).
(The pretreatment parameters, the chemical compositions of pretreated corn stover samples and the pretreatment effect were set as X variables, and the enzymatic hydrolysis
was set as Y variables. Alkaline, S/L ratio, T, H time, PFI, Glucan, Xylan, Extractive, Lignin, Solid Yield, D-lignin, R-glucan, R-xylan were set as X variables, meanwhile, E-glucan,
E-xylan, Y-glucan, Y-xylan, Y-total sugar were set as Y variables.)
Note: a: the scores t1 and t2, were summarizing the varibales. The scores are orthogonal, completely independent of each other. the score t1 (first component) explains the
largest variation of the X space, followed by t2. b: w*c[1] (for first component) and w*c[2] (for second component) showed the weights combining the original variables. c:
the VIP (Variable Importance for the Projection) plot summarized the importance of the variables both to explain X and to correlate to Y. VIP plots are sorted by order of
importance.
 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318 315

Fig. 5. PLS model 4 analysis (a: score scatter plot, b: loading scatter plot, c: Variable Importance for the Projection).
(Glucan, xylan, Extractive, Lignin, Solid Yield, D-lignin, R-glucan, R-xylan were set as X variables, E-glucan, E-xylan, Y-glucan, Y-xylan, Y-total sugar were set as Y variables).
Note: a: the scores t1 and t2, were summarizing the varibales. The scores are orthogonal, completely independent of each other. the score t1 (first component) explains the
largest variation of the X space, followed by t2. b: w*c[1] (for first component) and w*c[2] (for second component) showed the weights combining the original variables. c:
the VIP (Variable Importance for the Projection) plot summarized the importance of the variables both to explain X and to correlate to Y. VIP plots are sorted by order of
importance.
316 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318

Fig. 6. Influence of independent variables on the variables.

(a: E-glucan. b: E-xylan. c: Y-glucan. d: Y-xylan. e: Y-total sugar)
(The influence degree of all the independent variables X on the each dependent variables Y.)
 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318 317

could improve the hydrolysis efficiency, improve the yield of glucan The overall VIP analysis results showed that in order to get high
and xylan, which could led to increase the final total sugar yield. enzymatic effect, the lignin of raw corn stover samples should be
removed. Lignin removing could help to take over the main obsta-
cle of enzymatic hydrolysis after pretreatment. At the same time,
3.4. The PLS model 3 analysis for the whole pretreatment process
during the alkaline pretreatment, the carbohydrates (glucan and
xylan) should be reserved as much as possible, which could supply
As shown in Fig. 4, in the PLS model 3, there were 18 variables,
the fermentable sugar during the subsequent enzymatic hydrolysis
which including 13 X variables and 5 Y variables. The pretreatment
process.
parameters, the chemical compositions of pretreated corn stover
In this PLS model, the influence degree of all the independent
samples and the pretreatment effect were set as X variables, and
variables X on the each dependent variables Y could be seen in
the enzymatic hydrolysis was set as Y variables. Alkaline, S/L ratio, T,
the analysis graph. For example, the hydrolysis efficiency of glucan
H time, PFI, Glucan, Xylan, Extractive, Lignin, Solid Yield, D-lignin,
and xylan mainly had strong positive correlation with the glucan
R-glucan, R-xylan were set as X variables, meanwhile, E-glucan,
content, xylan content, and removal rate of lignin. Meanwhile, they
E-xylan, Y-glucan, Y-xylan, Y-total sugar were set as Y variables.
had negative correlation with solid recovery rate and lignin content,
In this PLS model, after two principal components analysis, the
and had no obvious relations between recovery rate of glucan and
R2X(cum) was 0.544, the R2Y(cum) was 0.643, Q2(cum) was 0.585.
xylan. The reasons were as follows: extractive had little effect on the
This PLS model could explain the 64% information of the process,
process mainly because most of the extractive had been removed
and had 59% predictive ability.
after alkaline pretreatment.
As can be seen in the model graph, the variables which could
As shown in Fig. 6, based on the PCA and PLS model analysis
provide great impact on the enzymatic efficiency were as follows:
above, in the sodium hydroxide based alkaline pretreatment of
the total alkaline charge, removal rate of lignin, lignin content,
corn stover, in order to improve the subsequent enzymatic sac-
glucan content, xylan content, solid recovery, pretreatment tem-
charification effect after pretreatment, the alkaline charge could
perature, extractive, the ratio of liquid to solid. Wherein, the total
be increased, the maximum pretreatment temperature should be
alkaline charge, removal rate of lignin and lignin content had the
decreased, the holding time of the maximum pretreatment tem-
similar importance degree, which showed that the lignin removing
perature should not be longer (more than 30 min in this paper).
had very important influence on the subsequent enzymatic sac-
The analysis results also showed that PFI treatment could improve
charification after alkaline pretreatment. The lignin removing was
the final total sugar yield in the whole process.
mainly got by the reaction of lignin and alkaline chemicals during
the pretreatment.
For example, when the removal rate of lignin was 21.1%, the 4. Conclusions
E-glucan, E-xylan, Y-glucan, Y-xylan and Y-total sugar were 58.9%,
42.4%, 59.1%, 41.2%, 52.51%, respectively. When the removal rate of For the pretreatment of corn stover, 18 variables were analyzed
lignin increased to 51.7%, the E-glucan, E-xylan, Y-glucan, Y-xylan by PCA and PLS analysis. The PCA analysis results provided the rela-
and Y-total sugar were increased to 72.4%, 52.9%, 70.0%, 55.9%, and tionship between the 18 variables. By setting different independent
64.8% respectively. When the removal rate of lignin was further variables and dependent variables, the PLS model analysis provided
increased to 77.54%, the E-glucan, E-xylan, Y-glucan, Y-xylan and the influence degree of the dependent variables on the indepen-
Y-total sugar were increased to 78.05%, 51.90%, 77.38%, 53.19%, and dent variables. The final analysis results showed that total alkaline
68.48% respectively. However, it should be noted that the increase charge, and removal rate of lignin had greatest impact on the pre-
of the removal rate of lignin could not necessarily correspond to treatment and enzymatic hydrolysis. The ratio of liquid to solid and
the increase of the enzymatic efficiency. That was because the enzy- the extractive content had a little impact on the final experimental
matic hydrolysis was a complex process, which was subject to many results. The PFI treatment had obviously positive correlation with
factors. When the removal rate of lignin was too high, there could be the final enzymatic hydrolysis effect.
a great loss of carbohydrates which could decrease the total sugar After the PCA and PLS analysis, the relationship between the
yield in the enzymatic hydrolysis process. pretreatment condition parameters, composition of materials, pre-
treatment effect and enzymatic effect could be provided, and the
correlations in-between could play a significant role in guiding the
3.5. The PLS model 4 analysis for the whole process whole process of pretreatment of corn stover. The PCA and PLS
analysis could help to optimize conditions to improve the final
In the PLS model 4, there 13 variables, which including 8 X enzymatic saccharification efficient after alkaline pretreatment of
variables and 5 Y variables for this analysis. The chemical com- lignocellulosic biomass.
position of pretreated corn stover, pretreatment effect were set as
X variables (Glucan, xylan, Extractive, Lignin, Solid Yield, D-lignin,
Acknowledgements
R-glucan, R-xylan), and the enzymatic hydrolysis effect variables
were set as Y variables (E-glucan, E-xylan, Y-glucan, Y-xylan, Y-
This work was also partially supported by the National Natural
total sugar). After the two principle components analysis of the PLS
Science Foundation of China (No. 31470609), the Qingdao Insti-
model, the R2X(cum) was 0.739, the R2Y(cum) was 0.564, Q2(cum)
tute of Bioenergy and Bioprocess Technology Director Innovation
was 0.514. This PLS model 4 could explain the 56.4% of information
Foundation for Young Scientists (No. Y37205110C), the Qing-
of the experimental data, and its predictive ability was 51.4%.
dao University of Science and Technology Foundation for Young
As can be seen in the VIP graph of Fig. 5, the X independent vari-
Teachers (No. 010022706), and the Open Foundation of CAS Key
ables which had great influence degree on the overall Y dependent
Laboratory of Bio-based Materials (QIBEBT-CAS).
variables were as follows: removal rate of lignin, lignin content,
glucan content, xylan content, solid recovery rate, glucan recovery
rate, xylan recovery rate and extractive content. The results showed Appendix A. Supplementary data
that the main four components of corn stover (glucan, xylan, lignin
and extractive) had great influence on the final total sugar yield, Supplementary data associated with this article can be found,
especially the lignin had the greatest influence and the extractive in the online version, at http://dx.doi.org/10.1016/j.resconrec.2016.
had the lowest influence. 12.007.
318 B. Li et al. / Resources, Conservation and Recycling 122 (2017) 307–318

Reference: Luo, X.L., Zhu, J.Y., 2011. Effects of drying-induced fiber hornification on enzymatic
saccharification of lignocelluloses. Enzyme Microb. Technol. 48, 92–99.
Agbor, V.B., Cicek, N., Sparling, R., Berlin, A., Levin, D.B., 2011. Biomass Luo, X.L., Zhu, J.Y., Gleisner, R., Zhan, H.Y., 2011. Effects of wet-pressing-induced
pretreatment: fundamentals toward application. Biotechnol. Adv. 29 (6), fiber hornification on enzymatic saccharification of lignocelluloses. Cellulose
675–685. 18, 1055–1062.
Ayamga, E.A., Kemausuor, F., Addo, A., 2015. Technical analysis of crop residue Miura, T., Lee, S.-H., Inoue, S., Endo, T., 2012. Improvement of enzymatic
biomass energy in an agricultural region of Ghana. Resour. Conserv. Recycl. 96, saccharification of sugarcane bagasse by dilute-alkali-catalyzed hydrothermal
51–60. treatment and subsequent disk milling. Bioresour. Technol. 105, 95–99.
Bichraoui-Draper, N., Xu, M., Millerb, S.A., Guillaumea, B., 2015. Agent-based life Oh, S.Y., Dong, I.Y., Shin, Y., Hwan, C.K., Hak, Y.K., Yong, S.C., Won, H.P., Ji, H.Y.,
cycle assessment for switchgrass-based bioenergy systems. Resour. Conserv. 2005. Crystalline structure analysis of cellulose treated with sodium hydroxide
Recycl. 103, 171–178. and carbon dioxide by means of X-ray diffraction and FTIR spectroscopy.
Ding, S.Y., Liu, Y.S., Zeng, Y., Himmel, M.E., Baker, J.O., Bayer, E.A., 2012. How does Carbohydr. Res. 340, 2376–2391.
plant cell wall nanoscale architecture correlate with enzymatic digestibility? Paranhos, R.S., Petter, C.O., 2013. Multivariate data analysis applied in Hot-Mix
Science 338, 1055–1059. asphalt plants. Resour. Conserv. Recycl. 73, 1–10.
Godoy, J.L., Vega, J.R., Marchetti, J.L., 2014. Relationships between PCA and Segal, L., Creely, J., Martin, A., Conrad, C., 1959. An empirical method for estimating
PLS-regression [J]. Chemom. Intell. Lab. Syst. 130, 182–191. the degree of crystallinity of native cellulose using the X-ray diffractometer.
Himmel, M.E., Ding, S.Y., Johnson, D.K., Adney, W.S., Nimlos, M.R., Brady, J.W., Text. Res. J. 29 (10), 786–794.
Foust, T.D., 2007. Biomass recalcitrance: engineering plants and enzymes for Sluiter, A., Hames, B., Ruiz, R., Scarlata, C., Sluiter, J., Templeton, D., Crocker, D.,
biofuels production. Science 315, 804–807. 2008. Determinatioin of Structure Carbohydrates and Lignin in Biomass.
Khoo, H.H., Wong, L.L., Tan, J., Isoni, V., Sharratt, P., 2015. Synthesis of 2-methyl Laboratory Analytical Procedure (LAP). NREL/TP-510-42618. National
tetrahydrofuran from various lignocellulosicfeedstocks: sustainability Renewable Energy Laboratory, Golden, Colorado, USA.
assessment via LCA. Resour. Conserv. Recycl. 95, 147–182. Sun, Y., Cheng, J.Y., 2002. Hydrolysis of lignocellulosic materials for ethanol
Li, B., Bandekar, R., Zha, Q., Alsaggaf, A., Ni, Y., 2011. Fiber quality analysis: OpTest production: a review [J]. Bioresour. Technol. 83 (1), 1–11.
fiber quality analyzer versus L&W fiber tester. Ind. Eng. Chem. Res. 50 (22), Van Dyk, J.S., Pletschke, B.I., 2012. A review of lignocellulose bioconversion using
12572–12578. enzymatic hydrolysis and synergistic cooperation between enzymes-factors
Li, B., Liu, H., Xu, H., Pang, B., Mou, H., Wang, H., Mu, X., 2014. Characterization of affecting enzymes, conversion and synergy[J]. Biotechnol. Adv. 30, 1458–1480.
the detailed relationships of the key variables in the process of the alkaline Von Schenck, A., Berglin, N., Uusitalo, J., 2013. Ethanol from Nordic wood raw
sulfite pretreatment of corn stover by multivariate analysis. BioResources 9 (2), material by simplified alkaline soda cooking pre-treatment [J]. Appl. Energy
2757–2771. 102, 229–240.
Lin, W.S., Lee, M.S., Huang, Y.C., Den, Walter., 2015. Identifying water recycling Yang, L.F., Cao, J., Mao, J.Y., et al., 2013. Sodium carbonate-sodium sulfite
strategy using multivariate statisticalanalysis for high-tech industries in pretreatment for improving the enzymatic hydrolysis of rice straw [J]. Ind.
Taiwan. Resour. Conserv. Recycl. 94, 35–42. Crop Prod. 43, 711–717.
Liu, C., van der Heide, E., Wang, H., Li, B., Yu, G., Mu, X., 2013. Alkaline twin-screw Yu, G., Li, B., Liu, C., Zhang, Y., Wang, H., Mu, X., 2013. Fractionation of the main
extrusion pretreatment for fermentable sugar production. Biotechnol. Biofuels components of corn stover by formic acid and enzymatic saccharification of
6, 97. solid residue. Ind. Crop Prod. 50, 750–757.
Ljungqvist, C.H., Lyng, R., Thuvander, F., 2005. Evaluation of PFI beating on the Zhu, J.Y., Pan, X.J., Wang, G.S., Gleisner, R., 2009. Sulfite pretreatment (SPORL) for
strain to failure of spruce fibres using single fibre fragmentation. Nord Pulp robust enzymatic saccharification of spruce and red pine. Bioresour. Technol.
Pap. Res. J. 20 (4), 370–377. 100 (8), 2411–2418.