Subscriber access provided by Gothenburg University Library
Article
A new quantitative structure-activity relationship model for Angiotensinconverting enzyme inhibitory dipeptides based on integrated descriptors Baichuan Deng, Xiaojun Ni, Zhenya Zhai, Tianyue Tang, Chengquan Tan, Yijing Yan, Jinping Deng, and Yulong Yin J. Agric. Food Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jafc.7b03367 • Publication Date (Web): 06 Oct 2017 Downloaded from http://pubs.acs.org on October 7, 2017
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Agricultural and Food Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 33
Journal of Agricultural and Food Chemistry
A new quantitative structure-activity relationship model for angiotensin-converting enzyme inhibitory dipeptides based on integrated descriptors 1
Baichuan Deng,†,ǁ Xiaojun Ni,†,ǁ Zhenya Zhai,† Tianyue Tang,† Chengquan Tan,†
2
Yijing Yan,† Jinping Deng,*,† Yulong Yin*,†,‡
3
†
4
Guangdong, P.R. China
5
‡
6
Poultry Production, Key Laboratory of Agro-ecological Processes in Subtropical Region, Institute
7
of Subtropical Agriculture, Chinese Academy of Sciences, Changsha 410125, Hunan, P.R. China
College of Animal Science, South China Agricultural University, Guangzhou, 510642,
National Engineering Laboratory for Pollution Control and Waste Utilization in Livestock and
8
1
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 2 of 33
9
ABSTRACT: Angiotensin-converting enzyme (ACE) inhibitory peptides derived
10
from food proteins have been widely reported for hypertension treatment. In this
11
paper, a benchmark dataset containing 141 unique ACE inhibitory dipeptides was
12
constructed
13
relationships (QSAR) study was carried out to predict half-inhibitory concentration
14
(IC50) of ACE activity. 16 descriptors were tested and the model generated by
15
G-scale descriptor showed the best predictive performance with the coefficient of
16
determination (R²) and cross-validated R² (Q²) of 0.6692 and 0.6220, respectively.
17
For most other descriptors, R² were ranging from 0.52-0.68 and Q² were ranging
18
from 0.48-0.61. A complex model combining all 16 descriptors was carried out and
19
variable selection was performed in order to further improve the prediction
20
performance. The quality of model using integrated descriptors (R2 0.7340±0.0038,
21
Q² 0.7151±0.0019) was better than that of G-scale. An in-depth study of variable
22
importance showed that the most correlated properties to ACE inhibitory activity
23
were hydrophobicity, steric and electronic properties and C-terminal amino acids
24
contribute more than N-terminal amino acids. Five novel predicted ACE-inhibitory
25
peptides were synthesized and their IC50 values were validated through in vitro
26
experiments. The results indicated that the constructed model could give a reliable
27
prediction of ACE-inhibitory activity of peptides and it may be useful in the design
28
of novel ACE-inhibitory peptides.
through
database
mining
and
quantitative
2
ACS Paragon Plus Environment
structure–activity
Page 3 of 33
Journal of Agricultural and Food Chemistry
29
KEYWORDS: ACE-inhibitory peptides, QSAR, Variable selection, Variable
30
importance, Amino acid descriptors
31
INTRODUCTION
32
Nowadays, inhibitors of angiotensin-converting enzyme (ACE) have been
33
considered as first-line therapy for hypertension.1 ACE is a zinc- and chloride-
34
dependent metallopeptidase (EC. 3.4.15.1)2 and plays a dual role in regulating
35
renin-angiotensin system (RAS) and kallikrein-kinin system (KKS). It catalyzes the
36
conversion of inactive angiotensin I (decapeptide) to generate strongly
37
vasoconstrictive angiotensin II (octapeptide) as well as inactivates the vasodilator
38
bradykinin.3 Therefore, ACE has become an appropriate target for antihypertensives.
39
The inhibition of ACE would lead to the reduction of angiotensin II production and
40
consequently the decrease of blood pressure.4 Various synthetic ACE inhibitors,
41
such as captopril, enalapril, ramipril and lisinopril, have been developed for the
42
clinical treatment of hypertension.5 However, synthetic ACE inhibitors inevitably
43
cause adverse side effects such as cough, allergic reactions, taste disturbances, and
44
skin rashes.6 Thus, numerous ACE-inhibitory peptides have been identified from
45
hydrolytic products of food-derived proteins and could be used as a potent
46
functional food additive and represent a healthier and natural alternative to
47
ACE-inhibitory drugs. The origin of these peptides were from milk,7 porcine 3
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
48
skeletal muscle,8 bovine collagen,9 bovine blood,10 egg,11 soybean,12 rapeseed,13 oat
49
(avena sativa),14 marine,15 etc.
50
Among approaches for studying bioactive peptides, many ACE-inhibitory peptides
51
have been discovered by the classical approach, involving peptides production,
52
isolation, purification and identification.16 Then, these newly discovered
53
ACE-inhibitory peptides will be collected and deposited in related databases. Based
54
on databases or literatures, bioinformatic approach has become a more efficient and
55
economical tool for peptide research and discovery of new bioactive peptides when
56
compared with the classical approach.17 Particularly, quantitative structure–activity
57
relationship (QSAR) is a crucial tool for bioinformatic approach and plays an
58
important role in the study of bioactive peptides. In recent years, a number of
59
experimentally validated ACE-inhibitory peptides were used to build QSAR models
60
and in certain cases to predict novel and potent ACE-inhibitory peptides.5, 18 Among
61
these, di and tri-peptides were most frequently studied because they have excellent
62
biological properties such that they can be intact absorbed into blood circulation and
63
they are usually resistant to gastrointestinal proteolysis.19 A classical dataset of
64
dipeptide sequences of 58 ACE inhibitors20 are often utilized to test effectiveness of
65
diverse kinds of amino acid descriptors in QSAR studies. A database consisting of
66
168 dipeptides, in which 95 sequences are unique, was constructed from published
67
literatures to study the QSAR of ACE-inhibitory peptides.21 Besides, most of 4
ACS Paragon Plus Environment
Page 4 of 33
Page 5 of 33
Journal of Agricultural and Food Chemistry
68
previously studies only use a single amino acid descriptor to construct QSAR model,
69
which may result in the loss of descriptive information and neglecting of the
70
connection between different descriptors.
71
Databases such as BIOPEP,22 ACEpepDB (http://www.cftri.com/pepdb/index.php)
72
and PepBank23 contain ACE-inhibitory peptides, but the number is limited. The
73
records with experimentally validated IC50 values are even fewer. Recent years, new
74
ACE-inhibitory peptides are continuously reported in literatures. Kumar et al.
75
established a specific and new database for antihypertensive peptides, AHTPDB,
76
which contains 5978 peptide entries.24 Among these, 3364 entries have provided
77
information of IC50 values of peptides and 1694 were unique peptides.24 Moreover,
78
this database contains 1878 records of dipeptides, including 141 unique dipeptides
79
sequences with IC50 values.
80
In this study, we used the 141 unique ACE-inhibitory dipeptides from AHTPDB to
81
construct a dataset. It is, to our knowledge, the largest number of unique dipeptides
82
ever used in a single QSAR model. 16 different descriptors were used to construct a
83
sophisticated QSAR model in order to use more comprehensive information to
84
describe amino acids. We also used outlier elimination and variable selection
85
methods to optimize the model and improve the prediction performance. The newly
86
predicted ACE-inhibitory peptides were synthesized and their IC50 values were
87
validated through in vitro experiments. The objectives of this study were to 5
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
88
construct a reliable QSAR model for ACE-inhibitory dipeptides prediction and it
89
may be useful in the design of novel ACE-inhibitory peptides.
90
MATERIALS AND METHODS
91
Chemicals. Angiotensin-converting enzyme (ACE) from rabbit lung and
92
hippuryl-histidyl-leucine (HHL) as a substrate of ACE were purchased from
93
Sigma–Aldrich (St. Louis, MO, USA). The chemically synthesized (purity >95%)
94
peptides were obtained from DGpeptidesCo., Ltd. (Hangzhou, China). All the other
95
reagents used in this study were of analytical pure.
96
Assay for ACE-inhibitory activity. ACE-inhibitory activity was assayed by the
97
method of Cushman and Cheung (1971) with slight modifications.25 The peptide
98
solution (50 µL) was mixed with 5 mM HHL solution (150 µL), followed by
99
pre-incubation for 5 min at 37 ℃. Afterwards, 50 µL of a 25 mU/mL ACE solution
100
(prepared in a 0.1 M sodium borate buffer containing 0.3 M NaCl at pH 8.3) was
101
added, the reaction mixture was further incubated for 30min at 37 ℃. The enzymic
102
reaction was terminated by adding 250 µl of 1 M HCl and the liberated hippuric acid
103
was extracted with 1.5 ml of ethyl acetate by vortex mixing for 30 sec. After
104
centrifugation (4000 × g, 10 min), 1 ml aliquot of the upper layer was transferred
105
into a glass tube, and evaporated by heating at 120 ℃ for 30 min. The hippuric acid 6
ACS Paragon Plus Environment
Page 6 of 33
Page 7 of 33
Journal of Agricultural and Food Chemistry
106
was redissolved in 3 ml of distilled water. The absorbance was measured at 228 nm
107
using UV-spectrophotometer (UV-2501PC, Shimadzu, Tokyo, Japan). The IC50
108
value was defined as the concentration of the inhibitor required to inhibit 50% of the
109
ACE-inhibitory activity.
110
Data collection. Datasets for the antihypertensive peptides (AHTPs) were manually
111
collected from AHTPDB (http://crdd.osdd.net/raghava/ahtpdb/), which is a database
112
of experimentally validated antihypertensive peptides and most of the peptides
113
belong to the family of angiotensin I converting enzyme inhibiting peptides.24
114
Source of information of this database were mainly collected from three databases,
115
i.e.
116
(http://www.uwm.edu.pl/biochemia/index.php/pl/biopep)
117
database (http://erop.inbi.ras.ru/), and published literatures. First of all, we selected
118
all the dipeptides in this database, including information about sequence, half
119
maximal inhibitory concentration (IC50), IC50 determination assay, source and
120
molecular mass. IC50 represents the concentration that inhibits 50% activity of ACE.
121
Data processing. During data collection from the database, it was noticed that many
122
of the identical peptides have exhibited the same or different IC50 values. For the
123
peptide with multiple IC50 values, the median value was retained to remove
124
duplicates. A total of 1878 dipeptides were obtained before merging. Then, the total
ACEpepDB
(http://www.cftri.com/pepdb/),
7
ACS Paragon Plus Environment
and
BIOPEP EROP-Moscow
Journal of Agricultural and Food Chemistry
125
number of unique peptides included in QSAR model is 141. All the IC50 values were
126
log-transformed prior to modeling.
127
Building the QSAR model. QSAR is defined as a relationship linking structural
128
characteristics of molecules to their biological or physicochemical properties. Data
129
sets for the processed dipeptides are presented in Table S1. The peptide sequences
130
were transformed into X-matrix by means of 16 descriptors, respectively, while
131
dependent variable Y represents activity values (IC50) of peptides. These descriptors
132
were collected from published articles which can well represent the structural
133
characteristics of the amino acids for QSAR models, including Z-scale,26 5Z-scale,27
134
DPPS,28 MS-WHIM,29 ISA-ECI,30 VHSE,31 FASGAI,32 VSW,33 T-scale,34 ST-scale,35
135
E-scale,36 V,37 G-scale,38 HESH39 and HSEHPCSV.40 For a set of peptides analogues,
136
the structures would be characterized by describing each varied amino acid position
137
with the descriptor’s parameter values. For example, the G-scale descriptor
138
including eight kinds of parameters, if we used it to describe dipeptides, the
139
chemical structures of dipeptides would be described by 16 (8 parameters × 2 amino
140
acids) variables. Thus, a set of peptide sequences varied in n positions can be
141
described by 8×n variables. The amino acid at the N-terminus was designated as n1,
142
and its properties were described as n1G1, n1G2, n1G3, n1G4, n1G5, n1G6, n1G7
143
and n1G8 of the G-scale model. The C-terminus was designated as n2 and so on.
8
ACS Paragon Plus Environment
Page 8 of 33
Page 9 of 33
Journal of Agricultural and Food Chemistry
144
After that, a combination of 204 variables for each dipeptides was undertaken, and
145
204 predictor variables were defined with the above descriptors express as:
146
Z-scale(D1-D6); 5Z-scale(D7-D16); DPPS(D17-D36); MS-WHIM1(D37-D42);
147
MS-WHIM2(D43-D48);
148
FASGAI(D69-D80);
149
ST(D119-D134);
150
HSEHPCSV(D181-D204).
151
Partial least square (PLS) regression41 was used to build the correlation between
152
amino acid descriptors (predictors, X) and log-transformed IC50 (dependent, Y) and
153
it was implemented using MATLAB R2015a software. All variables were
154
auto-scaled to unit variance prior to the analyses. The data set was validated by
155
cross-validation as internal validation, the number of significant PLS components
156
was chosen automatically by using various rules based on a statistic called Q²,
157
which is the cross-validated R², referred to as the predictive ability of the model. R²
158
is the coefficient of determination, which is also an important parameter in PLS
159
analysis and provides an estimate of the model fit.
160
Model population analysis (MPA). MPA is a general framework for chemical
161
modeling which uses random resampling and statistical analysis techniques to
162
extract important information from the data.42 Generally, it contains three steps: (1) a
163
random resampling procedure to obtain sub-datasets, (2) a model building procedure
ISA-ECI(D49-D52); VSW(D81-D98);
V(D135-D140);
E(D99-D108);
G(D141-D156);
9
ACS Paragon Plus Environment
VHSEA(D53-D68); T(D109-D118); HESH(D157-D180);
Journal of Agricultural and Food Chemistry
Page 10 of 33
164
to build sub-models, (3) and a statistical analysis procedure to extract information
165
from the outcome of sub-models. In this study, MPA was applied for outlier
166
detection and variable selection.
167
Outlier detection. In an attempt to obtain a robust and highly predictive model, it is
168
crucial to identify and remove outlying samples from measured data before
169
modeling. The MPA-based method was used to detect outliers of the data.43 To begin
170
with, a number of (e.g. 5000) sub-datasets were generated by applying random
171
resampling procedure in sample space. Each sub-dataset contains 80% of random
172
selected samples from the pool of samples. Then, for each sub-dataset, a PLS
173
regression were built. Thus, a number of (e.g. 5000) were built. In the next step, the
174
sub-models were used to predict the IC50 value of remaining samples separately and
175
the prediction errors for each sample were recorded. Finally, for each sample, a
176
statistical analysis was applied on the prediction errors. The average of prediction
177
errors (MEAN) and standard deviation of prediction errors (STD) were used as the
178
basis for outlier detection. In this study, 3-sigma rule was applied and the samples
179
which exceed the range of mean±3*standard deviation for MEAN (or STD) were
180
considered as outliers. This method eliminated outliers one by one until all samples
181
were within the range.
182
Variable selection. Variable selection was carried out after excluding the outliers. In 10
ACS Paragon Plus Environment
Page 11 of 33
Journal of Agricultural and Food Chemistry
183
the present study, a bootstrapping soft shrinkage (BOSS) method was applied for
184
variable selection.44 It is also based on the idea of MPA.42 Firstly, a number of
185
sub-models were generated using bootstrap resampling in sample space. Then, for
186
each sub-model, the regression coefficients were extracted. The regression
187
coefficients for sub-models were summed up to obtain weights for variables. In the
188
next step, weighted bootstrap resampling45 was applied to build new sub-models,
189
where variables with larger weights had larger probabilities to be selected into the
190
sub-models. The resampling procedure was repeated and the less important variables
191
were eliminated gradually. This variable selection method used multi-model instead
192
of single model for comparison and considered random combination of variables,
193
which had advantages in selecting optimal variable combination compared with
194
previous methods.44 The selected variable is represented as n1/n2-descriptor’s
195
name-the parameter number, where n1 denotes N-terminus and n2 denotes
196
C-terminus. For example, ‘n1-G-1’ means that the selected important variable is the
197
first parameter of the G-scale to describe the amino acid at N-terminus.
198
Statistical Analysis. All statistical analyses were performed by using MATLAB
199
software (Version R2015a, the Mathworks, Inc).
11
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
200
Page 12 of 33
RESULTS AND DISCUSSION
201
(Please insert Table 1)
202
QSAR study on ACE inhibitory dipeptides. Modeling of these dipeptides was
203
conducted according to the data sets in the Table S1. Table 1 summarizes the most
204
important statistical parameters of the model based on dipeptides dataset using the
205
above 16 kinds of descriptors. After elimination of outliers, the final sizes of
206
calibration dataset were slightly different and resulted in substantially improved
207
models. According to the three-sigma rule, each descriptor excluded 2-6 outliers.
208
Figure 1 shows the process of outlier elimination on the model built with G-scale
209
descriptor. The outlier numbers were displayed in each figure of Figure 1a-e, the
210
elimination order was 130, 80, 127 and 125, respectively. All samples were within
211
the range according to the three-sigma rule (dashed line) after removing outliers
212
(Figure 1e). The process of removing outliers for QSAR model with other
213
descriptors is the same as G-scale descriptor. After eliminating outliers, all models
214
established and descriptors were presented in Table 1. It can be seen clearly that the
215
model derived from G-scale descriptor has the best predictive performance.
216
Modeling of the G-scale descriptor with activities has the higher Q² (0.6220) and
217
could explain 66.92% of the sum of squares in Y-variance (R2) after excluding
218
outliers. The Q² value of G-scale, HSEHPCSV, 5Z-scale models are larger than 0.6.
219
For most of the other descriptors, Q² values are between 0.5 and 0.6. Only the 12
ACS Paragon Plus Environment
Page 13 of 33
220
Journal of Agricultural and Food Chemistry
models of MS-WHTM1 and ISA-ECI show Q² values of lower than 0.5.
221
(Please insert Figure 1)
222
To further improvement the model, we built the model using integrated variables,
223
where variables of 16 descriptors were combined. And a large dataset with 204
224
variables were obtained. It was followed by outlier elimination and variable
225
selection on PLS regression models. The aim of this process is to integrate the
226
information of different descriptors together to make a better model. The process of
227
outlier elimination on integrated model is displayed in Figure 2. The order of outlier
228
elimination is 127, 130, 125, 124 and 80, respectively. All samples were within the
229
range according to the three-sigma rule (dashed line) after getting rid of all outliers
230
(Figure 2f). Table 1 shows that the Q²and R² values obtained by using integrated
231
descriptors are 0.6205 and 0.7110, respectively. It is comparable to the result of the
232
best performed single descriptor (G-scale descriptor). Moreover, the integrated
233
model leaves room for further improvement of the model, superior to any
234
single-descriptor models.
235
(Please insert Figure 2)
236
Variable selection. The bootstrapping soft shrinkage (BOSS) method was applied
237
for variable selection on the integrated descriptors model.44 The effective of BOSS
238
has been proved elsewhere.46 In the present study, BOSS was run 100 times and the
239
results were shown in Table 1. Compared the model with all descriptors, variable 13
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 14 of 33
240
selection not only reduced the number of variables, but also improved the prediction
241
performance of the model. The Q² values after variable selection is 0.7151. It has a
242
distinct improvement compared to the full-variable model, of which the Q² value is
243
0.6205 (Table 1). On average, 48 variables were selected from 204 variables by the
244
variable selection method. Researches showed that not all molecular descriptors
245
were related to biological activity, so it is necessary to delete redundant descriptors
246
to improve the prediction performance of the QSAR model.39 Moreover, it also
247
emphasizes the importance of removing outliers and variable selection method in
248
QSAR modeling.
249
(Please insert Table 2)
250
Comparison with the reported models. QSAR studies have been carried out on 58
251
ACE-inhibitory dipeptides using T-scale, G-scale and HESH descriptors, with the Q2
252
values 0.784, 0.831 and 0.838, respectively (Table 2).38 The Q2 value of integrated
253
descriptors combined with BOSS is 0.910, which is larger than previous reports. Wu
254
et al. carried out QSAR study of 168 ACE inhibitor dipeptides with Z-scale with the
255
Q² of 0.711 and R2 of 0.732.21 Fu et al. further improved the model to obtain Q2
256
0.716 and R2 0.746.9 By using integrated descriptors the Q2 and R2 were further
257
improved, which are 0.804 and 0.816, respectively (Table 2). The comparison of the
258
results with the previous reports showed that our method can give higher prediction
259
accuracy on the same datasets.
260
It should be noted that the dipeptides in Wu’s study contained 72 duplicated 14
ACS Paragon Plus Environment
Page 15 of 33
Journal of Agricultural and Food Chemistry
261
sequences (only 94 unique dipeptides). The existence of duplicated sequences may
262
result in an over-optimistic Q2. In the present study, duplicated sequences were
263
eliminated and the median of IC50 value for a unique sequence was retained. As a
264
result, 141 unique dipeptides were used for modeling. It is, to our knowledge, the
265
largest number of unique dipeptides ever used in a single QSAR model. Thus, the
266
prediction performance of our model (Q2=0.7151) is better or comparable with the
267
previous studies.
268
(Please insert Figure 3)
269
Evaluate the importance of variables. For the 16 single descriptor models, the
270
importance of amino acid properties in each position is evaluated using the value of
271
PLS regression coefficients and variable importance in project (VIP).47 Figure 3
272
shows the evaluation of variable importance of G-scale model. Through the PLS
273
regression coefficients values (Figure 3a), it is observed that variables of G1, G5, G6
274
and G7 are important for the bioactivities of ACE-inhibitor dipeptides. For the
275
position n1, G1, G2, G4 and G5 are negatively related to the log values, while G3,
276
G6, G7 and G8 are positively related to the log values. For the position n2, G1, G2,
277
G3, G5 and G6 are negatively to the log values, while G4, G7 and G8 are positive. It
278
is evident that position n2 is more relevant to biological activities than position n1.
279
G-scale descriptor including eight kinds of parameters were derived from 457 kinds
280
of physicochemical properties of the amino acid index database, which was 15
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 16 of 33
281
classified into three sorts of parameters including hydrophobic, steric and electric
282
properties.38 The eight parameters were encoded as G1∼G8 which represented
283
hydrophobicity, STERIMOL minimum width of the side chain, loss of side chain
284
hydropathy by helix formation, optical rotation, side chain molecular volume,
285
frequency of the 4th residue in turn, amino acid composition of EXT of
286
multi-spanning proteins and net charge index, respectively. For the ACE-inhibitory
287
dipeptides, amino acid residues with information of hydrophobic and stereo
288
characteristics are most important to biological activities. The importance of the
289
amino acid residue in position n2 is mainly decided by G1 followed by G5, which
290
represents hydrophobicity, side chain molecular volume, respectively. For both
291
positions, amino acid residues with large bulk chain as well as hydrophobic side
292
chains are preferred, such as phenylalanine, tryptophan, and tyrosine. VIP plots of
293
the PLS models using G-scale descriptor are summarized in Figure 3b. VIP reflects
294
the size of the contribution for variables to activity. For the QSAR model using
295
G-scale descriptor, the most influential property parameters to ACE- inhibitory
296
dipeptides are G1, G5 and G7, and the properties contributing to the model are
297
Hydrophobicity > Steric Property > Electric Properties according to the VIP value. It
298
is obvious that the position n2 is most influential to biological activities and the
299
order of the important variables in n2 is G5 > G1 > G7 > G3 > G2 (VIP value >1).
300
For position n1, eight parameters are arranged in a proper order as G2 > G1 > G8 >
301
G5 > G7 > G6 > G3 > G4. According to these results, the results of PLS regression 16
ACS Paragon Plus Environment
Page 17 of 33
Journal of Agricultural and Food Chemistry
302
coefficients and VIP are similar.
303
The variable importance evaluation of E-scale, HSEHPCSV, 5Z-scale, VHSEA and
304
Z-scale showed that hydrophobicity and steric properties were important for the
305
bioactivities. For HESH, the significant properties contributing to the model was
306
hydrophobicity, especially for the C-terminus. The regression coefficients of
307
FASGAI show that the vital parameters of the bulky properties may be conducive to
308
enhancing bioactivities of ACE inhibitors. The properties of the important variables
309
of DPPS were steric and electronic properties, while for 3D descriptors, relative
310
importance of the X-variables of ISA-ECI in the QSAR model was isotropic surface
311
area, MS-WHIM descriptor was primarily of electrostatic potential. According to the
312
regression coefficients and VIP values of all the descriptors, it could be seen that
313
position n2 (C-terminus) of the dipeptide played an important role in ACE-inhibitory
314
activity. For the important properties of variables, hydrophobicity, steric properties,
315
and electronic properties were crucial, in addition to hydrogen bonding.
316
There were many studies speculated that the amino acid with hydrophobic property
317
on C-terminus was positively highest correlated with ACE inhibitors bioactivity.18, 20,
318
31
Dipeptides with aromatic side chains and proline on C-terminus and branched
319
aliphatic side amino acids on N-terminus were essential for high inhibitory
320
activity.48 Our results are in agreement with the previous findings.
321
(Please insert Figure 4) 17
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 18 of 33
322
The BOSS method includes some randomness in its algorithm, so that the selected
323
variables in each run can be slightly different. This property gives BOSS a new way
324
of evaluating variable importance, i.e. the frequency of selected variables. The
325
variables which have higher frequency of being selected by BOSS show higher
326
variable importance. Figure 4 displayed the frequency of variables selected by
327
BOSS method on ACE dipeptide dataset. Based on BOSS method, after running 100
328
times, the frequency of the selected variables was shown. Among 16 descriptors,
329
G-scale and HSEHPCSV have the highest frequency, followed by ST-scale,
330
FASGAI, MS-WHIM and VSW. Back to the parameters of each descriptor, the top 8
331
variables with descending order are as follows: ST-4 > G-1 > FASGAI-6 >
332
HSEHPCSV-11 > G-3 > G-8> HSEHPCSV-4 > VSW-5 > MS-WHIM2-1. They may
333
have high correlated with ACE inhibitors bioactivities. In other words, properties of
334
hydrophobicity, steric, electronic and hydrogen bonding were more relevant to the
335
biological activities of ACE-inhibitor dipeptide.
336
It can be seen from Figure 3 that the variables with high frequency comes from
337
different descriptors, and the combination of descriptors greatly improved the
338
prediction ability of the QSAR model. Most of the highly selected descriptors, such
339
as G-scale, HSEHPCSV, FASGAI, have good performance when applied
340
separately in QSAR models. However, some descriptors, such as ST-scale,
341
MS-WHIM, have poor predictive performance when modeled separately. They also 18
ACS Paragon Plus Environment
Page 19 of 33
Journal of Agricultural and Food Chemistry
342
have high frequency when applied in variable selection. Coupled with the fact that
343
the model obtained by BOSS has improved prediction performance, we may
344
conclude that these descriptors are also important in QSAR model building. It is
345
suggested that BOSS can extract additional information from the poorly performed
346
descriptors and have considered the interaction with highly performed descriptors.
347
(Please insert Table 3)
348
Prediction and validation of potential ACE-inhibitory dipeptides. According to
349
the constructed QSAR models, the ACE-inhibitory activities of the remaining
350
dipeptides were predicted. Inevitably, there is a certain degree of variation based on
351
QSAR models, therefore, in vitro experiments is required to further validate the
352
activity of peptides predicted.8 In this study, five predicted dipeptides, which had the
353
lowest predicted IC50, were synthesized chemically to determine the IC50 values.
354
Table 3 displays the comparison between predicted and experimental values of
355
dipeptides. The predicted logIC50 values of CW, TW, HW, QW and CY were 0.98,
356
1.20 1.24, 1.24 and 1.35, respectively. The experimental values were 0.54, 1.15, 1.09,
357
2.03 and 1.63, respectively. It can be seen that all these 5 predicted dipeptides are
358
verified to have ACE-inhibitory activities and all the prediction errors are lower than
359
0.5, except for QW. Among the five dipeptides, CW has the lowest predicted
360
logIC50 and measured logIC50, which means the highest ACE-inhibitory activity.
361
Besides, HW, TW, CY and QW also show strong ACE-inhibitory activities. Based 19
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 20 of 33
362
on structure-activity relationship, it has been suggested that high ACE-inhibitory
363
activity of the peptides should have hydrophobic amino acids, especially on
364
C-terminus. The C-terminus of the five predicted dipeptides contains tryptophan or
365
tyrosine, showing strong hydrophobicity. These results indicated the validity of the
366
prediction models, which could provide a reliable prediction on ACE-inhibitory
367
activity of peptides.
368
In conclusion, we constructed a benchmark dataset for QSAR study of ACE
369
inhibitory dipeptides, which contains 141 unique dipeptides. It is, to our knowledge,
370
the largest number of unique dipeptides ever used in a single QSAR model. Among
371
the 16 amino acid descriptors, G-scale descriptor has the best predictive
372
performance and can be selected to describe the structure of ACE-inhibitory
373
dipeptides. Meanwhile, further improvement on the predictive ability of the QSAR
374
model was obtained using integrated descriptors combined with variable selection
375
method. The newly predicted ACE-inhibitory peptides were validated through in
376
vitro experiments, which verified the reliability of the model. The QSAR model we
377
built may be useful in the design of novel ACE-inhibitory peptides.
378
AUTHOR INFORMATION
379
Corresponding Authors
380 381
*
(J.D.) Fax: +86 84615285 Tel: +8613787136677 E-mail:
[email protected] *
(Y.Y.) Fax: +86 84615285 Tel: +8613974915255 E-mail:
[email protected] 20
ACS Paragon Plus Environment
Page 21 of 33
382 383
Journal of Agricultural and Food Chemistry
Author Contributions ǁ
Baichuan Deng and Xiaojun Ni contributed equally.
384
Funding
385
The authors gratefully thank the National Natural Science Foundation of China for
386
support of the projects (Grant Nos. 31330075, 31572420 and 31110103909). The
387
studies meet with the approval of the university’s review board.
388
Notes
389
The authors declare no competing financial interest.
390
ABBREVIATIONS USED
391
ACE, angiotensin-converting enzyme; QSAR, quantitative structure-activity
392
relationship; PLS, Partial least square regression; IC50, half maximal inhibitory
393
concentration; VIP, variable importance in project; BOSS, bootstrapping soft
394
shrinkage method.
395
Supporting Information
396
A table listing the sequence and IC50 values of 141 unique ACE-inhibitory
397
dipeptides.
21
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 22 of 33
REFERENCES 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436
(1) Hanif, K.; Bid, H. K.; Konwar, R., Reinventing the ACE inhibitors: some old and new implications of ACE inhibition. Hypertens. Res. 2010, 33, 11-21. (2) Sturrock, E. D.; Natesh, R.; van Rooyen, J. M.; Acharya, K. R., Structure of angiotensin I-converting enzyme. Cell. Mol. Life Sci. 2004, 61, 2677-2686. (3) Skeggs, L. T., Jr.; Kahn, J. R.; Shumway, N. P., The preparation and function of the hypertensin-converting enzyme. J. Exp. Med. 1956, 103, 295-299. (4) Lonn, E. M.; Yusuf, S.; Jha, P.; Montague, T. J.; Teo, K. K.; Benedict, C. R.; Pitt, B., Emerging role of angiotensin-converting enzyme inhibitors in cardiac and vascular protection. Circulation. 1994, 90, 2056-2069. (5) Jahangiri, R.; Soltani, S.; Barzegar, A., A review of QSAR studies to predict activity of ACE peptide inhibitors. Pharm Sci. 2014, 20, 122-129. (6) Cooper, W. O.; Hernandezdiaz, S.; Arbogast, P. G.; Dudley, J. A.; Dyer, S.; Gideon, P. S.; Hall, K.; Ray, W. A., Major congenital malformations after first-trimester exposure to ACE inhibitors. N. Engl. J. Med. 2006, 354, 2443-2451. (7) Nakamura Y, Y. N., Sakai K, Okubo A, Yamazaki S, Takano T, Antihypertensive Effect of Sour Milk and Peptides Isolated from It That are Inhibitors to Angiotensin I-Converting Enzyme. J. Dairy Sci. 1995, 78, 1253–1257. (8) Castellano, P.; Aristoy, M. C.; Sentandreu, M. A.; Vignolo, G.; Toldra, F., Peptides with angiotensin I converting enzyme (ACE) inhibitory activity generated from porcine skeletal muscle proteins by the action of meat-borne Lactobacillus. J. Proteomics. 2013, 89, 183-190. (9) Fu, Y.; Young, J. F.; Løkke, M. M.; Lametsch, R.; Aluko, R. E.; Therkildsen, M., Revalorisation of bovine collagen as a potential precursor of angiotensin I-converting enzyme (ACE) inhibitory peptides based on in silico and in vitro protein digestions. J. Funct. Foods. 2016, 24, 196-206. (10) Zhang, T.; Nie, S.; Liu, B.; Yu, Y.; Zhang, Y.; Liu, J., Activity prediction and molecular mechanism of bovine blood derived angiotensin I-converting enzyme inhibitory peptides. PloS one. 2015, 10, e0119598. (11) Majumder, K.; Wu, J. P., A new approach for identification of novel antihypertensive peptides from egg proteins by QSAR and bioinformatics. Food Res. Int. 2011, 44, 1371-1378. (12) Wu, J.; Ding, X., Hypotensive and physiological effect of angiotensin converting enzyme inhibitory peptides derived from soy protein on spontaneously hypertensive rats. J. Agric. Food Chem. 2001, 49, 501-506. (13) He, R.; Malomo, S. A.; Alashi, A.; Girgih, A. T.; Ju, X.; Aluko, R. E., Purification and hypotensive activity of rapeseed protein-derived renin and angiotensin converting enzyme inhibitory peptides. J. Funct. Foods. 2013, 5, 781-789. (14) Cheung, I. W. Y.; Nakayama, S.; Hsu, M. N. K., Angiotensin-I Converting Enzyme Inhibitory Activity of Hydrolysates from Oat (Avena sativa) Proteins by In Silico and In Vitro Analyses. J. Agric. Food Chem. 2009, 57, 9234-9242. (15) He, H. L.; Liu, D.; Ma, C. B., Review on the angiotensin-I-converting enzyme (ACE) inhibitor peptides from marine proteins. Appl. Biochem. Biotechnol. 2013, 169, 738-749. (16) Capriotti, A. L.; Cavaliere, C.; Piovesana, S.; Samperi, R.; Lagana, A., Recent trends in the analysis of bioactive peptides in milk and dairy products. Anal. Bioanal. Chem. 2016, 408, 2677-2685. (17) Udenigwe, C. C., Bioinformatics approaches, prospects and challenges of food bioactive peptide research. Trends Food Sci. Technol. 2014, 36, 137-143. (18) Nongonierma, A.; Fitzgerald, D., Learnings from quantitative structure activity relationship (QSAR) studies with respect to food protein-derived bioactive peptides: A review. RSC Adv. 2016, 6, 75400-75413. (19) Miner-Williams, W. M.; Stevens, B. R.; Moughan, P. J., Are intact peptides absorbed from the healthy gut in the 22
ACS Paragon Plus Environment
Page 23 of 33
437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477
Journal of Agricultural and Food Chemistry
adult human? Nutr. Res. Rev. 2014, 27, 308-329. (20) Hellberg, S.; Eriksson, L.; Jonsson, J.; Lindgren, F.; Sjöström, M.; Skagerberg, B.; Wold, S.; Andrews, P., Minimum analogue peptide sets (MAPS) for quantitative structure-activity relationships. Int. J. Pept. Protein Res. 1991, 37, 414-424. (21) Wu, J.; Aluko, R. E.; Nakai, S., Structural requirements of Angiotensin I-converting enzyme inhibitory peptides: quantitative structure-activity relationship study of di- and tripeptides. J. Agric. Food Chem. 2006, 54, 732-738. (22) Minkiewicz, P.; Dziuba, J.; Iwaniak, A.; Dziuba, M.; Darewicz, M., BIOPEP database and other programs for processing bioactive peptide sequences. J. AOAC Int. 2008, 91, 965-980. (23) Shtatland, T.; Guettler, D.; Kossodo, M.; Pivovarov, M.; Weissleder, R., PepBank--a database of peptides based on sequence text mining and public peptide data sources. BMC Bioinf. 2007, 8, 280. (24) Kumar, R.; Chaudhary, K.; Sharma, M.; Nagpal, G.; Chauhan, J. S.; Singh, S.; Gautam, A.; Raghava, G. P., AHTPDB: a comprehensive platform for analysis and presentation of antihypertensive peptides. Nucleic Acids Res. 2015, 43, D956-D962. (25) Cushman, D. W.; Cheung, H. S., Spectrophotometric assay and properties of the angiotensin-converting enzyme of rabbit lung. Biochem Pharmacol 1971, 20, 1637. (26) Hellberg, S.; Sjöström, M.; Skagerberg, B.; Wold, S., Peptide quantitative structure-activity relationships, a multivariate approach. J. Med. Chem. 1987, 30, 1126-1135. (27) Sandberg, M.; Eriksson, L.; Jonsson, J.; Sjöström, M.; Wold, S., New chemical descriptors relevant for the design of biologically active peptides. A multivariate characterization of 87 amino acids. Journal of Medicinal Chemistry 1998, 41, 2481-91. (28) Tian, F.; Yang, L.; Lv, F., In silico quantitative prediction of peptides binding affinity to human MHC molecule:an intuitive quantitative structure-activity relationship approach. Amino Acids. 2009, 36, 535-554. (29) Zaliani, A.; Gancia, E., ChemInform Abstract: MS‐WHIM Scores for Amino Acids: A New 3D‐Description for Peptide QSAR and QSPR Studies. J. Chem. Inf. Model. 1999, 39, 525-533. (30) Collantes, E. R.; Rd, D. W., Amino acid side chain descriptors for quantitative structure-activity relationship studies of peptide analogues. J. Med. Chem. 1995, 38, 2705-2713. (31) Mei, H.; Liao, Z. H.; Zhou, Y.; Li, S. Z., A new set of amino acid descriptors and its application in peptide QSARs. Biopolymers. 2005, 80, 775–786. (32) Liang, G.; Yang, L.; Kang, L.; Mei, H.; Li, Z., Using multidimensional patterns of amino acid attributes for QSAR analysis of peptides. Amino Acids 2009, 37, 583-591. (33) Tong, J.; Liu, S.; Zhou, P.; Wu, B.; Li, Z., A novel descriptor of amino acids and its application in peptide QSAR. J. Theor. Biol. 2008, 253, 90-97. (34) Tian, F.; Zhou, P.; Li, Z., T-scale as a novel vector of topological descriptors for amino acids and its application in QSARs of peptides. J. Mol. Struct. 2007, 830, 106-115. (35) Yang, L.; Shu, M.; Ma, K.; Mei, H.; Jiang, Y.; Li, Z., ST-scale as a novel amino acid descriptor and its application in QSAM of peptides and analogues. Amino Acids. 2010, 38, 805-816. (36) Venkatarajan, M. S.; Braun, W., New quantitative descriptors of amino acids based on multidimensional scaling of a large number of physical–chemical properties. J. Mol. Model. 2001, 7, 445-453. (37) Lin, Z. H.; Long, H. X.; Bo, Z.; Wang, Y. Q.; Wu, Y. Z., New descriptors of amino acids and their application to peptide QSAR study. Peptides. 2008, 29, 1798-1805. (38) Wang, X.; Wang, J.; Lin, Y.; Ding, Y.; Wang, Y.; Cheng, X.; Lin, Z., QSAR study on angiotensin-converting enzyme 23
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
Page 24 of 33
inhibitor oligopeptides based on a novel set of sequence information descriptors. J. Mol. Model. 2011, 17, 1599-1606. (39) Shu, M.; Mei, H.; Yang, S.; Liao, L.; Li, Z., Structural Parameter Characterization and Bioactivity Simulation Based on Peptide Sequence. QSAR Comb. Sci. 2009, 28, 27-35. (40) Dan-Qun; LIANG; Gui-Zhao; ZHANG; Zhi-Liang, New Descriptors of Amino Acids and Its Applications to Peptide Quantitative Structure-activity Relationship. Chin. J. Struct. Chem. 2008, 27, 1375-1383. (41) Wold, S.; Sjöström, M.; Eriksson, L., PLS-regression: a basic tool of chemometrics. Chemom. Intell. Lab. Syst. 2008, 58, 109-130. (42) Deng, B. C.; Yun, Y. H.; Liang, Y. Z., Model population analysis in chemometrics. Chemometrics & Intelligent Laboratory Systems 2015, 149, 166–176. (43) Cao, D. S.; Liang, Y. Z.; Xu, Q. S.; Li, H. D.; Chen, X., A new strategy of outlier detection for QSAR/QSPR. Journal of Computational Chemistry 2010, 31, 592–602. (44) Deng, B. C.; Yun, Y. H.; Cao, D. S.; Yin, Y. L.; Wang, W. T.; Lu, H. M.; Luo, Q. Y.; Liang, Y. Z., A bootstrapping soft shrinkage approach for variable selection in chemical modeling. Anal. Chim. Acta 2016, 908, 63-74. (45) Barbe, P.; Bertail, P., The Weighted Bootstrap. Lecture Notes in Statistics 1995, 98. (46) Lin, Y. W.; Deng, B. C.; Wang, L. L.; Xu, Q. S.; Liu, L.; Liang, Y. Z., Fisher optimal subspace shrinkage for block variable selection with applications to NIR spectroscopic analysis. Chemometrics & Intelligent Laboratory Systems 2016, 159. (47) Wold, S.; Johansson, E.; Cocchi, M., PLS: Partial Least Squares Projections to Latent Structures, 3D QSAR in drug design. 1993; Vol. 1, p 523-550. (48) Cheung, H. S.; Wang, F. L.; Ondetti, M. A.; Sabo, E. F.; Cushman, D. W., Binding of peptide substrates and inhibitors of angiotensin-converting enzyme. Importance of the COOH-terminal dipeptide sequence. J. Biol. Chem. 1980, 255, 401-407.
24
ACS Paragon Plus Environment
Page 25 of 33
Journal of Agricultural and Food Chemistry
Table1. Comparisons among different QSAR models for ACE inhibitory dipeptidesa Before outlier elimination
After outlier elimination
Descriptors
Q2
R2
optPC
Q2
R2
optPC
Outlier
G
0.5331
0.5619
1
0.6220
0.6692
2
130, 80, 125,127
E
0.4589
0.5202
1
0.5965
0.6490
1
130, 80, 125,127,124
HSEHPCSV
0.4824
0.5920
4
0.6087
0.6809
3
80,127,130,125
5Z_scale
0.4132
0.5181
5
0.5629
0.6294
5
130, 125,127, 124,123
HESH
0.4306
0.4823
1
0.5419
0.6776
8
127, 130, 125,124
FASGAI
0.4521
0.5074
1
0.5918
0.6436
2
127, 130, 125, 80
VHSEA
0.4980
0.5396
1
0.5650
0.6033
1
127, 125, 130
V
0.4420
0.4715
1
0.5501
0.5827
2
127, 80, 125, 130
T
0.4943
0.5585
6
0.5716
0.6271
6
127, 124, 130, 125
ST
0.4918
0.5859
4
0.5755
0.6457
4
127, 130, 124
Z_scale
0.4753
0.5149
2
0.6028
0.6387
3
130,125, 127, 124,123, 80
DPPS
0.4704
0.5449
2
0.5501
0.6155
2
125, 127, 130
VSW
0.5170
0.6101
2
0.5430
0.6185
1
127, 125, 130
MS_WHTM2
0.4294
0.4782
2
0.5348
0.5808
5
127, 124,130, 125
MS_WHTM1
0.4069
0.4540
1
0.4809
0.5238
1
127, 125
ISA_ECI
0.4323
0.4709
4
0.4851
0.5239
4
130, 127
0.5095
0.5528
1
0.6205
0.7110
2
127, 125, 130, 124, 80
0.7151±
0.7340±
2±
0.0019
0.0038
0.4976
Integrated descriptors BOSS a
R² is the coefficient of determination; Q² is the cross-validated R²; optPC is optimal principal components for PLS regression model; the results of BOSS are shown in the form of mean value ± standard deviation in 100 runs. 502
25
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 26 of 33
Table 2. The comparison between QSAR models of ACE-inhibitory dipeptides using reported datasets Dataset
Descriptor
Modeling Method
Q2(CV)
R2
Ref
58 dipeptides 58 dipeptides 58 dipeptides 58 dipeptides 168 dipeptides 168 dipeptides 168 dipeptides
T-scale G-scale HESH Integral + BOSS Z-score 5Z-scale Integral + BOSS
PLS PLS PLS PLS PLS PLS PLS
0.784 0.831 0.838 0.910±0.002 0.711 0.716 0.804±0.001
0.868 0.870 0.877 0.937±0.004 0.732 0.746 0.816±0.002
34
26
ACS Paragon Plus Environment
38 39
21 9
Page 27 of 33
Journal of Agricultural and Food Chemistry
Table3. Prediction and experimental validation of potent ACE-inhibitory dipeptidesa logIC50 peptides predicted observed error CW 0.98 0.54 -0.44 TW 1.20 1.15 -0.05 HW 1.24 1.09 -0.15 QW 1.24 2.03 0.79 CY 1.35 1.63 0.28 a Predicted activity refers to the values obtained from PLS regression model; observed activity refers to the experimentally determined activity using synthetic dipeptides; logIC50 refers to the logarithmic form of IC50 (µM). 503
27
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 28 of 33
Figure captions
Figure 1. The process of outlier elimination on the model built with G-scale descriptor. The dashed line is defined as the threshold for outliers, which is mean±3*standard deviation for MEAN (or STD). (a) No outlier was eliminated, (b) sample no. 130 was eliminated, (c) sample no. 80 was eliminated, (d) sample no. 127 was eliminated, (e) sample no. 125 was eliminated and all outliers were removed from the model.
Figure 2. The process of outlier elimination on the model built with integrated descriptors. The dashed line is defined as the threshold for outliers, which is mean±3*standard deviation for MEAN (or STD). (a) No outlier was eliminated, (b) sample no. 127 was eliminated, (c) sample no. 130 was eliminated, (d) sample no. 125 was eliminated, (e) sample no. 124 was eliminated, (f) sample no. 80 was eliminated and all outliers were removed from the model.
Figure 3. (a) PLS regression coefficients and (b) VIP of the G-scale model of the ACE -inhibitory dipeptides. The larger value of VIP and the larger absolute value of regression coefficients denote higher variable importance.
Figure 4. The frequency of variables selected in BOSS method on ACE dataset (100 runs). The higher frequency denotes higher variable importance.
28
ACS Paragon Plus Environment
Page 29 of 33
Journal of Agricultural and Food Chemistry
Figure 1
29
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Figure 2
30
ACS Paragon Plus Environment
Page 30 of 33
Page 31 of 33
Journal of Agricultural and Food Chemistry
Figure 3 504
31
ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Figure 4 505
32
ACS Paragon Plus Environment
Page 32 of 33
Page 33 of 33
Journal of Agricultural and Food Chemistry
Graphic for table of contents.
33
ACS Paragon Plus Environment