Identification of protein S-Nitrosylation sites using composition of amino acid frequency Abstract S-nitrosylation is the covalent addition of a nitrogen monoxide group to the thiol adjacent chain of cysteine which is a form of post-translational protein modification, which can associate with stroke, cancer and a number of chronic degenerative diseases, neurodegeneration, including Parkinson’s and Alzheimer’s disease and Amyotrophic Lateral Sclerosis.
Identification of S-nitrosylation site is important for understanding the function of S-nitrosylation. However, the S-nitrosylation site is mainly difficult to detect because the experimental approaches are often time-consuming and labor intensive. Thus, an accurate computational method for identifying S-nitrosylation sites may help researchers to design their experiments and to understand the molecular mechanism of S-nitrosylation. In this study, a computational method has been developed to predict protein S-nitrosylation sites by including CKSAAP and binary encodings .Then, a random forest classifier was trained with these features. Our proposed method achieves an AUC score of 0.782 in the 10-fold cross-validation set for 1:1 ratio of positive and negative samples and performs significantly improve in both cross-validation and independent test.
Thus, the proposed method would be helpful computational resource for predicting S-nitrosylation sites. Keywords: S-nitrosylation, CKSAAP encoding, Binary encoding, Random forest, Feature selection. Introduction Nitric oxide (NO) and related nitrogen species which are considered reactive can mediate various post translational modifications (PTMs), for example metal nitrosylation, tyrosine nitration, and cysteine S-nitrosylation. Cysteine S-nitrosylation term is used to describe the covalent binding of an NO group to a protein cysteine (Cys) residue. S-Nitrosylation is a form of post-translational protein modification with similarities to phosphorylation.
This PTM is important molecular mechanisms by which NO regulates protein functions and has been shown to alter protein activities, protein-protein interactions, and subcellular localization under both normal and pathological conditions 1–4. MS-based proteomics approaches have been developed for theidentification of S-nitrosylated proteins and their modification sites from complex biological samples 5, 6. By trans-nitrosylation this molecule can transfer its NO moiety to protein cysteine residues. GSNO is often used to generate S-nitrosylated proteins in removes for the subsequent isolation and identification of S-nitrosylated proteins 7–11.
The identification of redox-sensitive cysteine residues is the most important to understad the regulatory functions of NO. Latest reports have suggested that S-nitrosylation can modulate protein stability 12 and trafficking 13,14, and play an important role in a variety of biological processes, including transcriptional regulation 12, apoptosis 15, and chromatin remodeling 16. Moreover, aberrant S-nitrosylation has been implicated in numerous diseases 15,17,18. S-Nitrosylation meets the criteria for validation as a cell signaling mechanism in that it is stimulus evoked19, precisely targeted20, reversible21, spatiotemporally restricted2223 and necessary for specific cell responses24. The first protein whose activity was shown to be regulated by S-nitrosylation in this fashion was the NMDA-type glutamate receptor in the brain2526.
S-Nitrosylation has since been shown to be ubiquitous in biology, having been demonstrated to occur in all phylogenetic Kingdoms2728 and has been described as the prototypic redox-based signalling mechanism29, hypothesized to have evolved on primordial Earth30. The reverse process of S-nitrosylation is termed denitrosylation, which in addition to occurring spontaneously in the presence of metal ions and under conditions of photolysis31, has recently been demonstrated to be an enzymically controlled process. S-Nitrosoglutathione reductase (GSNOR), which accelerates the decomposition of S-nitrosoglutathione (GSNO) and other SNO-proteins, is an alcohol dehydrogenase class III isoenzyme which has been shown to be conserved from bacteria to humans32.
Similarly, the thioredoxin/thioredoxin reductase system catalyzes the denitrosylation of a number of S-nitrosoproteins333435. Aberrant or dysregulated denitrosylation or S-nitrosylation has been associated with stroke (cerebral ischemia)36 and a number of chronic degenerative diseases, including Parkinson’s and Alzheimer’s disease37-40 and Amyotrophic Lateral Sclerosis (ALS)41. There is an emerging role of S-nitrosylation in cancer biology42. S-Nitrosylation of EGFR and Src activates an oncogenic signaling network in human basal-like breast cancer 43. Some research groups have attempted to define consensus motifs for S-nitrosylation by relating the amino acid sequences around identified target cysteine residues. Such analyses have shown that the target cysteine residues often lie within an acid-base or hydrophobic motif 45.
In difference, other studies have discovered that the primary sequence of the surrounding amino acid residues has no significant effect on the reactivity of cysteines to S-nitrosylation at the peptide level 46. Recently, 70 known S-nitrosylated sites have been used to identify general structures associated with S-nitrosylation. The results obtained shown that proximal acid– base motif, Cys pKa, sulfur atom exposure, and Cys conservation or hydrophobicity in the vicinity of the modified cysteine do not predict S-nitrosylation specificity. In its place, this analysis identified a revised acid-base motif that is located farther from the cysteine and in which the charged groups are exposed 47.
By the way, experimental identification of S-nitrosylated proteins together with their sites would serve as a foundation to understand the molecular mechanisms and regulatory roles of S-nitrosylation. Currently, computational studies of post-translational modifications (PTMs) are attracting considerable attention. In difference with time-consuming and expensive experimental methods, certain of the accurate and convenient computational approaches have been shown to be able to rapidly generate helpful information for further experimental verification.Materials and Methods Data collection First, 1356 amino acid sequences were downloaded from the most recent version of the Mus Musculus information resource Data Base of cysteine s-nytrosylation (http://140.138.
144.145/~dbSNO/download.php) In this collected data have 2641 experimentally verified S-nitrosylation sites in 1356 proteins. The number of S-nytrosylated proteins 3351 and number of cysteine S-nytrosylation instances 3259 in this data. Data preparation In this study ,the experimentally validated s-nytrosylation sites (Cysteine residues) were considered as positive samples, while all the remaining cysteine residues that have not been verified as s-nytrosylation sites in these proteins were considered as negative samples (i.e. non- s-nytrosylation sites). Each site was represented as a sequence fragment with cysteine in the center, which includes 1356 s-nytrosylation proteins covering 2641 positive and 10394 negative sites.
We are taken training data as 80% data and test data as 20% . The numbers of positive and negative samples are highly imbalanced in the original training dataset. Therefore, a relatively balanced dataset with a 1:1 ratio of positives to negatives (i.
e. 2641 positive sites and 2641 randomly selected negative sites) was compiled to train our predictor. It can be noted that for the independent test set, the corresponding proteins with all positive and negative samples were used. However, for the data set, only a 1: 1 ratio of the positive vs. negative samples was assessed.Fig. 1: The s-nytrosylation predictor pipeline. The datasets were collected from several published database.
At first total data set divided by two set between training and independent data set. For the remaining proteins, a 1:1, 1:2, 1:3 ratio of s-nytrosylation vs. non- s-nytrosylation sites was selected to construct the dataset. Then we select window and feature.
After encoding two types of features, the RF was utilized to build the classifier. Then, the best model was built after parameter optimization and performance evaluation. DATABASE Downloaded dataset Positive vs. negative Window size selection Feature selection Encoding Encoding CKSAAP Binary Random forest Cross-validation Parameter optimization Performance (Sn, Sp, Ac, Mcc and AUC) Statistical learning model Independent data set Training datasetPerformance assessment To investigate the performance of the proposed s-nytrosylation predictor, we used four measurements, including sensitivity (Sn), specificity (Sp), accuracy (Ac), and Mathew Correlation Coefficient (MCC) to evaluate the prediction performance of GPS-SNO. The four measurements were defined as below: ���=������+��� ���=����+���� ����=���+�����+����+��+��� �����=(���×���)?(���×���)?(���+���)×(���+���)×(���+���)×(���+���) where TP represents the number of correctly predicted s-nytrosylation windows, TN is the number of correctly predicted non s-nytrosylation windows, FP is the number of incorrectly predicted s-nytrosylation windows and FN is the number of incorrectly predicted non- s-nytrosylation windows. The values of all of these measurements lie between 0 and 1, and a higher value represents a better prediction. In addition, we also used the receiver operating characteristics (ROC) score for calculating the performance of the proposed prediction 49, 50. To evaluate the performance of the proposed predictor using the performance measures (Sn, Sp, Ac and MCC) as early mentioned earlier, a five-fold cross validation (CV) was used to investigate the performance of the proposed method based on the two encoding methods (CKSAAP and Binary).
In addition, at last we used 5-fold, 10-fold and 15-fold CV to investigate the performance of the proposed method based on the feature selection and without feature selection. For the 5-fold CV, the original datasets were randomly and equally divided into 5 subgroups. Among the 5 subgroups, one subgroup was singled out as the test dataset and the remaining 4 subgroups were considered as the training dataset. Then, we computed all the performance measures for each predictor. We replicated this procedure 5 times by changing datasets from the 5 subgroups.
Finally, we computed the average value for each performance measure for each predictor. CKSAAP encoding CKSAAP is one of the most classical encoding methods and was initially developed by Chen et al51. It has been widely used in several bioinformatics tasks 52–58. The procedure of CKSAAP is briefly described as follows. If a sequence fragment was composed of a window size 2r + 1 and 21 types of amino acids (including the gap (O)), it may contain (21× 21) = 441 types of amino acid pairs (i.
e. AA, AC, AD,. . ., OO) for every single k (k denotes the space between two amino acids). For example, ”AXXA”, whose k-space number is equal to 2.
In this study, the optimal kmax was set at 5, indicating that 21× (kmax + 1) × 21 = 2646 different amino acidpairs were collected to calculate the corresponding feature vector for each sequence. The feature vector was then calculated using the following equation : (����������������, ���������������, ���������������, … , ����������������)441 Where Ntotal is the length of the total composition residues (for example, if the fragment residue with a length L is 29 and k = 0, 1, 2, 3, 4, 5 then Ntotal = L – k -1 will be 28, 27, 26, 25, 24 and 23, respectively). �����, �����,. . ., ��� represent the frequency of the amino acid pair within the fragment. Binary encoding In order to make a robust predictor, binary amino acid encoding was considered to calculate the positional information from the corresponding sequence fragments. In this study, 21 (including gap (O)) amino acids were transformed into numeric vectors by adopting a binary vector.
The 21 types of residues were ordered as ACDEFGHIKLMNPQRSTVWYO. For adopting binary vector, in query proteins, A was represented as 1000000000000000000000 and C as 01000000000000000000, and so on. The selected window size of surrounding s-nytrosylation sites was 27. For the query proteins of s-nytrosylation sites, the center position was always C. Thus, it was not considered to be taken into account. Finally, the feature vectors with a dimensionality (21×26) = 546 were obtained from the binary encoding Random forest classifier RF classifier is a collection of decision tree classifiers, where in each tree is trained with a randomly selected subset of samples.
The decision tree is grown as follows. Suppose N samples are randomly selected with replacement from the F features, then the best split node is selected from F features. Finally, the decision tree is grown as large as possible without cutting. In the construction of the forest, it is generalized based on most votes given by all the individual trees; within the post for the error estimate it does not produce bias. It is relatively robust to noise and outliers 59. As a supervised learning algorithm, it has been widely used in protein bioinformatics 60–62.
The predicted result of the RF was decided by voting among the number of trees, which contains two classes, either positive samples (s-nytrosylation sites) or negative samples (non- s-nytrosylation sites). In this study, the RF algorithm was implemented using the ‘Random Forest’ R package. Feature encoding In statistics, feature selection, also known as variable selection, attribute selection or variable subset selection, is the process of selecting a subset of relevant features (variables, predictors) for use in model construction. In order to build an effective prediction model, we encoded each sequence fragment into a numeric vector, which was the crucial step to present the classifier and group architecture. Thus, a high-quality sequence encoding method for keeping the generated code compact in dimensionality was necessary. Instead of employing a simple binary representation, three types of amino acid feature encodings were adopted, including CKSAAPand binary .
As mentioned in ”Feature encoding”, each investigated s-nytrosylation or non s-nytrosylation fragment was encoded into as a high dimensional vector. Therefore, they may not equally contribute to determine the surrounding s-nytrosylation or non s-nytrosylation sites. Results and discussion Prediction capabilities of the different encoding features with a RF classifier In nature, the s-nytrosylation and non- s-nytrosylation datasets are highly unbalanced. It has been established that statistical learning algorithms become computationally intractable and the accuracy is strongly affected due to the nature of the unbalanced datasets.
To address this issue, many PTM site prediction studies employ a relatively balanced ratio between the positive and negative samples during the training of the classifiers (e.g. the ratio of positives versus negatives is controlled at 1:1 or 1: 2) 53, 58, 61.
After evaluating the different ratios of s-nytrosylation and non- s-nytrosylation peptides, a comparatively balanced training dataset (1:1 ratio of positive vs. negative) was used to develop the RF-based predictor of s-nytrosylation. In order to evaluate the performance capability of the different encoding features, two single feature encoding-based (CKSAAP and Binary) table1. The performance of these sensitivity (Sn), specificity (Sp), accuracy (Ac), and Mathew Correlation Coefficient (MCC) was assessed using a 5-fold cross-validation. To obtain the sensitivity (Sn), specificity (Sp), accuracy (Ac), and Mathew Correlation Coefficient (MCC) score we use 21, 27, 31 window size for the CKSAAP and binary encodings. The AUC performance values on the datasets are given in Table 1.
Table 1: Performance of the ratio and CKSAAP and binary encodings on the training dataset CKSAAP Window size ratio fold Sn Sp Ac Mcc AUC 21 1:1 5 0.6435397 0.677635 0.6587776 0.3193598 0.726 21 1:2 5 0.6821918 0.6873276 0.
6870404 0.1800967 0.708 21 1:3 5 0.7288136 0.7633444 0.7624081 0.1845999 0.
690 27 1:1 5 0.6548745 0.6959917 0.6730239 0.3484487 0.743 27 1:2 5 0.
6971279 0.6893409 0.6897978 0.1926642 0.724 27 1:3 5 0.6766917 0.7634511 0.7607996 0.
1749583 0.710 31 1:1 5 0.6605617 0.7081794 0.681296 0.3656536 0.753 31 1:2 5 0.
6958525 0.6924844 0.6927083 0.2052253 0.
736 31 1:3 5 0.6654412 0.7634013 0.7603401 0.172317 0.720 BINARY Window size ratio Fold Sn Sp Ac Mcc AUC 21 1:1 5 0.
5031915 0.5008792 0.5013787 0.003350293 0.500 27 1:1 5 0.5826087 0.5022422 0.5043658 0.
02721921 0.503 31 1:1 5 0.6148148 0.5039127 0.5073529 0.
04116815 0.503 In this table indicate that the highest score of the AUC 0.726(Sn=0.6435397,Sp=0.677635,Ac=0.
6587776,Mcc=0.3193598) for 21 window size, 0.743(Sn=0.6548745,Sp=0.6959917,Ac=0.6730239,Mcc=0.3484487) for 27 window size and0.753 (Sn=0.
6605617,Sp=0.7081794,Ac=0.681296,Mcc=0.3656536) for 31 window size in CKSAAP encoding . So when we are taken 1:1 ratio then the AUC score is better than the ratio 1:2 and 1:3.So the 1:1 ratio provides most accurate predictions of s-nytrosylation sites. And another issue is the encoding, in above result we see that the AUC score for CKSAAP encoding is better than binary encoding.
Therefore, CKSAAP encoding provides most accurate predictions of s-nytrosylation sites. Another issue is the optimal size of the sequence windows flanking the s-nytrosylation and non- s-nytrosylation sites. Table2: Performance of the window size on the training dataset Window size ratio fold Sn Sp Ac Mcc AUC 19 1:1 5 0.6402789 0.
6786834 0.6571691 0.3166418 0.724 21 1:1 5 0.6435397 0.677635 0.6587776 0.3193598 0.
726 23 1:1 5 0.6433306 0.6785346 0.6590074 0.3199341 0.732 25 1:1 5 0.6552873 0.6912821 0.
6714154 0.344695 0.740 27 1:1 5 0.
6548745 0.6959917 0.6730239 0.3484487 0.743 29 1:1 5 0.6670797 0.
709477 0.6858915 0.3741623 0.754 31 1:1 5 0.6605617 0.7081794 0.681296 0.3656536 0.
753 The optimal window size was decided based on the analysis table2. AUC value of the consecutive CKSAAP encoding methods mentioned above. The performance of the different window size is shown (table 2) .We observed that the model with a window size of 29 was optimal to discriminate s-nytrosylation and non- s-nytrosylation sites. Sequence specificity of S-nitrosylation site The proposed s-nytrosylation combines the positional amino acid encoding properties. For the dataset, the amino acid propensities of surrounding s-nytrosylation sites compared to the non- s-nytrosylation sites were displayed by Two Sample Logos software64 (Fig. 1). Briefly, in the two sample logo, only over- or under-represented residues at each position are plotted above and under the X-axis, respectively.
The height of the letter was in proportion to the percentage of positive (if over-represented) or negative samples (if under-represented) harboring the corresponding residue. The Y-axis reports the cumulative percentage of these over-/ under-represented residues. We can see that some amino acids are over-/underrepresented in the specific positions (Fig.1), which indicates that the positional amino acid encoding was an efficient method to identify the s-nytrosylation sites. But, positional based encoding is not enough to accurately predict the s-nytrosylation Sites, as demonstrated by the performance of binary encoding in s-nytrosylation .
.Fig. The amino acid propensities of surrounding s-nytrosylation sites compared to non- s-nytrosylation sites, as displayed with the Two Sample Logos software. It also shows that the position between the compositional amino acids of the s-nytrosylation and non- s-nytrosylation peptides had a wide difference, especially those located in the positions from -13 to -1 and +1 to +13.
Feature optimization The feature selection method was performed to analyze the most informative features of the surrounding s-nytrosylation substrates. Although the feature selection method did not result in an improvement in performance, it allowed us to collect the most important features from the corresponding encoding methods according to importance. Table 3 (a): Performance of the gain and correlation feature selection method for 1:1 ratio and 29 window size on the training data set. Gain Feature selection Features sizes ratio fold Sn Sp Ac Mcc AUC 100 1:1 5 0.6349745 0.6593186 0.6461397 0.
2932846 0.707 200 1:1 5 0.6677686 0.7101449 0.6865809 0.
3755301 0.759 300 1:1 5 0.6845804 0.7309881 0.705193 0.4129691 0.777 400 1:1 5 0.
6800821 0.7287428 0.7015165 0.
4059187 0.776 500 1:1 5 0.6799678 0.
7386424 0.705193 0.4144777 0.778 600 1:1 5 0.6845528 0.7399577 0.7086397 0.4208795 0.
772 700 1:1 5 0.6811594 0.7258647 0.
701057 0.4045616 0.767 800 1:1 5 0.6770665 0.
7319533 0.7008272 0.4053203 0.775 900 1:1 5 0.6706587 0.7314564 0.6964614 0.
3974924 0.771 1000 1:1 5 0.6782219 0.73 0.7008272 0.
4049248 0.770Correlation Feature selection Features sizes ratio fold Sn Sp Ac Mcc AUC 100 1:1 5 0.672089 0.
6994048 0.6847426 0.3704882 0.749 200 1:1 5 0.6814664 0.7348445 0.7047335 0.4128747 0.
771 300 1:1 5 0.6857143 0.7392219 0.7090993 0.4215539 0.
780 400 1:1 5 0.6821327 0.7361478 0.7056526 0.4147781 0.777 500 1:1 5 0.
6810943 0.7330531 0.7038143 0.4108751 0.778 600 1:1 5 0.6837782 0.
7334377 0.7056526 0.41425 0.778 700 1:1 5 0.688317 0.7421218 0.7118566 0.
4270628 0.776 800 1:1 5 0.676045 0.7361634 0.
7017705 0.407902 0.772 900 1:1 5 0.6818923 0.
7347368 0.7049632 0.4132642 0.782 1000 1:1 5 0.
6696574 0.7189474 0.6911765 0.3854662 0.761 From above tables we see that all features are very important to show better performance. But when we take top 500 features in this study, we get better result(AUC score =0.
778, Sn=0.6799678, Sp=0.7386424, Ac=0.
705193, Mcc=0.4144777) than other number of features in gain feature selection method for 29 window size and when we take top 900 features in this study, we get better result(AUC score = 0.782, Sn=0.6818923,Sp=0.7347368, Ac=0.7049632, Mcc=0.4132642) than other number of features in correlation feature selection method for 29 window size.
So , we get from the result that the AUC score is improved by feature selection and correlation feature selection method for 1:1 ratio ,29 window size and 5 fold cross validation is important to identify s-nytrosylation sites. Table 4 : Performance of the 5-fold, 10-fold, 15-fold cross validation in 1:1 ratio and 29 window size for before feature selection and after feature selection on the training data set From above tables we see that without feature selection in 15–fold (AUC score =0.767, Sn=0.
6766763, Sp=0.7235815, Ac=0.6973805, Mcc=0.3974999) are very important to show better performance and after feature selection in 10–fold (AUC score =0.782 , Sn=0.6871921, Sp=0.7379958, Ac=0.7095588, Mcc=0.
4221419) are very important to show better performance. feature ratio window fold Sn Sp Ac Mcc AUC Before feature 1:1 29 5 0.6670797 0.709477 0.6858915 0.3741623 0.754 Before feature 1:1 29 10 0.
6672151 0.7116085 0.6868107 0.3762135 0.762 Before feature 1:1 29 15 0.6766763 0.7235815 0.
6973805 0.3974999 0.767 After feature 1:1 29 5 0.6818923 0.7347368 0.
7049632 0.4132642 0.782 After feature 1:1 29 10 0.6871921 0.7379958 0.7095588 0.
4221419 0.782 After feature 1:1 29 15 0.6894156 0.7433071 0.7130055 0.4293537 0.780Table 5: Performance of the 15-fold cross validation in 1:1 ratio and 29 window size for before feature selection and 10-fold cross validation in 1:1 ratio and 29 window size after feature selection on the test data set feature ratio window fold Sn Sp Ac Mcc AUC Before feature 1:1 29 15 0.7222222 0.
7629108 0.7408602 0.4834237 0.834 After feature 1:1 29 10 0.7698413 0.8192488 0.7924731 0.5870145 0.
870 From above tables we see that without feature selection in 15–fold (AUC score =0.834, Sn=0.7222222, Sp=0.7629108, Ac=0.7408602, Mcc=0.4834237) CV are very important to show better performance for both test and training dataset and after feature selection in 10–fold (AUC score =0.
870, Sn=0.7698413, Sp=0.8192488, Ac=0.7924731, Mcc=0.5870145) are very important to show better performance for both test and training dataset. Conclusions In this article, we designed a simple and efficient predictor s-nytrosylation for identifying s-nytrosylation sites. For class prediction in the independent dataset, we observed that our proposed predictor performed better than the existing predictors. Moreover, a feature selection was carried out to identify the significant features, which helps to better understand the important rules that underlie the s-nytrosylation proteins.
The data analysis results demonstrated that the proposed method might be helpful to understand s-nytrosylation as well as the mechanisms of protein s-nytrosylation. In addition, we also provide a new s-nytrosylation sites database that contains 1356 experimentally verified s-nytrosylation proteins with 2641 cysteine s-nytrosylation sites. If we take correlation feature selection method, 1:1 ratio, 29 window size and 10-fold cross validation then we get optimal performance to identify s-nytrosylation sites. References 1.
Hess DT, Matsumoto A, Kim SO, Marshall HE, Stamler JS (2005) Protein S-nitrosylation: purview and parameters. Nat Rev Mol Cell Biol 6: 150–166. 2. Cho DH, Nakamura T, Fang J, Cieplak P, Godzik A, et al.
(2009) S-nitrosylation of Drp1 mediates beta-amyloid-related mitochondrial fission and neuronal injury. Science 324: 102–105. 3. Guo CJ, Atochina-Vasserman EN, Abramova E, Foley JP, Zaman A, et al.(2008) S-nitrosylation of surfactant protein-D controls inflammatory function. PLoS Biol 6: e266.4. Hao G, Xie L, Gross SS (2004) Argininosuccinate synthetase is reversibly inactivated by S-nitrosylation in vitro and in vivo.
J Biol Chem 279: 36192– 36200. 5. Jaffrey SR, Snyder SH (2001) The biotin switch method for the detection of Snitrosylated proteins. Sci STKE 2001: pl1. 6. Hao G, Derakhshan B, Shi L, Campagne F, Gross SS (2006) SNOSID, a proteomic method for identification of cysteine S-nitrosylation sites in complex protein mixtures.
Proc Natl Acad Sci U S A 103: 1012–1017. 7. Lindermayr C, Saalbach G, Durner J (2005) Proteomic identification of Snitrosylated proteins in Arabidopsis. Plant Physiol 137: 921–930.
8. Abat JK, Mattoo AK, Deswal R (2008) S-nitrosylated proteins of a medicinal CAM plant Kalanchoe pinnata- ribulose-1,5-bisphosphate carboxylase/oxygenase activity targeted for inhibition. FEBS J 275: 2862–2872. 9. Palmieri MC, Lindermayr C, Bauwe H, Steinhauser C, Durner J (2010) Regulation of plant glycine decarboxylase by s-nitrosylation and glutathionylation. Plant Physiol 152: 1514–1528. 10.
Lindermayr C, Sell S, Muller B, Leister D, Durner J (2010) Redox regulation of the NPR1-TGA1 system of Arabidopsis thaliana by nitric oxide. Plant Cell 22: 2894–2907. 11. Begara-Morales JC, Sanchez-Calvo B, Chaki M, Valderrama R, Mata-Perez C, et al.
(2013) Dual regulation of cytosolic ascorbate peroxidase (APX) by tyrosine nitration and S-nitrosylation. J Exp Bot. 12. Li F, Sonveaux P, Rabbani ZN, Liu S, Yan B, et al. (2007) Regulation of HIF- 1alpha stability through S-nitrosylation. Mol Cell 26: 63–74. 13. Hernlund E, Kutuk O, Basaga H, Linder S, Panaretakis T, et al.
(2009) Cisplatin-induced nitrosylation of p53 prevents its mitochondrial translocation. Free Radic Biol Med 46: 1607–1613. 14. Ozawa K, Whalen EJ, Nelson CD, Mu Y, Hess DT, et al. (2008) S-nitrosylation of beta-arrestin regulates beta-adrenergic receptor trafficking. Mol Cell 31: 395–405.
15. Tsang AH, Lee YI, Ko HS, Savitt JM, Pletnikova O, et al. (2009) S-nitrosylation of XIAP compromises neuronal survival in Parkinson’s disease. Proc Natl Acad Sci U S A 106: 4900–4905.
16. Nott A, Watson PM, Robinson JD, Crepaldi L, Riccio A (2008) S-Nitrosylation of histone deacetylase 2 induces chromatin remodelling in neurons. Nature 455: 411–415 17. Foster MW, Hess DT, Stamler JS (2009) Protein S-nitrosylation in health and disease: a current perspective. Trends Mol Med 15: 391–404. 18.
Foster MW, McMahon TJ, Stamler JS (2003) S-nitrosylation in health and disease. Trends Mol Med 9: 160–168. 19.
Hoffmann, J, Dimmeler, S, Haendeler (2003) Shear stress increases the amount of S-nitrosylated molecules in endothelial cells: important role for signal transduction. J. FEBS Lett.
551:153-158 20. Sun JH, Xin CL, Eu JP, Stamler JS, Meissner G (2003) Cysteine-3635 is responsible for skeletal muscle ryanodine receptor modulation by NO. Proc. Natl. Acad.
Sci. U S A 98:11158-11162 21. Padgett CM, Whorton AR(1995) S-nitrosoglutathione reversibly inhibits GAPDH by S-nitrosylation.Am. J. Physiol. 269:739-74922. Fang M, Jaffrey SR, Sawa A, Ye K, Luo X, Snyder SH(2000) Dexras1: a G protein specifically coupled to neuronal nitric oxide synthase via CAPON.
Neuron 28:183-193 23.Iwakiri Y, Satoh A, Chatterjee S, Toomre DK, Chalouni CM, Fulton D, Groszmann RJ, Shah VH, Sessa WC(2006) Nitric oxide synthase generates nitric oxide locally to regulate compartmentalized protein S-nitrosylation and protein trafficking. Proc. Natl. Acad. Sci. U S A 103:19777-19782 24. Hess DT, Matsumoto A, Kim SO, Marshall HE, Stamler JS(2005) Thioredoxin is required for S-nitrosation of procaspase-3 and the inhibition of apoptosis in Jurkat cells.
Nat. Rev. Mol. Cell.
Biol. 6:150 166 25.Lei SZ, Pan Z-H, Aggarwal SK, Chen H-SV, Hartman J, Sucher NJ, Lipton SA(1992) Effect of nitric oxide production on the redox modulatory site of the NMDA receptor-channel complex. Neuron 8(6):1087-1099 26. Lipton SA, Choi Y-B, Pan Z-H, Lei SZ, Chen H-SV, Sucher NJ, Singel DJ, Loscalzo J, Stamler JS(1993) A redox-based mechanism for the neuroprotective and neurodestructive effects of nitric oxide and related nitroso-compounds. Nature 364(6438):626-632 27.Seth D, Hausladen A, Wang YJ, Stamler JS(2012) Endogenous protein S-Nitrosylation in E.
coli: regulation by OxyR. Science 336(6080):470-473 28. Malik SI, Hussain A, Yun BW, Spoel SH, Loake GJ(2011) GSNOR-mediated de-nitrosylation in the plant defence response. Plant Sci.
181(5):540-544 29. Stamler JS, Lamas S, Fang FC(2001) Nitrosylation. the prototypic redox-based signaling mechanism. Cell 106(6):675-683 30. Derakhshan B, Hao G, Gross SS(2007) Balancing reactivity against selectivity: the evolution of protein S-nitrosylation as an effector of cell signaling by nitric oxide. Cardiovasc. Res.
75(2):210-219 31. Singh RJ, Hogg N, Joseph J, Kalyanaraman B(1996) Mechanism of nitric oxide release from S-nitrosothiols. J. Biol. Chem. 271(31):18596-603 32. Liu L, Hausladen A, Zeng M, Que L, Heitman J, Stamler JS(2001) A metabolic enzyme for S-nitrosothiol conserved from bacteria to humans.
Nature 410(6827):490-4 33.Stoyanovsky DA, Tyurina YY, Tyurin VA, Anand D, Mandavia DN, Gius D, Ivanova J, Pitt B, Billiar TR, Kagan VE(2005) Thioredoxin and lipoic acid catalyze the denitrosation of low molecular weight and protein S-nitrosothiols. J. Am. Chem. Soc. 127:15815-23 34. Sengupta R, Ryter SW, Zuckerbraun BS, Tzeng E, Billiar TR, Stoyanovsky DA(2007) Thioredoxin catalyzes the denitrosation of low-molecular mass and protein S-nitrosothiols. Biochemistry 46:8472-83 35. Benhar M, Forrester MT, Hess DT, Stamler JS (2008) Regulated protein denitrosylation by cytosolic and mitochondrial thioredoxins. Science 320:1050-4 36. Gu Z, Kaul M, Yan B, Kridel SJ, Cui J, Strongin A, Smith JW, Liddington RC, Lipton SA(2002) S-nitrosylation of matrix metalloproteinases: signaling pathway to neuronal cell death. Science 297(5584):1186-90 37. Yao D, Gu Z, Nakamura T, Shi Z-Q, Ma Y, Gaston B, Palmer LA, Rockenstein EM, Zhang Z, Masliah E, Uehara T, Lipton SA(2004) Nitrosative stress linked to sporadic Parkinson’s disease: S-nitrosylation of parkin regulates its E3 ubiquitin ligase activity. Proc. Natl. Acad. Sci. U S A 101(29):10810-438. Uehara T, Nakamura T, Yao D, Shi Z-Q, Gu Z, Masliah E, Nomura Y, Lipton SA(2006) S-nitrosylated protein-disulphide isomerase links protein misfolding to neurodegeneration. Nature 2441(7092):513-7 39. Benhar M, Forrester MT, Stamler JS(2006) Nitrosative stress in the ER: a new role for S-nitrosylation in neurodegenerative diseases. ACS Chem. Biol. 1(6):355-8. 40. Cho D-H, Nakamura T, Fang J, Cieplak P, Godzik A, Gu Z, Lipton SA (2009) S-nitrosylation of Drp1 mediates beta-amyloid-related mitochondrial fission and neuronal injury. Science 324(5923):102-5 41.Schonhoff CM, Matsuoka M, Tummala H, Johnson MA, Estevéz AG, Wu R, Kamaid A, Ricart KC, Hashimoto Y, Gaston B, Macdonald TL, Xu Z, Mannick JB(2006) S-nitrosothiol depletion in amyotrophic lateral sclerosis. Proc. Natl. Acad. Sci. U S A 103(7):2404-9 42.Aranda E, López-Pedrera C, De La Haba-Rodriguez JR, Rodriguez-Ariza A(2012) Nitric oxide and cancer: the emerging role of S-nitrosylation. Curr. Mol. Med. 12(1):50-67 43. Switzer CH, Glynn SA, Cheng RY, Ridnour LA, Green JE, Ambs S, Wink DA(2012) S-nitrosylation of EGFR and Src activates an oncogenic signaling network in human basal-like breast cancer. Mol Cancer Res.Sep; 10(9):1203-15. 44.Hasan MM, Zhou Y, Lu X, Li J, Song J,Zhang Z (2015) Computational Identification of Protein Pupylation Sites by Using Profile-Based Composition of k-Spaced Amino Acid Pairs. PLoS ONE 10(6): e0129635. doi:10.1371/journal. pone.0129635 45. Stamler JS, Lamas S, Fang FC (2001) Nitrosylation. the prototypic redox-based signaling mechanism. Cell 106: 675–683. 46. Taldone FS, Tummala M, Goldstein EJ, Ryzhov V, Ravi K, et al. (2005) Studying the S-nitrosylation of model peptides and eNOS protein by mass spectrometry. Nitric Oxide 13: 176–187. 47. Marino SM, Gladyshev VN (2010) Structural analysis of cysteine S-nitrosylation: a modified acid-based motif and the emerging role of trans-nitrosylation. J Mol Biol 395: 844–859 48. Md. Mehedi Hasan, Shiping Yang, et,al SuccinSite: a computational tool for the prediction of protein succinylation sites by exploiting the amino acid patterns and properties. Mol. BioSyst., 2016, 12, 786 49. M. Gribskov and N. L. Robinson(1996) Use of receiver operating characteristic (ROC) analysis to evaluate sequence matching.Comput. Chem., 20, 25–33. 50. R. M. Centor(1991) Signal detectability: the use of ROC curves and their analyses. Medical Decision Making, 11, 102–106. 51. K. Chen, L. Kurgan and M. Rahbari(2007) Prediction of protein crystallization using collocation of amino acid pairs. Biochem. Biophys. Res. Commun., 355, 764–769. 52 K. Chen, L. A. Kurgan and J. Ruan(2007) Prediction of flexible/rigid regions from protein sequences using k-spaced amino acid pairs.BMC Struct. Biol., 7, 25. 53. Z. Chen, Y. Z. Chen, X. F. Wang, C. Wang, R. X. Yan and Z. Zhang (2011) Prediction of ubiquitination sites by using the composition of k-spaced amino acid pairs. PLoS One, 6, e22930.54. Y. Z. Chen, Y. R. Tang, Z. Y. Sheng and Z. Zhang (2008) Prediction of mucin-type O-glycosylation sites in mammalian proteins using the composition of k-spaced amino acid pairs. BMC Bioinf., 9, 101. 55. Z. Chen, Y. Zhou, Z. Zhang and J. Song (2015) Towards more accurate prediction of ubiquitination sites: a comprehensive review of current methods, tools and features.Briefings Bioinf., 16, 640–657. 56 K. Chen, L. A. Kurgan and J. Ruan(2008) Prediction of protein structural class using novel evolutionary collocation-based sequence representation.J. Comput. Chem., 29, 1596–1604. 57.Z. Chen, Y. Zhou, J. Song and Z. Zhang(2013) hCKSAAP_UbSite: improved prediction of human ubiquitination sites by exploiting amino acid pattern and properties. Biochim. Biophys. Acta, 1834, 1461–1467. 58. M. M. Hasan, Y. Zhou, X. Lu, J. Li, J. Song and Z. Zhang(2015) Computational Identification of Protein Pupylation Sites by Using Profile-Based Composition of k-Spaced Amino Acid Pairs .PLoS One, 10, e0129635. 59. L. BREIMAN(2001) SNP-based analysis of genetic substructure in the German population.Machine Learning, 45, 5–32. 60. C. Li, X. F. Wang, Z. Chen, Z. Zhang and J. Song(2015) Computational characterization of parallel dimeric and trimeric coiled-coils using effective amino acid indices. Mol. BioSyst., 11, 354–360. 61. Y. Li, M. Wang, H. Wang, H. Tan, Z. Zhang, G. I. Webb and J. Song(2014) Accurate in silico identification of species-specific acetylation sites by integrating protein sequence-derived and functional features. Sci. Rep., 4, 5765. 62. Y. Zhou, S. Liu, J. Song and Z. Zhang (2013) Structural Propensities of Human Ubiquitination Sites: Accessibility, Centrality and Local Conformation. PLoS One, 8, e83167. 63. V. Vacic, L. M. Iakoucheva and P. Radivojac(2006) Two Sample Logo: a graphical representation of the differences between two sets of sequence alignments.Bioinformatics, 22, 1536–1537.
