Candidate Single Nucleotide Polymorphism Markers for Arsenic Responsiveness of Protein Targets

Abstract

Arsenic is a toxic metalloid that causes skin cancer and binds to cysteine residues—a property that could be used to infer arsenic responsiveness of a target protein. Non-synonymous Single Nucleotide Polymorphisms (nsSNPs) result in amino acid substitutions and may alter arsenic binding with cysteine residues. Thus, the objective of this investigation was to identify and analyze nsSNPs that lead to substitutions to or from cysteine residues as an indication of increased or decreased arsenic responsiveness. We hypothesize that integration of data on molecular impacts of nsSNPs and arsenic-gene relationships will identify nsSNPs that could serve as arsenic responsiveness markers. We have analyzed functional and structural impacts data for 5,811 nsSNPs linked to 1,224 arsenic-annotated genes. In addition to the identified candidate nsSNPs for increased or reduced arsenic responsiveness, we observed i) a nsSNP that results in the breakage of a disulfide bond, as candidate marker for reduced arsenic responsiveness of KLK7, a secreted serine protease participate in normal shedding of the skin; and ii) 6 pairs of vicinal cysteines in KLK7 protein that could be binding sites for arsenic. In summary, our analysis identified non-synonymous SNPs that could be used to evaluate responsiveness of a protein target to arsenic. In particular, an epidermal expressed serine protease with crucial function in normal skin physiology was prioritized on the basis of abundance of vicinal cysteines for further research on arsenic-induced keratinocyte carcinogenesis.

Introduction

Arsenic (As) is recognized as an environmental toxicant of concern for global public health and a leading cause of toxicity and carcinogenicity.^1,2 Arsenic targets the human skin and long-term exposure to arsenic, principally through drinking water, has been correlated with increased risk of skin cancer.^3–5 The cellular toxicity of arsenic has been well documented from case studies of poisoning incidents and medicinal use.^2,6 However, due to increased epidemiological reports of arsenic related cancers in places such as Southeastern Michigan (USA), Taiwan, China, India and Bangladesh, public health concerns about long-term exposure have arisen.^2,6–8 Inorganic arsenic is classified by the United States Environmental Protection Agency (U.S. EPA) as a Group A carcinogen based on sufficient evidence of carcinogenicity in humans.⁹ Chronic oral exposure to inorganic arsenic can have adverse effects on tissues in the human body systems.¹⁰ However, the human skin is the critical organ of arsenic toxicity because arsenic has a strong affinity for the keratin proteins which are rich in the sulphur containing cysteine residues.^11–13 Chronic exposure to arsenic induces sequential changes in the skin epithelium, proceeding from hypopigmentation to hyperkeratosis which may eventually lead to skin cancer.¹¹ Arsenic-induced skin lesions are early warning markers for development of cancers in internal organs.^14,15 A well-known beneficial use of arsenic is that arsenic trioxide is used for treatment of relapsed or refractory acute promyelocytic leukemia.^16–20 However, side effects of treatment with arsenic trioxide include significant adverse cardiac effects.²¹

The avalanche of genome sequences combined with genome-enabled datasets from high-throughput gene expression, genotyping, haplotyping and protein assays is making it possible to gain biological insights into previously unknown gene-toxicant interactions. Arsenicogenomics, an aspect of toxicogenomics, therefore, provides a means to i) understand how various genes respond to arsenic and ii) how arsenic modifies the function and expression of specific genes in the genome. Early physical manifestations of arsenic toxicity in endemic areas are skin lesions including melanosis and keratosis. However, not everyone exposed to arsenic in an endemic region would develop skin lesions.²² Therefore, future research on arsenic-induced cancers and, in particular, skin lesions should consider the impact of genetic variation in individual susceptibilities to arsenic toxicity.

Single nucleotide alterations in the DNA sequence represent a major source of genetic heterogeneity²³ and the most common type of genetic variation in the human genome. The diversity of single nucleotide polymorphisms (SNPs) derived from arsenic responsive genes in different populations could provide biomarkers for an individual’s susceptibility to arsenic-induced diseases. Genomic and bioinformatics techniques now exist to identify and analyze the presence of SNPs in populations.²⁴ Furthermore, the dense distribution of SNPs across the genome makes them ideal markers for large-scale genome-wide association studies to discover genes in common complex diseases, such as cancer. A SNP-induced amino acid substitution in the coding region can be broadly divided into synonymous (no change in amino acid) or non-synonymous (change in amino acid).²⁵ Furthermore, the functional impact of the SNP on protein function has been described as deleterious (disruptive) or non-deleterious (benign/neutral).²⁶ Non-synonymous substitutions could lead to missense or nonsense mutations in the encoded polypeptide. In particular, nonsense mutations that generate premature termination codons (PTCs) are responsible for approximately one-third of human genetic diseases.²⁷ In addition, substitutions in one or two amino acids of a protein sequence can alter the quantity of encoded protein during expression in mammalian cells.²⁸ Considering that arsenic trioxide is also used for treatment of acute promyelocytic leukemia,^16,18,19,29 genetic variation may also affect response to therapy.

Arsenic binds to sulfhydryl (SH) groups of cysteine (Cys) residues to form arsenic-thiol linkages, a property that could be used to infer arsenic responsiveness of a protein target as well as contribute to oxidative and protein folding stresses.^30–32 In the human arsenic (+3 oxidation state) methyltransferase (hAS3MT) sequence, Cys residues at positions 156, 206 and 250 play important roles in the enzymatic function and structure.³³ Mutation of the arsenic-sensing Cys151 in Kelch-like ECH-associated protein 1 (Keap1) abolished arsenic activation of nuclear factor erythroid 2-related factor 2, a transcription factor responsible for induction of antioxidative cytoprotective genes.³⁴

Non-synonymous Single Nucleotide Polymorphisms (nsSNPs) result in amino acid substitutions and may alter the number of cysteine residues available to arsenic for binding to a protein cellular target. Therefore, the objective of this investigation was to identify and analyze nsSNPs that lead to substitutions to or from cysteine residues as an indication of increased or decreased arsenic responsiveness of a protein. We hypothesize that integration of data on molecular impacts of non-synonymous single nucleotide polymorphisms and arsenic-gene relationships will help identify nsSNPs that are candidate arsenic responsiveness markers.

An integrative approach combining results from selected web-based toxicogenomics and genomics databases as well as bioinformatics tools was used to prioritize candidate nsSNPs markers for arsenic responsiveness of protein targets. In the first step, a list of genes annotated to interact with arsenicals was retrieved from the Comparative Toxicogenomics Database.³⁵ Subsequently, the nsSNPs linked to these arsenic-annotated genes were extracted from SNPs3D³⁶ and analyzed for functional and structural impacts data on protein isoforms. Significant amino acid substitutions to or from cysteine residues were then prioritized based on structural effect resulting in breakage of a disulphide bond as well as function in skin cells. Furthermore, structural homology modeling was used to identify vicinal (neigboring) cysteines in prioritized protein targets of arsenic. In order to facilitate additional investigations on prioritized SNPs, protein targets and molecular mechanisms of arsenic action, we have constructed a collection of over 100,000 sentences from over 16,000 PubMed³⁷ abstracts on arsenic. Finally, a web resource Arsenic Sentence Database was developed to enable web-based search of the sentences by keywords and PubMed identifiers.

Methods

Functional and structural impacts of single nucleotide polymorphisms on arsenic annotated genes

The molecular functional effects of non-synonymous SNPs based on sequence and structure analysis were retrieved from the SNPs3D web resource and database³⁶ for genes curated in the Comparative Toxicogenomics Database (CTD)³⁵ to have a relationship with arsenic. In the CTD, the term gene also includes mRNA and proteins. We described relationship in terms of arsenic modifying the function and/or expression of genes. Furthermore, we referred to the genes as arsenic-responsive genes or proteins. In SNPs3D, the classification into in vivo functional impact categories of the SNP was based on two Support Vector Machine (SVM) models: protein sequence conservation profiling and protein structure stability. In both machine learning models, an SVM is trained using 5 sequence profiles and 15 protein stability features. Additional details on the methods are available at the SNPs3D website http://www.snps3d.org/help/method.html. The nsSNPs were ranked according to SVM score. For both sequence and structure SVM scores, a nsSNP with negative score was classified as deleterious while a nsSNP with a positive score was classifed as non-deleterious. We observed that some nsSNPs in SNPs3D were assigned a SVM score of −0.00. However, they were not tagged as deleterious. Thus, these nsSNP were classified in this investigation as non-deleterious. Furthermore, High Confidence (HC) SVM scores were greater than 0.50 or less than −0.50.³⁸

The computational workflow consisting of a suite of customized Perl and Unix scripts was developed to process results obtained from CTD and SNPs3D. The Entrez Gene³⁷ Identifiers and Gene Symbols were extracted from XML formatted results of arsenic-gene interactions from CTD. Furthermore, the Entrez Gene identifiers were then used to remotely download the SNP Analysis page in SNPs3D for each CTD arsenic-annotated gene. For example, the SNP Analysis page for a known arsenic-interacting gene Glutathione S-transferase Omega 1 (Gene Symbol: GSTO1 and Entrez Gene Identifier: 9446) in SNPs3D is http://www.snps3d.org/modules.php?name=SnpAnalysis&locus_ac=9446.

The collection of html files was processed to mine for relevant data to construct a dataset. The fields of the dataset were the dbSNP identifier,³⁷ RefSeq protein isoform identifier,³⁷ mutation, SVM sequence profile score, SVM structure score and description of impact of nsSNP on protein stability. There were instances that no data were predicted for the SVM structure score and the protein impact. However, scores were predicted for all the SVM sequence profile. The gene symbols were also extracted from the CTD and combined with the dataset from SNPs3D. The final integrated dataset consisted of the Entrez Gene Identifier, Gene Symbol, the SNP Identifier, the Amino Acid Substitution, the SVM score computed for the sequence and structure profile models and the structural consequence of the SNP. In SNPs3D, we preferred to remotely download the SNP Analysis pages so as to extract data and links to additional information such as i) sequence alignment evidence of tolerance of the amino acid position to mutation and ii) values associated with the 15 protein stability factors. These additional datasets were not available in the files available for download.

Arsenic responsiveness based on substitution to or from cysteine residues

We conjectured that substitution to or from a cysteine residue could lead to increased or decreased responsiveness of a protein isoform to arsenic. In order to identify these cysteine substitutions, a suite of customized Perl and Unix scripts was developed to extract records that met the criteria from the integrated CTD and SNPs3D dataset. For example, in SNPs3D SNP Analysis page for GSTO1, nsSNP rs45529437 is linked to substitution from cysteine to tyrosine in position 32 (C32Y). Furthermore, rs11509436 is linked to substitution to cysteine from serine in position 86 (S86C).

Structural homology modeling

Structural models of protein mutants link to candidate SNPs that alter arsenic responsiveness were generated using MODELLER 7v7³⁹ with appropriate homologous high resolution X-ray crystal structure templates from the Protein Data Bank (PDB).⁴⁰ SYBYL (Tripos Inc) was used to identify vicinal cysteines following a quick minimization routine using AMBER force field.

Construction of sentences collection from PubMed abstracts on arsenicals

In order to identify descriptors of interest in sentences and cluster sentences with identical descriptors, we implemented a sentence splitting algorithm on a collection of PubMed³⁷ abstracts annotated with at least one of the Medical Subject Heading (MeSH) terms: arsenic or arsenicals. The sentence splitting algorithm implemented uses Perl regular expressions to enhance the Comprehensive Perl Archive Network (CPAN) Text: Sentence splitter module (http://search.cpan.org/) to achieve a high accuracy in sentence disambiguation. A web interface for searching the catalog of sentences was also developed.

Results

Functional and structural impacts of single nucleotide polymorphisms on arsenic annotated genes

The set of genes for predicting potential responsiveness to arsenic was obtained from the Comparative Toxicogenomics Database (CTD).³⁵ A total of 1,604 genes consisting of 1,492 human genes and 112 non-human genes documented to have a relationship with arsenicals (Medical Subject Heading Identifier [MeSH ID]: D001152) were retrieved on May 10, 2010 (Supplementary Data). We describe the gene set as a list of arsenic-annotated genes and their protein isoforms as arsenic-annotated proteins. The functional impacts of SNPs on protein function, as predicted by support vector machine (SVM), were retrieved from SNPs3D.³⁶ Support Vector Machines (SVM) are supervised machine learning techniques that have been applied to numerous classification tasks to predict the class of an example based on training examples. The SNPs3D database provides an SVM profile score for the functional impact (deleterious or non deleterious) of an nsSNP as well as 3-dimensional protein structure of the impact of the nsSNP on protein stability. According to Yue et al³⁶ the SVM used in SNPs3D is trained on monogenic disease data, thus deleterious is defined as “sufficiently damaging to protein function in vivo as to be consistent with a monogenic disease outcome”. A screenshot of a section of SNPs3D page for a gene is presented in Figure 1. The pages of the arsenic-annotated genes available in the SNPs3D database were the data source for extracting relevant data. The protein stability impacts in SNPs3D were i) Four classes of electrostatic interaction: reduction of charge–charge, charge–polar or polar–polar energy, or introduction of electrostatic repulsion; ii) three solvation effects: burying of charge or polar groups, and reduction in non-polar area buried on folding; iii) and two terms representing steric strain: backbone strain and over-packing; cavity formation (affecting van der Waals energy); iv) and loss of a disulfide bridge.

Figure 1.

Screenshot of a SNPs3D page for a gene. The functional and structural impacts, molecular effect and frequency of non-synonymous SNPs associated with the protein isoforms (RefSeq accession) is documented on the page. The negative SVM score (value in red) indicates a deleterious substitution.

From SNPs3D, we extracted the Entrez Gene Identifier, the SNP Identifier, the Amino Acid Substitution and the SVM score computed for the sequence and structure profile models. The dataset constructed consisted of 5,811 nsSNPs linked to 1,224 arsenic-annotated genes. Furthermore, a total of 8,992 nsSNP-induced substitutions (3,700 deleterious, 5,292 non-deleterious) were linked to 1,872 protein isoforms in the National Center for Biotechnology Information (NCBI) Reference Sequence Database.³⁷ The SVM scores observed for the substitutions ranged from −4.73 to 6.59 with 743 unique scores (Fig. 2). Of the 3,700 nsSNP-predicted substitutions, there were 2,739 high confidence deleterious substitutions (SVM sequence profile score <−0.5) linked to 745 genes, 1,785 nsSNPs and 1,094 protein isoforms. Furthermore, there were 4,191 high confidence non-deleterious substitutions (SVM sequence profile score >0.5) linked to 964 genes, 2,829 and 1,459 nsSNPs. A summary of the dataset is presented in Table 1. In order to facilitate selection of nsSNPs according to confidence of SVM score, we classed substitutions into categories with a 0.5 interval (Table 2).

Figure 2.

Plot of frequencies of unique Support Vector Machine (SVM) scores for dataset of non-synonymous SNPs linked to arsenic-annotated genes. The SVM score predicted for each nsSNP substitution was extracted from the SNPs3D page.

Table 1.

Summary of datasets.

Dataset	Count
Arsenic-annotated genes Retrieved from CTD	1,604
Arsenic-annotated genes with nsSNPs in SNPs3D	1,224
Non-synonymous SNPs	5,811
Arsenic-annotated protein isoforms	1,872
Amino acid substitutions	8,992
Deleterious substitutions	3,700
Non-deleterious substitutions	5,292
High confidence deleterious substitutions*	2,739
High confidence non-deleterious substitutions**	4,191

SVM sequence profile score:

^*SVM <−0.5;

^**SVM >0.5

Table 2.

Distribution of SVM scores for nsSNP substitutions.

SVM score class*	Non-deleterious		Deleterious
	Frequency	Percent frequency	Frequency	Percent frequency
0.0–0.5	1083	20.46%	977	26.41%
0.5–1.0	1186	22.41%	831	22.46%
1.0–1.5	1170	22.11%	627	16.95%
1.5–2.0	826	15.61%	448	12.11%
2.0–2.5	560	10.58%	348	9.41%
2.5–3.0	246	4.65%	255	6.89%
3.0–3.5	126	2.38%	132	3.57%
3.5–4.0	49	0.93%	63	1.70%
4.0–4.5	25	0.47%	18	0.49%
4.5–5.0	12	0.23%	1	0.03%
5.0–5.5	2	0.04%
5.5–6.0	3	0.06%
6.0–6.5	3	0.06%
6.5–7.0	1	0.02%

Note:

^*Absolute values.

Arsenic responsiveness based on substitution to or from cysteine residues

A total of 196 nsSNPs linked to 144 genes and 225 protein isoforms were observed to cause substitutions to cysteine residues. In the case of substitutions from cysteine residues, 92 nsSNPs linked to 79 genes and 122 protein isoforms were observed. In the protein isoforms analyzed, the substitutions to or from cysteine residues were restricted to the following six amino acid residues: Phenylalanine (F), Glycine (G), Arginine (R), Serine (S), Tryptophan (W) and Tyrosine (Y).

Four classes of nsSNPs were identified on the basis of significant deleterious and non-deleterious effects on protein function as well as SNP-associated residue changes to or from cysteine (Table 3). The 111 nsSNPs that resulted in non-deleterious substitutions to or from cysteine are candidates for evaluating increased or decreased responsiveness to arsenic, respectively (Supplementary Data). A set of nsSNPs that mutates the residue to or from cysteine with significant impact on protein function and structure (with agreement of both SVM scores for sequence and structure profiles) are presented in Table 4 and Table 5, respectively. All the identifiers for SNP are from the dbSNP and begin with “rs”.

Table 3.

Categories of nsSNPs observed in analysis of arsenic-annotated genes.

Category	Gene count	SNP count	Protein isoforms
From Cys, deleterious	45	53	67
From Cys, non-deleterious	36	39	58
To Cys, deleterious	101	130	145
To Cys, non-deleterious	63	72	108

Table 4.

Candidate nsSNPs that increase protein’s arsenic responsiveness through amino acid change to cysteine.

Entrez gene	Gene symbol	nsSNP	Protein isoform	Mutation	SVM sequence score	SVM structure score	Impac on protein stability**
834	CASP1	3203613	NP_150635	S33C	2.76	1.33	PS
834	CASP1	3203613	NP_150634	S126C	2.36	1.33	PS
11200	CHEK2	28909981	NP_665861	S471C	2.20	1.13	PS; HBL
834	CASP1	3203613	NP_150636	S33C	1.99	1.33	PS
8644	AKR1C3	35575889	NP_003730	R170C	1.82	1.59
55713	ZNF334	41283032	NP_955473	R237C	1.71	1.08	PS
55713	ZNF334	41283032	NP_060572	R275C	1.68	1.08	PS
6389	SDHA	1041948	NP_004159	S346C	1.62	0.92	HBL
2877	GPX2	17880492	NP_002074	R146C	1.61	0.84	SBL
983	CDC2	8755	NP_203698	R59C	1.53	1.22	PS
10935	PRDX3	11554910	NP_054817	Y53C	1.19	0.99	PS
10935	PRDX3	11554910	NP_006784	Y71C	1.14	0.99	PS
5265	SERPINA1	1802962	NP_000286	S325C	0.88	1.37	PS
5265	SERPINA1	1802962	NP_001002235	S325C	0.88	1.37	PS
5265	SERPINA1	1802962	NP_001002236	S325C	0.88	1.37	PS
983	CDC2	8755	NP_001777	R59C	0.81	1.22	PS
4233	MET	34589476	NP_000236	R970C	0.77	0.99	PS

^*Abbreviations: PS, On the protein surface; HBL, hydrogen bond lost; SBL, saltbridge lost.

Table 5.

Candidate nsSNPs that reduce protein’s arsenic responsiveness through amino acid change from cysteine.

Entrez gene	Gene symbol	nsSNP	Protein isoform	Mutation	SVM sequence score	SVM structure score	Impact on protein stability*
23660	ZKSCAN5	28411998	NP_055384	C579W	−3.63	−0.84	BP; OP
23660	ZKSCAN5	28411998	NP_659570	C579W	−3.63	−0.84	BP; OP
7465	WEE1	17854721	NP_003381	C379R	−3.62	−1.18
9753	ZSCAN12	2232432	NP_001034732	C332R	−3.59	−1.11	ESR
9446	GSTO1	45529437	NP_004823	C32Y	−3.53	−1.20	OP
5650	KLK7	17855561	NP_644806	C226W	−2.44	−1.20	OP; BDB
5650	KLK7	17855561	NP_005037	C226W	−2.43	−1.20	OP; BDB
5879	RAC1	7673785	NP_008839	C157R	−2.06	−1.24	BC; OP
3945	LDHB	3575	NP_002291	C294Y	−1.83	−1.05	BP; OP
5879	RAC1	7673785	NP_061485	C176R	−1.77	−1.52	OP
5265	SERPINA1	8350	NP_000286	C256W	−1.76	−0.51	BP
5265	SERPINA1	8350	NP_001002235	C256W	−1.76	−0.51	BP
5265	SERPINA1	8350	NP_001002236	C256W	−1.76	−0.51	BP
2868	GRK4	35824641	NP_892027	C215R	−1.66	−0.79	ESR; OP
2868	GRK4	35824641	NP_001004057	C215R	−1.55	−0.79	ESR; OP
835	CASP2	11551881	NP_116764	C370F	−1.54	−0.99	OP
2868	GRK4	35824641	NP_001004056	C183R	−1.51	−0.80	ESR; OP
10465	PPIH	11550298	NP_006338	C131R	−1.28	−1.41	BC; ESR; OP
3107	HLA-C	41563216	NP_002108	C125R	−1.15	−1.28	BC; OP; BDB
3107	HLA-C	41543517	NP_002108	C188W	−0.81	−0.86	BP; OP; BDB

^*Abbreviations: BP, buriedpolar; BC, buriedcharged; BDB, breakage of a disulfide bond; ESR, electrostaticrepulsion; OP, overpacking; SBL, saltbridge lost.

Structural homology modeling

In SNPs3D, breakage of a disulfide bond is assigned to any mutation that replaces a cysteine residue in an S–S bond with a non-cysteine residue. Eight nsSNPs from 6 genes (9 protein isoforms) were identified to result in breakage of a disulfide bond (Table 6). The impact of these nsSNPs on protein stability has pointed us to potential regions of arsenic binding to vicinal (neigboring) cysteines. Since, we are interested in arsenic-induced skin cancer, we further analyzed the stratum corneum chymotryptic serine protease KLK7 (kallikrein-related peptidase 7) for annotated structural consequences of SNP marker for reduced responsiveness, the effects of arsenic on expression as well as distribution of cysteine residues. A screenshot of SNPs3D page on structural impact of nsSNP rs17855561 is presented in Figure 3. The nsSNP rs17855561 in both KLK7 protein isoforms is predicted to result in breakage of a disulfide bond and potentially reducing responsiveness to arsenic by changing the Cys in position 226 to Tryptophan (W) (Table 6). Furthermore, according to data extracted from Bae et al⁴¹ by CTD curators, sodium arsenite results in decreased expression of KLK7 mRNA in the virally immortalized human keratinocyte cell line RHEK-1. Structures of KLK7 from protein sequences NP_005037 and NP_644806 were generated using high resolution X-ray crystal structure of human kallikrein (PDB ID: 2QXI). Structural homology models of wild type KLK7 structure revealed six cysteine pairs 36–165; 55–71; 137–239; 144–211; 176–190 and 201–226 (Fig. 4).

Table 6.

Non-synonymous SNPs that predict potential region of arsenic-binding to vicinal cysteines.

Gene symbol	nsSNP	Protein isoform	Mutation
FST	1127760	NP_006341	C239S
FST	1127760	NP_037541	C239S
HLA-C	41563216	NP_002108	C125R
HLA-C	41543517	NP_002108	C188W
HLA-C	41562916	NP_002108	C188F
IL4	4986964	NP_000580	C27R
IL4	4986964	NP_758858	C27R
KLK7	17855561	NP_005037	C226W
KLK7	17855561	NP_644806	C226W
TFRC	9852079	NP_003225	C363W
TLR4	2770145	NP_612564	C306W

Figure 3.

Screenshot of predicted structural impact of nsSNP rs17855561 on human kallikrein-7 preproprotein (NP_005037). The nsSNP results in steric strain (over-packing) and breakage of a disulfide bond by changing Cysteine (C) residue in position 226 to a Tryptophan (W).

Figure 4.

Predicted structure of human kallikrein-7 preproprotein (NP_005037). The vicinal cysteine pairs are 36–165; 55–71; 137–239; 144–211; 176–190; 201–226. The homology structure shows that the Cys201 pairs with Cys226. Arsenic is known to reduce the expression of KLK7 in an epidermal cell line.

Construction of sentence collection from PubMed abstracts on arsenicals

In order to facilitate further studies on these identified genes, single nucleotide polymorphisms and other aspects of arsenic, we have segmented 16,057 PubMed abstracts into a collection of 108,235 sentences. The abstracts were selected based on annotation of the abstract in PubMed with at least one of the Medical Subject Heading (MeSH) terms: arsenic or arsenicals. An Arsenic Sentence Database that facilitates query of the sentences with keywords as well as retrieval of sentences for specific PubMed abstracts is available at http://compbio.jsums.edu/arsenic_pubmed.

The utility of the database was demonstrated by a search for “GSTO1” the symbol for gene encoding the enzyme glutathione S-transferase omega 1 which catalyzes the monomethyl arsenate reduction, the rate-limiting step for inorganic arsenic biotransformation in humans.^42,43 A cluster of 80 sentences containing the symbol were retrieved from the database. Furthermore, a subset of the GSTO1 sentences and containing the word “polymorphism” allowed us to identify 13 sentences from 7 PubMed abstracts (Table 7). These abstracts were on genetic variation observed in GSTO1 and other genes involved in arsenic metabolism.

Table 7.

Cluster of sentences from PubMed abstracts on GSTO1 polymorphism.

Sentence identifier*	Sentence text
18216717_10	The percent of inorganic arsenic in the urine of 205 Chilean participants showed a bimodal distribution that was not associated with the Ala140Asp, Glu155del or Ala236Val polymorphisms in GSTO1-1.
19686770_5	Genotyping of CYP2E1, GSTO1 and GSTO2 was determined by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP).
16638819_4	We identified 31 and 66 polymorphisms in GSTO1 and GSTO2, respectively, with four nonsynonymous-coding single nucleotide polymorphisms (cSNPs) in each gene.
12928150_5	We screened two genes responsible for arsenic metabolism, human purine nucleoside phosphorylase (hNP), which functions as an arsenate reductase converting arsenate to arsenite, and human glutathione S-transferase omega 1-1 (hGSTO1-1), which functions as a monomethylarsonic acid (MMA) reductase, converting MMA(V) to MMA(III), to develop a comprehensive catalog of commonly occurring genetic polymorphisms in these genes.
12928150_9	In hGSTO1-1, 33 polymorphisms were observed.
12928150_11	In contrast to hNP, in which the IA group was more polymorphic than the EA group, in hGSTO1-1 the EA group was more polymorphic than the IA group, which had only 1 polymorphism with a frequency >10%.
19635583_4	This study was conducted to investigate the relationship between polymorphisms in the GSTO1 and GSTO2 genes and arsenic metabolism and oxidative stress status in Chinese populations chronically exposed to different levels of arsenic in drinking water.
19635583_5	Two polymorphisms (GSTO1*A140D and GSTO2*N142D) with relatively higher mutation frequencies in the Chinese population were determined by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP).
19635583_8	Multivariate analysis revealed that there was no association between the urinary profile or oxidative stress status and the polymorphism of GSTO1*A140D or GSTO2*N142D.
19635583_9	Collectively, polymorphisms in GSTO1 or GSTO2 do not appear to contribute to the large individual variability in arsenic metabolism or susceptibility to arsenicosis.
14680363_3	To understand this variability, we studied the relationship between polymorphisms in the gene for human monomethylarsonic acid (MMA(V)) reductase/hGSTO1 and the urinary arsenic profiles of individuals chronically exposed to arsenic in their drinking water.
14680363_9	These two subjects had the same unique polymorphisms in hGSTO1 in that they were heterozygous for E155del and Glu208 Lys.
18414634_9	RESULTS: Among four candidate genes, PNP, As3MT, GSTO1, and GSTO2, we found that distribution of three exonic polymorphisms, His20His, Gly51Ser, and Pro57Pro of PNP, was associated with arsenicism.

Note:

^*Sentence identifier consists of PubMed identifier (PMID) and the location of the sentence in the abstract with abstract title as the first sentence.

Discussion

Since amino acid substitutions to or from cysteine residues attributed to SNPs might be a determinant of responsiveness of target proteins to arsenic and possibly arsenic-induced skin cancer, we undertook to prioritize genes and SNPs to understand keratinocyte carcinogenesis resulting from arsenic exposure. Our analysis of the functional and structural impacts of 5,811 nsSNPs associated with 1,224 putative arsenic responsive genes identified i) 196 candidate nsSNPs for increased arsenic responsiveness by substitutions to cysteine residues for 144 genes; ii) 92 candidate nsSNP for decreased arsenic responsiveness by substitutions from cysteine residues for 79 genes; iii) nsSNP rs17855561 that results in breakage of a disulfide bond, as candidate marker for reduced arsenic responsiveness in KLK7, a secreted serine protease that has been demonstrated to participate in normal shedding of the skin⁴⁴ and iv) 6 pairs of vicinal cysteines in KLK7 protein.

The bioinformatics analysis pipeline identified genes with evidence for potential SNP-induced increased or decreased responsiveness to arsenic. To the best of our knowledge this report is the first large-scale analysis of SNP-induced substitutions evaluating the abundance of cysteines in putative protein targets of arsenic. A recent large-scale analysis of the curation of chemical-gene relationships from biomedical literature has provided over 1,400 genes whose activity were perturbed by arsenic in a variety of conditions and/or cell types.³⁵ Our analysis extends the curation efforts by CTD by integrating SNP data that could help understand the molecular mechanisms of arsenic action in diverse cell types including keratinocytes.

We have used the structural impact annotation “breakage of a disulfide bond” as an evidence of the presence of potential arsenic-binding vicinal cysteines in a protein sequence. The function and structure of proteins are often determined by Cys residues since many proteins folding are dependent on disulfide bonds.³³ Arsenic binding to target protein depends on the number, accessibility and relative positioning of Cys residues.⁴⁵ Furthermore, the ability for trivalent arsenicals to bind to thiol groups of biomolecules is an accepted mechanism for being more toxic than pentavalent arsenicals. The mRNA from the tissue serine protease KLK7 that was identified by our pipeline was down-regulated by arsenic in human keratinocyte cell line RHEK-1.⁴¹ The proposed functions of KLK7 in the normal skin physiology include i) activating interleukin 1 beta (IL-1b) and ii) basal permeability barrier function of stratum corneum by degrading two major lipid processing enzymes beta-glucocerebrosidase and acidic sphingomyelinase.^44,46,47 The crucial function of KLK7 in normal skin function and potential perturbation by arsenic justifies a need to determine the potential energy of each Cys residues in KLK7 combined with their proximity to enzyme active sites or other functional regions of the protein. We hypothesize that arsenic binds to at least one of the 6 pairs of vicinal cysteines resulting in conformational changes that down regulate KLK7 function. Our hypothesis for KLK7 can be tested using similar experiments conducted to determine the role of the 8 Cys residues in arsenic binding for human beta-tubulin.³² In this investigation, we have verified the sequence-based prediction of vicinal cysteine with structural homology modeling.

The functional implications of SNP modified polypeptides of KLK7 warrant further investigation. Our text mining approach using the gene symbol GSTO1 as a search term in a collection of over 100,000 sentences from over 16,000 PubMed abstracts retrieved publications that could guide further research on the impact of arsenic as well as SNP-induced polymorphism on KLK7 function. In the case of the two enzymes Glutathione S-transferase omega 1 and omega 2 (GSTO1 and GSTO2) that catalyze monomethyl arsenate reduction, variant allozymes have been shown to degrade more rapidly than their respective wild type allozymes.⁴⁸ Similar protein degradation experiments for KLK7, could unravel the impact of molecular differences resulting in susceptibility of arsenic-induced skin cancer in arsenicosis-endemic populations.

Conclusions

Single nucleotide polymorphisms (SNPs) can alter the physico-chemical properties of proteins. The susceptibility to arsenic-induced diseases as well as response to arsenic-based drugs has been linked to single nucleotide polymorphisms. Our analysis identified non-synonymous SNPs that could be used to evaluate responsiveness of a protein target to arsenic. Furthermore, an epidermal expressed serine protease KLK7 with crucial function in normal skin physiology was prioritized on the basis of abundance of vicinal cysteines to understand arsenic-induced keratinocyte carcinogenesis.