Am J Transl Res 2019;11(5):2683-2705 www.ajtr.org /ISSN:1943-8141/AJTR0093089

Original Article The analyses of SRCR based on -protein interaction network in esophageal squamous cell carcinoma

Zepeng Du1,2, Qiaoxi Xia6, Bingli Wu6, Jiyu Ding6, Yan Zhao2, Ling Lin6, Mantong Chen6, Zhixiong Cai3, Shaohong Wang2, Liyan Xu7, Enmin Li6, Zhiyong Wu4, Yun Li1, Haixiong Xu5, Dong Yin1

1Guangdong Provincial Key Laboratory of Malignant Tumor Epigenetics and Genes Regulation, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, Guangzhou 510120, China; Departments of 2Pathology, 3Cardiology, 4Surgical Oncology, 5Neurosurgery, Shantou Central Hospital, Affiliated Shantou Hospital of Sun Yat-sen University, Shantou 515041, China; 6Department of Biochemistry and Molecular Biology, 7Institute of Oncologic Pathology, Shantou University Medical College, Shantou 515041, China Received February 27, 2019; Accepted April 29, 2019; Epub May 15, 2019; Published May 30, 2019

Abstract: The scavenger receptor cysteine-rich (SRCR) , with one to several SRCR domains, play important roles in human diseases. A full view of their functions in esophageal squamous cell carcinoma (ESCC) remain unclear. Sequence alignment and phylogenetic tree for all human SRCR domains were performed. Differentially- expressed SRCR genes were identified in ESCC, followed by protein-protein interaction (PPI) network construction, topological parameters, subcellular distribution, functional enrichment and survival analyses. The variation of con- served cysteines in each SRCR domain suggested a requirement for new classification of the SRCR family. Six genes (LGALS3BP, MSR1, CD163, LOXL2, LOXL3 and LOXL4) were upregulated, and four genes (DMBT1, PRSS12, TMPRSS2 and SCARA5) were downregulated in ESCC. These 10 SRCR genes form a unique biological network. Functional enrichment analyses provided important clues to investigate the biological functions for SRCR net- work in ESCC, such as extracellular structure organization and the PI3K-Akt signaling pathway. Kaplan-Meier curves confirmed that high expression of SCARA5, LOXL2, LOXL3, LOXL4 were related to poor survival, whereas high expres- sion of DMBTI and PRSS12 showed the opposite result. SRCR genes promote the development of ESCC through its network and could serve as potential prognostic factors and therapy targets of ESCC.

Keywords: SRCR gene, protein-protein interaction network, sequence analysis, functional enrichment, esophageal squamous cell carcinoma

Introduction teristics of the conserved cysteines in the SR- CR domains, SRCR genes are usually classified The scavenger receptor cysteine-rich (SRCR) into two types: type A genes, which contain six superfamily was first defined by Goldsteinet al. cysteine residues that are encoded by at least in 1979, describing their activity in the uptake two exons, and type B genes, which have eight of modified low-density lipoprotein (LDL) [1]. cysteine residues that generated from a single These SRCR genes were described upon the exon [3]. These cysteine residues participate in identification of a series of receptors contain- intra-domain disulfide bonds, creating the pro- ing one or several extracellular domains homol- tein structure for substrate recruitment. ogous to the cysteine-rich C-terminal domain of the type I class A scavenger receptor express- In the past few decades, new members of the ed in macrophages [2]. The SRCR domain is SRCR gene family have been identified and approximately 100-110 amino acids long and much has been learned about their biological characterized by the multiple conserved cyste- properties [4]. Accumulated research has re- ine residues. The number of SRCR domains in vealed that these SRCR genes/proteins have a SRCR genes varies, ranging from a single copy broad range of ligand-binding substrates, in- to up to 14 repeats. Depending on the charac- cluding polyribonucleotides, proteins, polysac- SRCR genes in ESCC charides, and lipids [5]. Nevertheless, many An identity matrix, which contained similar per- SRCR genes encode mosaic or multidomain centages of every pair of SRCR domain se- proteins proposed to function in the homeosta- quences, was generated by ClustalW after the sis of epithelia and the . Some alignment. To visualize the results, the sequ- SRCR genes are associated with a number of ence similarity matrix was log-transformed, clu- diseases and pathogenic states, such as au- stered by Cluster 3.0 software (http://bonsai. toimmune diseases, Alzheimer’s disease, ath- hgc.jp/~mdehoon/software/cluster/software. erosclerosis and cancer, thus exhibiting prom- htm) and viewed in TreeView (http://jtreeview. ising potential as targets for diagnostic and/or sourceforge.net/). The Neighbor joining meth- therapeutic intervention [5]. od in MEGA X software was applied to construct a phylogenetic tree with bootstrap replicates The esophagus is an open organ, connecting set as 1000 [8]. mouth and stomach, and easily suffers from stimuli, such as heat, food and reflux. Globally, Differentially-expressed SRCR genes in esoph- esophageal cancer is the seventh most com- ageal cancer mon malignant tumor, as well as the sixth most common fatal cancer [6]. Based on these char- At least three mRNA expression profiles for acteristics, the esophagus is a good model to esophageal squamous cell carcinoma (ESCC) study the expression pattern of SRCR genes. were obtained and analyzed in this study, whi- Much research in the past has reported the ch were available from GEO database (https:// function, structure and evolution of specific, www.ncbi.nlm.nih.gov/geo/), to detect the di- single SRCR genes in many species [7]. How- fferential expression of SRCR genes. Both GS- ever, human SRCR genes have not been a fo- E53622 and GSE53624 were provided from cus and fully considered in human disease, the same previous research, which analyzed especially for cancer. In this study, human the lncRNA and mRNA expression profiles by SRCR genes in esophageal cancer were col- microarray in paired ESCC tumor and normal lected, and their specific protein-protein inter- tissues [9]. In GSE53622, 119 paired patients action networks were established. Multiple bio- were divided randomly into training (n=60) and informatics and clinical survival analyses of test (n=59) groups. A prognostic signature was SRCR genes were performed to understand developed from the training group and valid- their roles and clinical significance in the back- ated in the test group, and further in an ind- ground of esophageal cancer. ependent cohort (n=60), which generated the GSE53624 dataset. In GSE23400, Su et al. Materials and methods profiled global of 53 paired ESCC tumor and matched normal samples to Human SRCR protein sequence collection better characterize molecular changes in ESCC [10]. Human SRCR proteins were searched and col- lected through a query in the UniProt protein These datasets were re-annotated and norm- database (http://www.uniprot.org/). “SRCR” alized by the robust multi-array average (RMA) was used as keyword to search in the UniProt method in the R language limma package database, with a search condition set as “HU- (http://www.bioconductor.org/packages/2.9/ MAN” and “Reviewed”. The FASTA format of the bioc/html/limma.html). Significantly different- protein sequences was retrieved from Uniprot ially expressed genes were identified by the database as well. Since these proteins conta- SAM (Significance Analysis of Microarray, assi- in other functional domains other than SRCR milating a set of gene-specific t-tests) method, domain, we categorized each of these SRCR with a fold-change >1.5 or fold-change < 0.67 proteins by different domains in order to study (FDR P.adjust < 0.01) being defined as a signifi- the amino acid sequences of SRCRs. Then the cant change [11]. Since three expression pro- sequence alignment of these SRCR units was files were analyzed in this study, only genes performed using ClustalW embedded in BioEdit that were significantly changed in 2 out of the software (http://www.mbio.ncsu.edu/bioedit/ 3 ESCC expression profiles were considered bioedit.html), and modified by Jalview (http:// for future study. Hierarchical clustering was www.jalview.org/) to indicate the conserved performed using the data of differentially-ex- amino acid residues. pressed SRCR genes from these three ESCC

2684 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC mRNA expression profiles to generate a heat- yzer, which can calculate network diameter, map. density, centralization, and clustering coeffi- cients, providing insight into the organization Protein-protein interaction network for differen- and structure of complex networks [14]. The tially-expressed SRCR genes degree distribution P(k) of a large-scale net- work was defined as the fraction of nodes in Updated human protein-protein interaction da- the network with degree k. The pattern of th- ta was acquired from HPRD (http://www.hprd. eir dependencies can be visualized by fitting a org/), BioGRID (http://thebiogrid.org/), and Int- line on the node degree distribution data. Act (http://www.ebi.ac.uk/intact/). These phys- NetworkAnalyzer calculates the positive coor- ical protein interactions were collected from dinate value for fitting the line, where the pow- published papers, which were confirmed by low er law curve is of the form y = bxa. The R2 value -throughput or high-throughput experiments, is a statistical measure of the linearity of the providing high confidence for future analyses, curve fit and used to quantify the fit to the such as disease research integrated with the power line. When the fit is good, the 2R value is human PPI network [12]. very close to 1. The power law of distribution of These three protein interaction datasets were node degrees, one of most important network integrated manually to reduce redundancy and topological characteristics, was analyzed as obtain a unique dataset containing all known we previously described [15]. Other important published human protein-protein interactions topological parameters, including shortest path so far. This unique PPI dataset has 20,492 length, neighborhood connectivity, and close- unique proteins and 505,143 interactions, and ness centrality, were also analyzed and shown was considered as the parental PPI network as they represent the network properties. from which new or child PPI networks were con- Functional enrichment analyses of the Full structed. Cytoscape is a powerful software for SRCR-PPIN visualizing complex networks and integrating many types of attribute data, such as expres- (GO) Biological Process (BP) sion level and co-expression coefficient [13]. enrichment analysis was carried out by the Differentially-expressed SRCR genes were ma- ClueGO app in Cytoscape using the hypergeo- pped to the parental PPIN, and the first class metric test, with significant results (Bonferroni- directly-interacting proteins were extracted to corrected p-value < 0.01) visualized as nodes construct a sub-network, which we called “Full in different colors depending on biological gr- SRCR-PPIN”. To reduce the redundant connec- ouping [16]. To indicate the relationship bet- tions between interacting proteins, only SRCR ween GO terms, a kappa score reflecting the genes and their connecting proteins were overlapping gene number between terms was extracted to construct a more direct PPI sub- set to 0.3 as the threshold in ClueGO, to link the network, which was referred as the “Concise significant terms to create a network of func- SRCR-PPIN”. tionally and hierarchically organized GO terms.

The fold-change in expression of each protein, An additional pathway analysis was carried out as well as the expression correlation coefficient using the Kyoto encyclopedia of genes and of each pair of proteins in the Concise SRCR- genomes (KEGG) pathway database, which is PPIN, were analyzed by a custom R program. also integrated in ClueGO, with a statistically Next, the fold-change and correlation coeffi- significant Bonferroni-corrected p-value less cient were integrated into the Concise SRCR- than 0.05. The statistical option was set as PPIN, as node attributes and edge attributes, enrichment/depletion (two-sided hypergeomet- to indicate the expression pattern of the net- ric test) with Bonferroni step down p-value work in the background of ESCC. correction.

Analyses of the topological parameters of the Subcellular distribution of the Full SRCR-PPIN Full SRCR-PPIN The subcellular locations for all proteins in the Analyses of topological parameters of the Full Full SRCR-PPIN were obtained from the HP- SRCR-PPIN were conducted by NetworkAnal- RD database (http://www.hprd.org/), and were

2685 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC imported into the network as the node attri- ed on an uncoated chamber. After 48 h (inva- bute. If some proteins were annotated by mul- sion assays) or 36 h (migration assays), the tiple locations, e.g. if a protein was able to membranes were fixed and stained with hema- translocate from the cytoplasm into the nucle- toxylin. Cell numbers were quantified by count- us, its position would be annotated as cyto- ing 10 random fields under a light microscope plasm/nucleus. The Cerebral plugin was appli- (200×). The mean value was calculated from ed to distribute the protein nodes in the Full data obtained from 10 random fields. Both mi- SRCR-PPIN network into different layers accor- gration and invasion assays were repeated ding to their subcellular locations, generating three times. a pathway-like graph with their interactions unchanged [17]. Colony formation assay

Survival analyses of differentially-expressed Colony formation was detected using a plate SRCR genes in ESCC colony formation assay [22]. Transfected cells were counted with TC-20 (Bio-Rad, CA, USA), The ESCC mRNA expression profile GSE53625, and then 3×103 cells were inoculated into each as well as its clinical data, were downloaded well of a six-well plate containing RPMI 1640 from GEO database (https://www.ncbi.nlm.nih. medium supplemented with 10% fetal bovine gov/geo/), which contains the expression of a serum. Cells were incubated for 14 days at total 179 pairs of ESCC clinical samples. The 37°C and 5% CO2. Colonies were fixed in 100% X-tile 3.6.1 program was applied to define the ice-cold ethanol and were visualized by hema- optimal cutoff point for the expression level for toxylin staining. Colonies (>50 cells) were co- differentially-expressed SRCR genes to classify unted, and the average number of colonies fr- ESCC patients into high expressing and low om three separate experiments was express- expressing groups, following the Kaplan-Meier ed. and log-rank test survival analyses (significant with P < 0.05) using GraphPad Prism7 [18]. Results

Differentially-expressed genes from LOXL2 The conservation and evolution of human overexpressing cells SRCR protein sequences

In the previous study, we had overexpressed Currently, 27 SRCR proteins have been identi- LOXL2 in the KYSE150 esophageal squamous fied in . The number of SRCR cell carcinoma cell line, followed by a microar- domains, Gene ID, and genomic location are ray analysis, which has been submitted to the shown in Table 1. DMBT1 (deleted in malignant GEO database (http://www.ncbi.nlm.nih.gov/ brain tumors 1 protein) has the largest number geo) under accession number of GSE53645 of SRCR domains, with 14, followed by CD163L1 [19]. The differentially-expressed genes were (scavenger receptor cysteine-rich type 1 pro- obtained by fold-change analysis as previously tein M160) with 12 SRCR domains. Thirteen reported, and used for functional enrichment out 27 genes have more than one SRCR do- analysis in DAVID (https://david.ncifcrf.gov/). main. To determine the variants and evolution- ary relationship among the SRCR domains fr- Cell migration and invasion assays in vitro om the different SRCR genes, each SRCR do- main from these proteins was separated into a The full-length and SRCR domain-deleted ex- single unit, named by the order in its parental pression plasmids of LOXL2 were constructed protein. in our laboratory as previously reported [20]. Migration and invasion assays were performed The protein sequence alignment of these SRCR using transwell chambers as described previ- domains is presented in Figure 1A. Most of the ously [21]. In invasion assays, 2×105 cells were SRCR domains are approximately 100 amino inoculated into the top chamber of a matrigel- acids in length. Two longer SRCR domains were coated membrane with 8-μm pores (BD Falcon, observed, TMPRSS13 with a single 131 amino USA), and the bottom chamber was filled with acid SRCR domain, and LOXL4 with a 129-am- medium with 10% fetal bovine serum. In the ino acid SRCR domain. The approximately eight migration assay, 1×105 cells were directly plat- conserved cysteines in the SRCR domain are

2686 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Table 1. Basic information for human SRCR genes Genomic SRCR Other Domain or Functional Gene Symbol Full Name Gene ID Location Number Region DMBT1 Deleted in malignant brain tumors 1 protein 1755 10q26.13 14 CUB, ZP CD163L1 Scavenger receptor cysteine-rich type 1 protein M160 283316 12p13.31 12 None CD163 Scavenger receptor cysteine-rich type 1 protein M130 9332 12p13.31 9 None SCART1 Scavenger receptor cysteine-rich domain-containing 619207 10q26.3 8 None protein SCART1 SSC5D Soluble scavenger receptor cysteine-rich domain- 284297 19q13.42 5 None containing protein SSC5D LOXL2 homolog 2 4017 8p21.3 4 Lysyl-oxidase like LOXL3 Lysyl oxidase homolog 3 84695 2p13.1 4 Lysyl-oxidase like LOXL4 Lysyl oxidase homolog 4 84171 10q24.2 4 Lysyl-oxidase like PRSS12 Neurotrypsin 8492 4q26 4 Kringle, Peptidase S1 SSC4D Scavenger receptor cysteine-rich domain-containing 136853 7q11.23 4 None group B protein CD5 T-cell surface glycoprotein CD5 921 11q12.2 3 None CD5L CD5 antigen-like 922 1q23.1 3 None CD6 T-cell differentiation antigen CD6 923 11q12.2 3 None CFI Complement factor I 3426 4q25 1 Kazal-like, LDL-receptor class A, Peptidase S1 CORIN Atrial natriuretic peptide-converting enzyme 10699 4p12 1 FZ, LDL-receptor class A, Peptidase S1 HHIPL1 HHIP-like protein 1 84439 14q32.2 1 None HPN Serine protease hepsin 3249 19q13.11 1 Peptidase S1 LGALS3BP Galectin-3-binding protein 3959 17q25.3 1 BTB, BACK MARCO Macrophage receptor MARCO 8685 2q14.2 1 Collagen-like MSR1 Macrophage scavenger receptor types I and II 4481 8p22 1 Collagen-like SCARA5 Scavenger receptor class A member 5 286133 8p21.1 1 Collagen-like TMPRSS13 Transmembrane protease serine 13 84000 11q23.3 1 LDL-receptor class A, Peptidase S1 TMPRSS15 Enteropeptidase 5651 21q21.1 1 SEA, LDL-receptor class A, CUB, MAM, Peptidase S1 TMPRSS2 Transmembrane protease serine 2 7113 21q22.3 1 LDL-receptor class A, Peptidase S1 TMPRSS3 Transmembrane protease serine 3 64699 21q22.3 1 LDL-receptor class A, Peptidase S1 TMPRSS4 Transmembrane protease serine 4 56649 11q23.3 1 LDL-receptor class A, Peptidase S1 TMPRSS5 Transmembrane protease serine 5 80975 11q23.2 1 Peptidase S1 indicated by the red block (Figure 1A). It These SRCR domains were clustered based on seemed that the first conserved cysteine has their protein sequence similarities to form a a high rate of variation, followed by the fourth similarity matrix (Figure 1B). The SRCR doma- conserved cysteine. The most conserved resi- ins can be approximately classified into four due is the third cysteine, which appeared in hierarchical clustering groups by gradual ch- all SRCR domains. Interestingly, many SRCR ange, suggesting SRCR domains are greatly dif- domains lack 3 or 4, or even 5, conserved cys- ferent, but rather conserved in the history of teines. For example, TMPRSS4 has only one evolution. It is worthy to point out that many SRCR domain, but it has only three conserved different SRCR domains from the same gene cysteines. For the single SRCR domain in CFI, were divided into different groups. Take LOXL2, HPN, TMPRSS3 and TMPRSS5, they lack 4 LOXL3 and LOXL4 for example, the 1st and 3rd conserved cysteines, respectively. We also SRCR domains were located in the second cl- observed the 4th SRCR domain in LOX family ustering group, while the 2nd and 4th SRCR members LOXL2, LOXL3 and LOXL4 have only domains were located in the fourth cluster- five conserved cysteines. A list of missing con- ing group. It is interesting to find that SRCR served cysteines in SRCR genes is presented domains 1-13 in DMBT1 proteins share the in Table 2, based on the order of eight con- most similar protein sequences, forming a sm- served cysteines. all cluster, suggesting the 14th SRCR domain in

2687 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

2688 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

2689 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

2690 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 1. Sequence analyses of SRCR genes. A. Protein sequence alignment of a total of 91 human SRCR domains, which were extracted from 27 human SRCR proteins and considered as a single unit. Identical and amino acid resi- dues with similar physical and chemical properties are indicated by the same color. The eight conserved cysteines are indicated in red blocks. B. Cluster based on sequence similarities. These SRCR domains are approximately classified into four groups, indicating a gradual variation. C. The phylogenetic tree based on the protein sequences of SRCR domains. the DMBT1 gene might have originated from To examine the expression pattern of differen- another gene and replicated 13 times in the tially-expressed SRCR genes, hierarchical clus- DMBT1 genomic locus during the evolution tering based on their expression levels in the progress. three ESCC expression profiles, was carried out and visualized in the heatmaps. The results To understand the evolution branch of SRCR demonstrated that ESCC cancer tissues and domains in different proteins, a phylogenetic adjacent normal tissues can be distinguished tree was constructed using MEGA X software by differentially-expressed SRCR genes appr- (Figure 1C). In the tree, several proteins (HRN, oximately, though not perfectly (Figure 2). CORIN, TMPRSS4, TMPRSS13, TMPRSS5, TM- PRSS2, TMPRSS3) that contained only one Three different PPI networks for differentially- SRCR domain were clustered under the same expressed SRCR genes root. Consistent with sequence similarity ma- trix, proteins with a number of SRCR domains A PPI network was constructed comprised of might cluster under different roots, suggesting 10 deregulated SRCR genes as the seed nodes even the SRCR domains from the same SRCR with all their known interacting proteins. We gene had a different or un-parallel evolution referred to this PPI network as the “Full SRCR- rate. PPIN”, as it also shows the interactions between every two proteins in the network, and con- Ten differentially-expressed of SRCR genes in tained 323 nodes (proteins) and 2,781 edges ESCC (interactions) (Figure 3A). All 10 deregulated Three ESCC mRNA expression profiles were SRCR proteins have reported interacting pro- analyzed in this study to investigate the change teins from the literature. This indicates dere- of SRCR genes in ESCC. In total, 12 differential- gulated SRCR proteins can link to hundreds of ly-expressed SRCR genes (CD163, DMBT1, LG- other proteins by cascade interactions to ex- ALS3BP, LOXL2, LOXL3, LOXL4, MSR1, PRSS12, pand their biological effects. The top three ge- SCARA5, TMPRSS2, TMPRSS4 and TMPRSS5) nes having the highest number of interacting were identified in the GSE53624 dataset, 9 of proteins were LGALS3BP (206 edges), LOXL2 which (DMBT1, LGALS3BP, LOXL2, LOXL3, LO- (33 edges) and MSR1 (20 edges). XL4, MSR1, PRSS12, SCARA5 and TMPRSS2) To better illustrate, another smaller PPI net- were also differentially-expressed with signifi- cance in GSE53622 dataset. Three SRCR ge- work, which we named “Concise SRCR-PPIN”, nes (CD163, LOXL2 and TMPRSS2) were identi- of 10 dysregulated SRCR proteins was also fied as differentially-expressed genes in the created, but excluded the interactions betwe- GSE23400 dataset. To avoid bias from differ- en their interacting proteins, thus containing ent patient groups, only those were significantly only their interactions with other SRCR prot- changed in 2 out of the 3 ESCC expression pro- eins. This “Concise SRCR-PPIN” contained 323 files were defined as differentially expressed. nodes and 322 edges, enabling us to easily By this strict criterion, 10 differentially-expre- show that many SRCR genes are linked throu- ssed SRCR genes, including six upregulated gh one interacting protein, indicated by the lig- (CD163, LGALS3BP, LOXL2, LOXL3, LOXL4 and ht pink color (Figure 3B). For example, LOXL2 MSR1) and four downregulated (DMBT1, PRSS- links with TMPRSS2 through HNRNPL. To our 12, TMPRSS2 and SCARA5) genes were ob- surprise, LGALS3BP links with many other tained and applied in the ensuing analyses. SRCR proteins through different interacting Details about the GEO dataset, official gene proteins, such as LGALS3BP→ATP4A→CD163, symbols, log2 (fold change), and adjusted p-val- LGALS3BP→RPL23→LOXL3, LGALS3BP→HS- ue of differentially-expressed SRCR genes are PA5→LOXL3, LGALS3BP→CAND1→SCARA5, shown in Table 3. and LGALS3BP→HSPA1A→MSR1. Another ex

2691 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Table 2. The variation or deletion of conserved cysteines in SRCR domains 1st 2nd 3rd 4th 5th 6th 7th 8th SCART1|SRCR6 CD163|SRCR8 None HHIPL1|SRCR CD6|SRCR1 SCART1|SRCR7 SCART1|SRCR6 CFI|SRCR HHIPL1|SRCR CD163L1|SRCR11 LOXL2|SRCR1 SCART1|SRCR7 TMPRSS3|SRCR SCART1|SRCR8 SCART1|SRCR7 LOXL2|SRCR1 SCART1|SRCR5 LOXL3|SRCR1 CD5|SRCR2 TMPRSS5|SRCR CD163|SRCR8 TMPRSS13|SRCR LOXL3|SRCR1 CD5|SSRCR2 LOXL4|SRCR1 TMPRSS3|SRCR TMPRSS15|SRCR CD163L1|SRCR11 TMPRSS4|SRCR LOXL4|SRCR1 PRSS12|SRCR3 TMPRSS5|SRCR TMPRSS4|SRCR CFI|SRCR CORIN|SRCR PRSS12|SRCR3 MARCO|SRCR TMPRSS15|SRCR CORIN|SRCR CD6|SRCR1 MARCO|SRCR SCARA5|SRCR TMPRSS13|SRCR HPN|SRCR SCART1|SRCR5 SCARA5|SRCR MSR1|SRCR TMPRSS4|SRCR TMPRSS2|SRCR LOXL2|SRCR4 MSR1|SRCR LOXL2|SRCR3 CORIN|SPCR LOXL3|SRCR4 LOXL2|SRCR3 LOXL3|SRCR3 HPN|SRCR LOXL4|SRCR4 LOXL3|SRCR3 LOXL4|SRCR3 TMPRSS2|SRCR SCART1|SRCR7 LOXL4|SRCR3 LGAL3BP|SRCR CD5|SRCR2 LGAL3BP|SRCR PRSS12|SRCR2 CD5|SRCR3 PRSS12|SRCR2 PRSS12|SRCR4 CD5|SRCR1 PRSS12|SRCR4 CFI|SRCR CFI|SRCR LOXL4|SRCR2 LOXL4|SRCR2 LOXL2|SRCR2 LOXL2|SRCR2 LOXL3|SRCR2 LOXL3|SRCR2 LOXL2|SRCR4 LOXL2|SRCR4 LOXL3|SRCR4 LOXL3|SRCR4 LOXL4|SRCR4 LOXL4|SRCR4 PRSS12|SRCR1 PRSS12|SRCR1 TMPRSS3|SRCR TMPRSS3|SRCR TMPRSS5|SRCR TMPRSS5|SRCR TMPRSS15|SRCR TMPRSS15|SRCR TMPRSS13|SRCR TMPRSS13|SRCR TMPRSS4|SRCR TMPRSS4|SRCR CORIN|SRCR CORIN|SRCR HPN|SRCR HPN|SRCR TMPRSS2|SRCR TMPRSS2|SRCR Note: The variation or deletion of conserved cysteines with more than three in the same SRCR domains is indicated by the same color. The italics indicate the deletion in that position. ample of a multiple step link is MSR1→CO- and the green node suggests downregulation LGALT2→SCARA5. These results suggested th- in ESCC. The non-significantly changed proteins at these differentially-expressed SRCR genes/ remained as blue. For each pair of proteins in proteins might connect or cross-talk through at the network, the red line indicates positive cor- least one partner protein. relations, while the green line suggests a ne- gative correlation. Thicker lines correspond to To illustrate the change of expression and the higher correlation coefficients. Some interest- close connection for each interacting protein in ing interacting pairs were found, such as for the “Concise SRCR-PPIN”, we applied the GS- LGALS3BP→HSPA1A→MSR1. These three ge- E53625 (containing both GSE53622 and GS- nes were upregulated and had a strong posi- E53624) to analyze the log2 (fold change) of tive correlation. A similar pattern was found proteins in the network, as well as the co- between MSR1, APOE and LOXL4. The down- expression Pearson correlation coefficient for regulated SCARA5 positively correlated with each pair of proteins. Then, the log2 (fold CAND1, but was negatively correlated with LG- change) and co-expression correlation coeffi- ALS3BP. These results suggested that some cient were integrated into the Concise SRCR- SRCR proteins might form or switch protein PPIN as node edge attributes, respectively, to complexes directly or indirectly by different in- illustrate the expression pattern in ESCC (Fi- teraction strengths for their different func- gure 3C). The red node indicates upregulation tions.

2692 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Table 3. The differentially-expressed SRCR genes from three datasets GEO accession Gene Symbol Ensembl_ID Fold Change Log2 (Fold Change) P.adjust GSE53624 CD163 ENSG00000177575 2.01 1.00 9.98E-07 DMBT1 ENSG00000187908 0.26 -1.96 0 LGALS3BP ENSG00000108679 1.93 0.95 0 LOXL2 ENSG00000134013 6.80 2.76 0 LOXL3 ENSG00000115318 1.90 0.93 0 LOXL4 ENSG00000138131 1.57 0.65 0.0005 MSR1 ENSG00000038945 2.56 1.36 0 PRSS12 ENSG00000164099 0.55 -0.86 0 SCARA5 ENSG00000168079 0.23 -2.12 0 TMPRSS2 ENSG00000184012 0.09 -3.48 0 TMPRSS4 ENSG00000137648 0.61 -0.71 1.97E-06 TMPRSS5 ENSG00000166682 0.63 -0.66 0 GSE53622 DMBT1 ENSG00000187908 0.34 -1.58 1.01E-06 LGALS3BP ENSG00000108679 1.77 0.83 0 LOXL2 ENSG00000134013 5.63 2.49 0 LOXL3 ENSG00000115318 1.92 0.94 0 LOXL4 ENSG00000138131 1.66 0.73 0.002 MSR1 ENSG00000038945 3.84 1.94 0 PRSS12 ENSG00000164099 0.66 -0.61 0.0007 SCARA5 ENSG00000168079 0.22 -2.16 0 TMPRSS2 ENSG00000184012 0.14 -2.81 0 GSE23400 CD163 215049_x_at 1.55 0.64 4.03E-06 CD163 203645_s_at 1.55 0.64 3.36E-06 LOXL2 202998_s_at 2.48 1.31 0 TMPRSS2 211689_s_at 0.24 -2.08 0

Topology parameters indicate the scale-free of or 3 steps (Figure 4B). These results also sug- the Full SRCR-PPIN gest that this PPIN is flexible for proteins to form different protein complexes or switch be- Many kinds of networks, including biological tween each other. The neighborhood connectiv- networks, are distinguishable from random or ity, which is the average connectivity of nei- other chaos networks by their distinguishing ghboring proteins of a protein, showed a sharp topological characteristics. A real network has linear decreasing trend in the Full SRCR-PPIN been characterized as scale-free, suggesting (Figure 4C). This indicates that only a few nod- that its degree distribution follows a power law es with high connectivity might serve as hubs [23]. As shown in Figure 4A, the distributions of in the network. Closeness centrality measures node degree followed an approximate power how fast information can spread from a given law, with the equation y = 60.463 X-0.908 and an node to other reachable nodes in the network, R2 = 0.779. Scale-free, as one of most impor- and could be viewed as the efficiency of infor- tant characteristics of true complex biological mation spreading in the network. Curve fitting network, is usually indicated by an R2 of more showed closeness centrality dropped slowly, than 0.5 in previous co-expression PPINs [24]. suggesting that many proteins are closely con- nected (Figure 4D). Since many of the two-protein pairs in Full SRCR-PPIN are usually not directly connected Proteins in the Full SRCR-PPIN distribute in to each other, but rather are connected throu- multiple subcellular layers gh one or more steps. The shortest path length (the number of edges from one node to ano- The subcellular location of a protein is critical ther node) showed that it was mainly involved 2 for its function, because it provides a physio-

2693 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 2. Heatmap of differentially-expressed SRCR genes in three microarray datasets. A-C. Hierarchical clustering for differentially-expressed SRCR genes based on their expression levels in GSE23400, GSE53622 and GSE53624, respectively. A gradient of red to yellow color indicates high to low expression. Red bars and green bars in the left panel represent ESCC clinical samples and matched normal tissues, respectively. Each row represents a clinical sample.

2694 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

2695 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 3. Three PPI networks for differentially-expressed SRCR genes. A. The “Full SRCR-PPIN” showing the in- teractions between each protein pair in the network. SRCR proteins are indicated by a deep blue color, and their interacting proteins are shown in light blue. B. “Concise SRCR-PPIN” indicates many SRCR genes are linked through one interacting protein, indicated by the light pink color. C. The log2 (fold change) and co-expression correlation coefficient for proteins integrated into the Concise SRCR-PPIN. The red or green node indicates their upregulation and downregulation in expression. Non-significantly changed proteins remain as blue. The red or green line between nodes indicates a positive or negative correlation, respectively. Thicker lines indicate higher correlation coefficients. logical environment for their performance, su- results may also reflect their complex biological ch as complex formation, protein modification, processes at the spatial level. signal transduction and transcriptional regula- tion [25]. Cerebral plugin was applied to build a GO and pathway enrichment analyses reveal layered network based on the subcellular loca- migration-related functions tion of proteins. Then, the Concise-SRCR net- In order to fully understand the functions in- work was separated into several layers, includ- volved by the dysregulated SRCR genes and ing secreted, membrane, cytoskeleton, cyto- how they influence cell activity through their plasm, cytoskeleton/nucleus, cytoplasm/nucl- interactions in the network, GO “Biological Figure 5 eus, and nucleus ( ). For differentially- Process” enrichment analysis of the Full SRCR- expressed SRCR genes, their proteins were PPIN was performed, resulting in 67 enriched mainly distributed extracellularly (LOXL2, LO- GO terms, forming a functional annotation map. XL3, LXOl4, LGALS3BP and PRSS12), in the In this map, the nodes were no longer proteins, membrane (CD163, MSR1, DMBT1, TMPRSS2) but instead were enriched GO terms, with the and in the nucleus (SCARA5). These results edges suggesting a significant overlap of en- suggest that SRCR proteins might be impor- riched proteins between two GO terms (Figure tant signaling molecules upstream of signaling 6A). These results indicate the biological func- pathways, influencing signal transfer from ex- tions for differentially-expressed SRCR genes tracellular environment into the nucleus. These in ESCC involve extracellular structure organi-

2696 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 4. Topological properties of the Full SRCR-PPIN. A. The node degree distribution of the Full SRCR-PPIN, in- dicating a power-law distribution character. B. Shortest path length of the Full SRCR-PPIN. The frequency suggests the number of protein pairs have a certain shortest path length. C. Neighbor connectivity of the Full SRCR-PPIN. The decreasing distribution indicates the edges between sparsely connected and highly connected nodes prevail in the network. D. Closeness centrality distribution of the Full SRCR-PPIN. The closeness centrality of each node varies between 0 and 1. zation, cell cycle process, mitotic cell cycle, pro- ESCC patients with lower expression of LOXL2, teasomal protein catabolic process and protein LOXL3, LOXL4 and SCARA5, or higher DMBT1 catabolic process. and PRSS12 expression had longer survival.

An annotation network based on 12 enriched SRCR domains are important for the integrity KEGG pathways was also created and visual- of LOXL2 function ized as a group of functionally organized net- work (Figure 6B). Several significant known or LOXL2 protein contains four SRCR domains at potential KEGG pathways were observed, in- the N-terminus and a copper-dependent cate- cluding ECM-receptor interaction, Protein pro- gorized domain at the C-terminus. To confirm cessing in endoplasmic reticulum, ubiquitin- the potential functional networks in Figure 6, mediated proteolysis and the PI3K-Akt signal- the differentially-expressed genes after LOXL2 ing pathway. overexpression in ESCC cells were obtained Differentially-expressed SRCR genes are corre- and functional enrichment analysis was perfor- lated with the survival time of ESCC patients med. Consistent with the GO enrichment re- sults of the Full SRCR-PPIN, “extracellular struc- To test the clinical significance of differentially- ture organization” and “ECM-receptor interac- expressed SRCR genes in ESCC, we performed tion” as shown in Figure 6, an interesting GO survival analyses using publicly available data term, named “GO:0030199~collagen fibril or- from GEO. The results show a significant cor- ganization”, was found. This term contains 6 relation between the expression level of 6 dif- genes (LOXL2, TNXB, TGFBR1, COL12A1, CO- ferentially-expressed SRCR genes (LOXL2, LO- L5A2 and COL5A1) with their absolute fold- XL3, LOXL4, DMBT1, PRSS12 and SCARA5) change more than 1.5 (Figure 8A). We have and survival time of ESCC patients (Figure 7). previously shown that LOXL2 promotes the

2697 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 5. Subcellular localization distribution of the Full SRCR-PPIN. Differentially-expressed SRCR genes are indi- cated by a black box. The Concise-SRCR network was spread into layers according to location of proteins, which looks like non-canonical signal pathways. metastasis and invasion of ESCC cells [26]. To metabolic disorders. Emerging evidence indi- investigate the functional importance of SRCR cates these molecules are important regula- domains for the whole LOXL2 protein, a trun- tors in tumor behavior [27]. Despite their name, cated LOXL2 expression plasmid was const- SRCR genes are involved in more than just ructed with the deletion of its four SRCR do- scavenging. They have been shown to carry out mains (Figure 8B). The ability for LOXL2-LTQ to several functions, including functioning as lipid promote the metastasis and invasion of ESCC transporters, chaperones that transport other cells was obviously reduced compared to the cellular proteins to their destination, and even full-length LOXL2 (Figure 8C, 8D). Similarly, the as chemokines. Over the last few years, accu- colony-forming ability of ESCC cells express- mulated evidence indicates that SRCR genes ing LOXL2-LTQ was reduced compared to full- act as important regulators of tumor progres- length LOXL2 in ESCC cells (Figure 8E). These sion and host immune response to cancer [27]. results suggest that SRCR domains are impor- tant for the complete function of LOXL2. Many previous reviews prefer to divide SRCR domains into two categories: one contains type Discussion A domains, which possess six cysteine resi- dues encoded by multiple exons, and the other SRCR genes constitute a large family of evolu- is type B domains, which contain eight cystei- tionally conserved proteins, which are structur- ne residues encoded by a single exon [3, 28, ally and functionally diverse. In addition to the 29]. However, these previous studies did not previously known crucial role in maintenance reveal the conserved position of cysteines in of host homeostasis, SRCR genes have been the SRCR domain, as well as the duplication implicated in the pathogenesis of diseases, of SRCR domains between or within SRCR ge- e.g. atherosclerosis, neurodegeneration, and nes. By the comparison of protein sequences

2698 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

2699 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 6. Functional enrichment analyses of the Full SRCR-PPIN. A. A functional-grouped network was generated from Gene Ontology (GO) “Biological Process” en- richment analysis. The 67 GO terms were linked when their overlapped kappa score was ≥ 0.3. Similar GO terms are labeled in the same or similar color. B. KEGG pathways involving proteins in the network were analyzed by the hyper-geometric distribution. Node sizes represent the term enrichment significance.

2700 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 7. The expression level of differentially-expressed SRCR genes correlates with survival time of ESCC patients. The left panel of each gene shows its relative expression level after optimized classification by the X-tile program. The right panel of each gene indicates the amount of classification, as well as the hazard ratio. of each SRCR domain, we found that some mes, including LOX, LOXL1, LOXL2, LOXL3 and SRCR domains might contain fewer than 6 LOXL4 [31]. They catalyze peptidyl-lysine oxida- conserved cysteines, e.g. TMPRSS4 has only 3 tion in the formation of crosslinks between col- conserved cysteines, and many other SRCR lagens and elastin, and thus remodel the extra- genes lack 3 or 4 conserved cysteines. Never- cellular matrix through the modification of col- theless, the new findings in this study are gre- lagen fibers [32]. LOXL3 associates with STAT3 atly different from previous descriptions, prob- in the nucleus to deacetylate and deacetylimi- ably because the data before the post-genome nate STAT3 on multiple acetyl-lysine sites th- era was incomplete or inadequate. These cys- rough its N-terminal four SRCR domains. Mo- teine residues are considered to create three reover, the SRCR repeats from other LOX family and four disulfide bonds in class A and class B, members can catalyze protein deacetylation/ respectively [30]. We believe that the variation deacetylimination, suggesting a new mechani- of the number of conserved cysteines in SRCR sm of protein modification [33]. In this study, domains, forming different numbers of disul- the truncated LOXL2 is weaker than its full- fide bridges, would provide more possible or length counterpart in promoting ESCC cell me- diverse functions for SRCR genes in normal or tastasis. These results indicate that the four pathological stages. Moreover, different SRCR SRCR domains are important for the function of domains from the multiple SRCR-containing LOXL2. Of course, we could not rule out the genes have endured diverse evolution pressu- other critical functions carried out by these four re, making them cross appear in the evolution SRCR domains, especially each SRCR domain tree. It also suggests that SRCR genes need a remain to be specified in the future study. new classification system that takes into con- sideration both the conserved cysteines and Protein function has been determined at the their biological functions. scale level in the post-genomic era [34, 35]. Our functional enrichment results for both GO The LOX family contains five members of a and KEGG pathway provide a full view, reveal- group of extracellular copper-dependent enzy- ing the known and potential functions and

2701 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

2702 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

Figure 8. SRCR domains are important for the integrity of LOXL2 function. A. The significant GO term “collagen fibril organization”generated from LOXL2 overexpression, which contains 6 gene with their absolute fold-change more than 1. B. Schematic presentation of LOXL2 functional domains and their protein expression. C, D. Comparison of the ability to promote migration and invasion for full-length and truncated LOXL2, respectively. The abilities of migra- tion and invasion of cells caused by truncated LOXL2 were much weaker than that with full length LOXL2. E. Colony formation ability of ESCC cells expressing truncated LOXL2 was less than full length LOXL2 in ESCC cells. mechanisms of the differentially-expressed SR- the perineuronal environment, maintaining ne- CR genes in ESCC. For example, extracellular uronal plasticity, and synergizing with proteas- structure organization from GO enrichment and es such as tissue plasminogen activator [40]. ECM-receptor interaction from KEGG enrich- The roles of these SRCR genes/proteins in the ment, which are consistent with the knowledge pathological state makes them potential mark- that LOXL proteins play key roles in extracellu- ers for disease diagnosis and prognosis. lar matrix stability and integrity, promoting cell migration and development. Given the number Conclusions of receptors in the scavenger receptor family In summary, the current findings in this study and their extensive function, this study pre- are of benefit for gaining a more compreh- dicts that scavenger receptors are involved in ensive understanding of human SRCR genes. the pathogenesis of a variety of diseases. There- The differentially-expressed SRCR gene PPI fore, it is not surprising that scavenger recep- network is integrated with expression data tors have been shown to activate a range of from human esophageal cancer, and annotat- diverse signaling pathways. So far, a systema- ed by both GO and KEGG enrichment. These tic investigation of the prognostic significance analyses would greatly expand our knowledge of SRCR genes in cancers has not been report- of SRCR gene family, provide a roadmap for ed. The results in this study demonstrate that how to analyze a protein family through high- the expression of a panel of SRCR genes signifi- throughput experiments as well. cantly correlates with the survival time of ES- CC patients, with lower expression of LOXL2, Acknowledgements LOXL3, LOXL4 and SCARA5, or higher DMBT1 and PRSS12 expression correlating with good This work was supported by grants from the prognosis. The important functions and clinical National Natural Science Foundation of China significance of these SRCR genes have been (No. 81672473, No.81502138), the Science elucidated in recent years. Kasashima et al. and Technology Program of Guangdong (No. found the expression of LOXL1, LOXL3, and 2014A030310390, No. 2017A030313181). LOXL4 is correlated with lymph node metasta- We thanked Dr. Stanley Lin to read through our sis, invasion, and lymphatic and venous inva- manuscript and correct the misspellings and sion of gastric cancer. These patients with grammatical mistakes. LOXL1-, LOXL3-, or LOXL4-positive tumors ha- Disclosure of conflict of interest ve poorer overall survival [36]. SCARA5 expres- sion is decreased in most breast tumors tis- None. sues, and is associated with hypermethylation of the promoter. A significant correlation was Address correspondence to: Dong Yin, Guangdong detected between SCARA5 expression and the Provincial Key Laboratory of Malignant Tumor Epi- histological grade of breast tumors [37]. DMB- genetics and Genes Regulation, Sun Yat-sen Mem- T1 is expressed differently in cholangiocarc- orial Hospital, Sun Yat-sen University, Guangzhou inogenesis and poorer survival is associated 510120, Guangdong, China. Tel: +86-20-81332405; with the absence of DMBT1 expression in non- E-mail: [email protected]; Haixiong Xu, Depar- neoplastic biliary tissue, suggesting a tumor- tment of Neurosurgery, Shantou Central Hospital, suppressive role of DMBT1 in early cholangio- Affiliated Shantou Hospital of Sun Yat-sen University, carcinogenesis [38]. Shantou 515041, Guangdong, China. Tel: +86-754- 88903592; E-mail: [email protected] Diffuse astrocytoma can show a homozygous deletion of DMBT1, which is associated with References poor clinical outcome [39]. The results from Mitsui et al. suggest that PRSS12 has multiple [1] Goldstein JL, Ho YK, Basu SK and Brown MS. functions, such as axon outgrowth, arranging Binding site on macrophages that mediates

2703 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

uptake and degradation of acetylated low den- tion: connecting genotype to phenotype in can- sity lipoprotein, producing massive cholesterol cer. Nat Rev Genet 2017; 18: 395-410. deposition. Proc Natl Acad Sci U S A 1979; 76: [13] Baryshnikova A. Exploratory analysis of biologi- 333-337. cal networks through visualization, clustering, [2] Freeman M, Ashkenas J, Rees DJ, Kingsley DM, and functional annotation in cytoscape. Cold Copeland NG, Jenkins NA and Krieger M. An Spring Harb Protoc 2016; 2016. ancient, highly conserved family of cysteine- [14] Doncheva NT, Assenov Y, Domingues FS and rich protein domains revealed by cloning type I Albrecht M. Topological analysis and interac- and type II murine macrophage scavenger re- tive visualization of biological networks and ceptors. Proc Natl Acad Sci U S A 1990; 87: protein structures. Nat Protoc 2012; 7: 670- 8810-8814. 685. [3] Sarrias MR, Gronlund J, Padilla O, Madsen J, [15] Wu B, Li C, Du Z, Yao Q, Wu J, Feng L, Zhang P, Holmskov U and Lozano F. The Scavenger Li S, Xu L and Li E. Network based analyses of Receptor Cysteine-Rich (SRCR) domain: an an- gene expression profile of LCN2 overexpres- cient and highly conserved protein module of sion in esophageal squamous cell carcinoma. the innate immune system. Crit Rev Immunol Sci Rep 2014; 4: 5403. 2004; 24: 1-37. [16] Bindea G, Galon J and Mlecnik B. CluePedia [4] Canton J, Neculai D and Grinstein S. Scavenger Cytoscape plugin: pathway insights using inte- receptors in homeostasis and immunity. Nat grated experimental and in silico data. Bioin- Rev Immunol 2013; 13: 621-634. formatics 2013; 29: 661-663. [5] Martinez VG, Moestrup SK, Holmskov U, Mol- [17] Barsky A, Gardy JL, Hancock RE and Munzner T. Cerebral: a Cytoscape plugin for layout of lenhauer J and Lozano F. The conserved scav- and interaction with biological networks using enger receptor cysteine-rich superfamily in subcellular localization annotation. Bioinfor- therapy and diagnosis. Pharmacol Rev 2011; matics 2007; 23: 1040-1042. 63: 967-1000. [18] Camp RL, Dolled-Filhart M and Rimm DL. X-tile: [6] Torre LA, Siegel RL, Ward EM and Jemal A. a new bio-informatics tool for biomarker as- Global cancer incidence and mortality rates sessment and outcome-based cut-point opti- and trends--an update. Cancer Epidemiol Bio- mization. Clin Cancer Res 2004; 10: 7252- markers Prev 2016; 25: 16-27. 7259. [7] Dzik JM. The ancestry and cumulative evolu- [19] Wu BL, Lv GQ, Zou HY, Du ZP, Wu JY, Zhang PX, tion of immune reactions. Acta Biochim Pol Xu LY and Li EM. Exploration of potential roles 2010; 57: 443-466. of a new LOXL2 splicing variant using network [8] Kumar S, Stecher G, Li M, Knyaz C and Tamura knowledge in esophageal squamous cell carci- K. MEGA X: molecular evolutionary genetics noma. ScientificWorldJournal 2014; 2014: analysis across computing platforms. Mol Biol 431792. Evol 2018; 35: 1547-1549. [20] Lv GQ, Zou HY, Liao LD, Cao HH, Zeng FM, Wu [9] Li J, Chen Z, Tian L, Zhou C, He MY, Gao Y, BL, Xie JJ, Fang WK, Xu LY and Li EM. Identifica- Wang S, Zhou F, Shi S, Feng X, Sun N, Liu Z, tion of a novel lysyl oxidase-like 2 alternative Skogerboe G, Dong J, Yao R, Zhao Y, Sun J, splicing isoform, LOXL2 Deltae13, in esopha- Zhang B, Yu Y, Shi X, Luo M, Shao K, Li N, Qiu geal squamous cell carcinoma. Biochem Cell B, Tan F, Chen R and He J. LncRNA profile study Biol 2014; 92: 379-389. reveals a three-lncRNA signature associated [21] Du ZP, Wu BL, Xie YM, Zhang YL, Liao LD, Zhou with the survival of patients with oesophageal F, Xie JJ, Zeng FM, Xu XE, Fang WK, Li EM and squamous cell carcinoma. Gut 2014; 63: Xu LY. Lipocalin 2 promotes the migration and 1700-1710. invasion of esophageal squamous cell carci- [10] Su H, Hu N, Yang HH, Wang C, Takikita M, noma cells through a novel positive feedback Wang QH, Giffen C, Clifford R, Hewitt SM, Shou loop. Biochim Biophys Acta 2015; 1853: 2240- JZ, Goldstein AM, Lee MP and Taylor PR. Global 2250. gene expression profiling and validation in [22] Zeng FM, Wang XN, Shi HS, Xie JJ, Du ZP, Liao esophageal squamous cell carcinoma and its LD, Nie PJ, Xu LY and Li EM. Fascin phosphory- association with clinical phenotypes. Clin Can- lation sites combine to regulate esophageal cer Res 2011; 17: 2955-2966. squamous cancer cell behavior. Amino Acids [11] Tusher VG, Tibshirani R and Chu G. Significance 2017; 49: 943-955. analysis of microarrays applied to the ionizing [23] Lima-Mendez G and van Helden J. The power- radiation response. Proc Natl Acad Sci U S A ful law of the power law and other myths in net- 2001; 98: 5116-5121. work biology. Mol Biosyst 2009; 5: 1482-1493. [12] Yi S, Lin S, Li Y, Zhao W, Mills GB and Sahni N. [24] Sengupta U, Ukil S, Dimitrova N and Agrawal S. Functional variomics and network perturba- Expression-based network biology identifies

2704 Am J Transl Res 2019;11(5):2683-2705 SRCR genes in ESCC

alteration in key regulatory pathways of type 2 [34] Gabaldon T. Comparative genomics-based pre- diabetes and associated risk/complications. diction of protein function. Methods Mol Biol PLoS One 2009; 4: e8100. 2008; 439: 387-401. [25] Hung MC and Link W. Protein localization in [35] Zhang C, Zheng W, Freddolino PL and Zhang Y. disease and therapy. J Cell Sci 2011; 124: MetaGO: predicting gene ontology of non-ho- 3381-3392. mologous proteins through low-resolution pro- [26] Zou HY, Lv GQ, Dai LH, Zhan XH, Jiao JW, Liao tein structure prediction and protein-protein LD, Zhou TM, Li CQ, Wu BL, Xu LY and Li EM. A network mapping. J Mol Biol 2018; 430: 2256- truncated splice variant of human lysyl oxi- 2265. dase-like 2 promotes migration and invasion in [36] Kasashima H, Yashiro M, Okuno T, Miki Y, esophageal squamous cell carcinoma. Int J Kitayama K, Masuda G, Kinoshita H, Morisaki Biochem Cell Biol 2016; 75: 85-98. T, Fukuoka T, Hasegawa T, Sakurai K, Toyokawa [27] Yu X, Guo C, Fisher PB, Subjeck JR and Wang T, Kubo N, Tanaka H, Muguruma K, Hirakawa K XY. Scavenger receptors: emerging roles in and Ohira M. Significance of the lysyl oxidase cancer biology and immunology. Adv Cancer members lysyl oxidase like 1, 3, and 4 in gas- Res 2015; 128: 309-364. tric cancer. Digestion 2018; 98: 238-248. [28] Rohrer L, Freeman M, Kodama T, Penman M [37] Ulker D, Ersoy YE, Gucin Z, Muslumanoglu M and Krieger M. Coiled-coil fibrous domains me- and Buyru N. Downregulation of SCARA5 may diate ligand binding by macrophage scavenger contribute to breast cancer via promoter hy- receptor type II. Nature 1990; 343: 570-572. permethylation. Gene 2018; 673: 102-106. [29] Yap NV, Whelan FJ, Bowdish DM and Golding [38] Goeppert B, Roessler S, Becker N, Zucknick M, GB. The evolution of the scavenger receptor Vogel MN, Warth A, Pathil-Warth A, Mehrabi A, cysteine-rich domain of the class a scavenger Schirmacher P, Mollenhauer J and Renner M. receptors. Front Immunol 2015; 6: 342. DMBT1 expression in biliary carcinogenesis [30] Resnick D, Chatterton JE, Schwartz K, Slayter with correlation of clinicopathological data. H and Krieger M. Structures of class A macro- Histopathology 2017; 70: 1064-1071. phage scavenger receptors. Electron micro- [39] Motomura K, Mittelbronn M, Paulus W, Brokin- scopic study of flexible, multidomain, fibrous kel B, Keyvani K, Sure U, Wrede K, Nakazato Y, proteins and determination of the disulfide Tanaka Y, Pierscianek D, Kim YH, Mariani L, bond pattern of the scavenger receptor cyste- Vital A and Ohgaki H. DMBT1 homozygous de- ine-rich domain. J Biol Chem 1996; 271: letion in diffuse astrocytomas is associated 26924-26930. with unfavorable clinical outcome. J Neuro- [31] Barker HE, Cox TR and Erler JT. The rationale pathol Exp Neurol 2012; 71: 702-707. for targeting the LOX family in cancer. Nat Rev [40] Mitsui S, Yamaguchi N, Osako Y and Yuri K. Cancer 2012; 12: 540-552. Enzymatic properties and localization of mo- [32] Li T, Wu C, Gao L, Qin F, Wei Q and Yuan J. Lysyl topsin (PRSS12), a protease whose absence oxidase family members in urological tumori- causes mental retardation. Brain Res 2007; genesis and fibrosis. Oncotarget 2018; 9: 1136: 1-12. 20156-20164. [33] Ma L, Huang C, Wang XJ, Xin DE, Wang LS, Zou QC, Zhang YS, Tan MD, Wang YM, Zhao TC, Chatterjee D, Altura RA, Wang C, Xu YS, Yang JH, Fan YS, Han BH, Si J, Zhang X, Cheng J, Chang Z and Chin YE. Lysyl oxidase 3 is a dual- specificity enzyme involved in STAT3 deacety- lation and deacetylimination modulation. Mol Cell 2017; 65: 296-309.

2705 Am J Transl Res 2019;11(5):2683-2705