INTRODUCTION

Early-onset gastric cancer (EOGC) is defined as any GC that occurs at the age of 50 years or earlier. EOGC comprises approximately 10% of all patients with GC, and their reported frequencies vary between 2.7% and 15% in various studied populations¹,². Germline pathogenic variants of the CDH1 gene are well-documented genetic factors associated with early-onset diffuse gastric cancer (EODGC)¹,². The CDH1 gene, located on chromosome 16q22.1, has 16 exons and a length of 98,250 bp. The most common isoform of the protein encoded by this gene is translated from a 4.5-kb RNA transcript. The protein E-cadherin, encoded by the CDH1 gene, is a cell adhesion molecule involved in the maintenance and homeostasis of normal epithelial tissue³.

Worldwide, GC is the fifth leading cause of cancer mortality, with 8.2 cases/100,000⁴. Although GC is one of the main causes of mortality by cancer in the world, few studies have investigated CDH1 variants in EODGC patients in the Mexican population⁵-⁹. To the best of our knowledge, the only variants in the CDH1 gene that has been reported so far in Mexican patients with EODGC are c.377del and the SNP rs16260. Our objective was to study CDH1 germline variants and their potential functional impact in patients with EODGC in a Mexican population.

METHODS

We studied seven patients (five men and two women) with EODGC. A diagnosis was made by a histopathologist who analyzed the histopathology of the gastric tumors obtained by endoscopy as part of their medical diagnosis (all patients had diffuse-type tumors and exhibited signet-ring cells). The patients were recruited by the Gastroenterology Department of the Hospital de Especialidades at Centro Médico Nacional de Occidente of Instituto Mexicano del Seguro Social located in Guadalajara City, Mexico. Patients were invited to participate in the study on a consecutive basis if they met three criteria: (i) the patients and their parents were born in Western Mexico; (ii) the patients were unrelated to each other; and (iii) they were of Mexican Mestizo ethnicity. Those patients who agreed to participate signed an informed consent letter. An Institutional Review Board and Ethics Committee approved the study. The mean age of the patients was 39.7 years with a range of 22-48 years (we chose an age < 50 years for this study, similar to Corso et al.¹). Three patients met the criteria for suspected hereditary diffuse gastric cancer because they were younger than 40 years of age and had tumors of diffuse histology according to the International Gastric Cancer Linkage Consortium¹⁰. Furthermore, two patients had a family history of cancer (the father of patient five died of prostate cancer and the mother of patient seven had lung cancer) (Table 1).

Genomic DNA samples were obtained from peripheral blood leukocytes by the salting out method. CDH1 variants were identified by polymerase chain reaction (PCR), followed by Sanger sequencing. Eighteen fragments were amplified, which included 16 exons, the promoter, and the 3’UTR region of the CDH1 gene. The primers used to amplify exons 2, 6, 7, 9, 10, 13, and 16 were previously described by Corso et al.¹ The remaining primers (the promoter region and exons 1, 3, 4, 5, 8, 11, 12, 14, and 15) were designed by our group (sequencing primers and conditions can be provided on request). Sanger sequencing is a robust testing strategy able to determine whether a point mutation or a small deletion/duplication is present. PCR amplification followed by sequencing is considered the diagnostic gold standard. A Ready Reaction Big Dye Terminator kit v. 3.1 (Applied Biosystems, Foster City, CA, USA) was used for sequencing. Multiplex ligation-dependent probe amplification (MLPA) analyses can reveal large deletions and rearrangements not detectable by sequencing, this technique was employed for the identification of large exonic deletions in the CDH1 gene using the SALSA® MLPA® Probemix P083-C1 CDH1 kit C1-0114 following the manufacturer’s recommendations (MRC-Holland, the Netherlands). An ABI 310 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA) was used for capillary electrophoresis; the ChromasLite v.2.6.6 and Coffalyzer programs were used for data analysis.

The prediction of regulatory single-nucleotide polymorphisms (SNPs) in the promoter region of the CDH1 gene was performed with DeepSEA and SNPClinic v.1.0 software. DeepSEA is a deep learning-based algorithm that can accurately predict the epigenetic state of a sequence, including transcription factor binding, DNase I sensitivities, and histone marks in multiple cell types and can further utilize this capability to predict chromatin effects of sequence variants and prioritize regulatory variants¹¹. SNPClinic v.1.0 software calculates the impact of SNPs on the alteration of transcription factor binding sites (TFBSs), according to the JASPAR database, when chromatin is accessible in the input cell line/tissue¹². To perform the SNPClinic analysis, the following 14 ENCODE cell lines potentially involved in inflammation, carcinogenesis and/or metastases (not only gastric cancer) were tested for DNase I HUP chromatin accessibility: Caco-2 (colorectal adenocarcinoma), H1 hesc and H7 hesc (embryonic stem cells), Hct116 (colon cancer), Hepg2 (liver cancer), Hpde6e6e7 (pancreatic duct), Hvmf (connective tissue), Osteobl (osteoblasts), Be2c (bone marrow), Medullo (brain), TH0, THH1, and TH2 (T helper lymphocytes), and chronic lymphocytic leukemia. To filter out SNPs not impacting TFBSs, only putative TFBSs that had relative binding scores (RBSs) ≥ 0.8 in the major allele were selected as binding TFs. Student’s t-test with p ≤ 0.05 on the null hypothesis was used to test whether the list of RBSs above the threshold was equal to those RBSs below the threshold. As an additional filtering step, only regulatory SNPs (rSNPs) with a functional impact factor (homotypic redundance weight factor × ΔRBS) ≥ an absolute value of ten were considered true positive rSNPs, according to our previous validation¹². Because SNPClinic v.1.0 was validated for proximal promoters for SNPs located in introns, exons, and 3’UTR regions, and due to the effect of insertions/deletions, we used the DeepSEA software¹¹ and both the Ensembl¹³ and ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/) databases. For the DeepSEA software, the most important chromatin features were selected by first applying a filtering threshold of an E-value < 0.01 and then by applying a threshold of log two-fold change of 0.02.

We corroborated the ancestry of Mexican EODGC patients by means of PCR genotyping of 22 autosomal short tandem repeat (STRs) with the PowerPlex® Fusion System (Promega Corp., Madison, WI, USA), followed by capillary electrophoresis in the ABI 310 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). Genotype assignment was performed with allelic ladders assisted by GeneMapper v.3.2 software (Applied Biosystems, Foster City, CA). The admixture analysis based on STR genotype was performed with Structure¹⁴,¹⁵. For this purpose, STR population databases that included Mexican Native Americans¹⁶, as well as Europeans and Africans from the USA, were employed as ancestral references¹⁷. The structure parameters employed herein offered consistent admixture estimates regarding those based on AIMs and genome-wide SNPs, as previously demonstrated in Mexican populations¹⁸.

RESULTS

The genetic admixture of the 7 Mexican EODGC patients mainly included European (73-87%) and Native American (8-22%) ancestries (Supplementary Figure S1). Because these results are in agreement with the previous descriptions of the Mexican population¹⁹, further discussion of this finding will be omitted.

All EODGC patients presented from 5 to 8 germline CDH1 gene variants, and all patients had at least one variant in the promoter region. A total of 12 different variants were identified in the CDH1 gene, all of which are already known as SNPs: six in the promoter regions c.–612_-611insA (rs34561447), c.−472delA (rs5030625), c.−285C>A (rs16260), c.−273G>A (rs1330727101), c.−197A>C (rs28372783), and c.−146C>G (rs942269593); two in exons 13 and 14 c.2076T>C (rs1801552) and c.2253C>T (rs33964119); three in introns 1, 13, and 15 c.48+6C>T (rs3743674), c.2164+17_2164+18insA (rs34939176), and c.2439+177delT (rs3556654); and one in 3’UTR c.*54C>A (rs1801026) (Table 2).

The most frequent variants were c.2076T>C (A692A) and c.2439+177delT, which were observed in all subjects. The variants c.−472delA, c.−285C>A, c. −273G>A, and c. −146C>G, located in the promoter region, were found in only one patient (Table 2). The allelic frequencies of these variants reported in other populations are listed in table 3.

Conclusive results of the MLPA analysis were obtained for only six subjects because one DNA sample was of poor quality (Table 1). No deletions or duplications of the CDH1 gene were observed in any of the six patients, since the amplified probes were observed within a radius of 1.

The putative functional impact of the identified variants in the EODGC patients is shown in table 2. rSNPs reveal which TFBSs are putatively affected, thereby decreasing or increasing the affinity of DNA to the TF (Table 1). Our results reveal that SNP c.-612_611insA (rs34561447) alters a TFBS for CTCF. In addition, flanking this SNP, we found two active chromatin signatures, H3K27Ac and H3K23Ac. The H3K27Ac signature also overlapped with the SNP c.-472delA (rs5030625) for which we found additional markers of active chromatin, such as H4K5Ac and H2AK5Ac. Notably, this signature was undetected by the SNPClinic software because it was designed to detect DNase I sites rather than histone acetylation.

The SNP c.−285C>A (rs16260) alters TFBS for myeloid zinc finger gene 1 (MZF1, formerly ZNF42 or MZF1B), interferon response factor 3 (IRF3), NF-YA, and c-Fos. Furthermore, we observed that CDH1 was downregulated in the 14 ENCODE cell lines in which c.−285C>A was found, including osteoblasts, Hct116 (colorectal cancer), PC9 (non-small-cell lung cancer), and NPC_2 (nasopharyngeal carcinoma) cell lines, as well as in the four cell lines observed in ENSEMBL. The SNP c.−273G>A (rs1330727101) modifies at least three TFBSs for IRF3 and specificity protein 2 (SP2) (Table 2).

DISCUSSION

The molecular basis for EODGC has not yet been completely elucidated. Although alterations in genes, such as CDH1, have been reported, germline mutations in the CDH1 gene are less frequent than somatic mutations²⁰. Pathogenic germline mutations occur in up to 8% of EODGC cases². However, in most cases, the germinal variants are nonpathogenic or are of uncertain significance¹,²⁰,²¹.

In our study, CDH1 pathogenic mutations in EODGC patients were ruled out; however, we identified 12 different germinal variants previously reported as SNPs (rs34561447, rs5030625, rs16260, rs1330727101, rs28372783, rs942269593, rs3743674, rs1801552, rs34939176, rs33964119, rs3556654 and rs1801026) that could contribute to the phenotypes of these patients. Some of them have been associated with pathological processes, for example: the variant rs34561447 has been found in patients with non-small cell lung cancer with a low frequency²²; the variant rs1801552 has been identified in various pathologies such as orofacial clefts²³, primary infertility²⁴, and colorectal cancer²⁵; and the rs1801026 was associated to poorer survival in breast cancer patients²⁶.

Regarding the identified variants in regulatory regions, the SNP c.−612_-611insA (rs34561447) alters a TFBS for the protein CTCF, which delimits 3D boundaries of insulators by mediating chromatin looping between its binding sites. Two active chromatin signatures were also found flanking this SNP: H327Ac, which is present in active enhancers²⁷ and H3K23Ac, which is recognized by both the oncoprotein TRIM24²⁸ and monocytic leukemic zinc finger-related factor²⁹.

The SNP c.−285C>A (rs16260) alters the TFBS for MZF1, which is reported as tumorigenic for GC; even the mRNA levels of MZF1 have been proposed as a prognostic marker³⁰. SNPClinic analysis revealed at least 8 TFBSs for MZF1 in the proximal promoter (−2 kb) of CDH1, providing stronger evidence that MZF1 regulates this gene. SNP rs16260 also alters the TFBS for the proto-oncogene c-Fos. One function of c-Fos is the induction of epithelial-mesenchymal transition (EMT), at least in colorectal cancer³¹. The transcription factor interferon response element 3 (IRF3) activates transcription of interferon-related genes with antiviral activity against double-stranded DNA and sRNA viruses in a toll-like receptor 3-dependent manner³². This result is interesting, because even though Epstein-Barr virus and hepatitis B virus have been associated with GC and precancerous lesions, respectively³³,³⁴, their genomes comprise double-stranded DNA, not sRNA. To date, the only sRNA virus known to be related to GC is the retrovirus human T-cell lymphotropic virus type I; surprisingly, infection by this virus diminishes the relative risk for this type of cancer³⁵. rs16260 has been reported as a risk factor for prostate cancer³⁶; in our study, this variant was observed in a patient whose father had prostate cancer (Table 1).

The SNP c.−273G>A (rs1330727101) modifies at least 3 TFBSs for IRF3 and specificity protein 2 (SP2). SP2 is a ubiquitous factor that binds to GC boxes and has been suggested as a regulator of T-cell antigen receptor α³⁷. SP2 is also important in the phenotype maintenance of T helper 17 cells (TH17), which play an important role in maintaining mucosal barriers and contributing to pathogen clearance at mucosal surfaces³⁸.

The SNP c.−197A>C (rs28372783) altered the TFBS for upstream stimulatory factors 1 and 2 (USF1 and USF2). USF2 and a truncated isoform were shown to have a dominant-negative effect on telomerase reverse transcriptase (TERT) expression and on overall telomerase activity during lymphocyte activation³⁹. Furthermore, USF1 and USF2 regulate other functions of the TERT enzyme, such as angiogenesis, inflammation, cancer cell stemness, and EMT. These extended functions are relevant in the dynamics and homeostasis of the tumor microenvironment⁴⁰. In addition, in the presence of c.-197A>C, nuclear factor erythroid 2 (NF-E2) is overexpressed in patients with myeloproliferative neoplasms⁴¹.

The SNP c.−146C>G (rs942269593) changes the TFBS for IRF3, NF-YA, and SP2. The impact of this SNP could be quite similar to that of the abovementioned c.−273G>A (rs1330727101). The additional feature of this SNP is that NFYA binding motifs (CCAAT boxes) are enriched (high homotypic redundancy) in the promoters of overexpressed genes in breast, colon, thyroid, and prostate carcinomas²⁰.

Regarding the analyzed SNPs in non-regulatory regions, these SNPs were in the ClinVar categories “benign” and “likely benign;” however, Ensembl showed that these SNPs overlap with sequences of transcripts involved in antisense-mediated decay and splicing alteration, including but not limited to intron retention and long non-coding RNA (lncRNA) alteration. A previous effort was made to understand the role of the lncRNA-miRNA-mRNA network in GC; however, the report did not consider the effects of SNPs⁴². To the best of our knowledge, there is no available in silico tool to quantify the effect of these SNPs in the overlapping of these long non-coding regions; therefore, additional research on this topic is needed.

The DeepSEA software gave at least two relevant predictions. First, the CTCF binding site is altered when the synonymous variant c.2253C>T (rs33964119) is present. When CTCF is not bound to its target DNA sequence, the RNA elongation rate is accelerated and can result in exon exclusion and alternative splicing⁴³. The Ensembl database includes two shorter CDH1 isoforms that lack exon 14, comprising 647 residues; therefore, it is possible that this variant could cause skipping of exon 14 by a previous alteration in the CTFC binding site. However, this was not confirmed in our studied EODGC patients because in their case, the sequence of exon 14 can be deleted in mRNA and protein but not deleted in DNA. The second relevant prediction was that SNP c.*54C>T (rs1801026) alters the TFBS for USF1 and USF2, resulting in similar mechanisms as the promoter SNP c.−197A>C (rs28372783).

In the Mexican EODGC patients investigated in this study, predictive bioinformatic analysis presented a plausible explanation of the potential differentiated molecular mechanisms for the phenotypes observed in each of the patients (Table 1). Our results indicate that pathogenic CDH1 germline variants are not common in EODGC, suggesting that the variants observed in these EODGC patients can contribute to the phenotypes in the patients, and we must consider that other genes, such as ARID1A or RHOA, can be carriers of pathogenic mutations, as has been previously suggested⁴⁴.

Regarding searching for deletions or duplications in the CDH1 gene, these germinal alterations were not identified in the studied EODGC patients. Other studies have also not found deletions in this gene, including a study of 25 Korean EODGC patients² and a study of 88 Brazilian EOGC patients⁴⁵.

Finally, the number of patients is quite limited for drawing conclusions about the exact molecular etiology of EODGC. However, these in silico results are interesting and have strong in vitro support, primarily because SNPClinic and DeepSEA are programs that use ENCODE, Roadmap Epigenomics, chromatin profiles and JASPAR databases as inputs, which are supported by a large number of in vitro experiments. The findings of this study will be validated in future in vitro and in vivo investigations.