Introduction

Colorectal cancer (CRC) remains a major source of cancer morbidity and mortality, as it accounts for the fourth most common cause of cancer deaths worldwide [1]. In Sweden 20% of new cases have at least one close relative with CRC and their family members are at increased risk of the disease [2]. Since CRC is considered to develop over precursor lesions detectable and removable in risk-individuals under surveillance, it is important to identify subjects at increased risk [3]. Less than 5% of all CRC represent high-penetrant CRC syndromes, most commonly familial adenomatous polyposis (FAP) or Lynch syndrome [2, 4].

High-risk genes have typically been identified via linkage analysis in large pedigrees or sequencing of candidate genes [5–7], and more recently also by next-generation sequencing (NGS) [4, 8–10]. Genome-wide association studies (GWAS) using chip-based techniques have been applied to study common genetic variants in order to find out the proportion of their contribution to the CRC risk. To date, less than 100 CRC susceptibility variants have been identified across less than 50 chromosomal regions [11–14], however, in total these SNPs and their surrounding regions represent no more than 2% of the heritability of CRC [15], and typically, each with a small increased risk (OR < 2).

Thus, a substantial risk is still to be explained since the genetic contribution to CRC is estimated to 35% [16]. The existence of rare high-risk alleles has been hypothesized [17]. In order to test the hypothesis that CRC risk was associated with variations within gene coding sequence, Illumina exome array was used in association studies. The studies used over 12 000 cases and more than twice as many controls from six different European populations, mostly from UK [18]. Some additional novel common CRC low-risk loci were suggested [18]. There is a substantial variation in the frequency of various single nucleotide polymorphisms (SNPs) between populations and even within populations [19, 20]. The vast majority of this association study included the heterogeneous population in England mixed with the more homogeneous population from Scotland, plus four other European populations [18]. In this work, a different study was designed in order to maximize the likelihood of finding rare high- or moderate risk genes. The new study used a more homogeneous population from Sweden and focused mostly on familial cases.

Materials and methods

Results

Post QC exome-wide association analysis was carried out on 1 388 cases and 12 108 controls using 221 642 genetic markers. 79 496 of the markers were monomorphic and 109 616 markers were rare variants (minor allele frequency (MAF) < 1%). The experiment was set up to search for rare alleles with an increased risk of CRC. Thirty-nine markers were observed with the minor allele associated with a statistically significantly elevated risk exceeding Bonferroni-corrected exome-wide threshold (p-value < 2.24E-07). Seven markers were removed due to cross-platform base-calling errors, and Sanger sequencing was performed on all suggested mutation carriers of the remaining 32 markers and confirmed the genotyping results of eight variants (Table 1). All the eight variants were rare (allele frequency < 1% in controls) and associated with moderate risks (OR > 2). Seven of the eight variants were missense variants in genes RAB11FIP5, POTEA, COL27A1, MUC5B, PSMA8, MYH7B, and PABPC1L, whereas the other was a downstream variant close to the gene NEU1 (Table 2). Six of the exonic variants were predicted to be pathogenic by at least one functional inference tool (Table 2). Three of the SNPs could be tested in available family members in four families and segregated more or less with CRC or adenomas in the families (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Pedigrees of the families tested for variant segregation.

Mutation carriers and non-carriers are indicated with red and green squares, respectively. Diagnosis and age (when available) is indicated under each individual in bold text, and sample IDs in Italic. CRC, colorectal cancer; HP, hyperplastic polyp; GaCa, gastric cancer; TVAHD, tubulovillous adenoma with high degree dysplasia; BrCa, breast cancer; TVA, tubulovillous adenoma; SALD, serrated adenomas with low degree dysplasia.

https://doi.org/10.1371/journal.pone.0193547.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Candidate risk variants identified in GWAS confirmed by Sanger sequencing.

https://doi.org/10.1371/journal.pone.0193547.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Functional prediction of validated risk variants.

https://doi.org/10.1371/journal.pone.0193547.t002

We compared our results with the findings in a previous study [18]. Among the 30 previously reported risk SNPs comprising of 15 statistically significant markers and 15 markers with less evidence, 27 were confirmed with an increased risk (OR > 1), including all 15 significant risk variants and 12 out of 15 variants that did not reach the significance threshold (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Comparison of previously reported risk variants [18] to the current study.

https://doi.org/10.1371/journal.pone.0193547.t003

To further study the possibility that the eight SNPs are not pathogenic per se, but rather associated with other pathogenic SNPs in adjacent regions, we selected eight additional markers among all SNPs located within 1 Mb distance from the eight original risk alleles based on their risk and significance (p-value < 3E-05) for further investigation. Sanger sequencing confirmed the genotyping results in three of the eight markers, located in genes ALMS1, COL27A1 and HNF4A, respectively (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Additional candidate risk variants located within 1 Mb from the eight original risk markers.

https://doi.org/10.1371/journal.pone.0193547.t004

Discussion

Our previous study, testing a heterogeneous population of unselected CRC patients and controls with the Illumina exome chip, confirmed some of the more common risk alleles and was able to suggest 30 new ones, with frequencies between 10–50% and ORs typical for the low-risk allele pattern of 1.1–1.3 [18]. The present study interrogating a more homogeneous population with a selection of mostly familial CRC patients demonstrated eight novel rare moderate-risk loci confirming the hypothesis of existing rare moderate-risk alleles associated with cancer (allele frequencies below 1% and ORs 2.0–17.6). Seven of the variants were missense mutations in seven different genes, RAB11FIP5, POTEA, COL27A1, MUC5B, PSMA8, MYH7B and PABPC1L, and one was in the downstream region of NEU1 (Tables 1 and 2).

None of these eight SNPs were found in our previous study [18]. This is likely due to the fact that the first study used unselected cases and the second mostly familial cases, thus more likely to demonstrate rare moderate-risk alleles. The suggested SNPs were mostly exonic and could constitute pathogenic SNPs. However, it is still possible that the SNP is only associated with another pathogenic SNP in its vicinity, and probably targeting the same gene and if so, it cannot be replicated easily using different populations. Findings from a study using mixed populations should be more likely able to be replicated in any population, and, in fact, we in the current study managed to confirm 27 of 30 suggested SNPs from our first paper (Table 3). Another reason for not observing the eight SNPs in the previous study could be that these eight SNPs were not successfully genotyped within the first study. Studies using very rare SNPs are prone to genotyping errors and artefacts, and in fact many more SNPs were suggested but ruled out in our study leaving eight out of 39 from primary analysis.

Both cases and controls in this study were analyzed using Illumina Infinium BeadChip, but on different platforms at different time points. We designed the study by only using SNPs successfully genotyped on both genotyping platforms. In general, genotypes were called by GenCall algorithm in both datasets, however, an additional algorithm zCall was integrated for rare variant calling in the control dataset but not the case dataset. In order to eliminate false risk variants due to incorrect calling, we sequenced all cases as carriers for the suggested risk SNPs. Calling artefacts were observed in most of the suggested SNPs, leaving only eight (25%) suggested moderate risk genetic variants validated. This implicates the challenging nature of rare variant calling.

Accurate estimation of population allele frequency for rare variants depends particularly on representative population and adequate sampling. Although combining data could be used to reduce variance and improve estimation, pooling of samples representing distinct populations can lead to biased estimates. Difference in MAF between European population and all populations in the 1000Genomes [19] and ExAC [20] databases indicates the divergence among populations (Table 1). The SweGen project [31] included 1 000 individuals reflecting a cross-section of the Swedish population, but its sample size could be insufficient for estimation of rare allele frequency. In this study, we used 12 108 unaffected individuals from the Swedish Twin Registry as controls, likely providing a better representation of the Swedish population. This could possibly explain that for some candidate SNPs, the risk allele frequency in CRC cases is not substantially higher than that in SweGen (Table 1).

The variant rs201302413, had the highest OR, 17.6, and was in a gene not well studied. The gene, PABPC1L, is known to be involved in transcription and has been shown to be down-regulated by preoperative radiotherapy in rectal cancer [32]. Single cell sequencing in one CRC showed one clone with a PABPC1 mutation together with a CDC27 mutation [33].

The variant rs148220987 in the RAB11FIP5 gene had an OR of 6.3. The Rab GTPases family regulates intracellular membrane trafficking in eukaryotic cells and is known to be involved in cancer signaling pathways [34]. Rab11 regulates JNK and Raf/MAPK-ERK signaling pathways [35] and Rab1 and Rab11 are playing a key role in Notch signaling via vesicular trafficking [36]. Recent discoveries have demonstrated a family of genes downstream of the Rab GTPase, the FIP family [37]. RAB11FIP5 or Rip11, is one of them. FIP function is not well known but shown to be involved in crucial cellular physiological processes such as cell division, and cell migration in various human cancers [37].

The SNP rs754511 in the MYH7B gene had an OR of 3.6. This gene is not well studied in CRC but has been studied in melanoma, and MYH7B SNPs have been associated with increased melanoma risk [38–40]. It did not segregate well in family 340 (Fig 1).

rs200554635 in the MUC5B gene had an OR of 3.5. Stromal genes such as mucins have been shown to affect carcinogenesis, and polymorphisms in microRNA binding sites of those genes have been suggested to predict clinical outcome in CRC patients [41]. Mutations in gene GALNT12 have been found in colonic cancers and also suggested to predict CRC and this enzyme is initiating mucin type O-linked protein glycosylation and may contribute to a subset of colon cancers [42]. Moreover, Cox2 is well known to be involved in CRC carcinogenesis and has been suggested to act by inducing Muc5B and Muc17 secreting cells in the pathogenesis of esophageal cancer [43].

The SNP rs12193504 is located close to the NEU1 gene, which has been studied extensively in cancer and suggested to have a profound effect in human cancers [44]. The gene plays a role in sialidase-mediated regulation of tumorigenesis including growth factor receptor signaling, control of TOLL-like receptor signaling and immune-mediated tumorigenesis, regulation of epithelial mesenchymal transition as well as acquired chemo-resistance and regulation of vascularization [44]. NEU1 has specifically been suggested to suppress metastasis in human colon cancer cells [45]. This NEU1 related SNP is located within transcription factor binding sites and DNase hypersensitivity clusters, suggesting its potential regulatory function. This marker could be tested and found in one additional patient with stomach cancer in family 1164 and in one other CRC patient in family 275, but not in another person with a high-risk adenoma from the same family (Fig 1).

We detected variants of rs202238834 in three of the cases, leading to an allele frequency of 0.001, which is higher than the MAF of 0.0001 from the ExAC project. None of the controls had this variant. This variant is located in the exonic region of POTEA gene, member of a highly homologous gene family expressed in a wide variety of human cancers (colon, lung, breast, ovary and pancreas) [46]. Biological function of this gene family is not clear, but there is evidence for its role in inducing programmed cell death [47].

The SNP rs34578955 has a MAF of 0.03%-0.1% in European people, compared to allele frequency 0.14% in our cases. This SNP is within the COL27A1 gene. Collagens are stromal genes known to be involved in carcinogenesis in experiment animal models [48] and suggested to influence epithelial cells and tumor growth by influences from stromal cells [49]. We have previously suggested that expression of COL11A1 and COL5A3 in CRCs could be associated with CRC carcinogenesis [50, 51]. This SNP segregated in three patients with CRC over three generations (family 1213) supporting the SNP to be involved in all family members with CRC in this family (Fig 1). Furthermore, another variant in the splice region was found when the locus was searched for additional markers with suggestive p-value (3E-05) (Table 4).

The risk allele of rs137990346 occurred with a 0.14% frequency among our cases and none in controls, which is reasonable considering the expected frequency in SweGen, ExAC and 1000Genomes. The SNP lies within the PSMA8 gene, a specific subunit involved in histone acetylation and could play a critical role in chromatin remodeling, DNA repair and epigenetic regulation of gene expression [52].

Moreover, we also searched for additional candidate markers located close to the eight risk variants. Three more SNPs, rs202114540, rs140849642 and rs201749293 were found within 1 Mb from rs148220987, rs34578955 and rs201302413, respectively, and verified by Sanger sequencing (Table 4). Each of these three variants was observed in two unrelated CRC cases, but not in any controls, giving a p-value of 3E-05 which is suggestive for risk variants. SNP rs140849642 is located in the splice region of gene COL27A1, confirming the involvement of COL27A1, whereas markers rs202114540 and rs201749293 are in the coding regions of ALMS1 and HNF4A. Mutations in ALMS1 (centrosome and basal body associated protein) gene are known to cause Alström’s syndrome [53, 54]. Several GWAS also reported variants in or near ALMS1 to be associated with chronic kidney disease [55, 56]. HNF4A (hepatocyte nuclear factor 4 alpha) encodes a nuclear transcription factor regulating metabolism, cell junctions, differentiation and proliferation in liver and intestinal epithelial cells. Expression profile of its isoforms has been demonstrated modified in many cancers including CRC [57], and it was implicated that the interaction between Src tyrosine kinase and HNF4α has important implications for colon and other cancers [58].

In conclusion, we suggested eight novel candidate CRC risk loci, within the Swedish population. The study suggest that low frequent risk alleles contribute to the risk of cancer in seemingly high-risk families and also that the total risk in these families are contributed by more than one risk factor such as in complex disease.

Supporting information

Acknowledgments

We are grateful to the patients for their contribution and Anna-Lena Kastman for technical assistance. We acknowledge The Swedish Twin Registry for access to data. The Swedish Twin Registry is managed by Karolinska Institutet and receives funding through the Swedish Research Council under the grant no 2017–00641.