Introduction

Hereditary Non-Polyposis Colorectal Cancer (HNPCC) is a familial syndrome with an increased incidence of colorectal (CRC) and other related cancers [1,2], defined by the Amsterdam I or II clinical criteria. It is well established that approximately half of HNPCC cases are explained by germline mutations in the DNA mismatch repair (MMR) genes, mainly MLH1, MSH2, MSH6 and PMS2. As a consequence, these cases present MMR pathway defects and microsatellite instability (MSI) in the tumors, and are referred to as Lynch Syndrome (LS) [3,4].

The universal screening for LS currently comprises two different stages. Firstly, the immunohistochemistry (IHC) of the MMR proteins and/or the MSI status is studied in the tumor of every CRC patient, as well as in some endometrial cancers [5,6]. When this result is positive (MSI/absence of MMR) or if the tumor is not available, patients with a family history of CRC are then screened for germline mutations in the MMR genes, which was previously performed by methods such as Denaturation Gradient Gel Electrophoresis (DGGE) or High Resolution Melting (HRM), followed by Sanger sequencing of samples with altered patterns. Initially, candidates for this LS genetic testing were identified based on the Amsterdam criteria [7,8] However, these algorithms may miss some individuals with LS [9], reason for which the more lenient Bethesda guidelines were created to identify high-risk families that should undergo genetic testing. Although all of these are effective screening tools, they may still miss a proportion of patients with LS. It is worth noting that the IHC and MSI tests have lower sensitivity for detecting MSH6 and PMS2 mutation carriers in particular [9], and that the screening of PMS2 is a challenge due to the high number of pseudogenes.

After this screening, only those families in which a germline pathogenic mutation is found in one of the MMR genes are diagnosed with Lynch Syndrome. Those cases that show MSI/MMR defects in the tumor but lack the corresponding germline MMR mutations are classified as unexplained MMR deficiency, whereas the other half of HNPCC families with no evidence of MMR deficiency has been designated Familial Colorectal Cancer Type X (FCCTX) [10–13]. FCCTX patients lack germline MMR mutations and their tumors are microsatellite stable (MSS). The lack of information about the hereditability of cancer risk in all these unexplained families makes it difficult to carry out an individualized genetic counseling.

Next-generation sequencing (NGS) has revolutionized cancer genomics research, and can be used to search for Mendelian disease genes in an unbiased manner by sequencing the entire protein-coding sequence of known predisposition genes [14]. The practice of genetic testing is rapidly evolving owing to the recent introduction of multigene panels for the diagnosis of hereditary cancer [15]. Multigene panels can be a cost and time-effective alternative to sequentially testing multiple genes. Virtually all multigene panels include high-penetrance genes that establish the risk of a particular type of cancer (such as breast or colon), but also many moderate and low-risk genes. This challenges the personalized management of guidelines when a pathogenic mutation is found, since the phenotypic spectrum and penetrance are less defined or unknown for the latter [16]. The TruSight Cancer Sequencing Panel has been developed by Illumina in collaboration with experts in cancer genomics, and targets a set of 94 well-known cancer-predisposing genes.

The purpose of the present study is the use of multigene panel testing for the diagnosis of hereditary cancer in individuals from high-risk colorectal cancer families.

Methods

Patient selection

A total of 1204 high-risk CRC families have been referred for genetic counseling and/or gene testing for Lynch Syndrome at the Cancer Genetic Clinic of Hospital Clinico San Carlos between the years 2000 and 2016. Among them, 393 families fulfilled the Amsterdam I/II or Bethesda clinical criteria and were molecularly characterized by the study of MSI and/or MMR protein expression in the tumor, and the screening of germline MMR gene mutations (MLH1, MSH2, MSH6, PMS2 and EPCAM, located upstream of MSH2) by HRM followed by Sanger sequencing [17,18]. Only 141 families in which a pathogenic germline MMR mutation was found were diagnosed with Lynch Syndrome, while the rest of the families could not be explained by these single-gene analyses. Other genes such as POLE, POLD1 and NTHL1 were also studied with no positive outcome.

For the present study, we selected 98 patients from those unexplained high-risk CRC families for the test of a multigene cancer panel by NGS (Fig 1). The prioritization of the families was based on the absence of MMR proteins, presence of MSI, lower age at diagnosis or higher number of cancer patients in the pedigree. An MLH1 mutation carrier was added as a positive control. Participants were asked to donate 10ml of blood at the time of their initial visit. Personal and family histories were obtained from the proband and participating relatives, and cancer diagnoses were confirmed by medical and pathology records. A written informed consent was signed by each participant, and the study was approved wirh an internal code n° 16/204-E_BS by the Clinical Investigation Ethics Committee (CEIC) from the Hospital Clinico San Carlos.

Download:

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

Fig 1. Next generation sequencing workflow using the TruSight cancer sequencing panel.

https://doi.org/10.1371/journal.pone.0203885.g001

Results

Patient samples

Germline DNA was analyzed from 98 patients belonging to high-risk CRC families using the Illumina TruSight Cancer Sequencing Panel, which targets a set of 94 genes known to play a role in cancer predisposition. All patients had been previously tested for Lynch Syndrome by IHC/MSI analyses in the tumor and/or germline MMR single-gene mutation screening, with a negative result for the latter. Characteristics of the studied cohort are detailed in Table 1. The studied families were predominantly affected by CRC; however, other malignancies were observed in some family members, including endometrial, gastric, ovarian, breast, renal and pancreatic cancers. The mean age at diagnosis was 49.1 years old. Most of the carcinomas had an MSS phenotype and presence of the MMR proteins in the tumor. Only 11 patients showed an MSI phenotype, 7 of which had absence of at least one MMR protein (but with no germline MMR mutations detected). The majority of CRCs were Dukes B or C, and nearly all were left-sided.

Download:

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

Table 1. Clinical and molecular characteristics of the 98 probands.

https://doi.org/10.1371/journal.pone.0203885.t001

Results from the NGS targeted sequencing

Among the 98 patients, we found 19 pathogenic or likely pathogenic variants in 18 patients (18.4%), all of which were validated by Sanger sequencing. Table 2 shows the clinical and molecular features of the patients with these mutations. Out of the 18 patients, 8 had MLH1/MSH6/PMS2 mutations and 10 carried non-MMR mutations. Four out of the 8 patients with MMR mutations had MSI tumors and absence of the corresponding MMR proteins, while another 3 patients had a discordant tumor screening: one (ID 820) with a frameshift mutation in PMS2, c.714_720dup (MSI/presence of PMS2), another (ID 7400) with a splicing mutation in MLH1, c.1731+4A>G (MSS/absence of MLH1/PMS2), and the last (ID 7934) with a missense mutation in MLH1, c.677G>T (MSI/presence of MLH1/PMS2). Patient ID 555 carried two MMR variants, one in MLH1 (c.2050T>G, p.Y684D) and another in PMS2 (c.825A>T, p.Q275H). This patient was diagnosed of CRC at 35 years old, belonged to an Amsterdam I family and had an MSI tumor with absence of MLH1/PMS2. In silico studies of both mutations showed that the PMS2 mutation was neutral while the MLH1 mutation was predicted to be highly damaging. On the other hand, all of the 10 families with pathogenic variants in non-MMR genes showed MSS tumors, and in one of the families (patient ID 1041) we found two pathogenic variants in MUTYH (Table 2). From the remaining patients, 55 (56.1%) were revealed to only carry variants of unknown significance (VUS) in 38 different genes (S2 Table), while 25 (25.5%) just carried polymorphisms.

Download:

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

Table 2. MMR status in tumors from patients with selected variants identified by the TruSight cancer sequencing panel.

https://doi.org/10.1371/journal.pone.0203885.t002

It is worth noting that, in total, only three splicing variants were not validated by Sanger sequencing (EZH2 c.1947+1G>T, MSH2 c.942+2T>G and MLH1 c.1059-1G>A) and were eliminated from the data. The CNVs were also analyzed as described in Materials and Methods (S3 Table). However, potential structural variations were discarded by the study of SNPs in the corresponding chromosome localization using IGV. It was not possible to analyze some CNVs due to the low coverage or absence of SNPs in the region. Three CNVs affecting BRCA2, MSH2 and EPCAM were not confirmed in 2 of the samples (786 and 2954) by MLPA.

Type, prediction and frequency of likely pathogenic mutations by gene

Out of the 19 germline pathogenic or likely pathogenic mutations detected, 8 (42.1%) were found in MMR genes (5 in MLH1, 1 in MSH6 and 2 in PMS2). The remaining 11 mutations were detected in other cancer predisposing genes, including BRIP1 (n = 1), CHEK2 (n = 1), SMAD4 (n = 1), MUTYH (n = 4), HNF1A (n = 1), XPC (n = 1), TP53 (n = 1), and APC (n = 1). The type, prediction and frequency of all the mutations can be observed in Table 3. Among them, there were 9 missense, 4 frameshift, 3 stop-gained and 3 splice site variants. All these variants were rare, and 11 had frequency data not available (NA) in ExAc nor in GnomAD (Table 3). 15 of these mutations were classified as pathogenic or likely pathogenic by the ClinVar and/or InSight databases, following the 5-tier classification system proposed by Plon and colleagues [22]. In addition, 4 of them did not appear in any of the variant databases mentioned earlier, but were considered likely pathogenic due to the type of mutation, in silico predictions and/or the molecular features of the tumor. These novel variants were: SMAD4 c.1194G>A, PMS2 c.714_720dup, MSH6 c.1635_1636del and MLH1 c.2050T>G.

Download:

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

Table 3. Pathogenic and likely pathogenic variants by gene identified by the TruSight cancer sequencing panel.

https://doi.org/10.1371/journal.pone.0203885.t003

Frequency of VUS in cancer susceptibility genes

All the variants found by NGS were analyzed by ClinVar, and those variants classified as class 3 were selected as VUS. Only those VUS with a very low frequency (<0.005) in ExAc are included in S2 Table. The VUS in cancer genes can be grouped by their functional effect; Fig 2A shows that most of the VUS selected were located in genes involved in DNA repair mechanisms, tumor suppressor genes and proto-oncogenes. Fig 2B shows that the relationship between the number of patients and the number of variants per patient is inversely proportional: 33 patients had 1 VUS, while only one patient (ID 1008) had 5 VUS, and was curiously a patient with a strong family history. Eleven patients (IDs 499, 763, 820, 987, 1041, 1144, 1564, 1652, 1936, 3775 and 7934) with VUS also had a concomitant deleterious mutation (data shown in Tables 3 and S2 Table). Among the 55 patients that carried only VUS (56.1%), 31 fulfilled the Amsterdam I/II clinical criteria.

Download:

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

Fig 2. Class 3 variants found in genes implicated in hereditary cancer and clinical experience.

A) Number of rare VUS (MAF<0.005) per gene or group of genes, in those cancer susceptibility genes with at least 2 filtered VUS. The different FANC, ERCC and SDH genes were grouped together for simplification. B) Number of unclassified variants per patient of the study cohort. C) Clinical practice experience with multigene panel study. VUS: variants of unknown significance.

https://doi.org/10.1371/journal.pone.0203885.g002

Discussion

Approximately half of HNPCC families carry a germline, pathogenic mutation in one of the MMR genes and are thus considered Lynch Syndrome families. The other half does not present any evidence of MMR deficiency and the genetic basis underlying their cancer predisposition remains unknown, reason for which they are called FCCTX [13,23,24]. However, unexplained families also comprise cases which present MMR defects and/or MSI in the tumors but lack the corresponding germline MMR mutations. All together, these families represent a significant number of cases at the Genetic Counseling Units and they are considered a problem in the clinic. Despite this, the genetic understanding of hereditary CRC syndromes has grown over the years, leading to an increasing request for genetic testing [25]. The limitation of the old screening methods due to their lower sensitivity and the reduced number of genes studied has led to the rise of multigene panel testing in oncology [16,26], and given the advantages of analyzing multiple genes, the benefits of its application in the clinical practice are obvious [15,27].

In this study, 98 unexplained families were reanalyzed using Illumina's NGS TruSight Cancer Sequencing Panel, which targets 94 genes known to play a role in cancer predisposition. The panel testing identified 8 MMR mutations in our cohort (5 in MLH1, 1 in MSH6 and 2 in PMS2), 7 of which were found in patients whose tumors showed an altered MMR status (MSI and/or absence of MMR). These cases were missed by our prior screening, and are thus a result from improved testing for these genes. Although these patients represent most of the Lynch-suspected families included in our study, there were another 6 families that were not resolved. The panel also allowed the identification of mutations in other well-known CRC high-penetrance genes, such as MUTYH (biallelic), APC, SMAD4 and TP53, as well as in moderate-penetrance genes like MUTYH (monoallelic), CHEK2, HNF1A, BRIP1 and XPC. In total, pathogenic or likely pathogenic mutations were found in 18.4% of our cohort, while high-penetrance mutations represented 12.2% of the studied patients.

Four of the identified variants were novel and had not been previously described in any of the variant databases checked (ClinVar, InSiGHT and UMD), although SMAD4 c.1194G>A had been reported at somatic level in COSMIC. There is enough evidence to claim that the 3 deleterious novel variants (SMAD4 c.1194G>A, PMS2 c.714_720dup and MSH6 c.1635_1636del) are pathogenic, so they will be added to the public databases for future reference. Regarding the missense variant (MLH1 c.2050T>G), it was present in a patient whose tumor was MSI with absence of MLH1/PMS2. In addition, in silico studies showed the change as likely pathogenic. However, the segregation studies performed were not very informative given the lack of affected living relatives, so more studies are needed in order to confirm the pathogenicity of this variant.

Among all the non-MMR genes, APC, MUTYH and SMAD4 are well known to be implicated in CRC, specifically in polyposis. It is worth noting that the patient in which the APC mutation was found had over 30 polyps. However, APC as well as other polyposis-associated genes had already been screened with no positive results (data not shown). For that reason, the family was added to our study on the grounds that it fulfilled the Bethesda criteria. Like the MMR mutations identified, this variant represents a false-negative of previous screenings. Regarding the biallelic MUTYH carrier, there was no information of the presence of polyps at the time of patient selection, but a deeper look into the family history revealed that the patient did present multiple polyps. On the contrary, neither the SMAD4 nor the 2 monoallelic MUTYH mutation carriers showed any sign of polyposis to our knowledge. Nevertheless, the risk that monoallelic MUTYH mutations confer is uncertain, so it is unlikely that they are the only cause of cancer in the corresponding families [27].

The family that carried the TP53 mutation was not a classic Li-Fraumeni family, but did fulfill the revised Chompret criteria. It had been classified as an Amsterdam II family because there was one member affected with ovary cancer and another 3 with CRC at early ages, but there were also 2 lung cancers within the family and a cousin of the proband had developed a sarcoma very young.

The remaining genes with pathogenic or likely pathogenic mutations had moderate or less defined cancer risks. In the first place, the BRIP1 gene encodes a member of the RecQ DEAH helicase family that interacts with the BRCT repeats of BRCA1. The bound complex is important in the normal double-strand DNA break repair and appears to be involved in breast and ovarian cancer, where it acts as a tumor suppressor [28]. Rafnar et al. showed that women with BRIP1 mutations have an increased risk for ovarian cancer that may be as much as 5 times higher than the risk in non-carriers [29]. Noteworthy, one of the BRIP1 mutations reported by Rafnar et al. (c.1702_1703delAA, p.N568WfsTer9) was also found in a Spanish CRC patient [30], and the tumor showed a loss of the wild type allele in both studies. Here we have identified a different frameshift BRIP1 mutation (c.903delG, p.L301FfsTer2) in a patient diagnosed with CRC at 63 years old. The family fulfilled the Bethesda criteria but was not very informative due to its reduced size and limited information. Therefore, we still cannot determine the risk for CRC that BRIP1 mutations confer.

Another gene was CHEK2, a tumor suppressor that is activated when DNA is damaged or when DNA strands break. The c.1100delC mutation of CHEK2 has been confirmed to confer an increased risk of breast cancer in women unselected by family history [31,32,33]. The lifetime risk of developing breast cancer among women with a CHEK2 mutation has been reported to be approximately 25% [34]. In our cohort, we have found CHEK2 c.349A>G (p.Arg117Gly), which was considered likely pathogenic by Shoda et al. and proved to produce a nonfunctional protein both by biochemical and bioinformatics analyses [35]. In addition, their results suggest that both of these mutations cannot act in a dominant-negative manner and that tumorigenesis associated with this mutation may be due to haploinsufficiency [35].

We also identified a mutation in XPC, a key component of the XPC complex that plays an important role in the early steps of global genome nucleotide excision repair and is involved in damage sensing and DNA binding. The XPC mutation found in our study (c.1001C>A, p.Pro334His) had been described as pathogenic in ClinVar by Johns Hopkins University. However, we have identified this mutation for the first time in a CRC patient, who was diagnosed at the age of 50 and did not have any cancer family history. The patient died soon after the diagnosis, not being able to get any additional information. Last but not least, HNF1A is a transcription factor that regulates tissue specific expression of many genes. This gene is implicated in diabetes and had been described in renal cancer, but its role in CRC is unknown. We are also describing the mutation HNF1A c.92G>A (p.Gly31Asp) in CRC for the first time.

An increasing number of studies have been published over the past few years addressing the benefits of NGS panel testing for the diagnosis of hereditary cancers as compared with the traditional targeted single-gene screenings. While all of them agree on the huge advantages of panel testing, especially due to its time and cost-efficiency and its higher sensitivity, the only question that remains open to debate is the selection of genes that should be included [36–40]. Here we have used a 94-gene panel, which is to our knowledge the highest number of genes reported in this kind of studies by far. However, this choice was made based on a number of reasons. On the one hand, we have learned from the Genetic Counseling Unit that there is a relatively frequent overlapping of phenotypes among the different hereditary cancer syndromes [36,37], since the tumor spectrum is sometimes wider than expected [27,41] and the information provided by the family is sometimes incomplete. Therefore, we believe that the best strategy is to group all the syndromes into one single cancer-predisposing multigene panel, as proposed by some other groups [27,36], instead of dividing patients based on their phenotypes.

On the other hand, we strongly support the incorporation of lesser-known genes to NGS panels on a research basis, since the additional cost of adding these genes is minimal [42], and with the intensity of current research the uncertainty of many emerging genes is likely to be resolved soon [38]. This means that some variants that are not informative at the moment may be actionable in the future [38]. In addition, by adding these genes to our panel we are contributing to the accumulation of international research data, which is the only way to continue improving our understanding of CRC genetics [39]. For this reason, we also believe it is highly valuable to include a detailed list of VUS (S2 Table), something that most published studies fail to do [38]. The number of VUS identified cannot be compared with other studies, though, since it is associated with the number of genes on the panel [40]. Despite the high number of genes included in the panel used for the present study, there were some genes that were left out, such as POLE, POLD1, NTHL1 or MSH3. Although some of them (POLE, POLD1 and NTHL1) had already been screened in our cohort, this is a limitation of our study. It goes without saying that we would definitely recommend that future panels used in clinical studies for colorectal cancer families should include those genes as well, for the same reasons that were discussed above.

The clinical practice experience obtained with this multigene panel is shown in Fig 2C. Among all the families that were screened only 18.4% were informative, although this group is underrepresented considering that only unexplained families from previous screenings were included in the study. Out of these, 66.7% carried likely pathogenic mutations in high-penetrance genes and could benefit from a true genetic counseling, taking measures such as reducing the surveillance in non-carriers, who would avoid the stress attached to the lack of awareness. Regarding moderate-penetrance genes, a study with a larger number of patients is needed in order to establish the exact risk they confer. The introduction of NGS panels in the clinical routine of the hospital will help us with this task. Those patients who were just informed of VUS (56.1%) would also take advantage of this measure, since the only thing we can do for now is to keep track of public databases, study the segregation and do functional studies when recommended in order to improve their genetic counseling in the future. The remaining patients (25.5%) were informed that no gene had been found to be involved in their cancer predisposition. Although data regarding lesser-known genes and VUS is highly valuable from a research point of view, participants should be always informed about the limited clinical actionability of testing for genes that are not associated with their phenotype or have moderate penetrance.

In conclusion, the detection of new pathogenic mutations in high-penetrance genes can contribute to the explanation of the cancer heritability in our families, changing the individual clinical management. The NGS panel approach has the advantage of analyzing multiple genes in multiple samples simultaneously, reducing costs and time and increasing the sensitivity in comparison to targeted single-gene screenings. Therefore, multigene panels should be included in clinical laboratories for the screening of all high-risk cancer families regardless of other analyses in the tumor. The number of genes to be included in these panels is debatable, though, and should fit the purposes of each study.

Supporting information

Acknowledgments

We are very grateful to the families for their cooperation, and we would also like to acknowledge Alicia Tosar, Isabel Diaz and Paula Diaque for their technical assistance. We acknowledge Jordi Camps for his advice about the CNVs.