Clinical applications of next-generation sequencing in colorectal cancers

Abstract

Like other solid tumors, colorectal cancer (CRC) is a genomic disorder in which various types of genomic alterations, such as point mutations, genomic rearrangements, gene fusions, or chromosomal copy number alterations, can contribute to the initiation and progression of the disease. The advent of a new DNA sequencing technology known as next-generation sequencing (NGS) has revolutionized the speed and throughput of cataloguing such cancer-related genomic alterations. Now the challenge is how to exploit this advanced technology to better understand the underlying molecular mechanism of colorectal carcinogenesis and to identify clinically relevant genetic biomarkers for diagnosis and personalized therapeutics. In this review, we will introduce NGS-based cancer genomics studies focusing on those of CRC, including a recent large-scale report from the Cancer Genome Atlas. We will mainly discuss how NGS-based exome-, whole genome- and methylome-sequencing have extended our understanding of colorectal carcinogenesis. We will also introduce the unique genomic features of CRC discovered by NGS technologies, such as the relationship with bacterial pathogens and the massive genomic rearrangements of chromothripsis. Finally, we will discuss the necessary steps prior to development of a clinical application of NGS-related findings for the advanced management of patients with CRC.

Core tip: Next-generation sequencing (NGS)-driven genomic analyses are facilitating the genomic dissection of various types of human cancers, including colorectal cancer (CRC). This review contains an up-to-date summary of recent NGS-based CRC studies and an overview of how these efforts have advanced our understanding of colorectal carcinogenesis with novel biomarkers for genome-based cancer diagnosis and personalized cancer therapeutics.

INTRODUCTION

Colorectal cancers (CRC) are the third most common human malignancy, and are also the leading cause of cancer-related deaths worldwide[1]. Early detection of premalignant lesions such as adenomatous polyps has decreased the risk of CRCs[2], however, cases which are initially undetected and progress to advanced CRC with distant metastasis are still unfortunately incurable[3]. The development of CRC is a complex and heterogeneous process arising from an interaction between multiple etiological factors, including genetic factors[4] and environmental factors such as diet and lifestyle[5]. Recently, significant progress has been made in the characterization of genetic and epigenetic alterations in CRC genomes in support of the genomic view of colorectal carcinogenesis. Like other types of human solid tumors, CRC genomes harbor various types of genomic alterations ranging from small-scale changes (i.e., point mutations or small indels) to large-scale chromosomal copy number changes or rearrangements. Some of these alterations may contribute to colorectal carcinogenesis as oncogenic drivers, but the full spectrum of driver genomic alterations in CRC genomes is still incomplete.

For decades, the genome-wide profiling of cancer genomes has been mainly conducted using hybridization-based microarray technologies (i.e., expression microarrays and array-based comparative genomic hybridization)[6,7] or low-throughput Sanger sequencing[8]. Recently, the advancement of DNA sequencing technologies - next-generation sequencing (NGS) - has revolutionized the speed and throughput of DNA sequencing[9,10]. Table 1 lists the NGS platforms widely used in the characterization of cancer genomes. Since the first attempt at cancer genome sequencing using NGS technology[11], successful sequencing by NGS has been accomplished in many major human cancer types[12,13] including gastrointestinal malignancies such as esophageal[14], gastric[15], colorectal[16], and hepatocellular carcinomas[17,18]. The NGS-based studies of CRC genomes are summarized in Table 2. These studies identified the unique mutational spectrum and novel targets of genomic alterations in respective cancer types with biological and clinical significance.

Table 1.

The next-generation sequencing platforms for cancer genome analysis

NGS types	Whole genome sequencing	Exome sequencing	Epigenome sequencing1	RNA-seq
Source	Genomic DNA	Genomic DNA (targeted)	Genomic DNA (targeted)	RNA
Alteration types	Point mutations and indels, rearrangements2, DNA copy number changes	Point mutations and indels	DNA methylation and posttranscriptional histone modifications	Gene fusions3, alternative splicing events, point mutations and indels

¹Epigenome sequencing can be classified into chromatin immunoprecipitation (ChIP)-based methods to detect genomic domains with epigenetic modifications[80] and more direct DNA methylation sequencing such as bisulfite-sequencing[59,60];

²From whole genome sequencing data, the genomic rearrangements and DNA copy number changes are generally detected by paired-end mapping[81] and read-depth based methods[82], respectively;

³Gene fusions can also be identified by paired-end, high-coverage whole genome sequencing, but the transcription-related events such as exon skipping or other alternative splicing events can only be identified by RNA-seq. NGS: Next-generation sequencing.

Table 2.

The list of next-generation sequencing-based studies of colorectal cancer genomes

Ref.	NGS types	Major findings	Alteration types and software used
Bass et al[16]	WGS (9 pairs; tumor-matched normal)	Oncogenic fusion (VTI1A-TCG7L2)	Point mutations (MuTect[83])
			Indels (Indelocator)1
			Rearrangements (dRanger)1
TCGA consortium	WGS (97 pairs; low-pass)	See main text	Point mutations (MuTect)
	RNA-seq (218 tumors)		Recurrent mutations (MutSig)1
	Exome-seq (254 pairs)		DNA copy numbers (BIC-seq[82])
			Rearrangements (BreakDancer[81])
Timmermann et al[84]	Exome-seq (2 pairs, one MSI-H and one MSS)	Comparison of mutation spectrum between MSI-H and MSS CRC genomes	Point mutation and indel (Vendor-provided GS reference mapper, Roche)
Zhou et al[85]	Exome-seq (1 series: normal-adenoma-adenocarcinoma)	Comparison of benign and malignant CRC genomes in the same patient	Point mutation and indel (Samtools[86])
Kloosterman et al[73]	WGS (4 pairs; primary-metastasis-matched normals)Targeted 1300 genes (4 pairs)	Comparison of primary or metastatic CRC genomes	Chromothripsis and mutations (Burrow-Wheeler aligner[87] based in-house tools)
Brannon et al[88] (Proceedings)	Targeted 230 genes (50 pairs: primary-metastasis-matched normals)	Comparison of primary or metastatic CRC genomes	IMPACT (integrated mutation profiling of actionable cancer targets)
Yin et al[89]	RNA-seq (2 pairs)	RNA-seq based mutation study	Point mutations and indels (Samtools)

¹Description of the software is available at https://confluence.broadinstitute.org/display/CGATools/. NGS: Next-generation sequencing; CRC: Colorectal cancer; TCGA: The cancer genome atlas; MSI: Microsatellite instability.

SOMATIC MUTATIONS IN CRC GENOMES

Like other solid tumors, CRC is thought to initiate and progress through a series of genetic and epigenetic alterations. The progression model of colorectal carcinogenesis (i.e., from adenomatous polyp to benign adenoma, eventually progressing to invasive adenocarcinoma) has been referred to as a classical cancer evolution model in which the CRC genome acquires somatic alterations in a progressive manner throughout several developmental stages. In this model, dysregulation of the APC/WNT pathway via the inactivation of APC occurs in the normal epithelium as an initiation process, while the loss of TP53 and TGF-β/SMAD4 gives rise to clonal expansion of tumor cells in the invasive adenocarcinoma (Figure 1)[4,19]. However, the genomic alterations associated with colorectal carcinogenesis may be more complicated than previously assumed. A complete and comprehensive catalogue of oncogenic drivers associated with colorectal carcinogenesis remains to be discovered.

Figure 1.

A classical progression model of colorectal carcinogenesis. A classical progression model of colorectal carcinogenesis is illustrated with genes whose alterations are responsible for each of the progressive steps. The right panel shows recent next-generation sequencing-based reports of the corresponding genes. CRC: Colorectal cancer; ATM: Ataxia telangiectasia mutated.

To extend the mutational spectrum in CRC genomes, the first exome-wide screening of approximately 13000 genes was conducted by Sanger sequencing[20]. This analysis identified approximately 800 somatic non-silent mutations in 11 CRC genomes. To distinguish oncogenic drivers from neutral passenger mutations, they identified the mutations whose frequency was significantly higher than random. The analysis revealed 69 potential oncogenic driver mutations in CRC genomes, including several well-known cancer-related genes (i.e., TP53, APC, KRAS, SMAD4, and FBXW7), and a large number of previously uncharacterized genes. Since they examined two distinct tumor types (CRC and breast cancers), they were able to identify the differences in the panel of candidate driver genes as well as identify the differences in the mutation spectrum between CRC and breast cancer genomes. The difference in the mutation spectrum (i.e., the predominance of C:G to T:A transitions over C:G to G:C transversions in CRC genomes) was confirmed by a subsequent kinase sequencing study across various cancer types[21] and by a recent whole-genome sequencing of nine CRC genomes[16].

NGS-BASED CRC STUDIES - LESSONS FROM THE CANCER GENOME ATLAS CRC STUDY

The advance in sequencing technologies has facilitated the use of genome sequencing for cancer genome studies, including CRC genomes. The largest NGS-based exome sequencing study of CRC genomes to date (approximately 200 CRC genomes) has recently been published as part of the Cancer Genome Atlas (TCGA) projects[22]. The platforms used in the multidimensional genomic characterization of CRC genomes were compared with those used for glioblastoma multiforme in 2008 (Table 3)[23]. Two important lessons from this large-scale multidimensional TCGA CRC analysis are as follows:

Table 3.

The platforms used in the Cancer Genome Atlas consortium

Alteration types	Glioblastoma multiforme (2008, TCGA)	Colorectal cancers (2012, TCGA)
Point mutations, indels	Sanger sequencing	Illumina GA and HiSeq DNA Sequencing1
		ABI SOLiD DNA Sequencing1
DNA copy numbers	Agilent Human CGH Microarray 244 A	Agilent CGH Microarray Kit 1 × 1 M and 244 A
	Affymetrix Genome-Wide SNP Array 6.0	Affymetrix Genome-Wide SNP Array 6.0
	Illumina Human Infinium 550 K BeadChip	Illumina Infinium 550 K and 1M-Duo BeadChip
DNA Methylation	Illumina Infinium DNA Methylation 27	Illumina Infinium DNA Methylation 27
		Illumina DNA Methylation Cancer Panel I
Transcriptome	Affymetrix Human Genome U133 Plus 2.0	Illumina GA and HiSeq RNA sequencing1
	Agilent 244 K Custom Array	Agilent 244 K Custom Array
	Affymetrix Human Exon 1.0 ST Array
MicroRNA	Agilent 8 × 15 K Human miRNA Microarray	Illumina GA and HiSeq miRNA sequencing1
Whole-genome sequencing	N/A	Illumina HiSeq DNA sequencing1

¹Next-generation sequencing-based platforms are noted. The platforms used in the genomic characterization of glioblastoma multiforme in 2008 (left) and colorectal cancer genomes in 2012 (right) are shown. TCGA: The Cancer Genome Atlas.

First, similar to previous findings[20], most of the significantly recurrent mutations were observed at known cancer-related genes, such as APC, TP53, KRAS, PIK3CA, FBXW7, SMAD4, and NRAS. The study also revealed frequent coding microsatellite instability (MSI) on ACVR2A, TGFBR2, MSH3, and MSH6 by manual examination of sequencing reads for 30 known MSI loci. Although the majority of recurrent mutations were previously known, a number of novel mutations were also identified, which may have functional implications on colorectal tumorigenesis. For example, the mutations in SOX9[24], FAM123B[25], and 14 other genes are known to be implicated in the altered WNT/APC pathway. Although the biallelic inactivation of APC and the activating mutation of CTNNB1 encoding β-catenin are two major events that occurred in about 74% of the total CRC genomes studied, the mutations and deletions of an additional 16 (about 18%) genes in the WNT/APC pathway were not negligible, leading to the conclusion that nearly all CRC genomes (about 92%) have an alteration in the WNT/APC pathway[22].

A study which assigned potential molecular functions to rare mutations in CRC genomes using the pathway-level convergence was previously reported[26]. Thus, pathway- or network-level information from available resources (i.e., Gene Ontology[27]) and other methodologies to predict the functional impacts of non-synonymous point mutations[28,29] may help determine the potential functions of rare mutations and distinguish oncogenic drivers in studies with small-sized cohorts. This issue is also related to the sample-size problem in study design. Due to the limits on sample availability and research budget, many of the cancer mutation studies use a small discovery cohort for the generation of candidate mutations that are subsequently validated in an extended set. Increasing the number of samples in the initial discovery set would be beneficial in identifying events that are not highly recurrent but are still clinically meaningful (i.e., gene fusions involving receptor tyrosine kinases with available inhibitors). For example, the frequency of gene fusions such as ALK[30] and RET[31] in lung adenocarcinomas and FGFR[32] in glioblastoma multiforme are less than 5%. Although the level of recurrence is still the generally accepted functional indicator of genomic alterations[33,34], the incorporation of knowledge from other resources may facilitate the identification of biological or clinically relevant mutations more efficiently in a moderate-sized cohort.

Second, the concordant and discordant relationships between alterations examined across the samples may reveal valuable functional insights. The concordant relationship adopts the concepts of co-expressed networks in which the genes with significantly correlated expression levels (measured by Pearson’s correlation coefficients or mutual information) across diverse cellular conditions may have a functional relationship[35-37]. Importantly, TCGA-related studies revealed that the exclusivity between the potential oncogenic drivers may be common. For example, TCGA ovarian cancer study showed that the alterations of BRCA1 and BRCA2 (including germline or somatic mutations and epigenetic silencing via promoter hypermethylation) are mutually exclusive to each other[38]. The method to identify the pairs of exclusive genomic alterations is formulated as a standard analysis pipeline in TCGA projects as Mutual Exclusivity Modules (MEMo) in cancer[39]. In CRC genomes, MEMo analysis revealed that nearly half of the TCGA CRC genomes showed an exclusive relationship between the up-regulation of IGF2 and IRS2, and between the mutation of PIK3 pathway genes (PIK3CA and PIK3R1) and the homozygous deletion of PTEN[22]. This suggests that the IGF2-IRS2 axis is a major signaling pathway upstream of the PI3K pathway in CRC genomes[22]. The mutual exclusivity between the mutations of TP53 and ATM was also identified in the TCGA CRC genomes[22].

GENOMIC REARRANGEMENTS AND GENE FUSIONS IN CRC GENOMES

Bass et al[16] reported whole-genome sequencing (sequencing coverage about 30-fold) of nine CRC genomes. By comparing them with matched normal genomes, they identified approximately 140000 putative somatic mutations per CRC genome, which included approximately 700 non-silent point mutations and indels in coding sequences. One advantage of whole-genome sequencing is that the genome-wide landscape of the mutation spectrum in CRC genomes can be obtained, such as the relative paucity of mutations in exons and higher mutation frequency in intergenic regions than introns. This phenomenon is probably due to the selection pressure and transcription-coupled repair, and is consistent with other types of cancer genomes such as prostate cancers[40] and multiple myelomas[41]. Since most of the non-synonymous nucleotide substitutions were observed at known cancer genes such as KRAS, APC, and TP53, they focused on novel aspects that can only be identified from paired-end whole-genome sequencing data such as chromosomal rearrangements. Among the approximately 700 candidate rearrangements, 11 events give rise to in-frame fusion genes. The extended screening further revealed that VTI11A-TCF7L2 fusion is recurrent (3 out of 97 primary CRC genomes) and the siRNA-mediated down-regulation of this fusion transcript reduced the anchorage-independent cell growth in vitro, indicative of their potential oncogenic activity.

The low-pass (sequencing coverage approximately 3-4-fold) whole-genome sequencing of 97 TCGA CRC genomes also identified three genomes harboring NAV2-TCF7L1 fusion[22]. The predicted protein structures of fusion proteins lacked the β-catenin binding domain of TCF3 (encoded by TCF7L1), which is similar to the fusion of VTI11A-TCF7L2 that lacks the β-catenin binding domain of TCF4 (encoded by TCF7L2). In addition, the inactivation of TTC28 by genomic rearrangements is of note since this event is recurrent (21 out of 97 cases) and involves multiple partners for rearrangements in TCGA CRC genomes[22]. Gene fusions can also be identified from transcriptome sequencing, so called RNA-seq[42,43]. A recent RNA-seq-based study revealed recurrent gene fusions involving R-spondin family members of RSPO2 and RSPO3[44]. These fusions were exclusive to APC mutations in the observed CRC genomes, suggesting their potential roles in activating the APC/WNT pathway in colorectal carcinogenesis. Cancer-related gene fusion events are gaining attention since there has been no effective gene fusion screening method other than NGS-based paired-end sequencing. More importantly, many of the fusion candidates discovered so far represent oncogenic drivers and clinically actionable events (i.e., the fusion activates potential oncogenes such as tyrosine kinases that can be inhibited by small molecule inhibitors) as shown in recent studies[30-32] including the C2orf44-ALK fusion in CRC[45].

In addition, it was proposed that the genomic rearrangements in individual cancer genomes can be used as personal cancer markers to trace the disease activity (i.e., to detect recurrences or to evaluate the tumor burden of residual diseases)[46]. The proposed method, personalized analysis of rearranged ends, was applied to four cancer genomes, including two CRCs in a pilot test. It was demonstrated that the PCR-based quantification of the rearranged DNA in the plasma correlated well with the treatment course of CRC[46].

MICROSATELLITE INSTABILITY IN CRC GENOMES

Microsatellites are short tandem repeat sequences present at millions of sites in the human genome[47]. MSI defined as the length polymorphism of microsatellite repeat sequences, can arise due to a defect in the DNA mismatch repair system[48]. MSI is common in hereditary nonpolyposis colon cancers, also known as Lynch syndrome, where germline mutations of MLH1 and MSH2 are commonly observed[49,50]. About 15% of sporadic CRC are microsatellite-unstable, where transcriptional silencing of MLH1 by promoter hypermethylation is common[51,52]. The microsatellite-unstable sporadic CRC has distinct clinical and genomic features (i.e., common in right-sided colons and elderly females, and nearly diploid, etc.) compared to microsatellite-stable, but aneuploid CRC genomes. The key genes targeted by somatic point mutations and MSI-induced frameshifting mutations are different between the microsatellite-stable and -unstable CRC genomes, as shown in the TCGA CRC study[22]. Since the MSI analysis in the TCGA CRC study was limited to a manual search of exome sequencing reads for about 30 known loci with frequent MSI (i.e., TGFBR2, ACVR2A, BAX, etc.), a question remains as to whether we can fully exploit the NGS technology to screen the locus-level MSI in an exome- or genome-wide scale. One interesting report by Wang et al[53] showed that pancreatic cell lines with a homozygous deletion of MLH1 (which is a frequent target of promoter hypermethylation in MSI CRC genomes) frequently harbors truncating indels in TP53 and TGFBR2. This suggests that whole genome- or exome-sequencing data may be used for large-scale MSI screening to identify novel MSI events targeting tumor suppressor genes in cancer genomes.

NGS AND CRC EPIGENETICS

For decades, DNA methylation has been studied as one of the major cancer-related epigenetic modifications. Until recently, it was recognized that cancer genomes are undermethylated overall, but some genomic loci have focal DNA hypermethylation[54,55]. Transcriptional silencing by focal hypermethylation, especially at the CpG islands of gene promoters, is among the putative inactivating mechanisms of tumor suppressor genes in cancer genomes and often preferred over the inactivation by irreversible nucleotide substitutions[56]. Yet, the landscape of cancer-associated DNA methylation seems more dynamic than previously anticipated, as revealed by genome-wide CRC methylome studies[57,58]. Two recent CRC methylome studies used NGS-based sequencing of bisulfite-treated genomic DNA for bp-resolution methylome profiling[59,60]. Both studies proposed the presence of large blocks of DNA hypomethylation that occupied almost half of the genomes. Additionally, they reported that such findings as the genome-wide methylation variability of the adenoma genome is an intermediate between those of normal epithelium and CRC[60] and the domains of DNA hypomethylation regionally coincided with those of nuclear lamina attachment[59]. In addition, DNA methylation profiling has been also proposed as a means of early CRC diagnosis using non-invasive resources (i.e., blood- or stool-based)[61,62], which can benefit from NGS technologies.

NOVEL ASPECTS OF CRC GENOMES BY NGS STUDIES

NGS-based genome analysis may facilitate the identification of previously unrecognized, novel features of CRC cancer genomes. For example, owing to its high-throughput nature, NGS analysis may be able to detect the presence of foreign DNA sequences originating from bacterial or viral pathogens. Although the clear association between pathogens and certain human tumor types has been demonstrated in limited cases such as hepatitis B or C viruses with hepatocellular carcinoma, there have been ongoing efforts to use the sequencing data for pathogen discovery[63,64]. For instance, Kostic et al[65,66] analyzed nine CRC whole genome sequencing data sets[16] using their algorithm of PathSeq to identify microbial sequences enriched in CRC genomes compared to those in matched normal genomes. They observed that the sequences of Fusobacterium are enriched in CRC genomes, which was also shown in transcriptome sequencing results by independent researchers[67]. Although the oncogenic role of Fusobacterium in CRC genomes is only beginning to be elucidated[68], these results highlight the possibility that NGS-driven sequencing data will be a valuable resource to identify novel pathogens associated with human cancers.

Chromothripsis is a unique cancer genome-associated phenomenon in which tens to hundreds of chromosomal rearrangements occur in a “one-off” cellular event[69]. This phenomenon involves one or a few chromosomes in which massive chromosomal fragmentation is followed by rejoining of the fragments[70]. This results in unique genomic signatures that can be identified by paired-end sequencing (i.e., massive intrachromosomal rearrangements in the affected chromosome that can be visualized in a Circos diagram[71]) or from copy number profiles (i.e., frequent oscillations between two copy number states indicative of retained and lost chromosomal fragments). After the first discovery of chromothripsis in one chronic lymphocytic leukemia patient by paired-end sequencing[69], Stephens et al[69] also examined the copy number profiles of 746 cancer cell lines, observing that 2.4% of them (18 cell lines) showed the genomic signatures of chromothripsis. Recently, Kim et al[72] reported the tumor type-specific frequencies of chromothripsis as measured from a large-scale copy number profile of about 8000 cancer genomes including CRCs. Six out of 366 CRC genomes (1.8%) in the database showed the signature of chromothripsis (i.e., significant frequent alternation between different copy number states[72]) and the frequency was not substantially different from the average across the database (1.5%). Of note, Kloosterman et al[73] reported the paired-end whole-genome sequencing results of four CRC genomes with liver metastases, observing that all cases harbored evidence of chromothripsis. In addition, the comparison between primary and metastatic CRC genomes revealed that most genomic arrangements are shared both by primary and metastatic genomes, indicating that metastasis occurs quite rapidly with few additional mutational events, which was also proposed in mutation-based CRC genome studies[74]. Along with chromothripsis, several unique features of cancer genomes have been reported in breast cancer genomes (kataegis; regional hypermutations near rearrangement breakpoints)[75] and in prostate cancer genomes (chromoplexy; chains of copy-neutral rearrangements across multiple chromosomes)[40], which may expand the mutational categories in CRC genomes.

CONCLUSION

We have discussed the recent NGS-based CRC studies in various genomic aspects. The progress of CRC genomic analysis (but not exclusive to CRC) can be summarized into three issues: (1) the screening of clinically actionable targets for personalized targeted medicine; (2) the advancement of pathway-level understanding in colorectal carcinogenesis using a large-scale cohort; and (3) the identification of novel features or mutation types in CRC genomes. In terms of the first issue, Roychowdhury et al[76] reported an advanced NGS-based cancer patient management protocol that includes low-pass whole-genome, exome, and transcriptome sequencing of cancer genomes. The notable aspects of the protocol are the timeline (< 4 wk after enrollment) and cost (approximately 3600 USD), as well as the presence of a multidisciplinary sequencing tumor board (STB) to evaluate the mutation profiles of the patients and make a clinical decision. In their pilot study, the STB evaluated the sequencing results from a patient with metastatic CRC harboring NRAS mutation and CDK8 amplification and concluded that BRAF/MEK inhibitors and PI3K and/or CDK inhibitors could be beneficial for the patient. Second, the more complete and comprehensive collection of CRC-related somatic genomic alterations will advance the pathway-level understanding of colorectal carcinogenesis and help distinguish the oncogenic drivers from neutral passengers, as seen in large-scale meta-analyses of cancer genome profiles[72,77]. Finally, NGS-driven genomic studies are already reporting novel features of cancer genomes beyond the traditional mutational categories. Besides the MSI and chromothripsis we discussed, some researchers used publicly available genome sequencing data (including those of CRC genomes[16]) and reported novel mitochondrial mutations[78] and the activity of human retrotranspositions in the cancer genomes[79]. Taken together, NGS technology will advance our understanding of CRC genomes and the obtained knowledge will lead to a better diagnosis and personalized targeted therapeutics for CRC management.