Laterally spreading tumors (LST) are colorectal adenomas that develop into extremely large lesions with predominantly slow progression to cancer, depending on lesion subtype. Comparing and contrasting the molecular profiles of LSTs and colorectal cancers offers an opportunity to delineate key molecular alterations that drive malignant transformation in the colorectum. In a discovery cohort of 11 LSTs and paired normal mucosa, we performed a comprehensive and unbiased screen of the genome, epigenome, and transcriptome followed by bioinformatics integration of these data and validation in an additional 84 large, benign colorectal lesions. Mutation rates in LSTs were comparable with microsatellite-stable colorectal cancers (2.4 vs. 2.6 mutations per megabase); however, copy number alterations were infrequent (averaging only 1.5 per LST). Frequent genetic, epigenetic, and transcriptional alterations were identified in genes not previously implicated in colorectal neoplasia (ANO5, MED12L, EPB41L4A, RGMB, SLITRK1, SLITRK5, NRXN1, ANK2). Alterations to pathways commonly mutated in colorectal cancers, namely, the p53, PI3K, and TGFβ pathways, were rare. Instead, LST-altered genes converged on axonal guidance, Wnt, and actin cytoskeleton signaling. These integrated omics data identify molecular features associated with noncancerous LSTs and highlight that mutation load, which is relatively high in LSTs, is a poor predictor of invasive potential.
Implications: The novel genetic, epigenetic, and transcriptional changes associated with LST development reveal important insights into why some adenomas do not progress to cancer. The finding that LSTs exhibit a mutational load similar to colorectal carcinomas has implications for the validity of molecular biomarkers for assessing cancer risk. Mol Cancer Res; 14(12); 1217–28. ©2016 AACR.
Colorectal carcinomas develop from benign intraepithelial lesions known as adenomas. Laterally spreading tumors (LST) are a morphologically heterogeneous group of adenomas that grow laterally within the colonic mucosa (1). LSTs are an interesting subtype of adenoma because they grow to extremely large sizes, often many centimeters in diameter, but rarely develop submucosal invasion (1).
Comprehensive profiling of colorectal carcinomas has shown that they often contain deranged genomes (2, 3) making it difficult to identify specific culprits of malignant transformation. The value of studying LSTs is that they rarely progress to cancer. Therefore, tracking their molecular etiology and comparing this with cancers offers the potential to understand why some adenomas do not become invasive.
LSTs can be endoscopically characterized as either granular or nongranular on the basis of their surface morphology or by the Paris classification system, which describes the macroscopic morphology of different types of adenoma (1). Granular LSTs (G-LST) consist of aggregates of mucosal nodules with an uneven surface, whereas nongranular LSTs (NG-LST) have a smooth even surface. Although cancer rarely develops from LSTs, the risk of submucosal invasion differs with morphology (4, 5). For example, in a prospective, multicenter, observational study of all patients referred for endoscopic mucosal resection (EMR) of large LSTs (≥20 mm), invasion was found in 15.3% (15 of 98) of NG-LSTs but in only 3.2% (10 of 311) of G-LSTs (6). The molecular basis of the extensive growth of LSTs in the absence of cancer and the increased invasive potential of NG-LSTs when compared with G-LSTs is unclear primarily because no studies have undertaken a systematic exploration of the molecular characteristics of this subtype of adenoma.
This study integrates multiple layers of “omics” data (genomic, epigenomic, and transcriptomic) from LSTs to achieve 3 aims: (i) to identify alterations in genes and pathways that coincide with the development of LSTs; (ii) to compare genetic alterations between LSTs and colorectal carcinomas; and (iii) to identify molecular differences between granular and nongranular LSTs, which differ in their malignant potential.
Materials and Methods
Patients and tissue samples
A discovery cohort (7 G-LSTs, 4 NG-LSTs, n = 11, Supplementary Table S1 and Supplementary Fig. S1) of LSTs and paired normal mucosal tissue was obtained from 9 patients (6 males; mean age, 70 years; range, 51–87 years) using EMR. This included 3 separate lesions (G1, G2, and NG) that were obtained from one individual (male, 70 years) during the same endoscopic procedure. Only non-polypoid (Paris IIa) components were used for genomic analyses. Validation cohort 1 (n = 44, Supplementary Table S2) consisted of large (≥20 mm) LSTs with mixed Paris classification and validation cohort 2 (n = 40, Supplementary Table S2) consisted of 20 granular and 20 nongranular large (≥20 mm) LSTs (Paris IIa or IIb). Histologic assessment indicated all lesions showed no evidence of submucosal invasion. All cohorts were obtained from fresh tissues selected (by M.J. Bourke) from a previously described consecutively collected cohort obtained by EMR between 2009 and 2011 at Westmead Hospital (ethics approval 2009/6/4.6 and 11194; ref. 7).
DNA methylation analysis
Methyl-CpG–binding domain (MBD) protein DNA enrichment and high-throughput sequencing (MBD-Seq) were performed as described in Supplementary Methods. Methylation status was validated using combined bisulphite restriction analysis (COBRA) and single-molecule bisulfite sequencing (12, 13). Methylation status of 413 colorectal carcinomas from The Cancer Genome Atlas (TCGA; ref. 2) was determined by calculating the mean β value for probes within the promoter CpG island (probes outside the CpG island were excluded) and classifying tumors as either methylated (β ≥ 0.25) or unmethylated (β < 0.25).
Whole exomes were captured from 10 LSTs (lesion 5 was omitted because of insufficient DNA) and normal mucosa samples using the Roche SeqCap EZ Choice HGSC (Catalogue number: 06465587001) and sequenced using the Illumina HiSeq 2500 platform (2 × 125 bp). Data were processed and aligned (hg19) using Picard MarkDuplicates (Broad Institute) and Bowtie2 (14). Somatic single-nucleotide variants (SNV) were identified using the intersection of Strelka (15) and MuTect (16) following comparison with normal mucosa. Somatic small insertions and deletions (indels) were identified using Strelka and GATK-HaplotypeCaller subtraction (17, 18). Only mutations with ≥10× coverage, present in ≥10% of reads and detected on ≥4 reads were considered as genuine. Only mutations predicted to be damaging to protein function (ANNOVAR; ref. 19) were considered when identifying candidate driver gene alterations.
Whole-genome sequencing analysis
Whole-genome sequencing (WGS) data from 33 colorectal carcinomas and normal mucosa from TCGA (2) were analyzed using the same approach described above for whole-exome sequencing (WES). Tumors were categorized as follows: 19 microsatellite-stable (MSS), 8 microsatellite-instable (MSI), and 6 with nonsynonymous mutations in the exonuclease domain of the POLE gene.
Copy number alterations
A total of 22 LSTs were analyzed for copy number alterations (10 discovery cohort and 12 validation cohort) using Illumina Human610-Quad SNP arrays or Comparative Genomic Hybridization (CGH) 720K whole-genome tiling array v3.0 (NimbleGen, 05520860001). SNP array data were analyzed using OncoSNP, as described previously (20). CGH data were analyzed using SignalMap (NimbleGen) with LogR ratio thresholds of >0.2 representing copy number gains (amplifications/duplications) and ←0.2 representing copy number losses (deletions). Deletions at 5q were assessed at the microsatellite markers D5S346, D5S2495, and D5S617 using PCR and capillary electrophoresis on an ABI 3500 instrument (Applied Biosystems).
Sanger sequencing mutation analysis
The mutation status of KRAS (codons G12 and G13), NRAS (codons G12 and G13), and BRAF (codon V600) were assessed using pyrosequencing as described previously (21). BRAF codon G469, KRAS codons D117 and A146 and, NRAS codon Q61 mutations were assessed by sequencing. The mutation status of APC and CTNNB1 was assessed using sequencing as described previously (22).
Real-time quantitative reverse transcriptase PCR
Quantitative reverse transcriptase PCR (qRT-PCR) was performed as described previously (23). Samples were analyzed in quadruplicate, and gene expression was normalized to GAPDH [NM_002046.5] gene. All primer sequences are available on request.
χ2 tests were used for categorical variables and one-way ANOVA for continuous variables. Statistical significance was defined at a level of less than 0.05. Analyses were carried out using IBM SPSS software (version 22; SPSS Inc., Chicago). Gene set enrichment analysis (GSEA) was performed as described previously using MSigDB (24). Ingenuity Pathway Analysis was performed as described in Supplementary Methods.
Mutation rates in LSTs are similar to those observed in MSS colorectal carcinomas
Using WES, we identified a median of 110 (mean, 100; range, 3–155) exonic SNVs (synonymous and nonsynonymous) per lesion in our discovery cohort of LSTs (mean sequence coverage of 55× in normal tissues and 134× in LST tissues). This represented a median of 2.4 mutations per megabase. We compared this mutation rate to that observed in 33 colorectal carcinomas from TCGA (2) using the same bioinformatics approach. Surprisingly, we found a similar rate of mutations in MSS colorectal carcinomas (median, 119 mutations; mean, 135; median, 2.6 mutations per megabase, Fig. 1A). Sanger sequencing of 10 SNVs and 2 indels confirmed that mutation calls in LSTs were accurate (Supplementary Fig. S2). As expected, significantly higher mutation rates were found in colorectal carcinomas with MSI (median, 1,648; mean, 1,774; median, 35.9 mutations per megabase, P ≤ 0.05, t test) and in colorectal carcinomas containing mutations to the exonuclease domain of the POLE gene (median, 6,078; mean, 6,852; median, 132 mutations per megabase, Fig. 1A). We compared the variant allele frequency (VAF), which measures the proportion of copies of a gene that contain a specific mutation, and found that the median VAF of mutations in LSTs was significantly lower than the median VAF in colorectal carcinomas without MSI (P ≤ 0.001, t test, Fig. 1B). CGH and SNP array data showed that copy number alterations were rare, with none detected in 3 LSTs and an average of only 1.5 whole chromosome duplications or deletions per lesion (Fig. 2). To confirm that copy number alterations were rare in LSTs, we examined a further 12 lesions (7 G-LSTs, 5 NG-LSTs) from validation cohort 1 using CGH and SNP arrays. Only 4 deletions and 12 whole chromosome duplications were detected across all 12 lesions (mean, 1.6 copy number alterations per lesion, Fig. 2). Deletions of chromosome 1p, 5q, 14q, and 18 and duplications of chromosome 7, 8, 13, 19, and 20 represent some of the most common changes seen in colonic and rectal adenocarcinomas (Fig. 2). Copy number changes that were specific to colorectal carcinomas, yet lacking in LSTs, included deletions to 8p, 15q and the TP53 gene at 17p. Collectively, these data show that the number of mutations in LSTs is similar to colorectal carcinomas without MSI; however, mutations in LSTs are not compounded by widespread copy number alterations.
Identification of genes altered in LSTs relative to paired normal mucosa
We followed the analytic strategy outlined in Supplementary Fig. S3 to enrich for genes showing multiple damaging alterations. This involved only considering nonsynonymous SNVs and indels that were predicted to be damaging to protein function and that were present in expressed genes, as determined by RNA-Seq analysis of corresponding normal mucosa. This identified a median of 44 damaging mutations per LST (mean, 39; range, 0–65, Fig. 3A). No nonsynonymous mutations were detected in one LST (specimen 82), despite a read coverage across exons that was comparable to the other LSTs (55× in normal, 142× in LST). Next, we identified which genes were affected by the copy number alterations described above (validation cohort LSTs only). Six deletions that encompassed a total of 807 separate genes on chromosomes 18p, 18q, 1p, 6p, or 5q were detected in 5 LSTs (Fig. 2). The only recurrent deletion was at chromosome 5q (lesions 81 and 85) encompassing the APC gene. Duplication of the whole of chromosomes 7, 8, 12, 13, 19, or 20 was detected in 4 LSTs; however, genes on these duplicated chromosomes were not considered further because of the large numbers of genes involved and the nonspecific nature of this type of copy number alteration. Finally, using MBD-Seq, we detected promoter CGI hypermethylation at a median of 711 genes (mean, 804; range, 189–1,837) per LST. However, on average, only 28.5% of these aberrations correlated with loss of gene expression, defined as >2-fold downregulation relative to paired normal mucosa (ref. 8; hereafter referred to as epigenetic inactivation). This corresponded to a median of 225 epigenetically inactivated genes (mean, 225; range, 53–564, Fig. 3A) per LST and a total of 1,389 separate genes across all 11 lesions profiled. Comparison with methylation data from TCGA colonic and rectal adenocarcinomas showed that most genes identified as frequently epigenetically inactivated in LSTs were also frequently methylated in colorectal carcinomas (Supplementary Fig. S4).
Integration of genomic, epigenomic, and transcriptomic data identifies novel genes with multiple alterations
Having identified genetic and epigenetic alterations, we next integrated these data to identify genes with multiple alterations across our discovery cohort. This involved identifying genes with damaging mutations in more than one patient (recurrent) or those with multiple alterations (damaging mutation, epigenetic inactivation, deletion) in a single lesion and at least one alteration in LSTs from other patients (see Supplementary Fig. S3 for a summary of analytical approach). This identified a total of 22 genes that included known colorectal carcinoma–related genes such as APC, KRAS, SOX9, and BRAF (Fig. 3B). However, although these genes were mutated, deleted, or epigenetically inactivated in multiple LSTs, all (with the exception of APC, KRAS, BRAF, and SLC2A10) were mutated, deleted, or showed focal high-level amplification in less than 10% of colorectal carcinomas (Fig. 3B). These included ANO5, MED12L, EPB41L4A, RGMB, NRXN1, SLITRK1, and ANK2. Many of the mutations in known driver genes such as APC, BRAF, KRAS, and SOX9 showed VAF > 0.25, which indicates that these are early clonal events (25). In addition, several genes identified as frequently altered in LSTs, such as ANO5, NRXN1, and SLITRK1, also showed VAF > 0.25 (Supplementary Table S3) suggesting these were also early clonal events in the development of these lesions.
We next determined the frequency of alterations to 30 genes previously described as frequently altered by mutation or copy number alterations during the development of colorectal carcinomas (n = 212; ref. 2; Fig. 3C). These genes separated into 5 signaling pathways: Wnt (10 genes), TGFβ (8 genes), PI3K (5 genes), MAPK (5 genes), and p53 (2 genes). In LSTs, frequent alterations were found in genes within the Wnt (APC, SOX9) and MAPK (KRAS, BRAF) pathways; however, significantly fewer alterations were detected in the TGFβ and p53 pathways (P = 0.039 and P = 0.046, respectively) and the TP53 gene (P = 0.019), with a trend toward significantly fewer alterations in the PI3K pathway (P = 0.069, Fisher exact test; Fig. 3C). To determine whether mutations in these pathways were present in only a small proportion of cells (subclonal), we identified driver gene mutations with a VAF < 0.1. Only 3 additional damaging mutations in driver genes were identified: a p.R2871T SNV in ATM (VAF, 0.064), a c.1494+1G>A splice site variant in APC (VAF, 0.082), and a p.R95H SNV in ERBB4 (VAF, 0.092). These findings reinforce that colorectal carcinoma driver gene mutations are rare in LSTs, aside from mutations to Wnt and MAPK genes.
Genomic, epigenomic, and transcriptomic changes in LSTs converge on the axonal guidance, Wnt, and actin cytoskeleton signaling pathways
We next investigated whether altered genes were enriched within specific biologic pathways. We took all genes identified as epigenetically inactivated or mutated in our discovery cohort (n = 1,728), as well as those inactivated in G-LSTs (n = 1,427) or NG-LSTs (n = 573), and performed Ingenuity Pathway Analysis (see Supplementary Methods). When all LSTs were considered, significant enrichments were observed in the axonal guidance, thyroid cancer, and Wnt/β-catenin signaling pathways (Fig. 4A). These pathways were targeted by recurrent alterations to genes encoding ephrins (EPHA4, EPHA6, EPHA7), neurotrophic tyrosine kinase receptors (NTRK1-3), and Wnts (WNT7A). We also identified frequent alterations to a family of predicted axonal guidance genes known as the SLIT and NTRK-like family, namely, SLITRK1 (mutated in 3), SLITRK2 (epigenetically inactivated in 3), and SLITRK5 (epigenetically inactivated in 7 LSTs, Supplementary Fig. S5). However, the marked difference in the number of gene alterations in G-LSTs (n = 1,728, mean 247 per lesion) versus NG-LSTs (n = 573, mean 143 per lesion) raised the possibility that the predominant signaling pathways altered in each subtype may be different. When G-LSTs were considered separately, we identified enrichment of colorectal carcinoma metastasis signaling (Fig. 4B and D), whereas NG-LSTs showed alterations to actin cytoskeletal and Rho GTPase signaling genes (Fig. 4C and D). Alterations to axonal guidance genes were heavily biased toward epigenetic inactivation and were markedly less frequent in NG-LSTs when compared with G-LSTs (Fig. 4D).
We reasoned that biologically relevant signaling pathways would also show transcriptional changes. To this end, we used DESeq (11) to identify 303 and 458 genes showing significant upregulation or downregulation relative to paired normal tissues (Fig. 4E). Upregulated genes showed significant enrichment for colorectal carcinoma metastasis, Wnt, and axonal guidance signaling and again we observed changes to ephrins (EPHB1) and Wnts (WNT5A) as well as matrix metalloproteinases (MMP1, 3, 7, 9, 10, 11, and 12), and cadherins (CDH3, Fig. 4E). Downregulated genes showed significant enrichment for tight junction and Rho GTPase signaling genes including claudins (CLDN3, 4, 8, and 23), MYLK, WASF3, and Rho's (RHOD, RHOF, Fig. 4E). Interestingly, downregulation of genes in the Rho GTPase signaling pathway, which is consistent with the deactivation of this pathway, was observed in all 11 LSTs (Fig. 4E). This demonstrates that while Rho GTPase signaling is one of the most significantly enriched pathways for gene alterations in NG-LSTs, it is in fact similarly altered across all 11 LSTs examined.
To further investigate the importance of axonal guidance signaling in the development of LSTs, we investigated gene expression microarray data from a previous study of 25 LSTs and 17 polypoid adenomas (26). Gene Set Enrichment Analysis (GSEA) confirmed that axonal guidance genes were significantly differentially expressed (>2-fold up- or downregulated) in LSTs relative to normal mucosa (FDR q = 0.006) but not in polypoid adenomas relative to normal mucosa (FDR q = 0.115).
Validation of genetic and epigenetic alterations
To explore the broader significance of the alterations we had identified, we investigated their prevalence in a cohort of 44 large (≥20 mm in diameter) noninvasive LSTs (validation cohort 1). We also investigated the recurrence of 5q deletions, mutation of the APC, KRAS, BRAF, and CTNNB1 genes, and the hypermethylation of 17 genes on the basis of their frequent occurrence in the discovery cohort of LSTs or because they were connected with axonal guidance, Wnt, or actin cytoskeletal signaling. Copy number analysis using microsatellite markers at 5q and sequencing across the APC gene identified deletions or mutations in 83% of 41 informative lesions (Fig. 5). CTNNB1 mutations were found in 5% of 44 lesions, whereas KRAS and BRAF mutations occurred in 51% of 43 and 16% of 44 lesions, respectively (Fig. 5). COBRA and bisulfite sequencing confirmed frequent methylation (≥50%) of 13 of the 17 genes investigated including SLITRK5 (methylated in 98%) and ANO5 (methylated in 63%, Fig. 5, Supplementary Fig. S5), 2 of the most frequently epigenetically inactivated genes identified in the discovery cohort. There was no correlation between tumor location, Paris classification, morphology, histologic type or dysplasia, and mutation or methylation at the genes examined.
Genetic, epigenetic, and transcriptional alterations correlate with LST morphology
A convergence of evidence suggested that the molecular landscape of granular and nongranular LSTs was likely to be different. For example, in our discovery cohort, epigenetic inactivation was significantly more frequent in G-LSTs (average, 288 genes; range, 70–564) compared with NG-LSTs (average, 115 genes; range 53–225; t test: P = 0.03). In support of this, genes with recurrent epigenetic inactivation specifically in one morphological subtype were also more common in G-LSTs (315 genes) than in NG-LSTs (9 genes, Supplementary Fig. S4 and Supplementary Table S4). Furthermore, pathway analysis of altered genes had revealed the enrichment of different signaling pathways in G-LSTs and NG-LSTs (Fig. 4B and C). Therefore, we further investigated the molecular differences between G-LSTs and NG-LSTs. In a previous study, we had shown that in a homogeneous cohort of flat LSTs (Paris classification IIa or IIb), KRAS mutations at codons G12 and G13 were significantly more frequent in G-LSTs than in NG-LSTs (22). However, the detection of KRAS D117N (n = 1) and A146V (n = 1), NRAS Q61K (n = 1) and BRAF G469V (n = 1) mutations in 4 of 10 LSTs in the discovery cohort prompted us to reinvestigate the relationship between morphology and mutations in these genes in a second validation cohort (validation cohort 2) consisting of 20 G-LSTs and 20 NG-LSTs (Paris classification IIa or IIb). KRAS mutations (at codons G12, G13, D117, or A146) occurred in 55% (22 of 40), NRAS mutations (at codons G12, G13, or Q61) occurred in 0% (0 of 40), and BRAF mutations (at codons G469 or V600) occurred in 5% (2 of 40) of lesions (Supplementary Fig. S6). Differentiation of lesions according to surface morphology revealed a trend toward KRAS mutations in G-LSTs (70%, 14 of 20) versus NG-LSTs (40%, 8 of 20, χ2 test, P = 0.055). Differentiation of lesions according to histologic type showed that mutation of KRAS was significantly more frequent in tubulovillous LSTs (77%, 17 of 22) than in those with tubular architecture (28%, 5 of 18, χ2 test, P = 0.002).
Next, we investigated transcriptional differences between G-LSTs and NG-LSTs. Principal component analysis (PCA) of gene expression showed that the discovery cohort of LSTs could be stratified on the basis of surface morphology (Fig. 6A) but not histologic type or tumor location. To further explore these transcriptional differences, we identified 3,345 genes that were significantly upregulated or downregulated in either G-LSTs or NG-LSTs (Fig. 6B). A small proportion of these genes (n = 258) was significantly upregulated in G-LSTs yet downregulated in NG-LSTs or vice versa (genes within gray-shaded boxes in Fig. 6B). We reasoned that these transcriptional changes represented the most extreme differences in gene expression between the 2 morphologic subtypes. The remaining 3,087 genes were upregulated or downregulated in both G-LSTs and NG-LSTs but to different degrees. Hierarchical clustering showed that the 258 differentially expressed genes robustly grouped granular lesions separately from nongranular lesions and that the 3 LSTs from the same patient clustered together (Fig. 6C). Pathway analysis of these 258 differentially expressed genes showed enrichment for cAMP-mediated and CXCR4 signaling (Fig. 4D). Genes within the CXCR4 signaling pathway were mostly upregulated in NG-LSTs, whereas the same genes were mostly downregulated in G-LSTs (Fig. 6D), which is consistent with the hyperactivation of CXCR4 signaling specifically in NG-LSTs. Collectively, these data indicate that the different morphologic subtypes of LSTs show distinct genetic, epigenetic, and transcriptional profiles at a subset of genes.
In this study, multiple layers of genome-wide data were integrated to generate molecular maps of LSTs. This showed that despite the low malignant risk of this subtype of adenoma they have an unexpectedly high mutation rate that is comparable with MSS cancers. These mutations are not compounded by copy number alterations and target different genes when compared with mutations observed in colorectal carcinomas. LSTs frequently show alterations to genes in the axonal guidance, Wnt, and actin cytoskeleton pathways but rarely show alterations to genes in the TGFβ (SMAD2, SMAD3, ACVR1B, ACVR2A, TGFBR1, TGFBR2), PI3K (PIK3CA, PIK3R1, PTEN, IGF2, IRS2), and p53 (TP53 and ATM) pathways, which are known to be altered in colorectal carcinomas. These data are consistent with a model in which the extensive growth of LSTs is driven by molecular alterations that impart continued proliferation, such as APC, SOX9, KRAS, and BRAF mutations, but are insufficient for malignant transformation. Comparison of molecular alterations in the 2 morphologic subtypes of LSTs, which differ in their invasive potential, identified subtle molecular differences, including a higher frequency of KRAS mutations and epigenetic inactivation in very low-risk G-LSTs and the hyperactivation of CXCR4 signaling in NG-LSTs, which by comparison are associated with an increased risk of cancer.
Determining the molecular events that drives an adenoma to become cancerous was recently highlighted as 1 of 5 crucial questions that need to be addressed to better understand how colorectal carcinoma develops (27). To address this, this study took a novel approach by specifically focusing on large (>20 mm in diameter) LSTs, which despite their size have a very low risk of progressing to cancer. At the outset, it was anticipated that the mutation rate in LSTs would be low, as described previously for other types of benign lesions. For example, parathyroid adenomas contain an average of only 3.6 nonsynonymous mutations per adenoma (28). The number of mutations in polypoid colorectal adenomas has also been reported to be much lower than in colorectal carcinomas with an average mutation rate of 0.63 per megabase or a median of 35.5 mutations (synonymous and nonsynonymous) per adenoma (mean, 40.8; range, 12–101; ref. 29). The relatively high mutation load found in LSTs in this study did not appear to be driven by alterations in any known DNA repair pathway, and we have shown previously that LSTs do not exhibit MSI (22). However, this study also shows that a high mutation load is not necessarily a feature of all LSTs, as we were unable to find a single nonsynonymous mutation in one lesion in our cohort despite high-coverage exome sequencing data.
The integration of multiple layers of genomic and epigenomic data from paired normal mucosa and neoplastic tissues was crucial in identifying genes with multiple alterations. Hypermethylation and loss of gene expression in tumors relative to normal tissue are cardinal features of epigenetic inactivation, whereas simply assaying methylation neither precludes nor dictates that the methylation is tumor-specific that it affects expression of the linked gene, nor that the gene was expressed in relevant normal tissue from the same patient. We excluded hypermethylated, mutated, or deleted genes not expressed (transcriptionally silent) in the corresponding normal tissue as well as hypermethylated genes that did not show transcriptional silencing in the LST. Finally, our operating logic was that genes that are most likely to be important would be altered by more than one mechanism. Consequently, this study has identified genes that are frequently altered by epigenetic inactivation, mutation, and/or copy number alteration that may have been overlooked in previous studies.
Occasionally, LSTs do progress to cancer (6), which raises the question what additional molecular alterations might drive this progression. The low incidence of copy number alterations in LSTs suggests that genomic structural changes would be important in gaining compound alterations in important colorectal carcinoma–related genes. This is supported by previous reports that copy number alterations are rare in adenomas relative to colorectal carcinomas (20). The absence of mutations in the PI3K, TGFβ, and p53 signaling pathways in LSTs shows that the mutation of PIK3CA, PTEN, SMAD2/3, or TP53 is also likely to be important in malignant progression. Previously, assessment of PIK3CA in LSTs detected mutations in 14% of 35 cases (30). Although Chang and colleagues also described invasive cancer in 14% of these LSTs, it was not clarified whether these were the same lesions with PIK3CA mutations.
A potential limitation of this study is the relatively small number of LSTs profiled in our discovery cohort. However, we purposefully chose to profile a small cohort of LSTs and normal tissues using multiple methods rather than acquire insufficient detail from large numbers using only one method and with inappropriate controls. This provided a much richer appreciation of the molecular landscape of LSTs. Our validation of genetic and epigenetic alterations in 84 additional lesions confirms that many of the alterations we identified are common in LSTs. While analysis of additional numbers will undoubtedly clarify the precise frequencies of molecular alterations in LSTs, it is already clear that their underlying biology is different from colorectal carcinomas.
A summary of the genes and pathways identified as altered in LSTs is provided in Fig. 7. The convergence of genetic, epigenetic, and transcriptional alterations in the axonal guidance and actin cytoskeletal networks was a key finding of this study. GSEA and data from a previous study (26) confirmed that transcriptional changes in axonal guidance genes were significantly overrepresented in LSTs, but not in polypoid adenomas, indicating that this pathway may be particularly important in LSTs. Axonal guidance signaling is not specific to neuronal tissues and it has been linked to the development of colorectal, breast, and pancreatic cancers (31–33). Alterations to axonal guidance signaling molecules can influence cancer cell migration and invasive through deregulation of the actin cytoskeleton (33, 34), which we also found was targeted by multiple alterations in LSTs. Extensive validation was performed on many of the genetic, epigenetic, and copy number alterations identified in this study, including the MAPK (3 genes), Wnt (3 genes), actin cytoskeletal (5 genes), and axonal guidance (5 genes) pathways. SLITRK5 is known to regulate neurite outgrowth through regulation of the actin cytoskeleton (35, 36). Its epigenetic inactivation in the vast majority of lesions tested suggests this may be an early event in colorectal neoplasia. Several additional members of the SLITRK (SLITRK1 and 2), SLIT (SLIT1), and NTRK (NTRK1, 2, and 3) gene families were also recurrently mutated or epigenetically inactivated in LSTs. RGMB (repulsive axon guidance molecule family member B) is an axon guidance molecule that regulates SMAD and Wnt signaling through bone morphogenetic protein (BMP) receptors (37). RGMB was deleted and showed different frameshift mutations in 2 LSTs (specimens 81 and 85) suggesting this gene is biallelically inactivated. MED12L (mediator complex subunit 12-like) was mutated in all 3 LSTs from the same patient (G1 and G2, p.V1694L; NG, p.V2042M) and was epigenetically inactivated in 3 LSTs, with both mutation and epigenetic inactivation in one LST. MED12L functions as a β-catenin–dependent transcriptional coactivator. The identified mutations flank the β-catenin–binding domain and could potentially impact on transcriptional coactivation of Wnt target genes.
By profiling both G-LSTs and NG-LSTs, we were able to explore the molecular differences between these morphologic subtypes. Our data support previous findings that both villosity and granular morphology are related (22) and that KRAS mutations are associated with villosity (38). Two studies have shown that CGI hypermethylation at several candidate genes was more frequent in G-LSTs than NG-LSTs, but that overall, CIMP is rare in LSTs (39, 40). Our data delve far deeper than these previous reports by integrating genome-wide gene expression and DNA methylation data from each LST and its paired normal tissue. In addition to showing that only a subset of hypermethylation events correlate with loss of gene expression, our study is the first to show that epigenetic inactivation is much more common in G-LSTs and the first to identify the specific genes (genome-wide) that contribute to epigenetic differences between the 2 morphological subtypes. Finally, we also show that the morphologic subtypes of LST can be distinguished on the basis of gene expression profile. We show this globally but also identify significant differences in the expression of specific genes and pathways, most notably in CXCR4 signaling, which is involved in chemotaxis and cell migration (41). Therefore, it may be possible to distinguish G- and NG-LSTs and consequently invasive potential, by assessing the genetic, epigenetic, and transcriptional status of a panel of genes.
In summary, this study represents the first comprehensive genome-wide survey of the molecular landscape of LSTs. Our approach of integrating 4 layers of omics data has allowed us to elucidate biologically relevant genes and pathways that would not otherwise have been evident. In doing this, we have discovered novel genes associated with LST development and have gained important insights into why some adenomas do not progress to cancer. In the broader context, the high mutation load in LSTs shows that adenomas with a low risk of progressing to cancer can exhibit as many mutations as colorectal carcinomas. This provides new perspectives on the established notion that colorectal cancer results from a progressive accumulation of mutations and has important implications for the validity of molecular biomarkers for assessing cancer risk.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: L.B. Hesson, R.L. Ward
Development of methodology: L.B. Hesson, J.W.H. Wong, B. Ng, S. Srivastava, M.A. Sloane
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.B. Hesson, B. Ng, S. Srivastava, P. Zarzour, C.-T. Kwok, D. Packham, A.C. Nunez, A. Dower, M.A. Sloane, N.J. Hawkins, M.J. Bourke, R.L. Ward
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.B. Hesson, B. Ng, P. Zarzour, S. Srivastava, D. Packham, A.C. Nunez, D. Beck, A. Dower, C.E. Ford, J.E. Pimanda, M.A. Sloane, J.W.H. Wong, R.L. Ward
Writing, review, and/or revision of the manuscript: L.B. Hesson, J.E. Pimanda, M.A. Sloane, J.W.H. Wong, R.L. Ward
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.J. Hawkins, M.J. Bourke, R.L. Ward
Study supervision: L.B. Hesson, M.A. Sloane, J.W.H. Wong, R.L. Ward
Other (methylation analysis, mutation analysis, gene expression analysis, etc.): L.B. Hesson, B. Ng, P. Zarzour, S. Srivastava, C.-T. Kwok, D. Packham, D. Beck, R, Ryan, A.C. Nunez, A. Dower, M.A. Sloane
J.W.H. Wong is supported by a Future Fellowship (FT130100096) from the Australian Research Council.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.