Lineage tracing reveals multipotent stem cells maintain human adenomas and the pattern of clonal expansion in tumor evolution

Significance

The organization of cells within human colorectal adenomas, and specifically whether the tumors are maintained by stem cells, is unclear. Furthermore, the patterns of clonal evolution leading to the development of a malignant tumor have not been determined. We performed lineage tracing in human adenomas using a combination of nuclear and mitochondrial DNA lesions and epigenetic markers. Our data identify a stem cell population within adenomas and suggest that new growth of intratumor clones occurs infrequently, not as a steady continual process as often is assumed. Our work offers a unique insight into human cancer development.

Abstract

The genetic and morphological development of colorectal cancer is a paradigm for tumorigenesis. However, the dynamics of clonal evolution underpinning carcinogenesis remain poorly understood. Here we identify multipotential stem cells within human colorectal adenomas and use methylation patterns of nonexpressed genes to characterize clonal evolution. Numerous individual crypts from six colonic adenomas and a hyperplastic polyp were microdissected and characterized for genetic lesions. Clones deficient in cytochrome c oxidase (CCO⁻) were identified by histochemical staining followed by mtDNA sequencing. Topographical maps of clone locations were constructed using a combination of these data. Multilineage differentiation within clones was demonstrated by immunofluorescence. Methylation patterns of adenomatous crypts were determined by clonal bisulphite sequencing; methylation pattern diversity was compared with a mathematical model to infer to clonal dynamics. Individual adenomatous crypts were clonal for mtDNA mutations and contained both mucin-secreting and neuroendocrine cells, demonstrating that the crypt contained a multipotent stem cell. The intracrypt methylation pattern was consistent with the crypts containing multiple competing stem cells. Adenomas were epigenetically diverse populations, suggesting that they were relatively mitotically old populations. Intratumor clones typically showed less diversity in methylation pattern than the tumor as a whole. Mathematical modeling suggested that recent clonal sweeps encompassing the whole adenoma had not occurred. Adenomatous crypts within human tumors contain actively dividing stem cells. Adenomas appeared to be relatively mitotically old populations, pocketed with occasional newly generated subclones that were the result of recent rapid clonal expansion. Relative stasis and occasional rapid subclone growth may characterize colorectal tumorigenesis.

The development of colorectal cancer along the adenoma–carcinoma pathway has become the archetypal model of solid tumor evolution (1). Both the genetic lesions and morphological features that evolve during colorectal carcinogenesis are well cataloged (2, 3), but remarkably little is known about the dynamics of the intratumor clones that bear these lesions. Furthermore, although mouse models point to the presence of stem cell compartments within adenomas (4), the cellular hierarchy of human colorectal adenomas is undetermined. The dynamics of these intratumor stem cell clones critically inform the search for effective biomarkers, provide a means to rationalize surveillance strategies, and potentially guide the choice of therapeutic interventions (5).

Human adenomas have relatively low malignant potential: Longitudinal studies have found that fewer than 1 in 10 adenomas become malignant within 10 y of first detection (6). Estimates of adenoma growth rates based on longitudinal endoscopic and barium observational studies suggest that adenomas remain relatively static in size for many years, with a large proportion of smaller lesions even regressing over time (7–9). Modeling of the relative mutation burden of colorectal cancers versus adenomas suggested that it takes 17 y for a large adenoma to become malignant (10). However, the clonal dynamics during this period of carcinogenesis are unclear. Intratumor clonal evolution may be characterized by the independent evolution of many different persistent subclones; alternatively, there may be extensive clonal replacement by newly generated mutant clones (selective sweeps).

Colorectal adenomas typically are composed of crypts (Fig. 1), self-contained structures that are morphologically similar to their nondysplastic counterparts in the normal colon. Mouse models suggest that the hierarchies of cell organization within adenomatous crypts are caricatures of the normal intestine in which rapidly cycling stem cells [expressing the leucine-rich repeat-containing G protein-coupled receptor 5 (Lgr5) gene] are located at the crypt base (4). However, the cellular hierarchy and location of stem cells within human adenomas is undetermined. Crypt fission—the bifurcation of a crypt into two daughter crypts—is the predominant mechanism by which a mutated crypt can expand clonally, both within the normal epithelium to form an adenoma and within early (noncancerous) adenomas themselves (11, 12). Concurrently, the proportion of branching crypts in an adenoma is increased ∼80-fold compared with the normal colonic epithelium (11).

Fig. 1.

Multipotent stem cells reside within clonal CCO⁻ adenomatous crypts. (A) (i) H&E staining showing tubular adenoma C157 with low-grade dysplasia. (ii) CCO enzyme histochemistry identifying two patches of multiple, blue, CCO⁻ crypts. (iii and iv) Laser-capture microdissection of areas in ii outlined in red and (v) mtDNA sequencing of single cells from multiple blue crypts and within the same blue crypts versus adjacent brown, wild-type crypts demonstrated that all blue crypts shared a common, clonal point mutation in their mtDNA that was not present in adjacent brown crypts. (B) Immunofluorescence staining of serial sections from adenoma 160. Clonal CCO⁻ crypts contained cells positive for markers of neuroendocrine cells (chromogranin A) and secretory cells [mucin2 (MUC2) and mucin5AC (MUC5AC)], indicating these crypts contained a multipotential stem cell that had produced these distinct cell types. Detection of CCO expression was conducted on the same section as chromogranin A after visualization for chromogranin A expression. Detection of MUC2 and MUC5AC expression was conducted on the same section simultaneously. Negative controls were isotype-matched at the same concentration as the corresponding primary antibody. Asterisks indicate the crypt enlarged in high-power images. (Scale bars: ∼50 μm in low-power images and 25 μm in high-power images. (C) LGR5 mRNA detection in FFPE tissue from patients with familial adenomatous polyposis (left pair) and sporadic adenocarcinoma (right pair). Expression is detectable in bases of unaffected crypts, in patches of adenomatous epithelium, and, in this example, extensively through invasive adenocarcinoma. Bright-field and dark-field reflected light image pairs with Giemsa counterstain.

Clonal expansion by fission produces contiguous patches of clonal crypts (13). Intratumor clones thus can be demarcated by identifying patches of crypts carrying a clone-defining genetic lesion (14). Additionally, deficiency of cytochrome c oxidase (CCO), readily detectable by histochemical staining, is a means of visualizing intratumor clones (15). CCO deficiency usually is attributable to a mutation of the mtDNA where the gene is encoded; thus the shared ancestry of a patch of CCO⁻ crypts can be demonstrated by their having a clonal mutation. Furthermore, the presence of multiple cell lineages within a CCO⁻ clone demonstrates that the clone contains a multipotential stem cell (16).

A serendipitous means to study dynamics and infer rates of clonal expansion in human tissues is via analysis of methylation patterns of CpG islands associated with nonexpressed genes (17). Methylation and demethylation at (some) nonfunctional loci occurs stochastically during DNA replication and is somatically inherited. Therefore, comparison of the methylation patterns between two somatic cells reveals their clonal relationship: Cells with a recent common ancestor will tend to have similar methylation patterns, whereas distantly related cells are unlikely to share similar methylation patterns. In the normal human colon, small clonal patches of crypts tend to have dissimilar methylation patterns, suggesting that clonal expansion rates are very slow in the normal gut (18). Accordingly, methylation patterns offer a means of infering the relative life history of tumor clones; epigenetically diverse tumors are likely to be relatively old clonal expansions, whereas relative homogeneity of the methylation pattern within a clone indicates de novo clonal expansion.

Here, we have shown that human adenomatous crypts are clonal populations maintained by multipotential stem cells. We have identified intratumor clones and characterized the evolution of these clones and of the tumor itself by analyzing methylation pattern diversity within these clones.

Results

Clonal and methylation analyses were performed on six snap-frozen sporadic human adenomas and one hyperplastic polyp. Aggregate DNA extracts from each tumor were screened for alterations in the genes frequently mutated in colorectal cancers (Table 1). Truncating mutations in adenomatous polyposis coli (APC) were detected in every adenoma but not in the hyperplastic polyp, and v-Ki-ras2 Kirsten rat sarcoma viral oncogene homolog (KRAS) mutations were detected in three adenomas. No adenomas had detected mutations in exons 5–8 of tumour protein 53 (TP53). No loss of heterozygosity (LOH) was detected on chromosome 5q close to APC or on 17p close to TP53 by LOH analysis of up to three microsatellite markers.

Table 1.

Bulk-genotypes, epigenetic description of each tumor, and patient details

Adenoma	Age, y	Sex	Histopathology	Size, cm	LINE-1 % methylation		APC	KRAS	BRAF	MSI	TP53	% methylation CSX	ACD	ICD
					Normal tissue	Adenoma
1 (148)	73	M	TVA, LGD	2.2	—	—	c.4348C > T	c.35G > A	WT	MSS	WT	76	1.09	1.42
			Ascending colon				p.R1450*	p.G12D
2 (157)	66	M	TA, LGD	2.2	76	72	c.3964 G > T	WT	WT	MSS	WT	57	1.69	3.29
			Sigmoid				p.E1322*
3 (158)	64	M	TVA, LGD	1.3	70	72	c.3956delC	c.38G > A	WT	MSS	WT	87	0.96	1.53
			Rectal				p.A1299fs*1	p.G13D
4 (160)	69	F	TVA, LGD	1.2	72	63	c.4118_4126delCTGAACAC	WT	WT	MSS	WT	94	0.68	0.87
			Sigmoid				p.P1371fs*10
5 (162)	62	F	TVA, LGD	2	73	70	c.4474_4475delGCinsT	WT	WT	MSS	WT	77	1.19	2.75
			Splenic flexure				p.A1492fs*14
6 (174)	60	M	TA, LGD	—	—	—	c.3925delGAAAA	c.35G > A	WT	MSS	WT	34	1.50	3.38
							p.E1309fs*4	p.G12D
7 (154)	61	M	Hyperplastic polyp Sigmoid	1.1	76	80	WT	WT	WT	MSS	WT	36	2.19	3.54

ACD, intracrypt distance (average for all crypts in adenoma); ICD, intercrypt distance (average for all pairs of crypts in the adenoma); LGD, low-grade dysplasia; MSS, microsatellite stable; TA, tubular adenoma; TVA, tubulovillus adenoma; –, no data available.

Microsatellite instability (MSI), as identified by BAT25 and BAT26 instability (not the full Bethesda panel), was not detected in any of the adenomas, and all had wild-type v-raf murine sarcoma viral oncogene homolog B1 (BRAF), suggesting that the CpG island methylator phenotype (CIMP) was unlikely in these adenomas (19–21). No significant difference in global genomic DNA methylation level, as measured by pyrosequencing of CpGs within the long, interspersed nuclear element 1 (LINE-1) elements, was seen between the normal and adenomatous epithelium of each patient (Fig. S1; paired Wilcox test, P = 0.5) suggesting that the rate of change in basal methylation within adenomas was comparable to that of the surrounding morphologically normal tissue.

Absence of CCO activity was used as a proxy to detect mtDNA mutations (15). The mtDNA from cells from contiguous patches of CCO-deficient (CCO⁻) adenomatous crypts then were PCR-sequenced to identify clonal mtDNA mutations. A clonal point mutation was sought and found in the CCO⁻ patch in adenoma 157 and in four patches in the hyperplastic polyp; mtDNA sequencing was not performed on other CCO⁻ patches.

Adenomatous Crypts Are Clonal Populations Maintained by Multipotent Stem Cells.

CCO⁻ crypts were identified by enzyme histochemistry (Fig. 1A). Multiple small areas, each approximating the area of a single cell, were microdissected from CCO⁻ crypts in the hyperplastic polyp (sample 154) and in adenoma 157 and were subjected to whole-mtDNA sequencing. The same mtDNA point mutation was present in all blue CCO⁻ areas of each crypt but was not present in the surrounding brown CCO-proficient (CCO⁺) crypts, confirming that the CCO⁻ crypts were a clonal population. Adenoma 157 is shown in Fig. 1A.

To demonstrate that the clonal CCO⁻ crypts contained multipotential stem cell(s), immunofluorescence for markers of different cell types (mucin 2 and mucin 5AC for secretory cells and chromogranin A for neuroendocrine cells) was performed on crypts from formalin-fixed, paraffin-embedded (FFPE) blocks of adenoma 160 and the hyperplastic polyp. All these cell types could be detected within single CCO⁻ crypts of adenoma 160, confirming that they were derived from the same multipotential ancestor stem cell (Fig. 1B). Goblet cells and neuroendocrine cells also were detected within single CCO⁻ crypts of the hyperplastic polyp (Fig. S2).

In addition, isotopic in situ hybridization (ISH) was used to detect LGR5 transcripts in an additional adenoma and also in a colorectal carcinoma. Adenomatous crypts contained LGR5-expressing cells and showed an expanded population of LGR5-expressing cells when compared with nonadenomatous crypts (Fig. 1C). In the carcinoma, LGR5 expression was significantly up-regulated: Nearly all malignant cells appeared to contain transcripts.

Adenomas Have Diverse Methylation Patterns.

Individual crypts were microdissected from each adenoma, and clonal bisulphite sequencing was used to determine the methylation patterns at CpG islands associated with the gene cardiac-specific homeobox [CSX (NKX2-5)] that is not expressed in the colon. Within each adenoma, the variation in methylation patterns between the crypts was marked (Fig. 2 and Figs. S3 and S4), suggesting that each adenoma was of sufficient somatic age to have evolved distinct intratumoral patterns of methylation. After correction for the degree of methylation (SI Materials and Methods and Fig. S5), the diversity of the intercrypt methylation pattern was found to differ significantly among adenomas (F-test, P < 0.001; Fig. 2 B and C), suggesting differences in the evolutionary history of each adenoma.

Fig. 2.

Epigenetic diversity in adenomatous crypts. (A) Topographical maps showing crypts sampled from adenoma 174 (wholly APC-mutant, KRAS-mutant subclone). A pairwise comparison of adenomatous crypts showed crypts had markedly different methylation patterns. Individual adenomatous crypts showed intracrypt epigenetic diversity. Boxes of dots represent methylation patterns; each row is a tag (molecule), each column a CpG site. An open circle denotes an unmethylated site; a filled circle denotes a methylated site. (B and C) Pairwise differences between each pair of crypts, sorted by adenoma. In some adenomas (e.g., 160 and 158), all crypts had similar methylation patterns, whereas in other adenomas (e.g., 162) some crypts had methylation patterns that were very different from those in the majority of crypts, illustrating the differences in the life-history of each adenoma. (B) ICD and (C) minimum distance between crypts. Colors indicate the dominant genotype within the adenoma; blue: APC–mutant KRAS–wild-type, red: APC/KRAS-mutant, green: APC/KRAS–wild-type).

Intratumor Clones Are Epigenetically Homogeneous.

Intratumor clones were identified by their sharing a somatic mtDNA mutation (adenoma 157 and the hyperplastic polyp; Fig. 1) not borne by the rest of the crypts in the adenoma, or, in the case of adenoma 174, a KRAS point mutation present only in focal regions of the tumor (Fig. 2A). For four of five mtDNA-marked clones, the diversity in the intercrypt methylation pattern in the crypts comprising the intratumor clone was significantly less than the diversity between the clone and the rest of the tumor (Wilcox test, P < 0.05 in four of five cases; Fig. 3), indicating that these clones represented recent clonal expansions. Interestingly, the pattern of methylation diversity within the KRAS clone within adenoma 174 was similar to the pattern in the bulk of the adenoma (Wilcox test, P = 0.8; Fig. 3), suggesting this KRAS-mutant clone was not a recent clonal expansion.

Fig. 3.

Intercrypt diversity of the methylation patterns from intratumor clones compared with the diversity of the bulk of the adenoma. Methylation patterns of subclones tended to be significantly less diverse than the bulk of the tumor cell population. C, intraclone diversity (red); CN, diversity between clone and nonclone regions of the tumor (blue); W, diversity of whole tumor (green); ***P <.001, **P < 0.01, *P < 0.05 for Wilcoxon test between clone and nonclone; numbers at the top of the plot indicate the adenoma along with the mutation denoting the subclone.

To look for further evidence of recent subclone growth in the adenomas, we considered the distribution of epigenetic distances between crypts in the adenoma. The distribution of intercrypt distances (ICDs) within each adenoma was skewed with either an extended right or left tail (Fig. S6). Therefore, most crypts within each adenoma were equally unrelated to one another. However, there were a few exceptional crypt-pairs in each adenoma that either were particularly epigenetically dissimilar to the rest of the adenomatous crypt-pairs, or were particularly epigenetically similar to one another. Finally, distance-based phylogenetic trees were constructed (SI Materials and Methods) using the ICD of methylation patterns to define the distance between each of the crypts. The trees had a nearly star-shaped structure, and the intratumor clones identified by genetic mutation tended to be closely related (Fig. S7).

Relationship Between Spatial and Epigenetic Distance.

To probe further the evolutionary history of each tumor, the relationship between the epigenetic distance and the spatial distance between crypts was examined. Straight-line distances between crypts were measured on a single slide for each adenoma. To facilitate comparisons among adenomas, the distances were expressed in units corresponding to the average diameter of the crypts within the adenoma (averaged over at least 75 crypts when the adenoma was of sufficient size; if the adenoma contained fewer than 75 crypts, all crypts in the section were measured). To compensate for the relatively sparse spatial sampling of the tumor, the average epigenetic distance, g(r), of crypts that were within a radial distance of [r,Δr] of one another was considered (as per ref (22) where the interval Δr was set at the average adenomatous crypt diameter, and the epigenetic distance between pairs of crypts was computed using the metrics described above. There was no correlation between the epigenetic and spatial distance of crypts in any of the adenomas (Fig. 4).

Fig. 4.

Correlation between spatial and epigenetic distance in adenomas. (A) Spatial distance versus epigenetic distance for whole adenomas. No correlation between spatial distance and epigenetic distance was observed. Spatial distances between crypts were binned into intervals of five units, with a unit corresponding to the average crypt diameter. The average ICD of all pairs of crypts within the adenoma that fell within the spatial interval was computed. Each colored line represents a different adenoma. (B) Spatial distance versus cumulative epigenetic distance. The straight lines suggest that physically close crypt-pairs are no more similar in their methylation patterns than physically distant pairs of crypts. If physically close crypts were more epigenetically similar on average than distant pairs of crypts, the cumulative epigenetic distance would increase exponentially. The cumulative epigenetic distance was computed from the data in A to compensate for the sparse spatial sampling of adenomas.

Tumor Growth Rate.

The intercrypt diversity in the methylation pattern of adenomatous crypts within each adenoma was compared with the diversity of the same number of crypts in localized patches of normal colon (data reproduced from ref. 18), using a case-resampling bootstrap method to compare equal numbers of crypts in each adenoma and the normal colon (SI Materials and Methods). After correction for the percentage methylation (SI Materials and Methods), pairs of crypts from adenomas appeared, on average, only slightly more similar than pairs of normal crypts [linear regression, P < 0.01; ratio percent-methylation corrected mean ICD in normal tissue vs. mean ICD in adenoma, 1:1.01), suggesting that average crypt growth rates were not markedly greater in the adenoma than in the normal colon.

To assess further the relationship between intercrypt epigenetic diversity and the ancestral relationship between crypts, the epigenetic diversity of a set of clonally derived crypts (e.g., a subset of crypts composing an adenoma) was simulated using a mathematical model (Materials and Methods). Two models of adenoma growth were considered: (i) the ”burst” model positing that all adenomatous crypts formed at the same time from a single ancestral crypt (T years before sampling), and there was no further crypt growth; and (ii) the “continual” model that posited steady growth of crypts (each crypt divides every f years, with the first division occurring T years prior sampling). In both models, the simulated adenomatous crypts were derived from a single ancestral (transformed) crypt. Epigenetic diversity was suppressed in the continual growth model as compared with the burst model (Fig. 5) because continual growth led to crypts having a more recent common ancestor than if the crypts were formed in a burst at the initiation of tumor growth. In the burst model, the epigenetic diversity of a population of crypts was strongly related to the time since the burst, or, equivalently to the time to the most recent common ancestor.

Fig. 5.

Modeling epigenetic diversity during clonal expansions. The diversity of methylation patterns between a patch of five clonally derived crypts was simulated with a mathematical model. Two modeling assumptions were tested: (i) that the patch formed as burst (with very rapid division of the ancestral crypt) at the onset of tumor growth, and (ii) that the crypts within a patch continued to divide every f years. The epigenetic diversity of the crypts was greater in the initial burst model (cyan line) than in the continual growth model (orange line, shown for crypt divisions every f = 2 y), and this diversity increased as the clone aged. Dashed lines represent 95% quantiles of the simulated values. Boxplots represent the mean ICD of samples of repeated five crypts taken from the CCO⁻ clone in adenoma 157 (red box) or from the nonclonal area (blue box). The diversity of the subclone was significantly less than that of the nonclone, suggesting that the subclone shared a much more recent common ancestor than the bulk of the adenoma.

The epigenetic diversity measured in adenoma 157, and the CCO⁻ subclone detected in this adenoma was compared with the model’s predictions. The comparison showed that a clone that had grown as a burst at least a decade previously could explain the diversity observed in random samples of five crypts from the subclone. The adenoma bulk was predicted to be significantly older. The same pattern was observed for the diversity of the intratumor clones in the other adenomas.

Stem Cell Dynamics Within Adenomatous Crypts.

The clonal architecture of individual adenomatous crypts was investigated by considering the diversity of the methylation pattern within individual crypts. Individual crypts typically contained multiple distinct methylation tags (Fig. 2). The diversity of the methylation pattern within individual adenomatous crypts was comparable to the diversity of the methylation pattern observed in morphologically normal crypts from nonpaired patients (Fig. 6A; normal crypt data reproduced from ref. 18). After correcting for the percentage of methylation, diversity in adenomatous crypts was slightly but not statistically significantly less than in normal crypts (linear model, P = 0.2), suggesting that adenomatous crypts have stem cell dynamics similar to those of their normal counterparts. Comparison of a stochastic model of stem cell turnover within crypts [previously proposed for normal crypts (17)] suggested that this level of methylation diversity is consistent with each crypt containing a number of stem cells that compete with one another for their place in the crypt (Fig. 6A). Further evidence for the existence of multiple long-lived stem cells in each crypt was derived from the observation of adenomatous crypts composed of a mixture of CCO⁻ and CCO⁺ cells (Fig. S8); the CCO⁺ and CCO⁻ cells derived from different stem cells.

Fig. 6.

Diversity of the methylation pattern of individual adenomatous crypts compared with the adenoma bulk. (A) Intracrypt diversity of normal and adenomatous crypts as a function of age is consistent with a model of neutral stem cell competition within crypts. Points are intracrypt diversity measured in normal crypts (blue circles) or adenomatous crypts (red triangles); the x-axis (age) jitter is artificial to allow all data points to be seen. Solid black line shows mean intracrypt diversity as predicted by the mathematical model with parameters as described in Materials and Methods; dashed lines are 95% quantiles for the simulation results. (B) Intracrypt distances (A) versus ICDs (I) for each adenoma. The mean ICD was greater than the intracrypt distance for all adenomas. Statistical comparison was performed using a Wilcox test: **P < 0.001, *P < 0.05. Unmarked pairs did not show a statistically significant difference.

The diversity of methylation patterns observed within crypts tended to be significantly lower than the diversity among all pairs of crypts within the adenoma (Fig. 6B), indicating that crypts themselves comprised a clone that had a more recent common ancestor than the most recent common ancestor of the tumor as a whole. In other words, the tumor population appeared mitotically older than any individual crypt.

Discussion

The clonal evolution of cancer remains poorly understood, despite extensive knowledge about the genetic and morphological changes that occur during the progression to malignancy (2). Here we investigated the clonal dynamics of adenomatous crypt populations within human colorectal adenomas. We found limited examples of clonal expansions within adenomas; instead, relative stasis of adenomatous crypt clones appeared to be the norm. These data are consistent with a model for tumor growth in which the time to the most recent common ancestor of each crypt-pair is greater than the time needed for the methylation patterns of two clonally derived daughter crypts to diverge. Assuming that the time needed for methylation patterns of daughter crypts to diverge is on the order of a decade [as appears reasonable in the normal colon (18)] and that the adenomas studied are around a decade old (10), our results suggest that the majority of the adenomatous crypt population is produced at around the same time, near the onset of tumor formation, and then this initial period of growth is followed by relative stasis (Fig. 7). Further supporting this notion is our observation that, on average, the methylation patterns of crypts that were located close together in the adenoma were no more similar than those of crypts that were disparately spaced (Fig. 4). For comparison, in a tumor composed of a mosaic of rapidly growing clones, individual clones would likely be identifiable as patches of crypts with relatively similar patterns of methylation (23). Interestingly, similar patterns of epigenetic diversity have been reported in advanced colorectal cancers: Typically the late-stage lesions showed little evidence of recent clonal expansions, and all pairs of cancer crypts appeared equally unrelated by their methylation patterns (24). Taken together, these data suggest that relative stasis, in which clonal outgrowth occurs at a slower rate than the rate of methylation pattern divergence may be characteristic of all stages of colorectal carcinogenesis.

Fig. 7.

Cartoon of tumor evolution. (A) A stepwise model of tumor evolution in which sequential mutations trigger extensive clonal expansions of the new mutant clone within the tumor mass. (B) A model for the relative stasis scenario of colorectal adenoma evolution. Here, intratumor clones can form near the onset of neoplasia but do not sweep through the tumor and so appear as spatially localized clones with divergent intraclone methylation patterns. Rare subclones form later in tumor development, have relatively homogeneous methylation patterns, and also occupy only focal regions of the tumor. Colors denote distinct clones. Within a clone, methylation patterns diverge as the clone ages. MRCA, most recent common ancestor.

It is instructive to consider the evolutionary selective forces that are acting on the tumor. Supposing that strong selection for new mutants would lead to rapid clonal expansion, growing intratumor clones would be expected to be relatively epigenetically homogeneous. The high diversity observed in the adenomas suggests only weak selection for new mutants (presuming that these new mutants are present) in early adenomas. Consequently, it is reasonable to suppose that the founder clone of the adenoma is sufficiently well adapted so that additional mutants confer relatively little fitness benefit (25, 26). Clonal evolution, analogous to tumor progression, therefore may occur in fits and starts (27), perhaps driven by the infrequent emergence of a particularly well-adapted clone that has undergone significant genetic evolution. Clearly, genome-wide approaches are required to test this hypothesis. The textbook model of carcinogenesis suggests that the progression to malignancy proceeds in a regular stepwise fashion, with repeated rounds of mutation and clonal expansion of the newly mutated clone (28, 29). Instead a punctuated model in which clonal expansion is rare (and perhaps slow), and clonal stasis is the norm, may explain our data. If the initial growth of an adenoma is rapid, perhaps because the resource that restricts growth in later adenomas has not yet become limiting, then the initial growth phase may represent a period when new mutations are likely to become established in the adenoma population. A mutant crypt produced in this initial growth phase will likely expand beyond the point where it is likely to go extinct by occasional (random) crypt extinctions. When the period of relative stasis is reached (and the phase of net adenoma growth is over), clonal expansion may be relatively more difficult, restricting the ability of new mutants to get a foothold within the tumor. This model explains the observation of the KRAS-mutant subclone in adenoma 174 being as epigenetically diverse as the rest of the adenoma bulk. Such a mode of clonal evolution may go some way to explaining the long interval [which may be as long as 17 y (10)] between the establishment of the adenoma and invasion. Moreover, if similar considerations apply to other preinvasive lesions, such as ductal carcinoma in situ, then the interval before invasion could be explained similarly (30). We note that it is challenging to prove this hypothesis conclusively with our data, because the expansion of clones within the tumor may be so slow as to be undetectable by our methylation-pattern molecular clock assay.

It is interesting that we have detected the presence not only of multipotential stem cells but also of clonal expansion within a hyperplastic polyp: Such solitary lesions have not been reported to show evidence of progression potential (31), and although clonal genetic changes are observed in some hyperplastic polyps within the context of hyperplastic polyposis (32) associated with BRAF mutations (33), this lesion was microsatellite stable and BRAF wild type. Nevertheless, it is clear that mutations driving a clonal expansion can be selected within such solitary lesions, possibly supporting the view that hyperplastic polyps may be fundamentally neoplastic rather than hyperplastic, even in the absence of microsatellite instability and/or BRAF mutation (34).

Mathematical modeling suggested that the significantly lower diversity typically observed in demarcated subclones was caused by the subclone’s having a common ancestor that was significantly more recent than the common ancestor of the bulk of the adenoma. The model was predicated on the assumption of steady rates of cell division (one division per day) and a constant rate of methylation change (2 × 10⁻⁵⋅CpG site⁻¹⋅division⁻¹). Because of the uncertainty surrounding these rate estimates, and indeed because of the constancy of the rates, coupled with uncertainties about the number and patterns of division of adenomatous crypt stem cells, the estimates of time derived from the model can be considered only illustrative. Indeed, methylation at nonexpressed loci tends to be higher in cancers than in age-matched normal mucosa, suggesting carcinogenesis-specific differences in the evolution of the methylation pattern (35). What is clear from these data, however, is that adenomas are mitotically old populations, and the observed clonal expansions are rare and relatively slow events. Only two scenarios of adenoma clone growth were considered in our model: continual steady growth or an initial burst followed by relative quiescence. Biological reality may a complex hybrid of these two growth patterns, but we believe that our data are insufficient to distinguish the precise clonal expansion rates.

Theoretical modeling of the relationship between spatial and genetic distance suggests that a tumor composed of a patchwork of competing clones would show an asymptotic relationship between the average genetic distance and spatial distance between crypts (22). This relationship would be the result of clonal competition whereby the growth of a clone is slowed by its having to compete with its comparably fit neighboring clones. Genetic distance between two clones was defined as the number of loci where the genotype of the two clones differed (the Hamming distance). In our study, we saw no relationship between the spatial distance and the average epigenetic distance of nonselected loci of individual crypts, and consequently it is tempting to conclude that the adenoma is not composed of a patchwork of competing clones. However, there are caveats to this conclusion. First, the relationship between the spatial distance and genetic distance of crypts is dependent on the mutation rate. When the mutation rate is high, the dependence occurs over only short time scales, because for high mutation rates, which better approximate the methylation error rate, records of ancestry are lost rapidly. Second, it is conceivable that the early adenomas may be nearly monoclonal populations composed a single clone driven only by an APC mutation (36). Clonal competition indeed may occur when a clonal outgrowth happens, but our data may record few occurrences of clonal outgrowth.

Both the percentage of methylation and the diversity of the methylation pattern varied significantly among tumors. These marked differences suggest that the pattern of growth differed among tumors: Diverse tumors likely were older neoplasms than their more genetically homogeneous counterparts. However, we cannot discount the hypothesis that the variation in the evolutionary history of each adenoma is explained by inherent differences between the clonal dynamics of the normal mucosa of each patient that are inherited by the adenoma. The apparent differences in the evolution of each tumor were notably not reflected by differences in stage and grade: All adenomas showed low-grade dysplasia. Thus, histopathological classifications may not represent precise measures of the state of tumor evolution. Interestingly, KRAS-mutant adenomas tended to be more hypermethylated than their KRAS wild-type counterparts (84 vs. 65%, respectively; Wilcox test, P < 0.01) although the robustness of this association is unclear because of the small size of our tumor cohort. The percentage of methylation presumably reflected the methylation status of the initial clone that formed the tumor. Methylation accrues with mitotic aging (17), and so it is tempting to suggest that KRAS-mutant tumors may have formed from a clone that was mitotically older than the initiating clone of KRAS wild-type tumors. Activation of oncogenic KRAS in the otherwise normal intestinal epithelium causes hyperplasia and increases proliferation with crypts (37) but is insufficient to initiate tumor growth (38). Thus, hypermethylated KRAS-mutant adenomas may have grown from a crypt that acquired an activating KRAS mutation before the gatekeeper APC mutation.

Next-generation sequencing provides a powerful means for studying tumor evolution and permits assaying many more loci in the genome, and at greater depth, than is possible with the clonal bisulphite sequencing technique that we applied. Indeed, such analysis has suggested punctuated evolution of breast cancers (39, 40) and found extensive genetic heterogeneity in renal carcinomas (41, 42) and B-cell chronic lymphocytic leukemia (43). Although our study lacks the extensive genetic characterization of those approaches, our methods have two benefits. First, by sequencing individual crypts, we retain spatial information and restrict ourselves to the study of clonal units, readily allowing the spatial extent of clonal proliferations to be determined. Second, the methylation patterns of nonexpressed genes are selectively neutral, removing the potential confounding imposed by selection in the interpretation of lesion accumulation at nonneutral loci.

The morphological basis of clonal evolution is not discussed often. If there is relative stasis within an adenoma, how is it manifested? Intestinal tumors are hyperproliferative at the level of individual cells (11), although this cellular proliferation is unlikely to relate to the growth of an adenoma as a whole, because it may signal only a rapid turnover of cells within individual crypts, not net growth. Correspondingly, apoptosis at the crypt base, the likely location of the stem cell compartment (4), is a prominent feature of small adenomas [those less than 5 mm in diameter (44)], supporting our conclusion of turnover within adenomatous crypt stem cells and perhaps within crypts also. Furthermore, because there is little change in the adenoma size over many years (8), any clonal expansion (via crypt fission) within a tumor must be approximately balanced by an equal amount of crypt extinction. Adenomas have a dramatically increased proportion of branching crypts compared with the normal colonic epithelium (11); we note that this increase does not necessarily imply an increased crypt growth rate, because the branched crypts could represent “stalled” fission events or there could be extensive crypt death leading to a high rate of crypt turnover and no net growth. Although directly observing crypt death events presents experimental challenges, we note that luminal apoptosis often is observed in both small and larger adenomas (44) and may be a hallmark of crypt death, and, in turn, clone extinction.

Adenomatous crypts were shown to be clonal populations containing multipotent stem cells. Thus, the clonal structure of human adenomatous crypts matches that of the adenomatous crypts in the mouse: Using their Lgr5/Apc/R26R-Confetti mice, Schepers et al. (4) were able, through a second tamoxifen injection, to perform lineage retracing of Lgr5⁺ stem cells in the adenoma, which were shown to be multipotential. Similarly, Fig. S8 shows adenomatous crypts partially occupied by CCO⁻ mutant cells (which later convert wholly to CCO⁻ cells, in which we have shown multilineage differentiation). Such partially mutant crypts also indicate the presence of multiple stem cells in each adenomatous crypt.

It has been proposed that LGR5, an accepted marker of intestinal stem cells in the mouse, also identifies human intestinal stem cells in both normal and neoplastic tissues (45–47). Here we have shown an expanded population of cells expressing LGR5 mRNA in an adenoma as compared with nonadenomatous crypts (Fig. 1C), similar to that reported in Apc^min adenomas (4). In the carcinoma, LGR5 expression was detected in nearly all cells; if indeed LGR5 marks human cancer stem cells, this observation would suggest that most, if not all, colon carcinoma cells are stem cells, with few in the non-stem cell compartment. The non–self-renewing non-stem cell population represents an evolutionary dead-end, and so the apparent up-regulation of stemness observed in the cancer may be a consequence of selection for cells capable of self-renewing during cancer evolution.

Stem cell dynamics were inferred by considering the diversity in intracrypt methylation and were found to be consistent with a model of neutral stem cell competition (drift) within each crypt. Neutral competition between stem cells appears to be a hallmark of normal intestinal crypts (17, 48), and our data suggest that the same dynamics persist in human adenomatous crypts. Indeed, the diversity of adenomatous crypts was comparable to that observed in normal crypts, implying that the dynamics of stem cell turnover in the normal and adenomatous crypt are comparable. Furthermore, analysis of the variance in the microsatellite pattern in mismatch repair-deficient adenomas (49) and in the methylation patterns in colorectal cancers (24) suggests that cell hierarchies may persist throughout carcinogenesis, albeit with altered numbers of stem cells.

In summary, we have shown that human colorectal adenomas are epigenetically diverse populations that show occasional signatures of recent clonal expansions. Colorectal adenomatous crypts contain actively dividing multipotent stem cells.

Materials and Methods

Tumor Selection, Histology, and Bulk DNA Extraction.

Seven neoplasms were obtained from seven patients according to institutional guidelines, and all patients gave informed consent (Research Ethics Committee). Half of each neoplasm was snap-frozen, and the other half was processed for diagnostic purposes as FFPE blocks. Paired histologically normal margin tissue for each tumor also was available in FFPE blocks. The frozen tumor tissue was serially sectioned onto PALM Membrane Slides (Zeiss) to facilitate laser-capture microdissection at a thickness of 10 µm; typically ∼100 sections were taken from each tumor. FFPE tissue was serially sectioned (6 × 5µm sections) onto plain glass slides for histochemical analysis. Histopathological classification of the tumor and margins was assessed using a serial H&E slide by pathologists (N.A.W., M.R.-J., and M.R.N.).

Two additional neoplasms (one tubular adenoma, one moderately well-differentiated adenocarcinoma) processed as FFPE blocks were used for in situ RNA detection (see below). Two further tubular adenomas were subjected solely to CCO staining (see below).

Bulk DNA extraction was performed by needle dissection. In the case of FFPE material, all six serial sections were used for DNA extraction, and the serial H&E staining was used as a guide to enrich epithelial cell extraction. For adenomas, three entire serial sections were used for DNA extraction. DNA extraction was performed by overnight incubation in Picopure Proteinase K DNA extraction buffer (Arcturus Bioscience).

CCO Histochemistry.

Frozen tumor sections were assayed for CCO activity using a two-color enzyme histochemical staining as described by Taylor et al. (15). CCO⁻ cells stained blue, and CCO-normal cells stained brown.

Immunofluorescence for Cell-Type–Specific Markers.

Adenoma FFPE sections were probed with mouse anti-human CCO subunit IV (2 μg/mL) (Invitrogen), mouse anti-human Muc5AC (82 μg/mL) (Binding Site), rabbit anti-human chromogranin A (46 μg/mL) (DAKO), and Muc2 (40 μg/mL) (AbCam) primary antibodies.

Antigen retrieval was performed in the microwave on rehydrated FFPE sections in boiling sodium citrate buffer, pH 6.0, for 10 min (CCO) or Tris-EDTA buffer, pH 9.0, for 15 min (Muc2 and Muc5AC). All sections were blocked using a serum-free protein block (DAKO) for 10 min followed by incubation with the primary antibody or isotype-matched negative control (DAKO) for 35 min at room temperature. Sections were washed three times for 5 min each washing in PBS followed by incubation with Alexa Fluor-conjugated secondary antibody for 35 min (goat anti-mouse 488 for CCO and Muc5AC; goat anti-rabbit 555 for chromogranin A and donkey anti-rabbit 594 for Muc2) (Invitrogen). Sections were mounted with VECTASHIELD HardSet mounting medium with DAPI (Vector Labs).

ISH for LGR5 RNA.

ISH for LGR5 expression was performed using a riboprobe corresponding to LGR5 that was located 566 bp from the 5′ UTR to exon5 (University of California, Santa Cruz chr12:70,120,102–70,233,231, introns excluded, cloned into pGEM3Z) (kind gift of Stefania Segditsas, Wellcome Trust Centre for Human Genetics, Oxford, United Kingdom). Hybridization and detection was as previously described (50).

Microdissection.

Laser-capture microdissection was performed using either PALM (Zeiss) or Leica microdissection systems. For mtDNA analysis, small areas of crypts, approximating a single cell, were microdissected from a single slide. For genomic and methylation analysis entire crypts were microdissected from three or more serial sections. DNA extraction was performed by incubation in Picopure Proteinase K DNA extraction buffer (Arcturus Bioscience) for a minimum of 3 h. Tubes with only digestion buffer served as controls for detecting contamination.

Mutation Detection.

Bulk DNA extracts were screened for somatic mutations in genes frequently mutated in colorectal cancers: the APC mutation cluster region (approximately codons 1,250–1,450), KRAS exon 1 (codons 12–13), BRAF (the region around codon 600), and TP53 exons 5–9. Primers and conditions for screening the mutation cluster region of APC are listed in Table S1; otherwise primers and conditions were as previously described (51). Sequences were obtained using BigDye 3.1 cycle sequencing and were run on an ABI 3100 DNA analyzer (Applied Biosystems).

Individual crypt DNA lysates were sequenced for mutations detected in the bulk-DNA sample from that adenoma.

Microsatellite LOH and MSI Analysis.

LOH analysis was performed on individual crypt lysates using up to six informative microsatellite markers close to the APC gene (D5S346, D5S2001, and D5S489) and markers on 17p (D17S1832) and 18q (D18S474 and D18S58). Amplification was performed using a multiplexed microsatellite PCR kit (Qiagen) using the primers previously described (51). Fragment analysis was performed on an ABI3100 system (Applied Biosystems), followed by analysis with the Peak Scanner software (Applied Biosystems). LOH was considered present if the area under one allelic peak was more than twice that of the other, after normalizing peak areas relative to constitutional DNA. MSI was assessed using multiplexed assay for the BAT-25 and BAT-26 mononucleotide repeats as previously described (51).

Bisulphite Sequencing.

Methylation pattern analysis was performed as previously described (18). Briefly, individual crypt DNA lysates were treated with bisulphite using an Epitect bisulphite conversion kit (Qiagen). After conversion, a nested PCR protocol was used to amplify the CpG island locus within CSX. PCR products then were cloned using the pGEM-T system (Promega) and were sequenced to obtain single-strand resolution.

Detecting mtDNA Mutations.

The whole mitochondrial genome was amplified using a nested two-stage PCR protocol as previously described by Taylor and colleagues (15). Only samples with a blank negative control were taken forward for sequencing.

LINE-1 Pyrosequencing.

Bulk DNA extracts of adenoma tissue and paired adjacent normal tissue (a histopathologically normal resection margin) were treated with bisulphite using an Epitect bisulphite conversion kit (Qiagen) as described above. Treated DNA was amplified using the PyroMark CpG LINE-1 kit (Qiagen) and pyrosequenced on a PyroMark Q96MD sequencer (Qiagen). The kit assays four CpG sites within the LINE-1 repeated element; the percentage of methylation at each CpG site was recorded and used in subsequent statistical analysis.

To check whether stromal tissue within the adenoma might be confounding the LINE-1 methylation assessment, for adenoma 157 both macrodissected adenoma and laser-capture microdissected (LCMD) adenomatous crypts only were run in parallel; the mean methylation level across the four CpG sites was 72.1% for the macrodissected tissue and 72.2% for the LCMD sample, suggesting that any stromal cells present within the macrodissected adenoma samples did not affect the pyrosequencing analysis significantly.

Statistical Description of Methylation Patterns.

A tag denoted the methylation pattern of one allele (one sequenced clone). To summarize the tags present in a single crypt, two summary statistics were used. The percentage of methylation was the percentage of methylated CpG sites averaged across all sites and tags. The intracrypt distance was the average pairwise distance between all pairs of tags in the crypt. The distance between two tags was defined as the number of CpG sites in which the methylation status of the two tags differed.

To characterize the degree of variability in methylation pattern between pairs of crypts within an adenoma, two summary statistics were used. The ICD is the average pairwise difference between all the tags of one crypt compared with the tags of the other crypt. The minimum distance is the average difference between each tag in the first crypt compared with its most similar tag in the second crypt.

Mathematical definitions of these metrics are as previously stated (18).

Mathematical Modeling of Methylation Pattern Dynamics.

Methylation pattern diversity within crypts and in patches of clonally derived crypts was simulated using a stochastic probabilistic model as previously described (17, 18). Briefly, each crypt was assumed to contain a fixed number of stem cells denoted by N. At each (synchronous) round of division, each stem cell could produce no, one, or two stem cell offspring, and a corresponding two, one, or no non-stem cell offspring with probability p₀, p_1, and p_2, respectively, constrained so that the total number N stem cells remained after division. Each cell contained two CpG island loci, each containing M = 8 CpG sites (denoting the two alleles of CSX). Changes in the methylation status (methylation or demethylation) of each site occurred with probability µ = 2 × 10⁻⁵ per division; otherwise daughter cells inherited the methylation patterns of their parent cell. Crypt fission was simulated by dividing the N stem cells equally between the two daughter crypts. The full stem cell complement of the new crypts was formed by instantaneous division of the N/2 cells. For simplicity, simulations began with loci being randomly 50% methylated to account for an accumulation of methylation during an occult initiation phase of tumor growth (assumed to occur because of the high percentage methylation observed in the adenomas). For each parameter set, at least 10,000 simulation repeats were performed, and averages and quantiles of these simulations were reported. Simulations were performed by proprietary C code that is available on request.

Statistical Modeling.

Statistics were performed in R using parametric or nonparametric tests as appropriate. Individual tests used are stated in the main text. Measures of methylation pattern diversity correlated with the average percentage of methylation of the tags being compared. To account for this confounding factor, diversity measures were corrected for the percentage of methylation when appropriate (see SI Materials and Methods for details).