The human small intestine is the longest segment of the gastrointestinal tract, approximately five to six meters long in adults. Although it is three to four times longer than the colon, small intestinal tumors are rare, forming about 1% of all gastrointestinal malignancies 1. Like the colon, the small intestine is composed of multiple crypts (of Lieberkühn), which also contribute cells to finger-like projections called villi (Figure 1). Stem cells appear to reside near the crypt bases, producing numerous differentiated progeny that die within a week as they migrate upwards 234. Stem cells also produce Paneth cells that migrate downwards, and subsequently survive a few weeks.

Although there are likely to be many stem cells per crypt 234, intestinal stem cells have not been directly isolated or identified, and their properties and numbers are uncertain. Stem cell lineages may be immortal if at least one daughter retains stem cell properties after each division. However, mouse studies are more consistent with periodic losses of crypt stem cell lineages. The primary observation is that heterogeneous crypts created through mutagenesis or with chimeras eventually become homogenous 5. If stem cell lineages were immortal, heterogeneous crypts would not become homogeneous.

Stem cell turnover by a niche mechanism 56 can explain how heterogeneous crypts become homogeneous (Figure 1). Niche stem cells divide but total niche stem cell numbers remain unchanged as daughter cells constantly leave the niche and differentiate. Usually a stem cell produces one stem cell and one differentiated daughter (~95% of the time in mice 7), but sometimes a stem cell lineage may become extinct when both daughter cells leave the niche, balanced by another stem cell lineage that expands when both daughters remain within the niche. Such a niche population-type mechanism (random stem cell loss with replacement) eventually results in the loss of all stem cell lineages except one, allowing heterogeneous crypts to become homogeneous (Figure 1).

Human stem cell studies are difficult because the powerful visible fate-marking strategies of animal models are impractical. Most human crypts are visually homogeneous and therefore consistent with either immortal or niche stem cell mechanisms. A study of human colon crypts after therapeutic radiation was consistent with stem cell niches because the frequencies of mutant heterogeneous crypts progressively declined after radiation 8. Similar studies have not been performed on human small intestines.

One method that can reconstruct the past without direct serial observations infers ancestry from present day sequences. The past is encoded in sequences because errors may accumulate in a clock-like fashion (the molecular clock hypothesis 9) and record relationships between cells and times since the most recent common ancestor or bottleneck. Just as the age of the human race may be inferred by comparing contemporary genomes, so the ages of crypt stem cell populations may in theory be inferred from their sequences. With immortal stem cell lineages, adult crypts will have greater sequence diversity compared to crypts maintained by stem cell niches that periodically "bottleneck" whenever all but one current stem cell lineage are lost. Therefore, crypt diversity measurements potentially reveal whether stem cell lineages are immortal (last bottleneck around birth) or turn over by a niche mechanism (more recent bottlenecks).

Mutations are rare in normal cells, but recent studies illustrate that stem cell ancestries may be automatically recorded by random somatic epigenetic errors that frequently accumulate with aging 10. Methylation increases with age at certain CpG-rich regions in the colon 1112. This age-related methylation does not resemble a stereotypic developmental process because different cells within the same intestine may have different methylation patterns (5' to 3' order of methylated sites). Such random somatic errors resemble the drift of genomes during species evolution, and by analogy, such drift can be used to reconstruct somatic cell ancestry because methylation exhibits somatic inheritance 13 and all cells are related. Therefore, somatic methylation patterns potentially encode numbers of divisions since birth, and ancestral relationships among cells within a single crypt or villus. Here we infer from methylation patterns that crypt niches containing multiple stem cells maintain the human small intestine.

Crypts of Lieberkühn and villus fragments were isolated from fresh human small intestine (Figure 1) and individual allele methylation patterns were analyzed by bisulfite sequencing. Examples of crypt CSX methylation patterns (5' to 3' order of methylated sites) or "tags" are illustrated in Figure 2. Methylation represents somatic errors because CSX tags, like most CpG-rich sites 13, are initially unmethylated at birth. Although the tag patterns were complex, with differences within crypts and between crypts in the same intestine, the average percent methylation generally increased with age (Figure 3A). Crypt diversity (unique tags per crypt and intracrypt tag distances) also exhibited age-related changes (Figure 3A). Duodenum and ileum tag patterns appeared similar.

Different tags within the same intestine are consistent with stochastic errors that accumulate independently in different cells. To extract ancestral information encoded by seemingly random tags within a single intestine, crypts were modeled assuming stochastic methylation errors and either immortal or niche stem cells (see Methods). The primary difference between the models is that the probability of symmetric division (yielding two stem cell (P2) or two differentiated daughters (P0)) is zero with immortal stem cells, whereas the probability of symmetric division is greater than zero with niche stem cells (Table 2). Greater inter-crypt variability is expected with stem cell niches because both methylation errors and stem cell survival are stochastic.

As in the experimental data, the simulated crypts contained tags with different patterns, summarized by averages and intervals including 95% of simulated crypt outcomes (Figure 4A). Average simulated crypt values were consistent with average experimental values for either two immortal stem cells per crypt, or 4 to 256 niche stem cells. However, niche stem cells produced more inter-crypt tag variation, which better matched the variability observed in the small intestines. Inconsistent with immortal stem cell lineages were the observations that unique crypt tag variances were greater than simulated average values for all 12 intestines, and only 4 of the 12 intestines were within 95% simulation intervals for immortal stem cells (Figure 4B). In contrast, experimental variances for 10 of 12 intestines fell within 95% simulation intervals for the niche scenarios. In addition, more than four unique sequences per crypt were often observed, which should occur only rarely with two immortal crypt stem cells (four possible alleles), but may occur more frequently with 4 to 256 niche stem cells (Figure 4A).

Tag data from human large 10 and small intestinal crypts were similar, except the percent methylation appeared to increase more slowly with age in the small intestine (Figure 3A). Small intestine percent methylation was generally lower than published 1017 measurements of human colon crypts, with a trend towards a slower increase with age (Figure 5), but this difference was not significant (p = 0.24, F-test). Potentially methylation error rates are lower in the small intestine, but assuming equivalent error rates (a molecular clock hypothesis 9), stem cell division rates are either equivalent or lower in the small intestine compared to the colon. For example, simulations with a mitotic rate 75% of the colon (0.75 divisions per day) and a methylation error rate identical to the colon (2 × 10^-5 per site per division) better fit the small intestine data (Figure 3A and Figure 5).

Crypt methylation patterns encode niche dynamics, but exact niche stem cell numbers and probabilities of symmetric divisions remain uncertain because different combinations yield similar outcomes (Table 3). These niche combinations (4–256 stem cells and P1 from 0.98 to 0.89) are the same in large 10 and small intestines. For example, assuming that 95% of stem cell divisions yield one stem cell daughter, the crypt of Lieberkühn niche size is 64 stem cells. As in the colon 10, loss of all niche stem cell lineages except one or a "bottleneck" would recur on average every 3,000 divisions (with a 95% interquantile range of 1,000–7,000 divisions). Small intestinal crypt diversity (unique tags per crypt) did not change with age (Figure 3A), suggesting that niche dynamics or numbers of niche stem cells do not change with aging.

MYOD tags sampled from six small intestines were also consistent with the same niche parameters inferred with the CSX tags (Figure 3B). For a further test of our model, CSX tags were sampled from villi of six intestines. Villi should exhibit more tag diversity than crypts because they receive contributions from four to ten surrounding crypts 18. Furthermore, if methylation reflects the lifetime accumulation of stem cell mitotic errors, percent methylation should be similar between villi and crypts because crypt cells are only a few divisions older than villus cells. Consistent with these expectations (Figure 3C), there were significantly more unique tags per villus than per crypt (average 4.8 versus 3.0, p < 0.001, two-tail t-test) whereas methylation differences were not significant (average 21.8% versus 22.9%, p = 0.76). Distances between tags within a villus were also significantly greater than tag intracrypt distances (2.5 versus 1.7, p <0.001). Villus and crypt comparisons illustrate the ability of methylation tags to encode simultaneously both mitotic age (numbers of divisions since birth) and distinctly different ancestries.

The adult small intestine is physically composed of millions of crypts, and cells that constantly divide and die. Underlying the homeostatic visual appearance of a normal small intestine is an ancestral somatic cell tree that grows with age (Figure 6). Starting with the zygote, millions of related cells create an intestine. Past and present cells are all connected through time by ancestral lineages or branches, which ultimately root in the zygote or first common ancestor. Although stem cell definitions may differ 234, a crypt common ancestor is equivalent to a stem cell because all cells in a clonal crypt unit are progeny of this single common stem cell ancestor.

Both immortal and niche stem cell mechanisms can theoretically produce appropriate numbers of differentiated cells consistent with the morphology of normal small intestine. However, the shape of an intestinal somatic cell tree differs whether stem cell lineages are immortal or maintained by niches (Figure 6). Immortal stem cell lineages result in multiple long crypt branches, with the most recent common crypt ancestor present around birth. In contrast, a niche mechanism constantly creates newer more recent common crypt ancestors as most niche stem cell lineages become extinct.

Methylation tags can distinguish between immortal or niche stem cell lineages because random somatic epigenetic errors will accumulate differently. Specifically, a niche mechanism produces more variability between crypts (wider variation in unique tags per crypt within a single intestine) and less intracrypt diversity (smaller average intracrypt distances) than immortal stem cells (Figure 4). Small intestine and colon crypt methylation patterns 10 were similar and more consistent with stem cell niches. The same stem cell dynamics (Table 3) fit both colon and small intestinal niches. There were no apparent differences between ileal and duodenal crypt niches, and niche stem cell numbers (like the colon 10) did not appear to change or decline with age.

Niche stem cells defined by ancestry are physically intangible because common ancestors are defined by past events and no longer exist. Stem cells with immortal lineages are more readily identified because their pasts and futures are predictable. However, niche stem cell fates are unpredictable because all may potentially become common ancestors but only one will. The inability to predict niche survival may help explain why some adult stem cells have been so difficult to isolate or characterize. A niche somatic cell tree has few branches because random stem cell turnover eventually "prunes" all niche lineages except one (Figure 6). Such bottlenecks are predicted by our analysis to recur on average after 3,000 divisions.

Other mechanisms may also be consistent with crypt methylation patterns and the assumptions of our model remain unverified 19. For example, the observed variability in methylation patterns could also be generated if stem cell numbers were different between crypts within an intestine 19. Although our bisulfite sequencing appears technically adequate (no evidence of nonconversion of C to T and only rare mutations (<1 per 1,000 bases) at non-CpG sites), it is difficult to check the accuracy and reproducibility of the data. Repeat sampling of a crypt typically yields similar but not identical methylation patterns (data not shown), which may reflect crypt heterogeneity or experimental artifact.

One strategy to test empirically a stochastic model that inherently yields scattered results is to examine similar but biologically distinct entities. Both small intestine and colon crypts are thought to be maintained by stem cell niches 56 and the successful application of our model to the small and large intestines with methylation patterns from CpG-rich sequences on different chromosomes (CSX and MYOD) is empirical support for its ability to infer stem cell dynamics. In addition, small intestinal villi are physically different from crypts and represent mixtures from multiple adjacent crypts. Villi should be more diverse than crypts, yet villi and crypts should have similar numbers of divisions since birth. As one would expect if methylation records somatic cell histories, villus and crypt tag percent methylation were not significantly different, and villi contained significantly more tag diversity than crypts. Therefore, methylation patterns and our model are consistent with colon and small intestine crypt niches, and small intestine villi. Genetic alterations also appear to modulate stem cell dynamics because certain germline APC mutations are associated with significantly more diverse tags in normal-appearing familial adenomatous polyposis colon crypts, consistent with increased niche stem cell survival 17.

Mouse studies suggest that stem cell division rates are higher in the small intestines than the colon 2320. However, it is difficult with most assays to distinguish between stem and non-stem cell proliferation, and there are few human small intestine stem cell studies. Of interest, neoplasia, as with Min mice 21, is more frequently observed in the murine small intestines relative to the colon, whereas human small intestinal tumors are a small fraction of colon tumors 1. In contrast to murine studies, human small intestinal stem cell division rates appeared to be slower or similar to those in the colon because age-related methylation appeared to increase more slowly in the small intestine (Figure 5), although the difference was not significant. Assuming a molecular clock hypothesis 9 or equivalent CSX methylation error rates throughout the lower gastrointestinal tract, the data are more consistent with equivalent or slightly lower small intestinal stem cell division rates relative to the colon.

There is no definitive evidence supporting a somatic cell molecular clock hypothesis, but a priori a slower small intestinal stem cell division rate is more consistent with human intestinal cancer epidemiology 1, because fewer small intestinal errors (genetic or epigenetic) would be expected to accumulate during aging. A slower small intestinal stem cell division rate could also help explain why crypt purification is faster in the murine colon relative to the small intestine 5. Moreover, an ability to count relative divisions since birth in different human tissues allows distinctions (Figure 7) between mitotic tissue ages (divisions since birth) and chronological ages (years since birth). The colon and small intestines within an individual have equivalent chronological ages but may have different mitotic ages.

Methylation patterns suggest that niches containing multiple stem cells maintain crypts throughout the lower gastrointestinal tract. Small intestine stem cells appear to divide at equivalent or slower rates relative to the colon, and niche dynamics remain stable during aging. Although methylation tags only indirectly track stem cell dynamics and their exact interpretations are uncertain, they potentially allow for the systematic investigation of any intestine without prior experimental intervention. A somatic cell tree must underlie all human tissues and studies based on the hypothesis that certain somatic methylation errors record somatic cell histories may better define how cells divide and die during normal and abnormal aging.

The author(s) declare that they have no competing interests.

JYK performed most of the experiments and helped write the manuscript. KDS and ST helped with the statistical and quantitative analysis. DS analyzed the data and wrote the manuscript.