Large chromosome deletions, duplications, and gene conversion events accumulate with age in normal human colon crypts.

The extent to which the individual somatic cells of multicellular organisms remain genetically similar is unknown. Mutations may accumulate with time (aging), especially in tissues undergoing active mitosis like the human colon, where cell division is frequent and most cells are shed within a week ( Humphries & Wright 2008 ). In the human colon, this life-long continuous cell production is maintained by a stem cell hierarchy, where multiple long-lived stem cell lineages located at colon crypt bases produce much shorter-lived differentiated progeny that migrate outwards and are lost ( Fig. 1 ). Stem cell niche architecture may limit mutation accumulation (see Discussion) ( Cairns 1975 ). However, the relative efficiency of stem cell architecture and other mutation avoidance mechanisms are uncertain because it has been difficult to document the numbers and types of mutations that accumulate during aging. Mutation detection in normal tissues is hindered by polyclonality and the fact that mutations are randomly scattered throughout the genome of each cell. Given that each cell has different mutations, sequence analysis of a tissue sample gives only the predominant sequence at each position, which is most commonly the sequence in the original fertilized egg that gave rise to that individual. One might think that mutation detection with age could be achieved simply by sampling individual cells over the lifetime of the organism and expanding these cells in tissue culture. However, selection in culture can enrich for mutant subclones ( Colgin et al. 2002 ), and primary cells may mutate during in vitro propagation. To overcome these problems, we employed high density SNP microarrays to scan for chromosomal changes in individual human colon crypts. The assayed SNPs are spaced on average every ~5,000 bps, which provides the ability to detect relatively small chromosomal changes throughout the genome. Individual cells with different genetic changes would be occult to SNP microarrays because detection of loss of heterozygosity (LOH) requires the loss to be present in a high proportion of sampled cells. However, progeny of a single mutant stem cell may eventually become the majority of cells within a crypt (~2,000 cells per crypt) through selection in vivo, or random stem cell turnover (neutral drift ( Lopez-Garcia et al. 2010 ; Snippert et al. 2010 ), which can be termed niche succession ( Yatabe et al. 2001 )). Therefore, accumulation of genetic change within normal human colon crypts may be similar to the clonal evolution of tumor populations ( Shibata 2012 ), except crypt niche succession is not neoplastic ( Fig. 1 ). The types of detectable genetic changes can indicate whether selection versus neutral drift more often determine mutation fixation. Here we demonstrate that detectable chromosomal structural changes become increasingly common with age in normal human crypts, and in normal crypts from individuals with ulcerative colitis (UC). The findings of chromosomal changes in individual crypts from older individuals can be interpreted as a) chromosomal changes increase with age, or b) single stem cell dominance does not occur until very late in life. There is clear evidence that clonal dominance occurs around 16 week in mice ( Lopez-Garcia et al. 2010 ; Snippert et al. 2010 ), and data suggest the equivalent occurs prior to age eight years of age in humans ( Shibata 2012 ). Therefore, it is more likely that our current observations are the result of accumulation of mutation with age. Regardless of this point, these findings indicate that many human somatic cells gradually diverge from one another over the course of an average human lifespan.

Single individual whole normal crypts were obtained from fresh colectomies at the University of Southern California Keck School of Medicine using an EDTA washout method (from Dr. Darryl Shibata, USC) ( Yatabe et al. 2001 ). Normal crypts were isolated from normal appearing colon obtained at least 10 cm away from one another or from any tumor in the specimen. DNA was extracted in 15 ul of TE with 1 ul of 20 mg/ml Proteinase K at 56°C for 4 hours. The sample DNA was whole genome amplified with phi29 DNA polymerase using the Illumina kit (WG-312-1120 HumanOmniExpress-12 v1.1 DNA Analysis BeadChip Kit), which is one of the routine steps for preparing DNA samples for SNP array hybridization. This is a linear amplification method (with a ~7-fold increase in DNA amount) with minimal context bias and no precedent of causing LOH changes. DNA samples from the colon crypts were assayed with three different high density Illumina SNP microarray platforms (610-Quad, 660-Quad, 730-OmniExpress, ~600,000 to ~730,000 SNPs) using standard Illumina protocols. Data were processed using Bead Studio with a quality threshold of 0.15. Call rates ranged from 52.5% to 99.8% (average 86.3%) and were lower than the typical >99% call rate. The lower and more variable call rates were most likely due to the suboptimal amount of DNA (~20 ng) in each crypt, versus the recommended 200 ng of DNA per microarray. Crypts with call rates less than 60% were not studied further by either of the two methods of data analysis (see below). As a test of whether the SNP array approach could detect a known deletion on small amounts of DNA, we made serial 3-fold dilutions from 200 ng down to 0.8 ng of DNA for SNP array hybridization in duplicate to determine whether the amounts of DNA that we are using are sufficient ( Suppl. Fig. 8 ). We find that the results are reliable even with DNA amounts down to 0.8 ng. The DNA from a single human colon crypt (~2000 cells, ~12 ng DNA) is less than what is recommended for the Illumina microarrays, but is clearly still adequate. A total of 179 crypts from 18 individuals were considered for analysis by two different methods. The first method of data analysis is called a pairwise method. All crypts from the same subject individual were treated equivalently, and any crypts with overall call rates below 90% were not included. Across all of the crypts from a given subject, we filtered out all SNP positions with no-calls (NC) and homozygosity because these are not informative. We then did a complete pairwise analysis that pairs all possible combinations of two crypts, noting any differences among the crypts of a given subject. We identified locations with more than 3 continuous adjacent SNPs with the differences as an indication of large regions of a chromosomal change (deletion, gene conversion, or duplication). The Log R plot permits determination of the copy number (diploid, haploid (hemizygous), or double deletion) and hence permits distinction of a deletion from a gene conversion event. The B allele frequency plots primarily were used to determine an increase in chromosome copy number (e.g., duplication of a region on one homologue), but the Log R plots provide some indication of this (e.g., Fig. 5 ). In the second method of data analysis, the data were filtered with pairwise comparisons between a “reference” crypt (generally one with a high call rate) versus a “subject” crypt from the same colon, by first eliminating all no call SNPs. A likely LOH region was identified as a string of adjacent heterozygous SNPs. The smallest LOH region had 5 adjacent heterozygous SNPs, and the probabilities that LOH < 1 Mb occurred by chance were all less than 0.01. The results from both methods were summarized in 2x2 contingency tables separating subjects into categories of young versus old and the presence of a deletion or no deletion in all of the crypts examined. P-values from 2-tailed Fisher’s Exact Test were computed based on the entries in the contingency tables. Statistical analyses were performed in STATA version 12. The pairwise method has the advantage of not having to choose one crypt as ‘correct’ (i.e., with no LOH regions). The reference method has the advantage that crypts with lower call rates are included, allowing a larger number of crypts to qualify for analysis. We were interested in the degree of concordance of the two methods of analysis for this and future studies of this type where the number of cells is small, and hence, the amount of DNA is limiting, resulting in some samples with lower than normally acceptable call rates. As described in the Results, sequencing was done to confirm that regions of LOH identified using the SNP arrays had indeed undergone LOH. Products of some of the PCR assays were ligated to a TA cloning vector and multiple clones were sequenced to confirm that the cryps are clonal in nature. A representative example is shown in Figure 8 . PCR primers for the WGA material for the chromosomes 9 and 16 LOH regions are available upon request. PCR conditions were optimized according to the primer sequences. WGA for the chromosome 9 and 16 analysis was performed using the REPLI-g Minikit from Qiagen.

Chromosomal changes were assayed with high density Illumina genome-wide SNP arrays. Typically these arrays require ~200–250ng of genomic DNA (gDNA) to obtain call rates (the percentage of assays with discernible genotype calls) of greater than 98%. In this study we used gDNA from a single human colon crypt (~2000 cells, ~12 ng DNA). The validity of the assay for obtaining consisted results from such low among of DNA was confirmed (see Experimental Procedures). Call rates from a single colon crypt for each of the 179 hybridizations ranged from 52.5% to 99.8% (average 86.3%). The 179 crypts assayed from 18 individuals were analyzed by two different methods. In the pairwise method of analysis, all crypts from the same subject individual are treated equivalently, and only 83 crypts with overall call rates higher than 90% were included. Across all of the qualified crypts from a given subject, we filtered out all SNP positions with no-calls (NC) and homozygosity because these are not informative. We then did a complete pairwise analysis that compares all combinations of crypts, noting any differences among the crypts of a given subject. We identified locations where more than three continuous adjacent SNPs lose heterozosity in each pair of the comparison as deletions or gene conversion events. The Log R plot indicates whether the location is normal for copy number (i.e., diploid) versus hemizygous (i.e., locally haploid) for the DNA in that region. If there is LOH, but the region is diploid, then the event is a gene conversion. If there is LOH, but the region is haploid, then the event is a deletion. We also analyzed the data with a reference method that includes an additional 92 crypts (total of 175 crypts) of lower call rates to assess the feasibility of using a less stringent analysis for samples of suboptimal amounts of DNA. In the reference method of analysis, one crypt is chosen as a reference, against which the other crypts from that subject are compared. Four crypts with call rates lower than 60% were not included in the reference method. The pairwise method requires comparison of all possible pairs of samples from the same individual, so it is much more demanding, especially if the call rate of the samples is low. The assumption in the reference method that the ‘reference’ crypt has no mutation is not a necessary assumption in the pairwise method. Along with the stated goals of the study, we also were interested in knowing whether the less stringent reference method can arrive at similar conclusions as the more computationally intensive pairwise method. A total of 139 crypts from 14 “normal” subject colons were analyzed ( Tables 1 and 2 ) by the reference method. The pairwise method excluded crypts with lower call rates, and hence, only 63 crypts from 12 subjects were analyzed. Despite the differences between the two methods of analysis, the essential observations are similar. A total of 12 regions with LOH were detected in 6 out of 14 subjects by the reference method and a total of 7 regions of LOH were found in 4 out of 12 subjects by the pairwise method. Only one “normal” subject had more than one crypt with LOH in the analysis by both the reference and the pairwise method. In this individual, 4 out of 4 crypts had regions of LOH by the pairwise method, and 6 out of 14 crypts by the reference method, and all of the LOH regions were independent events ( Fig. 2 and Table 2 ). Given that SNP microarray analysis has not previously been done on such a small number of cells, we wanted to confirm the LOH detected using an independent molecular approach. We amplified the DNA using the whole genome amplification (WGA) method for crypts from subject 14 to examine the deletions on chromosomes 9. The crypts from the same individual without the LOH were used as the normal control for that region. Using PCR, followed by sequencing for SNPs flanking the putative region of LOH, we confirmed the last site of heterozygosity on each outer boundary of the potential LOH zone, as determined in the Illumina microarray. Specifically, heterozygosity (wild type) at a given SNP position gives both the A and B allele DNA sequence, whereas homozygosity (loss of one allele by deletion or gene conversion) gives only one DNA sequence at the SNP position. We then examined 87 SNPs within the LOH region (UCSC Genome Browser and dbSNP) using PCR of DNA from the crypt with the LOH and the normal control crypt, followed by sequencing. If a SNP within the LOH region was found to be homozygous, then the region of LOH could potentially include this SNP (see Experimental Procedures for sequencing confirmation of SNP LOH events). If a SNP within the LOH region was found to be heterozygous, then the LOH boundaries could be moved internally, with this SNP being the new boundary of the LOH. We started with SNPs closest to the last heterozygous positions determined in the Illumina array and moved from both LOH edges toward the center to search for the outer boundaries of the LOH. We were able to narrow the region of LOH to within a few kb for the chromosome 9 event in subject 14. Repetitive DNA (LTRs) within the last few kb made it impossible to define the precise boundary of the LOH. These repetitive sequences may have undergone a single-strand annealing type of homologous recombination, which led to a deletion (see below for a corresponding molecular analysis of one other subject event). The Log R plot ( Suppl. Fig. 4B , red box region) shows that there is a decrease in the copy number within this region, and this is the basis for inferring that there is a deletion within this region rather than gene conversion. The range of sizes for the chromosomal LOH changes, based on the SNP arrays, spanned from 73 kb to 46 Mb by both the pairwise and reference methods of analysis ( Fig. 3 ). The boundaries of the chromosomal changes were similar or identical ( Table 2 ), with only minor differences due to assumptions in the reference analysis that the reference crypt cannot be corrected by data from the other crypts. The pairwise analysis does not make this assumption. Four of the largest chromosomal changes involved nearly the entire long arm of a chromosome (46.6 Mb of chromosome 17 in subject 12, Fig. 4 ), 24 Mb in the distal long arm of a chromosome 7 (subject 9), a deletion in the distal long arm of a chromosome 5 (3.5 Mb in subject 13), and a deletion in the long arm of a chromosome 9 (3.3 Mb in subject 14)( Table 2 ). The first two of these events were diploid in the involved region based on the Log R plots, indicating that these were gene conversion events (see Discussion below regarding breakage-induced replication (BIR)). The second two events were hemizygous in this region based on the Log R plots, indicating that the event is a deletion. Other than the chromosomal changes that extended to the end of the q arms in subjects 9 and 12, all other changes were interstitial. All of the chromosomal LOH changes identified by the pairwise method were in four of the five individuals of 78 years or older. A 57 year-old subject with two LOH regions and an 80 year-old subject with one deletion, in addition to these four subjects, were identified by the reference method. Fisher’s exact test indicates that there are significantly more patients 78 years or older with chromosomal changes (80%) than patients 72 years or younger (0%) (p = 0.010, Suppl. Table 1a ) in the pairwise analysis, indicating a tendency for more deletions with age. The same test indicated that chromosomal changes were more commonly detected in individuals 78 years of age or older (83%), than in crypts from individuals 72 years of age or younger (13%) (p=0.026, Suppl. Table 1b ) by the reference method. Therefore, both methods of analysis indicate an increase in chromosomal changes with age, and these are predominantly deletional events. In addition to the deletional events, one duplication event was identified in a single crypt of one individual based on the B allele frequency and Log R ratio plots ( Fig. 5 ). Normal colon crypts (N=36) were also analyzed from four patients with UC, which is characterized by inflammation with damage and regeneration ( O’Connor et al. 2010 ). Potentially, crypts from UC patients may have a higher frequency of chromosomal changes than similarly aged normal individuals due to the damage and regenerative nature of this disease (“accelerated aging”). All subjects analyzed by both the pairwise and reference methods were found to have LOH regions in at least one crypt ( Tables 1 & 2 ). A total of 8 deletions were found in 19 crypts from 3 subjects studied by the pairwise method and a total of 15 LOH regions were detected in 36 crypts from 4 subjects by the reference method. Unlike the individuals with LOH in normal non-colitis crypts, where 5 of the 7 LOH regions were over 1 Mb, only a single LOH region among the eight ulcerative colitis patients was larger than 1 Mb ( Table 2 and Fig. 2 ). Interestingly, the LOH regions detected in the crypts from UC patients were all deletions and primarily clustered at 16p13.3, with three independent crypts from subject 15 ( Suppl. Fig. 5 ) and four independent crypts from subject 16 ( Suppl. Fig. 6 ) with deletions in the same general region of 16p. It is noteworthy that two deletions detected in two independent crypts from one of the normal patients were also in the 16p13.3 region. The 16p13.3 region is commonly deleted in cancer cell lines (“16p 6 Mb unexplained”), and also appears to be a DNA fragile site ( Bignell et al. 2010 ). Further molecular analysis of 137 SNPs in the chromosome 16 deletion zone for one of the crypts from subject 17 confirmed the deletion and showed similar results as for the chromosome 9 deletion described above for a “normal” subject ( Suppl. Fig. 7 ).

A long-standing biological question has been whether the somatic cells of multicellular organisms diverge genetically over the lifespan of the organism, due to DNA damage or replication errors. From a human health perspective, the accumulation of mutations in normal tissues may influence aging and increase cancer risks. However, it has been difficult to detect somatic mutations because it has not been possible to reliably sequence the genomes of individual cells from organisms. Previous work has shown that each colon crypt becomes increasingly monoclonal at an early age in mice ( Lopez-Garcia et al. 2010 ; Snippert et al. 2010 ), and this is likely the case in humans as well ( Yatabe et al. 2001 ). Here we use individual human colon crypts to detect and demonstrate that chromosomal changes become increasingly detectable in the last decades of a normal human lifespan. About 14.3% of normal non-UC human crypts from subjects aged 78 years or greater had a detectable åchromosomal change (10 deletions and one duplication out of 70 crypts), based on the reference method of analysis here. By the pairwise method of analysis, 29.2% of normal non-UC crypts from subjects aged 78 or greater had a detectable chromosomal change (7 deletions and one duplication out of 24 crypts). Although none of the colons can be considered completely “normal” because they were removed for therapy (primarily cancer, but prior to any chemotherapy or radiation), the chromosomal deletions represent the background mutation spectrum in otherwise uninvolved morphologically normal areas distant from the neoplasm. The chromosomal changes reflect the dynamics of human mutation and normal crypt stem cell biology because several steps are required before the changes observed can become detectable ( Fig. 1 ). Typically, a double-strand break (DSB) must first occur in a stem cell because a mutation in non-stem cells will be lost within days due to normal differentiation and luminal migration ( Humphries & Wright 2008 ). The chromosomal changes appeared to occur through any of three general mechanisms. A single DSB followed by templated copying of the other chromosome appears to have occurred when the deletion extends to the end of one arm of a chromosome ( Fig. 7A ). This is termed breakage-induced replication (BIR). This mechanism operates during replicative DNA synthesis and is well-documented in yeast ( Llorente et al. 2008 ), but few examples have been documented in humans or other multicellular eukaryotes. A second pathway also begins with a DSB, which can be joined to an independent DSB site on the same chromosome, resulting in a join by nonhomologous DNA end joining (NHEJ)( Fig. 7B ). This mechanism results in deletion of information on one homologue. Hence, there is LOH for the region of the deletion, and the genome is hemizygous in that region. A third pathway begins with a DSB ( Fig. 7C ). One end of the DSB invades the same chromosome at a region of homology located several kb to either side of the DSB (e.g., at an Alu repeat or LINE element) and undergoes homologous recombination that results in deletion of the region between the two sites of homology. The deleted region is seen as LOH and the genome is hemizygous in that region, as in the second pathway above. The difference between the second and third pathways is that the edges of the deletion occur in a region of homology in the third pathway, whereas this is not the case in the second pathway ( Fig. 7B vs. C ). The SNP microarray detection approach is not likely to detect reciprocal chromosomal translocations, and we found none. We also found no complex rearrangements on the same chromosome, such as chromothripsis; the latter has only been found in neoplasms thus far ( Stephens et al. 2011 ). After DSBs are repaired, the mutant stem cell must survive and eventually attain clonal dominance in order for the chromosomal deletion to become detectable. Crypts contain multiple stem cells with estimates of about 8 to 14 in mice ( Barker et al. 2007 ; Lopez-Garcia et al. 2010 ; Snippert et al. 2010 ), and more may be present in larger human colon crypts ( Yatabe et al. 2001 ; Nicolas et al. 2007 ). Two general mechanisms may drive niche stem cell dominance. First, the mutation may confer a selective advantage, allowing its stem cell to outgrow its neighbors. This mechanism, which should increase selective advantage with aging, appears unlikely because the deletions were widely distributed on different chromosomes, and none were near common colorectal cancer tumor suppressor genes. Potentially, the chromosomal changes may confer some unknown selective advantages, but a second mechanism that can also fix passenger or even deleterious mutations, thereby decreasing selective advantage, is neutral drift or ongoing random niche stem cell turnover. Neutral drift is observed in normal murine crypts, and similar stem cell turnover has been inferred for human colon crypts from o-acetyltransferase gene mutations, mitochondrial mutations, and methylation patterns ( Campbell et al. 1996 ; Yatabe et al. 2001 ; Fellous et al. 2009 ), but its role in mutation fixation is uncertain. Drift has a major role in shaping SNP frequencies in macroscopic populations ( Kimura 1968 ), and can fix by chance even deleterious mutations in small populations ( Whitlock 2000 ) such as crypt niches. The present data indicate that somatic chromosomal changes can be randomly fixed by neutral drift in colon crypt niches. For most mutations, drift appears to be the more common mechanism of fixation. The requirement for a mutant stem cell to attain niche dominance results in a lag between the mutation and the ability to detect genetic changes. This lag (the time for neutral drift to occur) is about 32 weeks in mice ( Lopez-Garcia et al. 2010 ) and has been estimated from methylation patterns at about 8 years in human colon crypts ( Yatabe et al. 2001 ). With an eight-year neutral drift interval, ten niche succession cycles would recur on average for each crypt in an 80-year old colon. The colon crypt chromosomal changes were easily detectable by our methods, which may reflect that they were fixed earlier in life in prior drift cycles. Studies of point mutations and small insertions/deletions in colorectal cancer (CRC) genomes suggest that the numbers of such mutations may be consistent with normal mutation and cell division rates ( Wang et al. 2002 ). Most colorectal cancers contain many more deletions than observed in normal crypts ( Leary et al. 2008 ), illustrating that normal mutation mechanisms are not sufficient to account for the numbers of deletions. Generally, the degree of chromosomal instability (CIN) associated with cancers does not appear to be present in most colon crypts ( Lengauer et al. 1998 ). Frequencies of chromosomal changes were much higher in UC crypts compared to non-UC crypts, illustrating that this disease state can modulate mutation and fixation. Potentially UC crypts could have “accelerated” aging as manifested by increased LOH, because of the increased cell turnover caused by inflammation. The UC chromosomal deletion spectrum may be somewhat different from normal aging because there seemed to be a deletional propensity or hotspot at 16p13.3, given that this occurred in three UC patients (cases 15, 16, and 17) but only occurred once in a non-UC individual (case 13). The 16p13.3 region is a frequent deletion site in cancer cell lines, and its fragility may be limited to certain cell types or conditions because DSBs were not observed with lymphocyte cultures ( Bignell et al. 2010 ). Because DNA fragile sites are polymorphic in human populations, the relative frequencies of the 16p13.3 deletions may reflect the inheritance of a fragile site rather than UC disease severity. The 16p13.3 deletions were rarely found in non-UC colons suggesting that the increased replicative stress of UC colons is uncommon in non-UC colons. This UC DNA replicative stress does not appear to increase overall genomic instability as manifested by an increase in other types of deletions. In individual 16, where four different crypts showed slightly different versions of the same chromosome 16 region deleted, there is a possibility that these four variations are related for biological reasons. For example, some NHEJ events involve insertion of DNA at the junction ( Yu & Gabriel 2004 ), and this insertion could make the junction unstable and subject to further rounds of deletion. If the original NHEJ event occurred in early development, then this region could be unstable in this patient more generally. We also acknowledge, but do not favor, the technical possibility that small variation in the boundaries reflect limitations of use of the SNP array approach on limiting amounts of DNA in individual 16. Detection of chromosomal aberrations with array-CGH analysis ( Le Caignec et al. 2006 ; Fiegler et al. 2007 ) and mutation analysis with next-generation sequencing ( Gundry et al. 2012 ) have been successfully carried out using DNA amplified from single cella. However, the reliability of using PCR amplified DNA from single cells for copy number analysis with SNP arrays remains uncertain. Deep sequencing as carried out by Gundry et al ( Gundry et al. 2012 ) provides information in genomic regions that amplify adequately in the whole genome amplification step; however, poorly amplified regions remain a concern. Also, deep sequencing to cover the entire genome is still 10-fold more costly than the SNP array method at this point. Comparative genomic hybridization (CGH) compares the controls and unknowns in the same hybridization; therefore, the regions of DNA that intrinsically do not amplify well do not lead to false identification of gain or loss of materials. Also, hybridization targets in CGH arrays cover large regions; thus, the entire region provides information in the readout. Unlike CGH, the hybridization targets in the SNP array are small in size with no information between any two targets (~10 kb apart on average). Therefore, deletions can easily be falsely identified if any small amplification dropout occurs. As pointed out by Gundry et al. ( Gundry et al. 2012 ), amplification drop out is not uncommon in whole genome amplification of single cells. Study of a very focal region of the genome has been done in the murine intestine using a chromosomally integrated LacZ reporter gene ( Dolle et al. 2000 ). This system is very good for studying point mutations and small insertions or deletions at specific genomic locations. While our approach cannot reliably identify these small changes, it surveys the entire genome in a single assay and allows the reliable detection of large chromosomal changes. Another interesting study documented more mutations in neural cells grown in culture briefly from old mice relative to young mice ( Bailey et al. 2004 ). Our results are consistent with that study. Our approach has the additional advantage of making sure that all changes occur within the individual patient rather than in culture. The accumulation of chromosomal structural changes documents how human genomes change with age in mitotic tissues. The study also indicates that phenotypically normal human crypts are mutation tolerant, even to LOH of entire chromosome arms. Genetic changes, rather than driving stem cell clonal evolution through the selection of increasingly more fit stem cells, appear to be more often randomly swept to fixation by neutral drift in a natural niche stem cell turnover rhythm. Many passenger mutations found in cancer genomes may therefore first accumulate in normal tissues even before neoplastic growth, and sequencing results of acute myeloid leukemia genomes clearly support this view ( Welch et al. 2012 ). Similarly, small subpopulations of cells in peripheral blood that have chromosomal deletions or focal LOH have been found to increase with age ( Jacobs et al. 2012 ; Laurie et al. 2012 ). Our study is the first to examine putative clonal normal tissues, and it suggests that somatic cell genetic divergence increases with age.