An informative example of gene conversion is the IDS gene, located on the X chromosome. Mutations in IDS cause Hunter syndrome. There is a pseudogene IDS2 located 20 kb from IDS in an inverted orientation relative to IDS, with 88% overall homology to IDS [3]. 20% of Hunter syndrome mutations involve structural rearrangements induced by the interaction of the two nonallelic homologous regions [3,4]. The rearrangements appear to be independent events, indicating a recurrent mutation rather than common ancestry. Observed rearrangements include deletions, inversions, and gene conversion events [3,4]. Among the regions exhibiting gene conversion, a complex pattern of alternating sequence fragments from each of the duplicons is apparent. The IDS2 pseudogene is missing several IDS exons, but exhibits homology with IDS on each side of this 'gap'. Some of the deletion events observed in the IDS gene appear to represent conversion of IDS sequence by IDS2 in the vicinity of this gap, leading to the elimination of those exons [4]. A one kilobase recombinational hotspot has been identified for the IDS/IDS2 events; this hotspot exhibits 98% identity compared with the 88% overall identity of the duplicons [3]. Lagerstedt et al. [3] suggest that recombination is initiated in this high-identity region, and spreads through branch migration until a region of sufficient sequence divergence is reached. Lagerstedt et al. propose a model in which gene conversion leads to changes in both duplicons, and in which mismatched base pairs in the heteroduplex DNA may be corrected to generate additional conversion [3]. Figure ​Figure11 illustrates this model. These observations lead to two important conclusions. First, when looking for evidence of gene conversion, one should examine all duplicons for a given sequence. Second, one should examine the entire contiguous high-homology sequence in those duplicons, and not limit the analysis to the immediate neighborhood of a particular locus.

Somatic gene conversion can have multiple kinds of effects. Most obviously, conversion of a coding sequence by a non-identical homologous sequence may lead to a dysfunctional gene product, or an immunogenic novel amino acid sequence. Conversion of a regulatory, promoter, suppressor, or enhancer sequence may alter gene expression, either up or down. Since converted sequence usually retains the methylation status of the source sequence [5], conversion may result in either the methylation of previously unmethylated promoter sequence, suppressing gene expression, or in the demethylation of previously methylated sequence, enabling gene expression where it was not previously expressed. Gene conversion may also be correlated with other effects of nonallelic homologous pairing. Crossover and conversion occur in the same hot spot regions, and gene conversion appears to be preferred over crossover when interacting regions are short [6]. Non-allelic crossover may lead to insertions, deletions and/or inversions. Homologous pairing within a short region of DNA could create DNA loop structures that alter transcription patterns [7]. Conversion could potentially occur during DNA/RNA pairing [8,9]. Any of the effects mentioned above could have a major impact on cell function, and provide plausible causative mechanisms for disease.

I propose to examine somatic gene conversion in the context of disease using single nucleotide polymorphism (SNP) microarray data. Because conversion tracts are short, linkage disequilibrium (LD) between a gene conversion locus and nearby SNP markers is likely to be weak or nonexistent [10]. As a result, it becomes necessary to analyze single-SNP markers without expecting to see correlated patterns in nearby markers as one would expect in a traditional disease association study.

The Wellcome Trust Case Control Consortium (WTCCC) data set was obtained using an Affymetrix 500 K platform [11]. Genotyping was performed on two large British control populations (58C, NBS), in addition to disjoint populations for bipolar disorder (BD), Crohn's disease (CD), coronary artery disease (CAD), hypertension (HT), rheumatoid arthritis (RA), type-1 diabetes (T1D) and type-2 diabetes (T2D). The WTCCC data has been extensively analyzed using a traditional genomewide association study [11]. This previous analysis required the presence of three concordant SNP markers in order to identify a disease-associated haplotype. Such an analysis is likely to miss gene conversion events because of the weak LD. Further, by focusing the analysis at the called-genotype level, such an analysis is insensitive to somatic changes to the genome.

DNA samples in the WTCCC study are obtained from lymphocytes. One might be concerned that an analysis of somatic mutation in lymphocytes may not be informative about somatic mutation in other tissues more closely associated with the diseases in the WTCCC study. Fortunately, there is some evidence that a phenomenon related to gene conversion known as sister chromatid exchange (SCE) is informative about disease when measured in lymphocytes. SCE involves crossover between homologous sister chromatids mediated by the homologous recombination pathway [12], and has been interpreted as indicating general genome instability and/or a response to DNA damage [13]. SCE is elevated in lymphocytes of individuals with CD [14], CAD [15], T1D [16], and T2D [17], but not RA [18], although in some cases the elevation may be related to treatment rather than disease [19]. SCE is also elevated in multiple sclerosis [20], systemic lupus erythematosus [21], several cancers [14], and in individuals with viral infections [22].

Since SCE analysis using lymphocytes (rather than tissues directly affected by the disease) is informative, one might expect lymphocytes to also show disease-associated gene conversion behavior. Because blood cells are widely circulating, they are likely to encounter agents of double strand breakage such as viruses, and therefore exhibit gene conversion if conversion is occurring anywhere in the body. Further, a disease may be associated with damage to a particular tissue, for example by autoimmune processes, and the destroyed tissue is unavailable for analysis. Other cell types such as lymphocytes might therefore serve as useful proxies for damaged tissues. If the mechanisms of in-vivo gene conversion are sequence-specific rather than tissue-specific, then lymphocytes would exhibit the same conversion experienced by the damaged tissue, without eliciting the destruction response.

To identify somatic changes in a population I propose a novel data analysis technique. The technique takes advantage of the fact that a sample contains DNA from many cells of a single individual. If a significant proportion of those cells have undergone gene conversion at a locus, then the resulting change in the genotype of those cells should be measurable as a perturbation in the intensity for the two allele probes at that locus. An SNP with a distribution of perturbations specific to a disease population serves as a marker for a potential disease-associated locus. More details about how such perturbations are measured, and why such perturbations would have a signature different from other sources of variation such as paralogous sequence variants, can be found in the Methods section below.

Once a set of SNPs showing the signature of gene conversion is identified in a disease population, it would be desirable to validate those associations using an independent source of information that links the disease to those SNPs significantly more closely than to randomly chosen SNPs. As noted above, one needs to consider not just the SNP locus itself, but all regions with homology to the duplicon containing the SNP. The most direct form of association between a region and a disease is to find a gene in the region that is known to be associated with the disease, or that participates in a critical pathway known to be relevant for the disease. Additional evidence might include data showing that the gene is expressed in the relevant tissue with function related to disease pathogenesis. Most regions of high homology contain at most a few genes, and so the analysis can be relatively specific. One could also look for adjacent genes for which the duplicon could plausibly contain an upstream enhancer locus. I use 30 kb as a threshold for this type of adjacency.

When duplicons are nearby on the same chromosome, the intermediate region between them is an additional region of interest. Improper recombination between such regions could lead to inversions, insertions, or deletions of the intermediate sequence. In some situations, somatic deletion of a genomic region can generate patterns similar to those that would be generated by gene conversion. Deletion might be suspected when the duplicons occur in an aligned fashion nearby on the same chromosome, a configuration that could lead to misaligned recombination.

In the presence of an agent that induces genetic damage, a cell may respond by inducing the homology-directed repair pathway [23]. If this pathway is induced in each of many cells in response to the same agent, the same homology-biased mutations may happen in a variety of tissues. Mutations in stem cells will persist in lineages descending from those cells.

The damage-initiating agent may act locally or globally. A local agent, such as a virus that damages DNA in a position-specific manner, could induce gene conversion selectively in the region surrounding the target sequence. A global agent, such as a deficient or inactivated DNA repair pathway [24], would lead to DNA damage in a broad (but not necessarily random) fashion, inducing generalized gene conversion at many loci. Local gene conversion will be identifiable as a perturbation in the disease population that is absent in the control population and other disease populations. Perturbations due to global gene conversion may be present, to a lesser degree, in other populations whose diseases are caused by global agents. The perturbations should presumably be absent in the control population and in populations for diseases caused exclusively by local agents.

For a SNP locus with an A/B polymorphism, the microarray generates a pair of intensity values I_A and I_B. Each intensity value is the average intensity over a small number of oligonucleotide probes containing the allele together with some flanking sequence. The (I_A, I_B) point typically falls within one of three clusters corresponding to the three genotypes AA, AB, and BB.

Consider now an individual with an AA genotype. Suppose that 20% of the sampled cells of this individual have undergone gene conversion in which one of the A alleles has been converted into a B allele by a homologous sequence, while the flanking sequence has remained unchanged. The left example of Figure ​Figure22 shows this kind of conversion. (Conversion of both A alleles would be rare, and is ignored.) This individual will display an overall (I_A , I_B) intensity pair that is 20% of the way from the AA cluster to the AB cluster. In another individual with a heterozygous AB genotype, a 20% conversion rate at the same locus would yield an overall (I_A , I_B) intensity pair that is 10% of the way from the AB cluster to the BB cluster, since only the conversion of the A allele will cause a change in probe intensities. In an individual with a BB genotype, no change would be observed.

Because there is experimental variation in intensity measurements, it may be difficult to determine whether a small perturbation in a single measurement represents gene conversion or merely noise. However, it is possible to study the distribution of perturbations for a population at a locus. If a population has a significant spread of intensities between clusters, when control populations do not, then one can hypothesize that gene conversion at that locus is happening in a population-specific manner. See the cluster plot for RA in Figure ​Figure33 for an example. If the population is a disease cohort, then the locus may be associated with the disease phenotype.

Returning to the example above, consider the complementary situation in which the flanking sequence near an SNP probe has been converted. Whether or not the SNP locus is changed, the converted sequence will no longer match either probe sequence. The right example of Figure ​Figure22 shows this kind of conversion. If many cells in an individual are converted in this fashion, a reduced signal from this sequence will be measured by both probes of the microarray. For a locus at which this effect is associated with the disease phenotype, all clusters will shift radially towards the origin in the cluster plots for the disease population.

Calling algorithms attempt to identify the boundaries of clusters corresponding to the AA, AB and BB genotypes. For example, the Chiamo algorithm [11], considers all populations simultaneously, and estimates cluster boundaries in a way that allows for some population-dependent differences. The intensity distributions vary from SNP to SNP, and so clustering is performed separately for each SNP.

Based on the analysis above, gene conversion for a particular population should be accompanied by either (a) an increase in the spread of the two-dimensional intensity distribution relative to the control population, or (b) a translation of the clusters towards the origin, relative to the control population. In case (a), there should be an increase in the number of points that are either between clusters, or on the fringe of a cluster. In case (b), there should be a decrease in the distance between clusters, leading to an increase in the number of points whose cluster assignment is ambiguous. Either way, there will be an increase in the number of no-calls generated by the calling algorithm, relative to the control populations. This is one 'signature' of gene conversion that I will try to identify.

The Chiamo calling algorithm has been applied to the WTCCC data, and it is possible to use those calls to help recognize the signature of gene conversion. Chiamo generates a confidence score for a call; the authors of the Chiamo algorithm recommend that when this score is below 0.9, the genotype should be considered a 'no-call.' When clusters are more dispersed, their peripheries can begin to overlap with each other. In such a situation, the Chiamo algorithm will have less certainty about points falling in the intermediate regions. Chiamo will define cluster boundaries more tightly, resulting in an increase in the no-call rate for intermediate points [11]. An example of this phenomenon is given in Figure ​Figure3,3, where the orange points (that are particularly frequent in RA at this locus) are no-calls.

An increase in no-calls between two clusters can lead to a biased allele distribution in the called genotypes. For example, if there are many no-calls between the AA and AB clusters, then the A allele will be underrepresented among the subpopulation whose genotypes are called with high confidence. This bias is another possible signature for gene conversion. (See Additional file 1 for an extended discussion of no-calls.) Note that there may be cases of gene conversion that do not show this signature because the non-called points do not change the observed allele frequencies.

To identify gene conversion events, I take three complementary approaches. The first approach that I call the 'stringent' filter is designed to optimize precision, that is, to minimize the number of false positives while possibly missing some true positives. The second approach is designed to provide better recall, that is, to include more true positives at the risk of also including false positives. This second approach is called the 'relaxed' filter. The third approach, termed the 'no-call-only' filter, looks only for extreme no-call rates, since some gene conversion loci may not exhibit changes in called allele frequencies.

For the stringent filter, called SNPs with high no-call rates in a population relative to the union of the two control populations are initially selected. A chi-squared statistic is calculated for each SNP based on a 2 × 2 chi-squared test comparing calls/no-calls for both the disease population and the control population. Only SNPs with an increase in the no-call rate in the disease population and a chi-squared statistic corresponding to P < 5 × 10^-5 in a one-sided test are retained by this initial selection.

A further selection is applied to test for a bias in the genotype distribution in the disease population relative to controls. Bias is assessed in one of two ways; an SNP that displays bias according to either of these tests is retained. Only SNPs in which the control population has at least ten individuals for each of the AA, AB and BB genotypes are considered. First, the three genotype frequencies in the disease population are compared with the corresponding frequencies in the control population using a 3 × 2 chi-squared test to determine the likelihood that they have a common distribution. Only SNPs with a chi-squared statistic corresponding to P < 5 × 10^-4 in a two-sided test are retained. Second, the three genotype frequencies in the disease population and control population are separately assessed for departure from Hardy-Weinberg Equilibrium using a conventional 3 × 2 chi-squared test. Only SNPs with a chi-squared statistic corresponding to P < 5 × 10^-4 in a two-sided test in the disease population and a chi-squared statistic corresponding to P > 0.01 in the control population are retained.

Gene conversion appears to require at least 300 base pairs of homology in humans [1]. Among known gene conversion loci, the smallest degree of identity between the homologous regions is 88% [1]. One should therefore not expect newly discovered loci to have identity much below 88%. I will thus use 85% identity as a lower bound for the stringent filter.

The candidate SNPs were evaluated for homologous flanking sequence elsewhere in the genome. The UCSC database of segmental duplications [25] was used to identify genomewide duplications with at least 1,000 base pairs of homology (after elimination of low-complexity repeats) and at least 90% identity. Additionally, each SNP that met the other stringent filter conditions was subjected to manual analysis using the BLAST network service at NCBI to identify duplications that may not meet the thresholds of the segmental duplication database, but that may still be relevant for gene conversion. (I used the Megablast algorithm with default parameters. When a duplicon contains several almost-contiguous segments, the identity of the duplicon is the identity reported by BLAST for the segment containing the region that maps to the SNP under consideration.) The three filters are summarized in Table ​Table1.1. The relaxed and no-call filters use different homology criteria from the stringent test so that the segmental duplication database can be used to automate the analysis. Because the segmental duplication database excludes regions with low complexity repeats, some SNPs in regions with more than 90% homology (for example, rs9378249) are not in the segmental duplication database.

The analysis does not consider SNPs on the Y chromosome. For the X chromosome, the analysis is limited to the female subpopulation within each cohort. As a result, some statistical power is lost, particularly for cohorts such as CAD that have a relatively small number of female members.

Cluster plots for all SNPs mentioned in the text can be found in Additional file 2.

Confounding factors could perturb cluster plots, potentially leading to false associations. Loci that did not meet the WTCCC quality control requirements have been excluded. The WTCCC reports a disparity between the NBS/RA/CAD cohorts and the other cohorts for some SNPs [11]; such disparities are rare among the SNPs meeting the filter conditions (Table ​(Table9).9). Additional quality control issues not identified by the WTCCC are possible. Nevertheless, it is hard to imagine how a quality control artifact could lead to population-specific effects that correlate with disease related genes.

Copy-number variation can be discounted as a general explanation for the observed phenomena, since none of the stringent test SNPs (and only one of the no-call SNPs) showed more than three clusters. Further, few of the stringent filter SNPs are within known CNV loci (Additional file 1). Even if copy number variation was the mechanism responsible for some of the present results, the results would still be interesting as novel cohort-specific associations.

The present paper provides support for the hypothesis that many complex diseases are caused in part by somatic mutation in regions with homology elsewhere in the genome. Diseases such as cancer often display gross karyotypic changes that could be due to improper recombination between nonallelic homologous regions in somatic tissue. Because detection of somatic mutations is technically much more demanding than that of germline mutations, somatic gene-conversion events in cancer have probably been underestimated [1].

Some puzzling epidemiological features of autoimmune diseases are consistent with a somatic mutation hypothesis. Association with viruses can be explained by the mutagenic actions of those viruses. Associations of autoimmune disease with higher latitudes has been hypothesized to relate to lower vitamin D levels [191]; vitamin D is associated with lower rates of double-strand breaks [192] and with protection from viral infections [193]. Complex inheritance patterns spanning multiple diseases would result from a common underlying genetic susceptibility based on sequence homology, combined with stochastic effects such as tissue-specific viral infection.

In order to be identified as a conversion region in this study, the region must contain a locus that is within the SNP repertoire of the microarray chip. A substantial amount of somatic gene conversion might affect loci with alleles that are fixed in the population. If so, alternative platforms will be needed to detect such conversion. It is likely that there is additional disease-specific somatic gene conversion that the present study has not detected even among the covered SNPs. Spread in the cluster plots might not be apparent if a particular disease-causing somatic mutation was rare enough that the perturbation was small relative to experimental variation.

On the other hand, common gene conversion events might preferentially include SNP loci. If a conversion event is common in somatic tissues, it may also be relatively common in the germ-line. If the germ-line event is not deleterious, a polymorphism could result. The consequences of somatic and germ-line changes are different, and a somatic mutation may cause disease where a germ-line change does not. For example, a somatic mutation may result in a novel protein that is immunogenic. Alternatively, some of the loci associated with a conversion event may be phenotypically neutral, and these may lead to polymorphisms as a result of partial conversion events in the germ-line.

The phenotype of a somatic mutation is likely to be very different from the phenotype of a germ-line mutation. Outside of cancer, there is very little data about phenotypes associated with somatic mutations. It is therefore dificult to correlate the observations of this paper with existing knowledge about somatic mutation. Correlations with genomewide studies of disease associated polymorphisms are possible in principle. However, given the methodologies used in those studies (for example, requiring multiple concordant SNPs [11]), it is not expected that correlations will be found given the absence of linkage disequilibrium for gene conversion [10].

It may well be that somatic gene conversion is, in some cases, a normal adaptive phenomenon. Such effects might be detectable using SNP microarrays by examining the intensity plots directly without employing a calling algorithm. The quality control protocols of SNP array studies typically exclude loci where the called allele frequencies depart from HWE in the control population, which would exclude loci for which somatic gene conversion was common. It may be worth re-examining such loci, particularly those in duplicated regions.

The present report suggests that somatic gene conversion is associated with mutations and genomic rearrangements that lead to disease. Working backwards, one could generate hypotheses for further study by identifying genomic regions with high degrees of homology that contain disease-relevant genes. For example, the BRCA1 gene that is involved in DNA stability and repair pathways [194] itself contains a segmental duplication that includes part of the gene and its promoter region [195,196]. Some BRCA1-related cancers appear to be caused by gene conversion events in individuals carrying one mutant BRCA1 allele [197]. Once BRCA1 function is compromised, gene conversion and rearrangement at other loci may become more frequent.

Gene conversion could be a cause of the disease phenotype, or it could alternatively be a side-effect of an underlying disease-causing genetic disorder with no direct bearing on the phenotype. The fact that disease associations are found for most of the stringent filter SNPs is strongly suggestive of a causative link in which the specific conversion events are the proximate causes of the phenotype.

I have used the output of the Chiamo algorithm without modification. Spread is inferred from a high number of no-call results at a locus. While this method of inferring spread appears to have been effective, more effective methods might be possible. Methods could measure suitably defined 'spread statistics' given allele intensity distributions for several populations.

The success of the analysis supports the hypothesis suggested by the sister chromatid exchange studies that DNA in lymphocytes undergoes similar transformations to DNA in tissues affected by disease. In principle, it may be possible to test for various somatic mutations using a blood sample. Specialized microarrays could be developed to detect specific sequences resulting from common somatic mutations.

Several important questions remain. The present study does not allow the quantification of risk associated with any particular gene conversion locus. Even the identification of which individuals have substantial conversion at a locus is approximate. Locus-specific experimental studies of conversion frequencies in health and disease are needed.

The present study also does not quantify the degree of conversion necessary to cause disease. In lymphocytes, for example, mutations in a very small number of cells could cause disease if those cells undergo clonal expansion. In other tissues, many cells might need to be mutated before tissue function is compromised. Stem cell mutations (which may be relatively common due to frequent mitosis) could lead to a regular inflow of mutated cells.