Plants live in ever changing environments and must adapt using strategies that fundamentally differ from those employed by animals. Developmental plasticity is at the core of those strategies allowing plants to modify their growth and the organs they produce in response to different environmental signals. This type of open-ended modular development enhances survival because damaged or diseased units can readily be discarded without compromising viability. Furthermore, because plants are constrained to sessile life styles, a modular growth habit affords greater versatility allowing phenotypic and genetic variation between modules to be used to the plant’s advantage, aiding adaption to pathogen life cycles ¹ or to longer-term environmental perturbations such as climate change. As a consequence of this profound developmental versatility, even individuals composed of cell populations derived from different plant species are viable and can coordinate the growth and development of chimeric organs ² . In an elegant paper published in 1981, Whitham and Slobodchikoff ³ proposed that mosaicism offers a unique adaptive advantage for plants by allowing introduction of genetic variants into the gene pool either through vegetative propagation or through sexual reproduction. They further propose that mutations arising somatically have a greater probability of being incorporated into the gene pool than mutations that arise in the gametes ³ precisely because germ line cells are derived from somatic tissues that arise late in the developmental history of the plant ^{4, 5} .

The relatively frequent occurrence of mosaics among various plant species has been extensively utilized in the development of novel ornamentals and for the selection and maintenance of desirable traits in many cultivated crops. Any desirable cultivars that have arisen in this manner have been maintained through vegetative propagation and, to date, are responsible for a significant fraction of agriculturally important perennial plants. On the other hand, desirable traits in many important annual crops, such as rice, soybean, maize and wheat, have been introduced through classical genetic manipulations using directed breeding strategies. Once generated, annuals with good agronomic performance are usually maintained by inbreeding.

In recent years, concern has grown over the presumed loss of genetic diversity resulting from the application of modern horticultural and breeding practices. Therefore, the benefit of excellent performance may come with a significant cost ^{6, 7} . However, recent and surprising results suggest that even highly inbred species harbor unanticipated sources of intrinsic genetic variation. For example, highly inbred soybean cultivars have been shown to manifest significant phenotypic and genetic variation in the absence of sexual manipulation ^{8– 10} . Such high intrinsic genetic variation has also been demonstrated for a number of other crop plants ¹¹ .

In the natural world, inbreeding occurs in many highly successful flowering plant species including wild relatives of Arabidopsis thaliana ¹² . Therefore, in nature species that are highly inbred have persisted despite their predicted reduction in genetic diversity. Why would such inbreeding strategies be successful and what are the implications from an adaptive perspective? One possibility put forward by Barrett ¹³ is that such populations are very successful in their particular niches and benefit from producing large numbers of genetically identical offspring. Nevertheless, selection should favor plant species that can co-evolve on time scales reflecting particular environmental challenges such as fluctuations and variations in pathogen populations. In keeping with this view, it has been shown that sequence variation in 20 diverse strains of Arabidopsis is highly non-random. In gene families mediating biotic interactions, such as those implicated in pathogen defense, variation far exceeds that seen in families involved in basic biological processes ¹⁴ .

The underlying mechanisms driving phenotypic variation in highly inbred lines, whether domesticated or wild, have often been inferred and have had limited experimental verification. Nevertheless, relatively simple molecular approaches have provided insight into some of the genomic events coinciding with visible changes in phenotype. In flax, for example, molecular assays have demonstrated that heritable phenotypic changes induced by environmental shifts are accompanied by reproducible changes in genomic DNA including changes in total DNA content, non-random changes in DNA sequences or sequence rearrangements ^{15– 18} . In soybean, reproducible non-random DNA sequence changes induced by in vitro culturing of root explants have also been demonstrated using restriction fragment length polymorphic markers ¹⁹ . Genomic changes manifesting similar hallmarks of biased sequence alterations have also been described for banana ²⁰ and in rice hybrids ²¹ .

In the work described by Roth et al. ¹⁹ soybean root explants were shown to repeatedly give rise to particular alleles that were absent in the donor plants but had previously been found and characterized in other varieties of cultivated soybean. To account for the appearance of these particular allelic variants the authors proposed that these organisms had evolved "internal generators of genetic variation" that mediated genome changes through some type of recombination process. In 2005, Lolle and colleagues ²² described a genome-wide phenomenon in Arabidopsis hothead ( hth) mutants that was very reminiscent of that described by Roth et al. ¹⁹ . Based on the nature and genome-wide locations of the sequence changes detected, it was proposed that a template-directed process mediated these changes and that these cryptic but stable extra-genomic templates themselves had persisted since at least the grandparental generation. Not surprisingly, this proposal met with considerable skepticism and numerous alternative explanations for these data have since been published ^{23– 28} .

In this study we have employed presence-absence molecular markers to test for non-Mendelian inheritance and found that Arabidopsis plants can inherit novel insertion sequences that were absent in their immediate parents. Furthermore, we show that discordant DNA-based marker profiles can be found between tissues isolated from different parts of an individual plant. These experiments demonstrate that individual plants spontaneously produce somatic sectors and are genetic mosaics. Since genetic variation can occur in the same plant in the absence of sexual reproduction, we propose that these novel insertion sequences must originate from cryptic reserves intrinsic to the host plant itself. The data presented support the original contention that a previously unknown template-directed mechanism exists ²² and raise the encouraging possibility that other inbreeding species, including crop plants, may also harbor a cryptic reserve of genetic variation.

By employing classical genetic approaches in conjunction with low and high-resolution molecular methods, we show that one Arabidopsis plant can have multiple genotypes. We have found instances of intra-organismal variation in different genetic backgrounds, in plants reared in different growth chambers, at different developmental stages and under sterile growth conditions. Furthermore, the incidence of sectoring and genetic discordance appears to be in some way conditioned by the hth mutant background as we found a consistently higher frequency of genetic discordance within single hth plants as compared to HTH wt plants. This was also true for shoot and root systems compared between aseptically grown seedlings and for tissue samples taken from adult plants and subjected to qPCR. Of critical importance, in showing that single Arabidopsis plants are genetic mosaics, experimental error due to cross-pollination and seed contamination can be completely discounted. To the best of our knowledge, this is the first report that documents the spontaneous but targeted appearance of unique genomic insertions at multiple discreet loci in single plants.

Only two other cases of spontaneous genomic insertions have been reported in plants that similarly could not be explained by any previously known mechanism. In both cases the insertion was non-random and targeted a specific locus. In the case of flax, the insertion sequence was 5.7 kilobase pairs in size ¹⁶ while in rice the insertion was comparatively small, being only 34 base pairs in size ³³ . Our data suggest that these reported cases of spontaneous genomic insertion events, like the sequence changes reported here, occur by a process intrinsic to the plant. As before, we propose the possibility that Arabidopsis plants harbor a cryptic store of sequence templates that can overwrite the parentally contributed genomes by a template-directed mechanism ²² .

If intrinsic drivers of genetic variation exist in inbreeding plant species, have additional incidents of cryptic genetic variation been documented in other systems? We believe that in soybean and cauliflower such events have indeed been reported and presented as cases of enigmatic phenotypic variation ^{8, 9, 34} . In other studies, molecular data have been featured. Again in flax, for example, molecular assays have demonstrated that heritable phenotypic changes induced by environmental shifts are accompanied by reproducible locus-specific copy number changes in genomic DNA ^{16– 18} . In soybean, reproducible non-random changes in restriction length polymorphic markers induced by in vitro ^{35, 36} culturing of root explants have also been documented ¹⁹ . Genomic changes manifesting similar hallmarks of biased sequence alterations have also been described in rice ^{21, 33} and corn ³⁷ hybrids, as well as in Arabidopsis ^{38– 40} .

In long-lived arborescent plants, intra-organism genetic variation has been demonstrated in a variety of systems ^{3, 36, 41} . The fitness benefits have also been validated using models that test whether the production of genetically divergent modules is an effective strategy for achieving adaptive co-evolution with organisms that feed on or infect the plant ^{35, 42, 43} . Models testing fitness benefits of module-level selection show that this is an effective strategy for achieving adaptive co-evolution between long-lived trees and short-lived herbivores when individual tree branches diverge genetically ³⁵ . Furthermore, this held true across a range of assumptions, even when reproduction was predominantly asexual. However, the fitness benefits were only fully realized for sufficiently long-lived trees that experienced strong selection ³⁵ . This fitness paradox is not exclusive to plants but also is relevant to organisms outside of the plant kingdom that have remained evolutionarily robust even though reproduction is predominantly asexual ⁴³ .

For a short-lived organism such as Arabidopsis, what adaptive value would within-organism genetic variation have? One possibility is that this heterogeneity offsets the predicted decline in genetic variation that should result from inbreeding. Plant development is open-ended and reiterative, allowing for the continuous output of repetitive units or modules that function to support the growth and reproduction of the individual. When combined with developmental plasticity and the absence of a sequestered germ line, modular development may actually drive plants toward becoming genetically heterogeneous ^{41, 43– 45} . As posited by Whitham and Slobodchikoff ³ , somatic sector formation permits the introduction of genetic variants into the gene pool either through vegetative propagation or through sexual reproduction. As these authors point out, germ line cells are derived from somatic tissues that arise late in the developmental history of the plant and therefore somatic mutations are more likely to introduce genetic variation than mutations that arise in the gametes ^{3, 4, 46} . By expanding the window of tolerance for genetic variation, plants may be afforded a better adaptive strategy given lifestyle constraints. The versatility of modular development combined with tolerance for genetic variation may allow plants to adapt at rates tailored to pathogen life cycles ¹ or to relatively expanded time scales, such as those affecting climate change. Even though self-fertilization is thought to have evolved approximately one million years ago ¹² , Arabidopsis plants have not suffered the consequential genetic erosion but have continued to thrive.

In addition to benefiting from a natural tendency toward genetic heterogeneity, the plant genome itself is thought to buffer the cost of having limited genetic diversity. In wild relatives of Arabidopsis the genome is thought to be highly dynamic and to respond to changes in environmental conditions or other extrinsic factors ^{42, 47} . Genome responses include elevated rates of homologous recombination that persist for multiple generations ⁴⁸ , changes in copy number ⁴⁹ and modulation of epigenetic gene regulation ⁵⁰ . Pervasive genetic buffering ^{46, 51} ensures that phenotypes with potentially deleterious consequences are attenuated. In addition to the genome responses listed above, our findings suggest that an intrinsic source of genetic variation can be leveraged to enhance the diversity in genetic output achieved by Arabidopsis plants.

In considering alternate template-dependent mechanisms, such as gene conversion or homologous recombination, none can account for the de novo appearance of unique sequence insertions. Nevertheless, it is possible that the insertion or deletion of small DNA sequence tracts, as described here, could reflect the activity of transposable elements ^{52, 53} . However, numerous lines of evidence argue against this possibility. For instance, when novel amplicons were detected, they co-migrated with their corresponding insertion or deletion allele and did not show size heterogeneity, as would have been expected for transposon-driven excision or insertion events. Sequence data confirm that deletion events reproducibly eliminate a fixed length of sequence while insertion events reproducibly introduce a fixed sequence tract and both events repeatedly target precise genomic sites. Insertion and deletion events do not appear to produce obvious junction sites with altered nucleotides. Similarly, insertion events introduce sequences that share identity with the Columbia reference genome and do not appear to be chimeric gene or genome fragments. Furthermore, transposable element-mediated events cannot account for the fact that these insertion sequences appear to be generated de novo since no comparable conserved region of homology exists elsewhere in the host genome, as demonstrated by our qPCR data. Lastly, as determined by DNA database searches, none of the indel markers used in this study share significant sequence homology with annotated Arabidopsis transposable elements.

If the genome of an intensely studied model organism such as Arabidopsis is subject to modification by the template-directed mechanism we propose, why has this phenomenon not been described previously? Our research shows that target choice and methodological approach are critical in differentiating these genomic events from other processes that also modify DNA sequences. Based on our findings, the only genomic targets that are truly diagnostic of this phenomenon are deletions. To the best of our knowledge, deletion alleles have been used in genetic studies precisely because they are known to be stable and not to revert but have not been used to study phenomena related to epigenetic inheritance. There is no generalized precedent for genetic instability of deletions and assuming otherwise would go against an established biological paradigm. Polymorphic molecular markers such as single nucleotides, simple sequence repeats, or insertions that are subject to alterations by other processes will not provide sufficient resolution to differentiate mechanism, even though they are also likely targets for this process. In particular, our findings may explain why genome sequencing efforts have failed to register these sequence deviations or, if detected, why they may have been attributed to sequencing error and eliminated during curation. One possibility that immediately emerges from this prediction is that raw sequence data contained in existing genome database archives may already contain evidence of extra-genomic sequence information, revealed by features such as highly biased loci-specific "errors".

Collectively, our genetic and molecular data show that many, and perhaps most, insertion events occur somatically in both seedlings and adult plants. Sectoring may therefore be a constitutive process that takes place throughout development but may be limited such that, at any given time, only a few cells host these genetic changes. Importantly, this may explain why sequence changes seen in revertant hth progeny have rarely been found to affect both alleles. Although sexual transmission of non-parental markers clearly does occur ²² , the fact that we have not found HTH/HTH progeny among seed-derived offspring suggests that sectors populating the gamete forming lineages are unstable or very rare. The qPCR data are consistent with this supposition. However, it is also possible that mechanistic differences exist between somatic and germ line tissues or that insertion events remain dynamic, limiting sexually transmitted changes to those that stabilize. It is also possible that certain genetic backgrounds condition this process as suggested by the greater number of events detected in hth mutants.

In addition to validating our genetic and molecular data, the qPCR results extend those findings and suggest that the genetic make up of individuals can be surprisingly complex. Our data show that each plant can produce multiple discreet sectors, at many different growing points and each with unique marker profiles. This finding implies that sectoring may be a relatively common occurrence, even in wt genetic backgrounds. Since the adult plants used for these experiments were left largely intact and only a small proportion of the plant sampled, many more sectors may have been present than quantified. As such, it is possible that our current census underestimates the frequency with which these smaller islands of genetic variation arise. Although sectors are more readily detected using qPCR, this method cannot distinguish, for example, between copy number variation within a small cluster of cells versus multiple cells that remain strictly diploid and are clonally related. Similarly, it is not possible to distinguish whether one sector hosts the full complement of genetic sequence changes, whether independent events occur in multiple discreet sectors, or if sectors overlap. Visualization of sectors in living tissue or tissue sections should help distinguish between these possibilities.

In addition to models demonstrating the fitness benefits of module-level selection ³⁵ , computational models provide surprisingly strong support for an ancestrally based "error-correcting" mechanism such as the one we propose to exist in Arabidopsis plants ⁵⁴ . In these constrained-optimization simulations, the evolutionary benefit of "genetic repair" strategies was compared between populations that access repair templates derived either from parents, grandparents or great-grandparents. Interestingly, a grandparent- or great grandparent-based genetic repair strategy is strongly favored over parental repair strategies. Furthermore, simulation results show that using a randomly selected template consistently gave superior results to those achieved using templates from the fittest parent or grandparent. From a biological perspective, such a strategy has considerable merit. Retaining a cache of templates derived from grandparental lineages would guarantee greater allele diversity precisely because the reservoir of allele variants would be deeper and allele redundancy would be less likely to occur. Random selection of templates would be the most parsimonious strategy to affect genome repair, again because it would promote diversity across alleles and between individuals. Since only those individuals that survived in previous generations would contribute to these cached templates, represented alleles would be biased to those that have proven robust under a spectrum of selective pressures.

In summary, the research presented here brings to light five striking findings. First, individual Arabidopsis plants are capable of producing somatic sectors during the course of normal vegetative development. Second, those sectors can have distinct and unique marker profiles and can differ in single nucleotide composition, can acquire small DNA insertions or can experience DNA sequence loss. Third, the de novo appearance of genomic insertions supports our original contention that cryptic sequence templates drive some of these changes ²² . Fourth, this phenomenon can be detected in wt genetic backgrounds raising the possibility that many Arabidopsis lab strains may be genetic mosaics. Finally, this process is genome-wide, impacting all 5 chromosomes, whether or not the target loci reside within genes or between genes.

Our data expand on the ideas put forth by Whitham and Slobodchikoff ³ and suggest that sector formation, even in a short-lived organism like Arabidopsis, may be a normal part of development and, furthermore, that the formation of sectors serves to capture novel genetic variation, irrespective of the source of that variation. Models testing the benefit of within organism genetic heterogeneity suggest that the average fitness of the population increases if some individuals within that population are genetic mosaics ³⁵ . As our data show, not all individuals in the populations we tested showed evidence of genetically distinct sectors but for those individuals that did, the number of sectors varied greatly. Our findings raise the possibility that inbreeding plants and, perhaps other organisms that predominantly propagate asexually, may sequester cryptic sources of genetic variation that can be harnessed to promote greater genetic diversity.