Recent studies of P. falciparum demonstrated the widespread prevalence of CNVs in populations and their likely adaptive influence on important traits such as drug resistance [75, 76]. Large scale amplification and deletions have been known for several decades [39–42]. However, a precise understanding of genome plasticity, origins of CNVs and their stability, including transient and reversible fluctuations in a generational time-frame is deficient not only for the malaria parasite, but for other organisms as well . For example, little is known about the behavior of copy number variant regions, the rate of reversion to an original state, the rate at which new variants arise, and the uniformity of the distribution of new variants in a sibling population. The segregating population examined in this study provides an ideal context in which to view the inheritance and stability, and occasionally to infer the origin of a CNV. We report extensive plasticity and segregation complexity of CNVs within the progeny.
Three different classes of CNVs - segregating, singleton de novo and recurrent de novo - were prominent in this study and are contrasted here for their inheritance patterns among progeny clones (Table 1, Figure 2, Additional files 4 and 5). Among these three classes, we observed duplications, deletions and multiallelic complex loci, as has been described for CNVs in human [25, 49] and chicken  (Additional files 4, 5 and Figure 7). We observed many de novo CNVs (Table 1). Information on de novo CNVs has been scarce because previous studies did not examine parent-progeny populations. With the availability of suitable genetic systems along with high-throughput technologies which enable genome wide discovery of CNVs, it is clear that de novo CNVs are an important source of genetic variation [49, 53, 77]. Furthermore, de novo events are not unprecedented in P. falciparum. Duplication of subtelomeric sequence has been documented previously in progeny of different genetic crosses including the HB3 × HB3 self cross . Previous development of the MS linkage map revealed non-canonical MS markers in the HB3 × Dd2  and non-parental sequence products in the HB3 × 3D7  as well as the HB3 × Dd2  genetic crosses, further emphasizing the genome plasticity of the parasite both at smaller (< 1 kb) as well as larger (> 1 kb) scales of sequence.
Our data provide clear evidence for copy number differences from the parent lines within the segregating progeny population. Most of the previously known segregating CNVs exhibited a Mendelian segregation pattern at a broad scale and mapped to markers close to their genome positions (Additional file 9). However, finer scale scrutiny of two segregating CNVs implicated in drug resistance revealed unique structural changes resulting from meiotic recombination events. The Chr 5 Pfmdr1 amplification which has been associated with Mefloquine resistance [79, 80] and is widely detected in natural parasite populations , exhibited both loss and gain of copies compared to the parental state (Figure 5). This highlights that both amplification and deamplification mechanisms have affected the locus. Similarly, the gch1 locus, postulated to be associated with antifolate resistance  and widely detected in parasite populations , also exhibited complex multiallelic copy number within a single meiotic generation (Figure 7). These examples illustrate the highly dynamic nature of CNV regions during a single meiotic generation that would not be recognized in a standard population-based CNV survey.
Four mechanisms can generate CNVs and lead to fluctuation of copy number in the CNV regions: homologous recombination (HR), non-allelic homologous recombination (NAHR), non-homologous end joining (NHEJ) and the replication based mechanism, microhomology-mediated break-induced replication (MMBIR) - which includes Fork Stalling and Template Switching (FosTes) . The absence of factors in the malaria parasite genome required for NHEJ combined with evidence for HR and NAHR from both laboratory genetic crosses and field isolates argue that recombination mechanisms play a central role in generating genetic diversity in the parasite. Consistent with previous reports, we demonstrate that recombination generates amplifications and deamplifications of both segregating and de novo CNVs. We show evidence of recombination detected by local allelic changes that resulted in copy number loss (Figure 6) and gain (Figure 6 and 7) in segregating CNVs and gain of de novo CNV (Figure 8, Additional file 12). While Chr 5 CNVs in two progeny clearly indicate HR origins, lack of evidence for reciprocal allele exchange in other progeny implies that most CNVs may appear due to unequal HR between sister chromatids. Unequal sister chromatid exchange is postulated as a mechanism that generated the multiple independent events of the pfmdr1 CNVs within parasite isolates . MS allelic changes at the Chr 12 locus (gch1) in our data indicate copy number fluctuation by sister chromatid exchange, a double crossover or gene conversion. Gene conversion has been reported to generate diversity within multigene families in P. falciparum . Duplication of chromosomal segments by gene conversion, including duplicative translocation, has been described in genetic crosses  and parasite clones . Alternatively, complex multiallelic/mosaic regions can result from gene conversion which can change the CN profile from that of the parents , an observation consistent with the several examples of de novo CNVs described in this study (Figure 8 and Additional file 12).
In general, it is difficult to establish CNV origins. The steps involved in generating a genetic cross include many opportunities for both sexual and asexual (in meiosis and mitosis) genetic exchanges [42, 47, 78, 83]. A more precise inference of mechanisms would benefit from knowledge of the number of mitoses that each parent lineage underwent prior to the generation of gametes for the cross, as well as the number of mitotic replication cycles that the parent and progeny parasites underwent after meiosis. Although allelic marker co-inheritance can pinpoint homologous recombination as one origin of CNVs when sufficient sequence differences can distinguish the parental allele segments, this method cannot differentiate the CNVs generated in asexual replication or in genomic regions that are identical (or very similar) in the parents.
While unlikely, it cannot be ruled out that recurrent mutation reflects parent subclone populations (i.e. gamete mixtures). Although parasites were cloned by micromanipulation or limiting dilution, and it is generally accepted that this method would produce true single-clone parent lines, we are necessarily dealing with these 'individuals' as populations expanded in culture. Therefore, it is possible that genetic changes arising in these cultured lines in preparation of gametes for the cross could contain mixed genotypes that are represented in the gametes which segregate into some subset of progeny clones. We found some recurrent de novo CNVs residing in both parent allele backgrounds that suggested independent origin. Furthermore, we did not find evidence for simultaneous introgression of CNV, which should be readily apparent in the presence of two or more distinct parent subclones. Overlap of several single as well as recurrent de novo CNVs with CNVs reported to have arisen under culture adaptation and/or in vitro culture, suggests that several de novo CNV regions may have emerged in culture adaptation (Additional file 13) but cannot be precisely determined at CGH resolution.
We noted a preponderance of CNV breakpoints within narrow genomic regions, including recurrent de novo CNVs that impacted the same genome segments. Genomic regions that show a high propensity for segmental duplications also have been suggested in isolates  and laboratory lines  of P. falciparum. Additionally, previous work has also demonstrated extensive occurrence of deletions particularly in the subtelomeric sequences [44, 46, 48, 84, 85], indicating that the subtelomeric regions may be highly unstable and represent fragile sites [85, 86]. It has been postulated that specific sequence features may underlie the fragility of the subtelomeric regions . Recurrent structural mutation has been observed in mice  and humans  during inheritance. Similarly, recurrent duplication has been detected previously in P. falciparum; especially in association with the subtelomeric regions in progeny of both the HB3 × Dd2 and HB3 × 3D7 genetic crosses , while recurrent subtelomeric deletions have been detected in independent clones of a field isolate . Several recent studies have demonstrated recurrent mutations as a key mechanism by which gene copy number fluctuations take place within short generational time scales . These studies have emphasized that recurrent CNVs may be an important biological process in evolution, as well as human disease [7, 53].
Skewed inheritance was observed for a majority of the segregating CNVs. Skewed inheritance was expected to an extent, given that skewed inheritance of parental alleles were previously noted within this population for seven regions, mostly located in the sub-telomeres, during construction of the MS linkage map . Consistent with the expectation from MS linkage analysis, five of the CNV regions overlapped with the skewed allele distributed regions in the MS map, emphasizing the role of CNVs in parasite selection. The skewed regions overlap with genes associated with parasite pathogenicity , gametogenesis [44, 46] and drug resistance . Regions of skewed inheritance have been observed not only in the HB3 × Dd2 genetic cross [42, 65], but also in other independent genetic crosses [42, 88, 89]. It has been suggested that the skewed inheritance may be related to the selection of alleles beneficial for parasite viability, growth and proliferation in a splenectomized chimp during the generation of the genetic cross and/or in parasite growth under in vitro growth conditions . If deletions are eliminated by selection, populations that emerge in culture should carry more amplifications than deletions. This trend was observed in the progeny clones carrying more gains than losses (69%, Additional file 3).
The stability and fitness of CNV loci is postulated to play an important role due to their implication in resistance to antimalarials . Previous work supported a co-adaptive role of pfmdr1 copy number with the CQ resistance gene pfcrt. Inheritance of these loci in the progeny clones of the HB3 × Dd2 has suggested an influence on fitness due to the presence of specific combinations of alleles that exist among the progeny. It was observed that high pfmdr1 copy number is maintained only in the context of its co-selected mutant pfcrt partner and CQ sensitive pfcrt is never paired with 3 copies of pfmdr1 . Two groups indirectly evaluated the in vitro dynamics and possible fitness effects of CNV in P. falciparum [67, 90]. Both attempted to address the fitness effects at a single CNV locus, in the presence and absence of drug pressure, using a single strain of P. falciparum. Each proposed a fitness cost associated with carrying the multicopy CNV as indicated by the out-growing of the single copy over the multicopy parasite in a mixture of parasites. Mathematical modeling of in vitro based experimental data suggested a CNV emergence rate of 1 in 108 parasites . The rate of emergence in the population is ultimately a reflection of the rate of de-amplification as well as parasite growth dynamics due to fitness costs associated with carrying higher copy numbers.
Emergence of CNV under in vitro conditions have been reported widely in P. falciparum with laboratory adaptation [68, 91, 92], under long term laboratory culture [19–22] and under drug pressure [21, 23, 67, 90, 93]. It has been widely postulated that parasites have fewer constraints during in vitro culture conditions such that growth advantages can be gained from decreased investment in activities such as protein exportation, knob construction, display of cytoadhesive molecules and variant antigens, and production of gametocytes . The overlap we observed of de novo CNVs with some of these genes is consistent with the interpretation that culture adaptation and cloning could be associated with lost functions via deletions.
Along with extensive chromosomal size variation identified previously by PFGE [38, 42, 43], our data demonstrate a highly plastic genome with strong potential to influence function through gene dosage effects. We explored the potential functional impact of CNVs. Functional enrichment analysis of the de novo CNVs revealed genes involved in carbohydrate metabolism, recombination and gametogenesis; while segregating CNVs involved drug response, fat metabolism, aromatic compound biosynthetic processes and regulation of DNA replication in P. falciparum. In both segregating and de novo CNVs, functions of polymorphic gene families were represented. The presence of functional gene families has been taken as an indication of positive selection on gene duplications over time [25, 94]. Gene duplication is now recognized as an important mechanism for evolution of new biological functions in organisms . CNVs in humans are enriched for genes involved in molecular interactions to specific environmental stimuli including drug detoxification, immune response, cell surface integrity and surface antigens. It has also been postulated that CNVs could carry genes that contribute to inter-individual variation and can play a role in the differences in drug response and immune defense , but not in intracellular processes such as biosynthetic and metabolic pathways . The genome wide distribution of CNVs and the abundance and breadth of genes overlapping CNV regions, as well as their widespread involvement in local and distant gene regulation, indicate the extensive contribution of CNVs in phenotypic variation, similar to that observed in human studies .