Introduction
Traditionally, when considering what constitutes 'genetic disease' one has usually referred to inherited traits that segregate in a mendelian fashion and result from base pair changes that alter an encoded protein's structure, function or regulation. Nevertheless, the most common condition observed by clinical geneticists is sporadic in >97% of cases and is not due to any mutant genes, but instead results from genomic copy-number variation (CNV): Down syndrome resulting from trisomy 21. In fact, 2–3% of children are born with a major birth defect and most often these are sporadic in nature1, 2. Moreover, genomic rearrangements and gene CNV (that is, alterations from the normal copy number; usually two with one copy inherited from each parent) can be responsible for nervous system diseases that present as neurodevelopmental disorders at birth or during childhood, or present later in adult life as either neurodegenerative diseases or psychiatric illnesses3, and are often sporadic.
What are the molecular bases of sporadic disease? Sporadic disease could represent a chromosomal abnormality or a recessive trait, or be due to de novo dominant mutations. To what extent are new mutations (that is, de novo events) responsible for sporadic traits? Are locus-specific mutation rates (
) equivalent for different types of mutations, such as point mutations and genomic rearrangements? These are important questions to answer in order to determine what molecular approach to apply to the study of sporadic diseases. Do we need a $1,000 individual human genome sequence? A 1,000,000-probe SNP chip? Or just better arrays for high-resolution genome analysis and detection of copy-number changes to be used in conjunction with robust CNV and genotype-phenotype databases (for example, Database of Genomic Variants, http://projects.tcag.ca/variation/; and DECIPHER–database of chromosomal imbalance and phenotype in humans using Ensembl resources–http://www.sanger.ac.uk/PostGenomics/decipher/)?
Chromosomal abnormalities, including aneuploidies (chromosome number different from the normal karyotype of 46,XX or 46,XY), segmental aneusomies (chromosomal deletions and duplications), unbalanced translocations and small marker chromosomes (SMCs), result in genomic imbalances and can cause sporadic disease. Whereas nondisjunction causing trisomy 21 is a relatively frequent event, most aneuploidies do not survive to birth. From population-based registry data, chromosomal anomalies have been shown to be responsible for birth defects in 
0.2% of live births4. Recessive traits are thought to account for a similar proportion4. These values are estimates based on observed rates for a representative set of disorders of known genetic etiology.
De novo dominant point mutations can cause sporadic disease, and indeed this is a common mechanism for achondroplasia, neurofibromatosis type 1 and tuberous sclerosis. Direct estimates of human per-nucleotide rates for spontaneous mutations have been determined from the per-locus mutation rates and sequences of de novo nonsense nucleotide substitutions, deletions, insertions and complex events at eight loci causing autosomal dominant diseases and 12 loci causing X-linked diseases5. These direct estimates range from 0.5–3.7 
10-8 and show the combined rate of all mutations to be on average 1.8 10-8 per nucleotide per generation5. This average direct estimate agrees remarkably well with the indirect estimates of 2.5 10-8 determined by comparing pseudogenes in humans and chimpanzees6. For the indirect methods to determine point mutation rates, 18 processed pseudogenes were sequenced from both species, including 12 on autosomes and six on the X chromosome. In the context of the 3 109 bp haploid human genome the 2 10-8 (1.8–2.5) de novo simple base mutation frequency corresponds to about 60 new mutations per germ cell, with male germ cells having more than female germ cells as most point mutations presumably represent DNA replication or repair errors, or 120 new point mutations per diploid embryo. Approximately 2% (2.4) of them will affect exonic sequences, thus about two exons (genes) have a de novo base-pair change, of which at least one-third are likely to have little effect as they occur in the third position of a codon. As eloquently discussed in James Crow's review of the origins, patterns and implications of human spontaneous mutations for selected conditions due to point mutations, the paternal age affect is likely to be responsible for the observation that 'sporadic cases' are most often found among the last-born children of a sibship7, 8. Deleterious de novo dominant mutations are likely to undergo negative selection. However, some deleterious point mutations that are harmful to the organism may be advantageous in the cellular context of the testis and undergo positive selection9.
De novo structural changes of the genome can also cause sporadic disease, but precise frequency estimates are elusive. Anecdotal observations suggest that genomic rearrangements may be frequent mutational events. There are at least 35 genes or linked loci in which mutations can cause Charcot-Marie-Tooth (CMT) neuropathy. Nevertheless, about one-half of all patients with CMT10 and 70% of all patients with the demyelinating type (CMT1) have a 1.4-Mb duplication as the cause of their disease11, 12. Furthermore, 76–90% of sporadic CMT1 cases have a de novo duplication12, 13. Genomic rearrangements occur frequently enough that both inherited and de novo events can be observed in the same family. A pathologic de novo 3.7-Mb genomic duplication rearrangement of 17p11.2 has been observed in an individual with a more complex phenotype from a family segregating a different genomic rearrangement that was associated with a neuropathy (the hereditary neuropathy with pressure palsies, HNPP, deletion reciprocal to the CMT1A duplication) through several generations. The de novo event in combination with the inherited rearrangement was responsible for the more complex clinical phenotype14. It is only recently that high-resolution genome technologies have become available that enable the detection of genomic changes too small to be resolved by conventional cytogenetic techniques and too large to be observed by either DNA sequencing or standard agarose gel electrophoresis15 (Fig. 1). Locus-specific DNA sequencing technologies in general do not resolve genome structural changes. Duplication and deletion genomic rearrangements result in CNV, a topic of intense recent interest3, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25.
Figure 1: Genome analysis and locus-specific mutation rates (
).
![Figure 1 : Genome analysis and locus-specific mutation rates (|[mu]|).](/ng/journal/v39/n7s/thumbs/ng2084-F1.jpg)
Above is shown the 3 109 bp haploid human genome with the methods used to resolve changes in the genome of different sizes. Chromosomal banding (green) examines the whole genome at once, but cannot resolve changes <5 Mb (106–107 bp) in size. DNA sequencing (purple) can resolve single nucleotide changes and changes of several bases, but cannot identify CNV. Pulsed-field gel electrophoresis (PFGE) and FISH (yellow) extend the reach of conventional karyotyping and resolve changes from 104–106 bp in size. Array CGH can resolve changes causing genomic imbalance from 103–108 bp (including aneuploidies), simultaneously performing thousands of locus specific FISH as well as detecting imbalances seen by chromosome analysis. Note the two to four orders of magnitude greater locus-specific mutation rates () for genomic rearrangements than for single-nucleotide changes.
Diseases due to genome structural changes because of architectural features rendering a portion of the genome unstable have been referred to as genomic disorders26, 27. For autosomal dominant genomic disorders that are fully penetrant with fitness w, the per-locus per-generation rate () of loss-of-function mutation can be estimated by the birth prevalence B: = (1/2)(1 – w)B. The estimates of birth prevalence are 1/4,000, 1/10,000 and 1/25,000 for DiGeorge-Velo cardiofacial (DGVCFS) del22q11.2, Williams-Beuren (WBS) del7q11.23 and Smith-Magenis (SMS) del17p11.2 syndromes, respectively (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=search&DB=omim)28. Thus, the estimated per-locus mutation rates for genomic rearrangements, assuming fitness of zero, are between 2 10-5 and 1.25 10-4: about three to four orders of magnitude greater than for nucleotide substitution rates. The DGVCFS (3.0 Mb) and SMS (3.7 Mb) loci each represent 3 106 bp of the 3 109 bp genome. Extrapolating for the entire genome 1,000 times the size of the deletions, then in the diploid genome of a newborn there may be one segmental deletion in between 1 in 25 ((2 10-5) 2 1,000) and 1 in 4 ((1.25 10-4) 2 1,000) newborns. The extrapolation from these loci to the entire genome may overestimate , as the common recurrent deletions for WBS, SMS and DGS occur because of a mechanism involving the flanking low-copy-repeat genome architecture. Analogously to CpG dinucleotides that are hotspots for base substitutions, predominately transitions due to methyl-mediated deamination29, 30, perhaps flanking low-copy repeats can be considered the 'hotspots' for CNV.
One of the best-characterized genomic disorders is CMT1A. It results from a 1.4-Mb duplication including the dosage sensitive PMP22 gene31, 32. The CMT prevalence is 40/100,000 (refs. 33,34). It has been estimated that CMT1 constitutes 60–70% of CMT (the remaining 30–40% represent the axonal form or CMT2)34, and the frequency of duplication in CMT1 is 70% (refs. 11,12). Reports vary between 20% and 26.5% (refs. 35,36) representing de novo mutations within the previous few generations. Thus, the locus-specific mutation rate is between (0.5 (4 10-4) 0.6 0.7 0.2) and (0.5 (4 10-4) 0.7 0.7 0.265) = between 1.7 10-5 and 2.6 10-5: still at least three orders of magnitude greater than the single-base substitution frequencies. Extrapolating this 1.4 106 bp region to the 3 109 bp haploid genome requires a multiple of 2,200 and thus for a newborn (diploid genome), again under the assumption of an approximately equal mutation rate across the genome, between 7 and 11% of individuals will have a de novo duplication. Such genomic rearrangements as those causing CMT1A and SMS vary in size from 1.4 Mb (CMT1A duplication) to 3.7 Mb (SMS deletion) and may encompass several genes of which only a subset (often only one, as for example PMP22 and RAI1 in CMT1A and SMS, respectively) are dosage sensitive and confer a phenotype.
In one recent elegant calculation and extrapolation by Gert-Jan Van Ommen, the frequency of new CNV in humans was directly estimated to be 1 10-4 from molecular studies at the DMD locus37. As initially argued by J.B.S. Haldane in 1935 (ref. 38), for a disease with a stable prevalence, the rate of appearance of mutated alleles must equal removal. Thus, for X-linked lethal conditions, wherein one-third of all mutated alleles reside in males and are genetic lethals, if male and female gamete mutation rates are equal, then the new mutation frequency equals one-third of the mutated allele frequency38. The DMD mutation frequency is 1 10-4, as DMD occurs in 1/3,500 male newborns. Genomic deletions account for 65% (ref. 39) and duplications 9% (ref. 40). Thus, the de novo deletion frequency in this 2.1-Mb segment of the genome is estimated at 1:15,000 and the duplication frequency 1:100,000. Extrapolating to the entire diploid euchromatic genome, about 2,000 times the size of the DMD gene, this implies a de novo occurrence of one segmental deletion per eight newborns and one segmental duplication per 50 newborns37. However, this approach may underestimate de novo CNV because the DMD locus is not known to have a genome architecture prone to the genomic instability that occurs with regions associated with many genomic disorders (that is, flanking low-copy repeats of >97% sequence identity, in direct orientation, on the same chromosome and within 5 Mb of each other26). Such architecture enables predictions of potential regions of genomic instability. Using such predictive approaches through computational analyses of the human genome to elucidate regions of genomic instability, and incorporating such features into an array for screening cohorts of affected individuals, Eichler and colleagues have identified both CNV in normal individuals41 and previously unrecognized genomic disorders in affected cohorts42, 43.
There are two primary recombination mechanisms for generating genomic deletions and duplications that cause CNV that can be associated with genomic disorders: nonallelic homologous recombination (NAHR) and nonhomologous end joining (NHEJ)27, 44, 45. NAHR appears to be the more frequent mechanism when a specific genomic region has the architecture that favors genomic instability, and it has been proposed as a mechanism for CNV given the CNV proximity to low-copy repeats20, 41. In the SMS deletion and Potocki-Lupski syndrome (PLS)46 duplication region of 17p11.2, for which comprehensive breakpoint analyses have been performed on subjects harboring rearrangements, the ratio of affected individuals with clustered breakpoints (that is, NAHR) versus nonclustered breakpoints is 4:1 for deletions and 2.5:1 for duplications (Fig. 2). Thus, extrapolation from the DMD locus may underestimate at loci that predominantly utilize the NAHR mechanisms. However, extrapolating from DMD rearrangement may also overestimate because the DMD gene is unusual in that its size of >2 106 bp results in genomic rearrangements mostly occurring within the gene and deleting exons or duplicating them, causing frameshift mutations, as opposed to the phenotype being conferred by a 'gene dosage' effect or other rearrangement-mediated mechanism44.
Figure 2: Breakpoint mapping in 17p11.2: the SMS deletion and PLS duplication region of proximal 17p.
The horizontal structure in the middle of the figure depicts proximal chromosome 17p with low-copy repeats shown as rectangles (color-coded or like symbols for given repeats). Above, red horizontal lines represent deletions (Smith-Magenis rearrangements) with breakpoints depicted by arrow heads (if no arrow head, the given breakpoint is outside the region assayed). Below, green horizontal lines depict duplications (Potocki-Lupski rearrangements). CNVs are ascertained by virtue of the clinical phenotype conferred (SMS or PLS) because of the dosage-sensitive RAI1 gene (loss or gain of one copy, respectively). To the right are bar graphs depicting the frequency of recurrent (breakpoints clustered in low-copy repeats) and nonrecurrent rearrangements63, 64, 65, 66, 67, 68. Note that recurrent deletions are four times as common (occur 4:1) when compared with nonrecurrent deletions. Recurrent to nonrecurrent duplications occur 2.5:1. Nonrecurrent deletions are smaller whereas nonrecurrent duplications are larger. Deletions can occur by interchromosomal, intrachromosomal and intrachromatidal recombination27, whereas duplications do not occur by intrachromatid recombination. However, for at least some fraction of duplications a replication, not recombination, mechanism may be responsible for the rearrangements69.
Another method for determining locus-specific mutation rates is to directly measure them in sperm. Using single-sperm PCR, Jeffreys and colleagues have determined the frequency of deletion rearrangements at the 
-globin locus to be 4.2 10-5 (ref. 47). Of course, this determines the rate of occurrence in the male germline, but some of these rearrangement events may be selected against, and thus not be relevant to sporadic disease. Such caveats apply to all genomic rearrangements, in that mutation rates determined from disease prevalence estimates can potentially be underestimates because of selection. Deletions may be selected against more than duplication because of both loss of genetic material and the potential 'unmasking' of recessive mutant alleles on the remaining chromosome. In general, chromosomal deletions (those detected by conventional cytogenetic G-banding techniques) have more severe phenotypic consequences, are smaller in size and have been reported for less of the genome than chromosomal duplications (that is, greater haplolethal than triplolethal effects)48, 49 (Fig. 2). It is also important to consider the possibility that, like Apert-causing de novo point mutations9, some genomic rearrangements may undergo positive selection in germ cells. This may be relevant to male germ cells and mutational mechanisms involving DNA replication.
Sperm analyses have also been applied to the study of the de novo constitutional t(11;22)q23q11) balanced translocation. This structural genome rearrangement represents a recurrent non-Robertsonian translocation. The recurrent breakpoint occurs at an unusual genome architectural feature: a palindromic AT-rich region50, 51. The experimentally determined de novo rates for t(11;22) are between 1.24 10-5 and 9.46 10-5 in sperm from individuals aged 40 and 33 years, respectively52. From this small sample (4 men aged 31 to 40), as well as limited studies on other genomic disorders, there is no evidence for a paternal age effect on the de novo rate of genomic rearrangements, in contrast to the effect shown for point mutations7. Such observations are also consistent with a predominately meiotic (recombination) and not mitotic (replication?) mechanism for generating the rearrangement. Interestingly, as in other de novo structural changes53, genetic variation of the genomic architectural features (in this case the palindromic AT-rich region) affects de novo mutation rates54.
An alternative direct method for measuring the frequency of de novo genomic rearrangements would be to identify de novo genome structural changes using high-resolution genome analysis and comparing the findings in a child with that in the parents. New CNV in these trio analyses represents de novo structural changes. In a recent study of children with autism, it was determined that de novo CNVs were significantly associated with autism55. CNVs were identified using an oligonucleotide genome-wide tiling-path array comparative genomic hybridization (CGH) that could resolve changes >100,000 bp (three interrogating probes with an average 35 kb spacing)55. Interestingly, two out of 196 (1%) of controls were found to have a de novo CNV not observed in their parents. In that same study the average size of the rearrangements, or locus, identified in the autism cohort was 2 106 bp. Thus, in the 6 109 bp diploid genome there are 3 103 average such loci and the per-locus mutation rate () can be estimated as 3 103 = ½ (1 10-2) or = 1.7 10-6. This is likely to be an underestimate given that the arrays used can only detect genomic rearrangements >100 kb in size. Moreover, the average locus size used in the calculations was just approximated from the positive experimental findings. Further explorations of trios using higher resolution oligonucleotide arrays should enable robust determination of locus-specific mutation rates for genomic rearrangements.
In conclusion, whether calculated by prevalence rates of sporadic microdeletion syndromes (DGVCFS, WBS, SMS), common autosomal dominant (CMT1A) or X-linked (DMD) disease prevalence rates; or direct measurement by either sperm PCR at the -globin or t(11;22) recombination breakpoint sites, or genomic assays comparing child to parents in trios, the de novo locus-specific mutation rates for genomic rearrangements are between 10-6 and 10-4 (Table 1): at least 2–4 orders of magnitude (100 to 10,000 fold) greater than those for point mutations (Fig. 1). Perhaps some of these direct methods for determining could be supplemented by estimating the rate of fixation of CNVs by comparing the human and chimpanzee genome, thus providing an indirect measure for CNV, analogously to how this indirect method was applied to base substitutions for comparison with direct methods6. One caveat is the robustness and coverage of segmental duplications or low-copy repeats in nonhuman primate genome sequences, given the notorious difficulties of assembling such repeat sequences with whole-genome shotgun assemblies56.
) for genomic rearrangements![Table 1 - De novo locus specific mutation rates (|[mu]|) for genomic rearrangements](/common/images/table_thumb.gif)
Thus, for many apparent sporadic diseases, and perhaps for multifactorial and complex traits, one must consider the possibility of de novo CNV as a potential genetic mechanism. Furthermore, the possibility that sporadic disease might be found to result from a combination of two CNVs at a single locus (for example, analogously to the recessive diseases spinal muscular atrophy and nephronophthisis type I) or at different loci, one from each parent, in whom the uncombined structural variation did not provide a genetic burden that was great enough to cause disease25, should also be entertained. It is also important to note that, given what we have learned about the mechanisms for genomic rearrangement associated with genomic disorders, some members of the human population may be more prone to sporadic diseases than others, depending on the structural variants in their genome53, 54. Nevertheless, when attempting to unravel the molecular genetic basis of a given phenotype, perhaps whole-genome sequencing and SNP-based linkage and association studies should be supplemented with approaches that robustly measure CNVs25. High-resolution genome analysis may identify the cause of traits previously intractable to conventional genetic analysis, as was shown recently for autism spectrum disorders55, 57, 58.
Application of high-resolution genome analysis in the clinic has begun to uncover many new genomic disorders even using targeted arrays that assay a small portion of the genome59, 60, 61, 62. CNV discovery will also continue. Many of these CNVs are likely to represent benign variants, but proving they do not have phenotypic consequences or that they do not account for normal physical or behavioral traits can be a challenge. It is also a challenge to establish a cause-and-effect relationship for a specific genomic rearrangement and a given phenotype; and even more of a challenge to determine the dosage-sensitive gene or genes within the genomic rearrangement. Regardless, when analyzing sporadic traits, one should consider new mutation. Given that the frequency of de novo structural changes can be four orders of magnitude greater than that of base pair changes, perhaps CNV should be considered a potential significant cause of sporadic disease.