INTRODUCTION

The concept of genomic disorders, a group of human genetic diseases caused by genomic rearrangement, was solidified when a number of studies of such diseases began to identify common genomic features in the susceptible regions (121). These diseases represent a wide spectrum of unrelated clinical entities in which the common underlying molecular basis is rearrangement of specific genomic segments from a few kilobases (Kb) to several megabases (Mb) in length. These genome rearrangements cause loss or gain of dosage sensitive gene(s) specific to each disease. Genomic disorders apparently share a common mechanism for the recurrent genomic rearrangement; that is, nonallelic homologous recombination (NAHR), or ectopic homologous recombination, between low-copy repeats (LCRs) that flank the rearranged unique genomic segment (121).

DNA rearrangements associated with genomic disorders consist of deletions, duplications, and inversions of unique genomic segments at specific regions (42, 121, 199). Each locus for a genomic disorder has characteristic genome architecture that influences the recombination event; this architecture predicts the mechanism for NAHR associated with LCRs. NAHR between direct LCRs results in a duplication or deletion, whereas that between inverted LCRs causes inversion (121).

Genes within the recombined unique genomic segment are responsible for the specific clinical features of the diseases. When the LCRs involve a part of the gene structure, the gene may be interrupted or fused with another gene by the recombination event, resulting in a disruption of its function. This is generally the case when the recombined genomic fragment is small or the LCRs compose a functional gene and its pseudogene. When the recombined genomic region is large enough to encompass multiple genes, the recombinant allele may involve loss or gain of one or more dosage-sensitive genes located between the LCRs.

In contrast to nucleotide substitution and other point mutations in which alteration of the protein sequence is the major cause for the disease mechanism, large genomic rearrangements change the copy number of genes. In such cases, the mechanism for disease is not the change in the conformation of the encoded protein, but instead results from a gene (and protein) dosage effect.

Different aspects of genome architecture, evolution, and rearrangements associated with genomic disorders have been reviewed recently (39, 42, 121, 130, 183, 187, 198, 199). In this work we review observations from the molecular analyses of genomic disorders that (a) suggest common mechanistic features, (b) further delineate architectural features that result in susceptibility to genome rearrangements, and (c) describe gene-dosage effects that often mediate the phenotypic consequences of genomic rearrangements.

GENOMIC DISORDERS, EXPERIMENTAL OBSERVATIONS, AND UNDERLYING MOLECULAR MECHANISMS

Genomic Disorders are High-Frequency Events

Genomic rearrangements associated with genomic disorders are high-frequency recurrent events. Mutation rates for the NAHR in genomic disorders are mostly unknown. However, it is likely that they approximate the population frequency of the resulting phenotypes because of the high frequency of de novo mutation and reduced fitness observed in many genomic disorders, especially those with large genomic rearrangements (187). One of the most common genomic disorders, DiGeorge syndrome/velocardiofacial syndrome (DGS/VCFS) associated with a deletion on 22q11.2, has a prevalence at birth of 1.3–2.5/10,000 (20, 58, 109, 209). Williams-Beuren syndrome (WBS) has an incidence of 1/20,000–50,000 (60). Prader-Willi syndrome (PWS) and Angelman syndrome (AS) have incidences of 1/10,000–20,000. The frequency of Smith-Magenis syndrome (SMS) was calculated to be 1/25,000 (61). These estimates of ∼1 × 10⁻⁴ are at least one order of magnitude higher than that of de novo point mutations in individual human autosomal dominant traits.

Charcot-Marie-Tooth disease type 1A (CMT1A) and hereditary neuropathy with liability to pressure palsy (HNPP) are common autosomal dominant neuropathies resulting from reciprocal duplication and deletion, respectively. Many sporadic cases result from de novo rearrangement (73, 145). Even in families segregating autosomal dominant CMT, the new mutation event can often be traced to a previous generation (157). Based on population studies of CMT1A patients, the recombination frequency of de novo CMT1A duplication in gametes is estimated as ∼0.1–0.65 × 10⁻⁴ (13, 66, 145, 192, 236). This high frequency of new mutation has been confirmed by direct examination of recombination in sperm (65). These observations suggest that genomic recombination events resulting in genomic disorders represent highly recurrent events. The occurrence of genomic rearrangements is not random, but instead is likely due to susceptibility for NAHR.

Genomic Disorders Result from Genomic Architecture, Not Population-Specific Alleles

Genetic diseases are often more prevalent in some populations than in others. Sickle-cell anemia in individuals of African descent, cystic fibrosis in European and North American Caucasians, phenylketonuria in Celtic and Northern Europeans, and Tay-Sachs disease in Ashkanazi Jews are specific examples [see review in (147)]. Such population-based variation of disease and associated allele frequencies are often observed in single-gene disorders resulting from point mutations. These frequent occurrences of diseases in certain populations can be associated with founder alleles or, possibly, with heterozygous advantage.

In contrast, there are no known susceptible populations in recombination-based genomic disorders. The ∼1.4 Mb CMT1A duplication (78) is present in each population in which it was examined for [(123) and references therein]. In addition, molecular analyses in distinct ethnic groups revealed that, in the majority of NAHR, the strand exchanges occurred preferentially at one common recombination hotspot (65, 92, 115, 120, 171, 173, 212, 241). Similarly, no population bias was observed in other genomic disorders. Patients with DGS/VCFS present with a common ∼3 Mb deletion, WBS results from a common ∼1.6 Mb deletion, and patients with SMS have a common ∼4 Mb deletion. Each of these genomic abnormalities is recurrent in distinct world populations with no particular bias in ethnic background [(186) and references therein]. Whether genome polymorphisms may eventually be shown to account for some specific population-prevalence differences for select genomic disorders remains to be determined. However, to date, for each genomic disorder studied, there are no apparent founder-effect or population-specific alleles. To the contrary, these disorders result from specific architectural features of the human genome with susceptibility for recurrent genomic rearrangement at the disease loci.

Low-Copy, Region-Specific Repeats Provide Substrates for Homologous Recombination

Molecular studies of recombination breakpoints in each genomic disorder have revealed the presence of LCRs flanking the recurrently recombined genomic segment [reviewed in (42, 121, 187, 198, 199)]. This finding suggests that the rearrangement might be mediated by NAHR between highly homologous LCR substrates. Unequal crossover between two homologous syntenic but nonallelic LCRs results in loss, gain, or inversion of the unique genomic segment flanked by LCRs. The identification of unique junction fragments of identical size in multiple patients with the same disease suggests a precise and recurrent recombination event (28, 122, 151, 160, 161, 164).

The LCR units can be present in a direct (tandem) or inverted order. Each orientation yields distinct recombination products (Figure 1). Utilizing direct repeats as substrates, unequal crossover between two chromosomes (interchromosomal), within a chromosome (intrachromosomal), or within a chromatid (intrachromatid) results in deletion and/or duplication of the genomic segment that is flanked by the recombined LCR units [see review in (199)]. In this case, rearrangement does not disrupt the genes within the region unless they are located at the recombination breakpoints. Instead, rearrangement either decreases (deletion, monosomy) or increases (duplication, trisomy) the copy number of the genes (Figure 1A). Two inverted LCR units aligned as recombination substrates between interchromosomal and intrachromatid pairing result in an inversion of the genomic fragment (Figure 1B). Intrachromosomal pairing may result in structural changes of chromosomes (199). An inversion rearrangement does not alter the copy number, but it may disrupt a gene that spans the recombination breakpoints. A combination of direct and inverted segments within an LCR constitutes a complex repeat unit (Figure 1C). In such cases, recombinantion products may vary, depending on which portion of the unit is utilized in the strand exchange. In general, relatively small LCRs (<∼1–10 Kb, e.g., between gene and pseudogene) mediate NAHR that results in small genomic rearrangements, whereas larger LCRs (10 Kb to 100s Kb) often mediate NAHR that results in a structural change involving large genomic segments (>1 Mb) that contain multiple contiguous genes (121).

The human genome contains a number of duplicated genes and pseudogenes that share significant sequence identity. Large segmental genomic duplication events have also created significant regions of paralogy in the genome (distinct nonallelic genomic segments with highly similar primary DNA sequence that are derived from a duplication event) (104). These paralogous segments can serve as substrates for NAHR.

A well-characterized example of a genomic disorder resulting from paralogous genes acting as recombination substrates is α-thalassemia (70, 100). The α-thalassemia trait usually results from a homozygous deletion of an ∼4-Kb fragment that is flanked by two functional copies of highly homologous α-globin genes (including introns) repeated in direct orientation on chromosome 16p13.3 (α1 and α2). NAHR between these two copies of the α-globin genes results in a deletion of one functional copy. This represents one of the first and best-characterized examples of unequal crossover in humans.

Red-green color blindness is another example of gene-gene recombination. Red and green pigment genes (OPN1LW and OPN1MW), which share 98% identity in exons, introns, and 3′ flanking regions, are located on Xq28 in a tandem orientation. Individuals with normal color vision, or euchromates, have one copy of the red pigment gene, OPN1LW, and one or more copies of the green pigment gene, OPS1MW. The majority of red-green color blindness in males is caused by deletions or fusion genes created by NAHR between OPN1LW and OPN1MW (141, 228). Thus, from these examples, one recognizes that genomic disorders can be responsible for common traits [e.g., α-thalassemia affects 5%–40% of the population in Africa and 40%–80% in South Asia (232), and color blindness affects ∼4%–5% of males (139)].

NAHR between two directly oriented homologous (95% identity) genes, CYP11B1 and CYP11B2, results in a chimeric gene that contains the recombination junction. This chimeric gene represents a gain-of-function allele, which results in glucocorticoid-remediable aldosteronism (GRA), an autosomal dominant form of early-onset hypertension (106, 107).

In ∼13% of Hunter syndrome patients, an X-linked recessive disorder caused by defects of the IDS gene, a recurrent recombination between IDS and IDS pseudogene (IDSψ) was identified (15). IDSψ is located ∼20 Kb away from IDS and contains two homologous segments (96.4% identity) to exon 2—intron 3 and intron 7 in an inverted orientation. NAHR between IDS and IDSψ generates an inversion of flanking genomic fragment, resulting in a disruption of the functional IDS gene (15, 213).

More than 80% of patients with incontinentia pigmenti, an X-linked dominant disorder, have a consistent deletion of an ∼11-Kb genomic segment. This recurrent deletion occurs by NAHR between two direct 870-bp LCRs (int3h), with one copy in intron 3 of the disease gene, NEMO, and the other ∼4 Kb distal to NEMO. Thus, exon 4 and all downstream exons are eliminated by the deletion rearrangement (6, 7, 82A).

Twenty-five percent to 30% of patients with hemophilia A harbor a recurrent inversion in Xq28. This inversion, spanning ∼400 Kb, is mediated by inverted LCRs, termed int22h. One copy is located in intron 22 of the factor VIII gene (int22h-1), and two copies (int22h-2 and int22h-3) are located at ∼400 Kb telomeric in an inverted orientation. NAHR between these inverted LCRs results in an inversion of the 400-Kb genomic fragment, which disrupts the factor VIII gene (99, 102, 143, 144).

Large (>10 Kb to 100s Kb) duplicated genomic segments are present in the human genome. These LCRs can act as substrates for homologous recombination (199). The sizes of recombined genomic fragments are generally large (Mb) and thus may contain multiple genes. In some cases, the resulting rearrangement is large enough to be detected by routine karyotype examination, whereas at other times it is submicroscopic and requires molecular technologies to identify.

CMT1A and HNPP are caused by duplication and deletion, respectively, of an ∼1.4-Mb genomic fragment on chromosome 17p12, which is flanked by direct LCRs, termed CMT1A-REPs. Meiotic NAHR between proximal and distal CMT1A-REPs generates reciprocal recombination-specific duplication and deletion (24, 25, 122, 160, 173).

A recurrent deletion of ∼1.6 Mb in length that results in WBS is flanked by two complex region-specific LCRs spanning ∼320 Kb. Most segments within these LCRs are located in inverted orientation, but a few segments are in tandem orientation (35, 151, 161, 221). The common deletion appears to be mediated between the tandem segments within the LCR (151, 161, 221).

Deletion of NF1 on 17q11.2 accounts for 5%–22% of patients with neurofibromatosis type 1 (30, 168, 217, 222). The deletion, which spans ∼1.5 Mb, is flanked by two copies of LCR, NF1-REPs. These flanking NF1-REPs are ∼85 Kb in length and in direct orientation (36, 84); an additional copy is located on 17q24 (36).

DGS/VCFS is associated with an often submicroscopic deletion in 22q11.2. Multiple copies of a complex region-specific repeat unit, LCR22, were identified on 22q11.2. Most patients with DGS/VCFS have a common deletion of ∼3 Mb that is flanked by LCR22A and LCR22D, whereas a small number of patients have an ∼1.5-Mb deletion that is flanked by LCR22A and LCR22B. The largest LCR (LCR22D) spans ∼400 Kb (37, 38, 188, 189).

SMS is associated with a recurrent ∼4-Mb deletion located in 17p11.2. The deletion is flanked by two direct LCRs, proximal and distal SMS-REPs. A third copy, middle SMS-REP, is located between the proximal and distal copies in an inverted orientation. Each SMS-REP spans 176–256 Kb (28, 156). Similar to LCR22, some portions of the SMS-REP LCR are located at several loci on chromosome 17 (17p13.1, 17p12, 17q11.2, 17q12, 17q21.2, and 17q23.2) (156). Rare SMS patients with a smaller deletion appear to represent recombination between distal and middle SMS-REPs. A reciprocal recombination event, resulting in a duplication of the same ∼4-Mb genomic fragment, dup(17)(p11.2.p11.2), is associated with a relatively mild mental retardation phenotype (164).

The PWS/AS region on 15q11.2q13 is also prone to recurrent deletions of ∼4 or ∼4.5 Mb in length that are flanked by the region-specific LCRs BP1, BP2, and BP3. NAHR between BP1/BP3 (4.5 Mb) and BP2/BP3A (4 Mb) accounts for ∼95% of the deletions in PWS/AS. These LCRs also mediate the reciprocal recombination event, resulting in dup(15)(q11.2q13) (3, 23, 29, 179).

Deletion in the azospermia factor c (AZFc) region on Yq11.2 is encompassed by a complex of three palindromic LCRs spanning 4.5 Mb (96, 227). The palindromes consist of six subfamilies of segments in both tandem and inverted orientations. Recurrent deletion of an ∼3.5-Mb fragment is observed in the majority of AZFc-deletion patients. This deletion is flanked by tandem repeat units 229 Kb in length (96).

Evolution of the Mammalian Genome May Result in Architecture that Predisposes it to DNA Rearrangements Causing Genomic Disorders

As it becomes clear that the unique genome architecture resulting from LCRs is associated with genomic disorders, questions arise as to how and when these disease-associated LCRs originated. Initial analyses of Southern blots demonstrated that CMT1A-REPs (160) and SMS-REPs (28) were not present in lower mammalian species (e.g., mouse and hamster), suggesting that human LCRs arose and have evolved recently (198). Molecular studies by FISH, Southern blotting (zoo blot), polymerase chain reaction (PCR), and sequence analyses using primate cell lines and DNA have gleaned information regarding the evolution of LCRs through segmental duplication during primate speciation.

The distal CMT1A-REP in the CMT1A/HNPP region appears to be the ancestral copy because the COX10 gene spans the distal CMT1A-REP, whereas the proximal copy contains only one COX10 exon (pseudoexonVI) (172). The proximal CMT1A-REP, transposed from distal CMT1A-REP, is found only in chimpanzee and human, but not in gorilla or other New World monkeys, suggesting that the duplication event occurred after speciation of chimpanzee from gorilla [∼3–6 million years ago (MYA)] (14, 89, 91).

In contrast, the evolution of SMS-REPs appears to have occurred much earlier. Primate FISH analyses identified three copies of SMS-REPs in all primates examined but not in the premonkey lemur cell line. Genome sequence comparison revealed various insertion/deletion and inversion events in the evolution of SMS-REPs. Based on these observations, Park et al. (156) proposed a three-step hypothesis for the evolution of SMS-REPs. First, an ancestral copy of the SMS-REP appeared ∼40–65 MYA. Second, distal SMS-REP was originated from the proximal SMS-REP. Finally, the middle SMS-REP was duplicated from proximal SMS-REP.

Similarly, evolution of LCR22s that are involved in the DGS/VCFS deletion and other chromosomal rearrangements has been elucidated. FISH analyses revealed multiple copies of LCR22s in human, pygmy chimpanzee, gorilla, and rhesus monkey (189). PCR and sequencing analyses identified duplicated LCR22 in owl monkey but failed to amplify sequences in galago. These observations suggest that segmental duplication events may predate the divergence of New World monkeys (∼40 MYA) (188).

Based on the sequence component analysis and zoo blotting of GBA (the gene responsible for Gaucher disease) and its pseudogene psGBA, investigators estimated the duplication event occurred between 25–40 MYA (234). This was further confirmed by phylogenetic sequence analysis of psGBA, which yielded an estimate for divergence of 27 MYA (127). In addition, recent studies of the entire chromosome 22 with FISH and computational analyses of finished sequence data identified several LCRs (up to 400 Kb in length and ∼99.8% identity) to be human-specific, pericentromeric transchromosomal duplications derived from chromosome 14 (9). An evolutionary study of tubulin loci in hominoid and Old World monkey genomes revealed recent (∼25 MYA) duplication events with strong positional bias within pericentromeric and subtelomeric regions in the genome (224). Furthermore, human genome sequencing analyses revealed that the human genome contains abundant LCRs (∼5% of the entire genome), many of which are relatively large (>10 Kb) with high sequence identity (>95%) (82). Taken together, these observations suggest that a large number of duplication events occurred relatively recently in the evolution of the primate genome.

Significant Regions of Homology Appear to be Required for Facilitating Rearrangement

Nucleotide sequence comparison studies of LCRs that act as substrates for NAHR resulting in genomic disorders at multiple disease loci have revealed remarkable sequence identity, usually >95% and often >98%. The inverted LCRs, int22h, in the hemophilia A region have sequence identity of ∼99.9% throughout the 9.5-Kb length (144). Two direct repeat units in the AZFc region that flank the 3.5-Mb common deletion revealed ∼99.9% identity spanning over 229 Kb. This repeat unit is embedded in an ∼4.5-Mb complex palindromic region with 99.97% identity (96). A pair of 2.2-Kb LCR units that flank GH1 revealed ∼99% identity (225, 226). CMT1A-REPs in the CMT1A-duplication/HNPP-deletion region share 98.7% homology over the 24-Kb paralogous region (172). The complex LCRs in the WBS region have ∼98% sequence identity spanning >320 Kb in length (161). The three SMS-REPs in the SMS region share ∼98% identity for ∼170 Kb (156). The four copies of LCR22 in the DGS/VCFS region also represent a complex repeat structure, sharing 97%–98% identity (189).

High sequence similarity between the LCRs was proven to be a requirement for homologous recombination in both prokaryotic and eukaryotic systems [(33) and references therein]. This may also apply to meiotic NAHR in human cells. The presence of LCRs with such high sequence identity in the human genome is likely to contribute to the pathological basis of NAHR and the high frequency of human genomic disorders.

Strand Exchange Occurs Preferentially in Regions of Perfect Identity (300–500 Basepair) within Repeated Sequence

Detailed molecular studies have narrowed down the region of recombination and defined the recombination breakpoints in a few genomic disorders. DNA sequence comparison between paralogous LCRs for these regions of recombination, and direct analysis of the strand exchange in junction fragments, showed that breakpoints are clustered within stretches that represent perfect sequence identity. These sequence-identity regions for strand exchanges have been proposed to represent the minimal efficient processing segments (MEPS) for human homologous recombination during meiosis (135, 171).

Vnencak-Jones et al. (225, 226) found perfect sequence identity within the growth hormone deficiency type 1A region in which 2.24-Kb direct LCRs mediate deletion of growth hormone 1 (GH1) gene. Ninety percent of recombination breakpoints were localized in a 536-basepair (bp) region with perfect sequence identity, which was the longest stretch within the LCR.

In the CMT1A/HNPP region, the recombination breakpoints were localized to a 1.7-Kb region within the ∼24-Kb CMT1A-REPs, which account for >78% of recombination events of more than 500 patients with CMT1A and HNPP (92, 115, 173, 212, 241). Further sequence analyses of recombination junction fragments narrowed the strand exchange to the 557-bp segment that contains >456 bp of perfect sequence identity (116, 117, 171). In addition, two other regions with clustered recombination breakpoints contain stretches (306–951 bp) of perfect sequence identity (171).

In the NF1 deletion on 17q11.2, most breakpoints for DNA strand exchanges occur in a 2-Kb segment within the ∼85-Kb NF1-REPs. There are two regions of perfect sequence identity for >600 bp within this 2-Kb segment. The recombination breakpoints are localized within these segments (120). The breakpoints of the IDS-IDSψ inversion that result in Hunter disease are clustered in a 1-Kb region of >98% sequence identity. Sequence analyses located the breakpoints in the regions with potential perfect sequence identity (98).

These observations suggest not only that substantial regions of homology are required to facilitate NAHR, but also that a certain length of perfect sequence match is necessary for efficient NAHR. Requirements for stretches of sequence identity or MEPS for homologous recombination have been shown in various organisms including bacteria (190, 231), yeast (85), and cultured mouse cells (114, 229). In bacteria, the length of identity for MEPS is between 23–27 bp and between 44–90 bp for the RecBC- and RecF-dependent pathways, respectively (190). In yeast, MEPS were estimated to be approximately 250 bp, regardless of the relative positions of the repeats in the yeast genome (85). In mouse L cells, the MEPS may be as short as 132 bp, but stretches of identity >232 bp provide more efficient substrates for homologous recombination (229). In each of these systems, the recombination rate rises linearly as the length of identity increases. These observations are consistent with the hypothesis that MEPS may be required for efficient homologous recombination in humans and play an important role in disease-causing genomic rearrangements (171). Based on the requirement for perfect sequence identity at crossovers in the few genomic disorders in which the recombinant junction has been sequenced, it appears that MEPS in human meiotic homologous recombination may be 300–500 bp in length (171). Whether these MEPS requirements differ between meiotic and mitotic recombination in human remains to be determined.

Double-Strand Breaks May be the Initiating Events for Recombination

The molecular mechanism of strand exchange in the NAHR in genomic disorders was unknown until sequence analyses of recombination junction fragments were available to compare the exchanged DNA fragments. Current thinking favors the double-strand break (DSB) repair model (205) or a synthesis-dependant strand annealing process (SDSA) [see (154) for review] as the mechanism for homologous recombination (Figures 2 and 3). In general, homologous recombination initiated by DSB may be accompanied by gene-conversion events, because it utilizes allelic homologues (or nonallelic paralogues in the case of NAHR) as the template to regenerate the sequence that was lost in the DSB and subsequent exonuclease excision.

Junction fragment sequences in patients with CMT1A and HNPP revealed a number of alternated proximal and distal CMT1A-REP sequences in the recombinant copy, an observation suggesting that gene-conversion events accompanied the strand exchange (117, 171). Similarly, sequence analysis of NF1 deletion junction fragments also revealed evidence for gene conversion adjacent to the recombination breakpoints (120). Sequence analysis of junction fragments of inversions between the IDS gene and its partial pseudogene (IDSψ) in Hunter syndrome patients also identified multiple alternating sequences of both the IDS and IDSψ copies adjacent to the recombination breakpoints (98). These observations of gene-conversion events associated with the crossover between tandem or inverted LCRs can be explained by the DSB repair mechanism (Figure 2). Such findings are consistent with NAHR in genomic disorders being initiated by DSB and followed by subsequent strand exchange.

An important role for homologous recombination initiated by DSB has been observed in both prokaryotic and eukaryotic cells. DSB repair appears important in molecular evolution of mammalian genes (10, 105), diversification of immunoglobin genes (11, 174), antigenic variation of the trypanosome (223), yeast-mating type switching (62, 63), faithful chromosomal segregation during meiosis in yeast (180), and DNA repair [see (51) for review]. The molecular basis of this DSB repair system seems highly conserved throughout evolution from bacteriophage T4 to humans, both at the level of sequence recombination mechanisms and the protein molecules involved [see (31) for review]. It is a fundamental system to maintain stability of DNA strands, but it is also likely to play an essential role in the molecular pathogenesis of genomic disorders. DSBs may initiate NAHR between paralogous LCRs, which in turn results in genomic rearrangements and genomic disorders. Thus, identifying DNA sequence features that may result in DSBs (e.g., complex secondary structures, VNTRs, transposable elements, etc.) is essential for a better understanding of how NAHR occurs.

Recombination Hotspot for Position Preference

Studies in two different LCR-mediated recombination breakpoints in CMT1A-REPs and NF1-REPs revealed the existence of hotspots for strand exchange (92, 115, 120, 171, 173, 212, 241) within ∼24-Kb and ∼85-Kb LCRs, respectively. In addition to the complete sequence identity at the recombination hotspots, some sequence motifs that could potentially promote DSB were identified in these LCRs, mariner transposase-like element in CMT1A-REPs and Chi-like sequence in NF1-REPs (120, 173). However, it is unclear whether the presence of hotspots indicates increased recombination rate at these specific locations. It is also unknown how frequently NAHR occurs in comparison to allelic homologous recombination at the same location. Han et al. (65) employed a sperm PCR assay to detect duplication-specific recombination junction fragments and measured the frequency of NAHR directly. They reported that it is almost identical to the average recombination rate for males in the human genome (65). A similar study was reported on the CYP21 region, in which recurrent deletion between CYP21 and its pseudogene results in congenital adrenal hyperplasia (216). The frequency of NAHR resulting in deletion was estimated to be approximately equal to the average recombination rate (216). In addition, no evidence for increased recombination ratio was found within the CMT1A-REPs by comparison of genetic and physical maps of the CMT1A duplication/HNPP deletion region. Therefore, at least in the CMT1A region, the hotspot should be defined as a hotspot for position preference, not for recombination frequency (78).

De Novo Rearrangements May Display a Parent-of-Origin Effect

Because the human genome architecture facilitates susceptibility to rearrangement, de novo mutations in genomic disorders reflect a relatively high frequency of events in the population. Investigations in families with de novo mutations enable studies of parental mutation differences, or parent-of-origin effects, in meiotic recombination events. In some genomic disorders, an apparent bias in the parental origin of the de novo recombination event has been observed. Maternal preference in parent of origin was observed in NF1 deletion (101, 119, 217), whereas paternal preference was identified in CMT1A duplication (but not in HNPP deletion) (13, 69, 116, 118, 153). There is no known parent-of-origin preference in either SMS deletion (61) or WBS deletion (162, 238). Deletion on 15q11.2q13 of paternal origin results in PWS, whereas that of maternal origin causes AS with similar population frequencies (186); thus no parent-of-origin effect for mutation is observed. The DGS/VCFS deletion on 22p11.2 revealed a maternal bias in one study (34), but other studies have shown no such preference (19, 47, 138).

Studies in de novo CMT1A duplication and HNPP deletion suggested possible differences in the parental origin and chromosome exchange among these reciprocal events. The majority of CMT1A duplications results from paternal interchromosomal misalignment, whereas studies of de novo HNPP deletions revealed apparently comparable numbers of maternal intrachromosomal rearrangement (118), although only limited HNPP patient samples were analyzed (n = 4). The molecular mechanisms for such bias in parental origin in genomic disorder is unknown. Recent genome sequence studies of the CMT1A/HNPP and SMS regions has raised two possibilities to explain this phenomenon.

The first hypothesis is based on sex differences in genetic recombination depending on the location in the chromosome. In general, males have higher recombination rates toward the telomeric ends, whereas recombination is suppressed toward the centromere. Females, on the other hand, have less variation in recombination rate along a chromosome (18). This is reflected in the sex difference for genetic distances from telomeric end in the human genetic map. In addition, individual genomic loci have specific characteristics for recombination rates. In the CMT1A/HNPP region, recombination frequency in females is more than eight times higher than that in males, whereas in the SMS region, no such difference was observed (12, 78). This sex difference in recombination rates on human chromosomes may contribute to the difference in the parental origin of genomic disorders, depending on where the disease loci are positioned.

A second possibility is that gene content in the rearranged genomic intervals can reflect the fertility of haploid sperm or egg after the meiosis in which they are deleted, which may possibly result in the sex differences in parental origin. TEKT3, which encodes a structural protein of sperm tail, maps within the CMT1A/HNPP region. Haploid sperm with the HNPP deletion result in the loss of TEKT3 and may, in turn, result in a reduced motility of sperms. This may not be the case for CMT1A duplication. Reduction in sperm motility may potentially explain the reduced paternal origin in HNPP when compared to its reciprocal genomic rearrangement, CMT1A duplication (78).

Reduced Recombination Frequency May Promote Misalignment Between LCRs

Comparison of genetic and sequence-based physical maps from the SMS and CMT1A/HNPP regions on 17p11.2-p12 revealed substantial differences between genetic and physical map lengths (Figure 4A). In the CMT1A/HNPP region, there is a large stretch of male-specific reduced recombination. The SMS-deletion genomic region, in contrast, reveals low recombination frequency in both males and females (12, 78). This reduced recombination frequency may result in an extended region of two allelic chromosomes without crossover or synapsis in meiosis. Such an absence of synapsis formation could potentially allow the chromosomes to slip on each other, if synapsis indeed precedes recombination, thus enabling unequal crossovers between the tandem paralogous LCRs. Alternatively, frequent interchromosomal equal crossovers may provide anchors to prevent chromosomal slipping and reduce the chance of unequal crossovers between paralogous LCRs. This may explain the fact that the frequency of de novo CMT1A duplication is 10 times greater in males than in females (118, 153).

A polymorphic inversion that includes the entire ∼1.6-Mb WBS deletion region on chromosome 7q11.23 was found in unaffected parents of WBS patients with the common deletion (151) (Figure 4B). Breakpoints of the inversion are probably located outside the tandem segments within the LCRs; thus the organization of the direct repeat units may not be affected. Instead, the inversion may inhibit the formation of chiasmata within the inverted region in meiosis, thus effectively reducing the recombination rate within the inversion and also the region near the breakpoints, as experimentally shown in fruit fly (142, 178) and mouse (244). This reduced recombination may increase the chance of misalignment between nonallelic REPs (discussed above); thus individuals carrying the inversion, as opposed to those who do not, may be of higher susceptibility to de novo deletion in meiosis.

ARCHITECTURAL FEATURES OF THE HUMAN GENOME

LCRs in the Human Genome

LCRs identified originally because of their association with genomic disorders account for only a portion of genome duplication in the human genome. Analysis of the draft human genome sequence suggests that at least 5% of the human genome contains segments that are duplicated (8, 82). Genome-wide FISH analysis, using a collection of large insert BAC clones from the tiling paths for genome sequencing showed that 5.4% of clones hybridize to more than one location (210). The assembly of draft sequence has been hampered by homologous duplicated genomic sequences (40, 82). It is expected that the proportion of LCRs identified in the human genome will increase with the completion of the Human Genome Project.

Analysis of the human genome sequence draft also revealed a remarkably complex organization for both ancient and recent genomic duplications. Three groups of duplication events can be identified based on the mechanism of origin: polyploidization (ancient entire genome duplication events), endoduplication (tandem duplication of local chromosomal regions through unequal crossover events), and segmental duplication (transposition of a genomic fragment from one location to another) (39). By comparison with other eukaryotic organisms such as yeast, fruit fly, and nematode, the human genome is significantly enriched for segmental duplication (82). These duplications are found throughout the genome as large blocks of genomic segments that share a high degree of sequence identity (>95%) and, thus, probably occurred recently during primate evolution. The size of segmental duplications, which often contains genes and consists of smaller modules, ranges from a few kilobases to a few megabases in length. LCRs generated by segmental duplication have been classified into two categories based on the distribution pattern in the genome (39): transchromosomal duplications (between chromosomes) and chromosome-specific duplications (within chromosomes).

Transchromosomal duplication defines a duplication of genomic material among different chromosomes. There appears to be a bias for location of transchromosomal duplication in pericentromeric and subtelomeric heterochromatic regions (8, 82). Analyses of the draft genome sequences from 10q11, 21q11, 2p11, and 16p11 reveal a dense appearance of transchromosomal duplications (82). Based on high sequence identity, most of these duplication events are likely recent (<12 MYA), but evidence of a broad range of sequence identity (90%–99.8%) as well as polymorphic structural variability (111, 176, 214) also suggest a possible contiguous occurrence of duplication events in primate evolution (39, 82). To date, no association between transchromosomal duplications serving as substrates for NAHR and a genomic disorder has been identified. However, recent observations suggest that the pericentromeric region may be associated with genomic instability resulting in chromosomal rearrangements, including chromosomal rearrangement in tumor cell lines (152) and pericentric inversion in chromosome 9 [inv(9)(p12q13)] and chromosome 2 [inv(2)(p11q13)] (75).

Chromosome-specific duplication is defined as a transpositional event within a chromosome, but some portions of LCRs are often duplicated to different chromosomes. Thus, the generation of such duplicated segments may require multiple steps (74, 75, 156). This class of duplications includes many of the LCRs that are associated with genomic disorders. Chromosome-specific LCRs appear to be enriched in centromeric regions. In chromosome 22, >90% of LCRs are localized within 10 Mb of the centromere (8). Many large LCRs associated with genomic disorders are localized within this interval (WBS at 7q11.23, PWS/AS at 15q11.2q13, SMS/dup(17)(p11.2p11.2) at 17p11.2, and DGS/VCFS at 22q11.2), whereas relatively small LCRs involved in Mendelian-trait genomic disorders do not appear to have this distribution (199).

Divergence and Discrepancy in Time of Origin of LCRs

Many LCRs share significant sequence identity with a peak at ∼97% (82). This high degree of sequence identity is equally conserved in exons, introns, and nongenic regions, suggesting that the duplication events occurred during recent evolution and probably during primate speciation. These observations are in sharp contrast to the genomic regions duplicated by ancient polyploidization, in which functional domains of genes preserve high sequence identity by selection pressure but other nonfunctional regions diverge (104).

The molecular-clock hypothesis posits a consistency between divergence of nucleotide sequences of two species at a neutral DNA segment and the timing of species division [(27, 95, 104) and references therein]. Based on the estimated calculation of the neutral rate of substitution with 1.3–1.5 × 10⁻⁹ substitutions per site per year, 2% sequence divergence is thought to correspond to 5 MYA (division of human and chimpanzee), 4% to 12 MYA (division of human and orangutan), 6% to 25 MYA (human and Old World monkeys), and 10% to 35 MYA (human and New World monkeys). This hypothesis can be also applied to duplicated DNA segments to calculate the timing of the ancient segmental duplication event (183).

LCRs that are involved in genomic disorders retain very high sequence identity, usually ∼98%. Based on the molecular-clock hypothesis, the genomic duplication events that generated these LCRs have been estimated to represent recent events in primate evolution. Evolutionary studies of each genomic disorder–associated LCR locus using zoo blotting, FISH and PCR/sequencing analysis, however, has sometimes revealed discrepancies between the time period of duplication estimated by the sequence variation–based molecular-clock hypothesis and that determined by molecular analysis on primate cell lines (39, 198). There can be a greater sequence conservation among LCRs than that expected from evolution divergence (Table 1).

TABLE 1

Discrepancies between estimation by molecular clock and that observed by molecular evolution studies for dating divergence

At least four possible explanations can be considered. First, the data could be biased by analyzing only disease-associated LCRs, which could select LCRs with greater sequence identity. This possibility is unlikely because the average sequence homology of the entire chromosome-specific LCRs is >97% (8). It is also unlikely that only the LCRs that are evolutionarily old, but remain highly homologous, are involved in the disease process. Second, there may be mechanisms that selectively preserve the nucleotide sequence identity between LCRs against the neutral nucleotide substitution rate. Usually sequences containing functional domains of genes maintain their identity during evolution; however, it is difficult to explain the sequence conservation of an entire LCR, which often consists of >100 Kb, because exons account only for a small portion. It is also unlikely that each copy of an LCR maintains a primary sequence feature that functions to conserve sequence identity because each LCR has a unique sequence and there is no sequence or structural similarity among LCRs. Third, the speed of the molecular clock may be different across species and between paralogous sequences in the human genome. This possibility requires further studies for evaluation. Fourth, gene conversion may be responsible for maintaining the sequence conservation within the LCRs (homogenization) (39, 198). Gene conversion is defined as a transfer of genetic information from one place to another without changing donor sequence. Therefore, gene conversion between paralogous LCRs can result in homogenization of sequence divergence.

Molecular Mechanisms for Gene Conversion that Potentially Homogenize LCRs

Technically it is difficult to detect gene-conversion events in meiosis because of the absence of coexisting genomic rearrangments. However, gene conversion becomes evident when it results in a disease phenotype. A number of examples of gene conversion that cause human diseases have been found. Gene conversion between paralogous LCRs often results in transfer of coding alterations from a pseudogene to a functional gene, causing a disruption of the protein function. The majority of mutations in CYP21 that result in congenital adrenal hyperplasia are converted from its highly homologous pseudogene CYP21P, which is located in a tandem orientation 30 Kb away. (197, 233). Spinal muscular atrophy (SMA) type II and III are caused by gene conversion replacing SMN1 with its low-activity pseudogene, SMN2, either in hemizygous (one deletion allele and one gene-conversion allele, thus three copies of SMN2) or homozygous (two gene-conversion alleles, thus four SMN2 copies) fashion (22, 235). Gaucher disease is caused by recessive mutations in GBA. A number of mutations including the most common allele L444P are found in GBA pseudogene, which maps 16 Kb downstream of GBA in tandem orientation (93, 202).

Gene conversion may also play a role in the homogenization of paralogous LCRs, although the molecular evidence to support this mechanism is yet to be unveiled. Potentially, gene conversion may take place both in meiotic and mitotic recombination. In an analysis of homologous recombination between CYP21 and CYP21P, however, gene conversion was mainly found in mitosis rather than in meiosis (216). This observation suggests that homologous recombination resulting in gene conversions may occur by a mechanism distinct from unequal crossover between paralogous LCR (216).

Some evidence suggests that the DSB repair model employing Holliday junctions is not applicable to the mitotic gene-conversion model (64). First, a small proportion of mitotic gene conversions is associated with crossover; disease-associated junction fragments in CMT1A duplication/HNPP deletion and CYP21 deletion were never identified in somatic cells of normal individuals (65, 171, 216), suggesting either that Holliday junctions are preferentially resolved without exchanges or that such structures do not form or are dismantled (64). Second, heteroduplex DNA derived from mismatches between the DSB-containing recipient locus and its donor template is rarely found in the donor but frequently found in the recipient. This is not expected in the reciprocal exchange model if the single-strand DNA invaded into the donor sequence remains uncorrected (64). Such examples have been found in yeast mating-type switching (132, 203), expansions and contraction of tandem repeats in yeast (155), P-element-induced gap repair in Drosophilia (56, 140), and trangenic experiments in rodent cells (175). One of the mechanisms consequently proposed was SDSA [(45, 140, 155); also see (154) for review].

In the SDSA model, the two semiconservatively replicated strands are unwound from their donor templates and annealed (Figure 3). This places all the newly synthesized DNA in the repaired recipient locus, eliminating the Holliday junctions and, thus, precluding crossover (64, 132).

Because of the difficulty in studying mitotic gene-conversion events in mammalian cells, the exact molecular mechanism underlying this recombination is largely unknown. The SDSA model may, however, explain the gene conversion that exchanges sequence information between paralogous LCRs in mitosis, resulting in conservation of sequence identity between LCRs. Although this model requires experimental confirmation, it has been shown that, in mammalian transgenic experiments, gene conversions are observed between sister chromatids with no reciprocal exchange (87, 175). Moreover, the role of gene conversion in conservation of sequence homology between paralogous LCRs was statistically examined in CMT1A-REPs (76). Comparison of sequence divergence between proximal and distal CMT1A-REP, and that of reference nonrepetitive genomic sequence between human and chimpanzee, revealed evidence for gene-conversion processes in CMT1A-REPs but not in the control orthologous sequences (76).

Mechanism Underlying the Transposition of LCRs

LCRs are located close, but not adjacent, to each other; they are usually separated by a less than 10-Mb interval of unique sequence. This observation suggests that the majority of LCRs are not the product of unequal crossover mediated by homologous recombination. A likely mechanism for their formation is transposition (8).

To date, no common sequence structures, such as short direct repeats or unstable simple repeat sequences, which can explain the susceptibility of LCR to transposition and integration, have been found within LCRs or flanking unique sequences. Samonte & Eichler (183) hypothesized that the positional preference of LCR transposition in pericentric and subtelomeric regions may be associated with the mechanism of the tranposition events because of the following: (a) Regions near centromeres and telomeres might have a greater tolerance for the incorporation of new genetic material without adverse effects to the organism, (b) a greater length of time is required to delete duplicated segments near centromeres because of the suppression of recombination in these regions in the genome, and (c) the unique sequence characteristics of pericentromeric and subtelomeric regions might contribute to the bias (183).

Genome Instability

Studies of the CMT1A region in 17p12 and the SMS region in 17p11.2 have revealed a number of architectural features suggesting that genomic rearrangements accumulated in this region and indicating that genome instability may facilitate genomic rearrangement and integration of LCR during genome evolution (156, 198) (Figure 5A). The most ancient LCR in this region appears to be the one adjacent to proximal CMT1A-REP, spanning ∼400 Kb (156, 160). This LCR, termed LCR17pA, was partially duplicated in the SMS region in 17p11.2. Proximal and middle SMS-REPs are integrated into or adjacent to these duplicated copies of LCR17pA. The distal CMT1A-REP also integrated into the original copy of LCR17pA, forming proximal CMT1A-REP (78, 156, 172). LCR17pA also contains the breakpoint for the evolutionary translocation in gorilla t(4;19), which corresponds to human HSA t(5;17) (201). LCR17pA is likely to be involved in rare SMS patients with unusual deletion (156) and was also found to contain a breakpoint for a marker chromosome 17 in a patient (200). Furthermore, breakpoints for isochromosome 17q [idic(17)(p11.2)], the most frequent chromosomal abnormality observed in the disease progression of the Philadelphia chromosome-positive chronic myeloid leukemia, cluster in a region between proximal and middle SMS-REPs (46). This suggests somatic instability of this region.

Similarly, 22q11.2 and Xq28 contain clusters of genomic segments that comprise evolutionary, recurrent, and sporadic rearrangement events (Figures 5B,C). These observations suggest that frequent chromosomal rearrangements may be associated with accumulation of LCR and genomic instability of the involved genomic regions. Such higher-order genome architecture may involve local genomic instability and possibly play an important role in the susceptibility to genomic rearrangement.

Evolution of New Genes by Transposition of LCRs

Evidence for genome duplication is found in a variety of organisms, from yeast to human, and is proposed to play a major role in evolution (149). Massive amounts of genome duplications have been repeated throughout the evolution of eukaryotes. A whole-genome-based comparative study revealed an average duplication rate of 0.01 per gene per million years (125). This is remarkably high and almost equal to the mutation rate per nucleotide site (125). Therefore, duplication could be just as frequent and as important a source of variation as nucleotide substitution. Moreover, duplicated genes are subjected to a rapid and nearly constant attrition, with an estimated half-life of 3–7 million years. This suggests that more than 90% of duplicates disappear within 50 million years (125).

Large LCRs often contain genes. Sequence analyses of genes within the large copies of LCRs highlight genomic features of positive Darwinian selection, representing an enforced divergence of coding sequences of some genes to modulate function of encoded proteins. One of the first identified examples for such positive selection is the primate ribonuclease genes, eosinophil cationic protein (ECP) and eosinophil-derived neurotoxin (EDN) (182, 243). These two genes were generated by duplication ∼31 MYA. A comparative analysis revealed an acceleration of mutations at rates exceeding those of any other coding sequences in other primate genes, whereas the structural and catalytic components required for ribonuclease activity were both conserved (182). The rate of nonsynonymous nucleotide substitution was significantly higher than that of synonymous substitution for the ECP gene, suggesting positive Darwinian selection during the evolution of the ECP gene (243).

Similarly, multiple copies of a novel hominoid gene, morpheus, were found in a recently duplicated LCR on chromosome 16, termed LCR16a. Sequence comparison between the paralogous copies revealed more sequence divergence in coding exons than noncoding introns (86). These observations indicate the presence of dynamic selection within a genomic sequence to effect functional divergence of encoded genes during primate evolution (183).

Another mechanism for the evolution of new genes during the primate speciation is gene fusion and fission at LCR integration sites. When a duplicated fragment contains pieces of genic information, mostly exons and introns, the transposition of an LCR can transport the information from one place to another. The exonic pieces of the integrated LCR can be fused to an existing gene at the integration site (this may be accompanied by disruption or fission of the gene present at the integration site), and a new chimeric gene can be created (Figure 6).

Such an example can be found in the CMT1A region. Approximately 3–6 MYA, the proximal CMT1A-REP, which contains a pseudoexon of COX10, merged into the region encoding the AGIP gene. This integration separated AGIP into three portions; the first half of the gene became a new gene, HREP, after fusion of the first five exons of AGIP and the complementary strand of COX10 within the proximal CMT1A-REP. The second portion, which contained one or more exons, was deleted probably when the proximal CMT1A-REP integrated. The remaining last exon of AGIP became a single exon gene, CDRT1. Each of the newly generated genes, HREP and CDRT1, contains a distinct promoter, and thus the expression profiles of these genes are different (78).

Similar examples have been identified in POM-ZP3 fusion gene on 7q11.23 (90) and CECR7, a putative transcript in the cat-eye syndrome critical region, on 22q11.2 (48). Furthermore, in silico analysis of the finished genome sequence from 157 copies of 78 modules of LCRs on chromosome 22 revealed evidence for 11 transcripts that were apparently created or modified by the duplication (9). Considering the number of recent duplication events occurring during primate genome evolution, a great number of potential gene fusion and fission events may exist and are probably ongoing in the human genome. A completed human genome sequence, as well as those of other primate species, will offer the opportunity to explore further the number and distribution of gene fusion and fission events.

Structural Polymorphism in the Human Genome May be Mediated by LCRs

Polymorphisms are benign variations of genetic components in the genome. In addition to the classic small nucleotide polymorphisms including simple tandem repeats (STRs), variable number tandem repeat (VNTR), and single nucleotide polymorphisms (SNPs), large genomic structural polymorphisms have been identified in the human genome. Such structural polymorphisms can be mediated by NAHR between LCRs. These polymorphic variations may have no phenotypic consequences but may result in susceptibility to a disease-related phenotype. Such polymorphic variations may also predispose the genomic loci to disease-related rearrangements.

Olfactory receptor (OR) genes comprise a large number of clusters distributed throughout the genome by massive duplication events during evolution (57, 215). Some of the clusters in subtelomeric regions of 3q, 15q, and 19p are present with extreme polymorphic variations caused by transchromosomal duplications (214), suggesting ongoing genomic divergence. In addition, one of the OR clusters on 8p encompasses two ∼400-Kb LCRs. Inversion of 8p23 that is flanked by these OR-containing LCRs, REPD and REPP, is present in 26% of the European population. Furthermore, this polymorphism was also found in 9/9 mothers of inv dup(8p) or der(8)(p23.1pter) patients. Breakpoint analysis in these patients revealed identical breakpoints that mapped to the polymorphic inversion, suggesting that this OR–8p inversion polymorphism may confer susceptibility to the formation of secondary genomic rearrangements (55).

Inverted LCRs on Xq28 flank the FLN1 and EMD genes. These 11.3-Kb LCRs with >99% homology act as substrates for NAHR that results in the inversion of the intervening genomic segment. Heterozygosity for inversion is observed in 33% of females in the general population (194). Double-crossover events, one at the LCR and the other within the internal genomic segment, between two X chromosomes with different orientation, result in deletion of EMD, which is observed in Emery-Drefuss muscular dystrophy (194, 195).

Interstitial duplication of chromosome 15q24-26 [dup (15)(q24q26)] was found in ∼7% of population and has been suggested to represent a susceptibility factor for panic and phobic disorders (59). This polymorphic duplication, termed DUP25, is flanked by chromosome-specific repeat structures, LCRs 15, spanning ∼17 Mb (59, 166). Furthermore, amazingly dynamic genomic and genetic variations appear evident, including de novo occurrence, mosaicism, reversion, and conversion. The presence of mosaicism and the absence of three alleles for the markers within DUP25 by haplotype analysis in any subject carrying DUP25 suggested that mitotic unequal crossover, by either intra- or interchromosomal recombination, may be the mechanism for duplication (59).

These observations indicate that instability resulting from the genome architecture can also be associated with polymorphisms and large-scale rearrangement in the human genome. Given the abundance of LCRs in human genome, there may be many such structural polymorphisms yet to be identified.

GENE-DOSAGE EFFECTS AND GENOMIC DISORDERS

Dosage-Sensitive Genes Can Mediate the Phenotype in Genomic Disorders and Other Haploinsufficiency Diseases

Genomic disorders represent a wide variety of clinical phenotypes, depending on the location, size, and type of the genomic recombination (199). Deletion, duplication, or inversion by recurrent recombination may result in either disruption of gene structure or change in gene dosage. The latter occurs when the rearrangement involves large genomic fragments encompassing multiple genes structurally intact with change in their copy number. Phenotypic consequences may result from the excess number of genes with duplication rearrangements or from gene deficiency (haploinsufficiency) with deletion rearrangements (Table 2). It appears that only a small fraction of genes are sensitive to dosage effects (43, 121).

TABLE 2

Examples of gene dosage effects

Gene dosage as a mechanism leading to phenotypic consequences in genomic disorders was explored systematically in the CMT1A/HNPP region. At the time of the discovery of genomic duplication in the majority of CMT1A patients (122), three hypotheses were proposed for the mechanisms by which the duplication caused CMT1A: (a) gene interruption, (b) position effect, and (c) increased gene dosage. Observations on patients with partial trisomy of chromosome 17 who have CMT as a part of their complex clinical phenotypes and on CMT1A patients with duplications encompassing a region smaller than the common ∼1.4-Mb duplicated region excluded the first two possibilities (83, 124, 153, 219). Subsequently, deletion of the same ∼1.4-Mb interval was found in HNPP patients (24, 25). PMP22, a gene encoding a major component of compact peripheral myelin, was mapped within this ∼1.4-Mb interval (128, 158, 211, 220), and a series of experiments supported the contention that the dosage-sensitive PMP22 is responsible for both CMT1A and HNPP peripheral neuropathy phenotypes. These observations include the following findings: (a) smaller recombination fragments, either duplicated or deleted, all involve PMP22 (26, 83, 124, 153, 219); (b) point mutations in PMP22 are responsible for CMT1A/HNPP without genomic rearrangements (146, 177, 218), whereas no PMP22 coding alteration was found in CMT1A patients with duplication (230); (c) homozygous duplication causes a more severe phenotype (122); and (d) PMP22 transgenic and knockout mice recapitulate features corresponding to human CMT1A and HNPP, respectively (1, 77, 126, 129, 185).

The proteolipid protein gene (PLP1) is another example of a dosage-sensitive gene. Duplication of the genomic segment that contains the entire PLP1 gene is responsible for the majority of patients with Pelizaeus-Merzbacher disease (41, 79, 80, 191, 237), whereas point mutations and deletion also cause the disease in a minority of patients [see (52) for a review]. In Plp1 transgenic mice the disease severity corresponds to the level of overexpression (5, 81, 88, 169). Interestingly, PLP1 and PMP22 share similarities with regards to protein structure, expression pattern, and, possibly, functional properties in the myelin lineage, thereby suggesting a common disease mechanism for pathological gene-dosage effects [(4) and references therein].

It has been a challenge to identify the genes responsible for phenotypic features within large genomic regions (>1 Mb) deleted or duplicated in other genomic disorders. The dosage of all the genes physically located within the region is equally altered. Based on this feature, these genomic disorders comprising chromosomal microdeletions or duplications have been referred to as contiguous gene syndromes (184). Recent studies, however, have revealed that the disease-specific phenotypes may be associated with a small number of genes, or even a single gene, that are sensitive to dosage alteration.

The common ∼1.6-Mb deletion for WBS has been defined recently by physical map construction (72, 133, 161). Identification of elastin gene (ELN) disruption in patients with isolated supravalvular aortic stenosis (SVAS), overlap of clinical features between SVAS and WBS, and the mapping of ELN within the WBS deletion indicated that haploinsufficiency of ELN is responsible for the vascular pathology observed as part of the WBS complex phenotype (32, 44). Analyses of patients with rare smaller deletions, however, led to controversial conclusions regarding the association of hemizygosity for the LIM-kinase 1 gene (LIMK1) and cognitive deficit of WSB (50, 208). Requirement for hemizygosity of other genes within the WBS deletion to contribute to the phenotype, or the influence of genetic background, may explain some discrepancies. Fourteen additional genes have been identified within the WBS common deletion region, for which an association with disease-phenotype remains to be demonstrated [see (49) for review].

Rubinstein-Taybi syndrome (RTS) in 10% of cases is associated with a submicroscopic deletion in 16p13.3 (17, 68) and may represent a contiguous gene syndrome (184). Molecular studies, however, identified a single dosage-sensitive gene CBP, encoding CREB binding protein, that is responsible for the complex disease phenotype. Translocation breakpoints in rare patients with RTS were localized at the CBP region, and point mutations were identified in RTS patients with no chromosomal rearrangements (163). Heterozygous CBP-deficient mice showed a phenotype mimicking RTS (150, 207).

JAG1, which encodes a ligand for the Notch family receptors, is responsible for Alagille syndrome (AGS) (103, 148). A minor portion (6%) of patients with AGS revealed interstitial deletion of chromosome 20p11.23, which contains JAG1. However, the majority (69%) of patients had point mutations, suggesting that AGS is caused by haploinsufficiency of the JAG1 gene (94). Interestingly, Jag1 heterozygous knockout mice only presented a partial AGS phenotype, whereas Jag1^+/− Notch2^+/− mice exhibited developmental abnormalities characteristic of AGS (131, 240).

LIS1 encodes a subunit of the brain platelet-activating factor acetylhydrolase (PAFAH) (67) that plays an important role in the neuronal migration in development by regulating dynein motor function (196). Heterozygous deletion on the genomic region of chromosome 17p13.3 that contains LIS1 results in isolated lissencephaly and Miller-Dieker lissencephaly syndrome (MDLS), suggesting that LIS1 is a dosage-sensitive gene responsible for the lissencephaly phenotype (170). Null and hypomorphic Lis1 mice revealed dosage-dependant neuronal migration defects, indicating that haploinsufficiency is the mechanism for the disease phenotype (71).

Haploinsufficiency of EXT1 and TRPS1 are partly responsible for Langer-Giedion syndrome [del(8)(q24.1q24.1)] (2, 110, 137, 239). EXT2 is likely causative for exostoses in Potocki-Shaffer syndrome [del(11)(p11.2p12)] (108, 165). Hemizygosity for WT1 and PAX6 results in WAGR syndrome [del(11)(p12p14)] (21, 53, 54, 181). Haploinsufficiency of NSD1 on 5q35 is the major cause of Sotos syndrome (97). In several of these latter disease phenotypes that result from haploinsufficiency of a specific gene located in a deletion interval, breakpoint analysis studies remain to be performed to determine whether they represent genomic disorders.

Animal Models to Study Gene-Dosage Effects

The mouse represents an excellent model to study the phenotypic consequences of genes involved in genomic disorders for the following reasons. First, methods to engineer chromosomal rearrangement have been established recently. Use of the Cre-loxP recombination system enables construction of deletion, duplication, and inversion of specific segments of mouse chromosomes (136, 167, 204, 242, 244). Second, gene order and genomic organization, as well as the function of the protein products for each gene, are well conserved between human and mouse. Employing chromosomal engineering technology in mice can facilitate modeling of genomic disorders and identification of the gene(s) responsible for the disease phenotype(s) (136, 242).

Chromosome-engineered mouse models have provided a model for DGS/VCFS and the dissection of dosage-sensitive genes involved in the phenotype. A genomic region on mouse chromosome 16 that is syntenic to the DGS/VCFS region (16) was deleted for ∼1 Mb using the Cre-loxP system (112). The heterozygous mice displayed congenital cardiovascular defects similar to those observed in human patients. The abnormality was traced to an embryonic defect of development in the fourth pharyngeal arch arteries. This defect could be complemented by crossing to mice harboring duplication of the same chromosomal region, thus rescuing the phenotype (112). In addition, a series of mouse chromosomes with smaller deletions were generated to narrow the critical region responsible for the defect of the fourth pharyngeal arch arteries to 200 Kb, where the gene for a member of the T-box transcription factors, Tbx1, maps. Subsequent creation of heterozygous Tbx1 knockout mice recapitulated defects in development of the fourth pharyngeal arch arteries observed with the 1-Mb deletion. The fact that PAC/BAC transgenics harboring Tbx1 complemented the phenotype in Tbx1^+/− mice suggested that haploinsufficiency of Tbx1 gene is responsible for the cardiovascular phenotype of the model mice and probably of human patients with DGS/VCFS (113, 134). Furthermore, neurobehavioral studies of the mouse model revealed sensorimotor deficits and learning and memory impairments, both of which can be observed in patients with DGS/VCFS (159). The influences of genetic factors distinct from deletion could be investigated through mating with different inbred strains. These latter studies revealed that reduced penetrance for cardiovascular defects are a function of genetic background (113, 206).

CONCLUSION AND FUTURE DIRECTIONS

When some of the first genomic disorders began to be elucidated a decade ago, it was not anticipated that genomic disorders would represent such a common cause of human genetic diseases. The number of recognized genomic disorders is continuously increasing, and a wide variety of size and complexity of LCRs associated with genomic disorders have been identified (199). The basic rules for generating genomic rearrangements have begun to emerge. NAHR between nonallelic LCRs results in duplication and deletion when the LCRs are tandem, and it generates inversion when the LCRs are in an inverted orientation. Furthermore, high sequence identity and the sometimes complex structure of LCRs, the involvement of DSB repair, MEPS for strand exchange, recombination hotspots for positional preference, and reduced recombination frequency and potential susceptibility for NAHR have been proposed as molecular mechanisms for NAHR.

In addition, massive nucleotide sequence information generated by genome sequence projects of vertebrates and nonvertebrates, especially the draft human genome sequence, has provided further insight into genome evolution by duplication, human genome architecture, and the enrichment for LCRs. From this perspective, humans may be the best “model organism” in which to study the genome rearrangements and associated phenotypes.

The clinical phenotype manifested in genomic disorders is a consequence of genes that display dosage effects. Identification of dosage-sensitive genes in each genomic disorder will likely allow an understanding of the association between clinical features of genomic disorders and molecular defects, as well as provide a molecular window into pathways that may provide for treatment possibility.

Because of the abundance of LCRs in the human genome and recombination mechanisms for genome rearrangement, it is likely that recurrent genomic rearrangements are a more common feature of human genetics than previously anticipated. In the near future, the association between many other human diseases and genomic rearrangement will likely be identified.

ACKNOWLEDGMENTS

We thank Drs. Bassem A. Bejjani, Philip J. Hastings, and Pawel Stankiewicz for their critical review and helpful discussion. This work is supported in part by The National Institute of Child Health and Development (P01 HD39420), the Baylor College of Medicine Mental Retardation Research Center (HD 2406407), the Texas Children's Hospital General Clinical Research Center (M01 RR00188), National Institute for Neurological Disorders and Strokes (R01 NS27042), and the Muscular Dystrophy Association. K.I. is a recipient of a development grant from the Muscular Dystrophy Association.

The Annual Review of Genomics and Human Genetics is online at http://genom.annualreviews.org