Intraspecific Evolution of Human RCCX Copy Number Variation Traced by Haplotypes of the CYP21A2 Gene

Abstract

The RCCX region is a complex, multiallelic, tandem copy number variation (CNV). Two complete genes, complement component 4 (C4) and steroid 21-hydroxylase (CYP21A2, formerly CYP21B), reside in its variable region. RCCX is prone to nonallelic homologous recombination (NAHR) such as unequal crossover, generating duplications and deletions of RCCX modules, and gene conversion. A series of allele-specific long-range polymerase chain reaction coupled to the whole-gene sequencing of CYP21A2 was developed for molecular haplotyping. By means of the developed techniques, 35 different kinds of CYP21A2 haplotype variant were experimentally determined from 112 unrelated European subjects. The number of the resolved CYP21A2 haplotype variants was increased to 61 by bioinformatic haplotype reconstruction. The CYP21A2 haplotype variants could be assigned to the haplotypic RCCX CNV structures (the copy number of RCCX modules) in most cases. The genealogy network constructed from the CYP21A2 haplotype variants delineated the origin of RCCX structures. The different RCCX structures were located in tight groups. The minority of groups with identical RCCX structure occurred once in the network, implying monophyletic origin, but the majority of groups occurred several times and in different locations, indicating polyphyletic origin. The monophyletic groups were often created by single unequal crossover, whereas recurrent unequal crossover events generated some of the polyphyletic groups. As a result of recurrent NAHR events, more CYP21A2 haplotype variants with different allele patterns belonged to the same RCCX structure. The intraspecific evolution of RCCX CNV described here has provided a reasonable expectation for that of complex, multiallelic, tandem CNVs in humans.

Introduction

Copy number variations (CNVs) occupy a small proportion of the human genome but contribute significantly to genetic diversity (Redon et al. 2006; Conrad et al. 2010), greatly influence cellular phenotypes such as gene expression (Stranger et al. 2007), and are responsible for a wide spectrum of diseases and disease susceptibilities (Zhang et al. 2009). Multiallelic CNVs (greater than 2 possible haploid copy number [Conrad et al. 2010]) constitute a sizeable fraction of large CNVs, are highly enriched with gene content, and are closely associated with segmental duplications by virtue of their prevalent duplicated alleles (Redon et al. 2006; Conrad et al. 2010). Multiallelic CNVs may be considered as recent duplications during fixation phase and under the effect of neutral or positive evolutionary processes (Innan and Kondrashov 2010; Teshima and Innan 2012): Consequently, they play a significant role in gene and genome evolution (Hurles et al. 2008; Marques-Bonet et al. 2009). Underlying this rapid evolution, CNV alleles (copy number on a chromosome) with large, homologous, and tandem repeats are prone to rearrangements via nonallelic homologous recombination (NAHR) mechanisms (Hastings et al. 2009) such as unequal crossover and gene conversion. Unequal crossover facilitates large structural rearrangements and copy number changes (Stankiewicz and Lupski 2002), whereas gene conversion mediates relatively short sequence transfers (Chen et al. 2007). By contrast, the relative contribution of rearrangement mechanisms to the emergence and maintenance of CNVs and their gene content is not well appreciated, in spite of the recent advances in our knowledge of the mutation mechanisms of genome-wide CNVs (Kidd et al. 2010). Furthermore, using genome-wide platforms, the multiallelic, tandem CNVs, as well as their duplicated gene contents, are very difficult to genotype directly and are poorly tagged by single-nucleotide polymorphisms (SNPs) (Conrad et al. 2010; Alkan et al. 2011; Campbell et al. 2011).

RCCX, often recognized by genome-wide CNV studies (Tuzun et al. 2005; Redon et al. 2006; Perry et al. 2008a; Conrad et al. 2010; Kato et al. 2010), is a complex, medium size (∼30 kb per module), multiallelic, tandem CNV in the major histocompatibility complex (MHC) class III region (Horton et al. 2004), and it commonly consists of monomodular, bimodular, and trimodular CNV alleles with the prevalence of approximately 15%, 75%, and 10% in Europeans, respectively (Blanchong et al. 2000; Vatay et al. 2003). Four genes—serine/threonine kinase 19 (STK19), complement component 4 (C4), steroid 21-hydroxylase (CYP21), and tenascin-X (TNX)—reside close to each other in each module. Considering all modules, each of these genes usually materializes in the form of one active gene and zero, one or two pseudogenes determined by the module number, except for C4, which has only active copies. There is a functional difference among C4 genes dividing them into C4A and C4B types, because five adjacent nucleotide substitutions cause four amino acid changes and immunological subfunctionalization (Szilagyi, Doleschall, et al. 2010). The retention of the C4A-C4B nucleotide differences is observed in great apes; hence, this specialization of duplicated C4 genes confers evolutionary advantage and provides a potential explanation for the emergence of RCCX CNV (Kawaguchi et al. 1992; Innan and Kondrashov 2010). In addition, each C4 gene contains a deletion CNV (0 or 1 haploid copy number [Conrad et al. 2010]) derived from the insertion of a human endogenous retrovirus K (HERV-K) sequence (Dangel et al. 1994; Tassabehji et al. 1994), and the prevalence of the insertion allele of this HERV-K (C4) CNV depends on the position of its harboring module in the RCCX (Blanchong et al. 2000). These variations in copy number and gene content result in a CNV with a highly complex structure, which is traditionally described by the copy number of RCCX modules, and, per module, by the deletion or insertion allele (the absence or presence of the insertion) of HERV-K CNV and the type of C4 gene (Yu et al. 2003), even though these features embody genetic polymorphisms that differ in nature and size. In this article, a haplotypic RCCX module is abbreviated with two letters, the first represents the alleles of the HERV-K CNV (L—long allele [insertion allele] or S—short allele [deletion allele], the use of L and S abbreviation follows the tradition of published works on RCCX CNV) and the second symbolizes the types of C4 gene (A or B). The multiplication of these two letters indicates the bi- and trimodular structures (see fig. 1 for some examples).

Fig. 1.—

Scale representation of the alignment of the RCCX variable region sequences from the external database and the localizations of the developed ASLR-PCRs. The names of cell lines and the schematic abbreviation of RCCX structures are indicated on the left side. A module is abbreviated with two letters, the first represents the alleles of HERV-K CNV (L—the long allele or S—short allele), and the second symbolizes the type of C4 gene (A or B). The duplication of these two letters indicates the bimodular structure. The alignment of the RCCX variable region has been generated from six MHC haplotype sequences of HLA-homozygous cell lines (NG_005163.2, NT_007592.15, NT_167245.1, NT_167247.1, NT_167248.1, and NT_167249.1). The alignment spans from the telomeric end of exon 4 of STK19 to the centomeric end of exon 28 of TNXB. Dashed line indicates sequence absent from the MCF cell line. The RCCX structures of cell lines are monomodular and bimodular. The variable region of bimodular RCCX contains two pairs of complete genes, complement component 4 (C4A and C4B), steroid 21-hydroxylase (CYP21A1P and CYP21A2), and two pairs of partial genes, serine/threonine kinase 19 (STK19 and STK19P) and tenascin-X (TNXA and TNXB). The CNV of the HERV-K virus sequence is located in the C4 genes. The module breakpoint of bimodular structures and the direction of the ends of chromosome 6 are indicated under the scale bar. The positions and names of ASLR-PCR primers and the length of PCR products are shown at the bottom. The names of ASLR-PCRs are abbreviated by the first letter or the C4 gene type of forward and reverse primers.

NAHR contributes substantially to the genetic diversity of RCCX. On the one hand, unequal crossover generates copy number changes (Yang et al. 1999; Blanchong et al. 2000) and very rare RCCX CNV alleles such as quadrimodular structures (Chung et al. 2002a; Koppens et al. 2002b), chromosomes with more than one active CYP21 gene (CYP21A2) (Koppens et al. 2002b), chromosomes with only CYP21 pseudogenes (CYP21A1P) (Koppens et al. 2003), and structures with chimeras of CYP21 genes (Tusie-Luna and White 1995; Koppens et al. 2002a; Lee 2004). On the other hand, the deleterious mutations of the CYP21A1P gene, such as an 8-bp deletion in exon 3 and four related substitutions in exon 6, can be transferred by nonallelic gene conversion, causing the majority of the point mutations in CYP21A2 (Collier et al. 1993; Tusie-Luna and White 1995; Concolino et al. 2010). CYP21A2 deficiency is by far the most common cause of congenital adrenal hyperplasia (CAH), the inherited inability to synthesize cortisol and aldosterone (White and Speiser 2000).

We assumed that RCCX structures were related to particular CYP21A2 alleles, and the primary aim of this study was to unravel the intraspecific evolution of RCCX CNV by means of the whole-gene haplotypes (the term haplotype was used to indicate "gene-based functional" haplotype [Hoehe 2003] in this study) of the polymorphic internal gene, CYP21A2. A molecular haplotyping technique has been developed for RCCX CNV based on the concept of allele-specific long-range polymerase chain reaction (ASLR-PCR) (Michalatos-Beloin et al. 1996), and full-length CYP21A2 haplotypes have been determined from the haplotypic products of ASLR-PCR in many cases. Bioinformatic haplotype reconstruction has followed the experimental work to resolve the experimentally indeterminable haplotypes from genotypic CYP21A2 sequences. Besides the intraspecific evolution, we also attempted to trace the NAHR events of RCCX CNV by the genealogical haplotype network. The characteristics of RCCX structure-CYP21A2 haplotype variants and NAHR events forming RCCX CNV described here have provided reasonable expectations for the intraspecific evolution of complex, multiallelic, tandem CNVs in humans.

Material and Methods

Subjects

Unrelated European subjects from Hungary who participated in a previous study on full-length CYP21A2 gene sequences (Blasko et al. 2009) were investigated initially, but original subjects with three copies of CYP21A2 (see later for the method of determination) and those who did not have sufficient quality (fragmented DNA is inappropriate for long-range PCR) or enough DNA for CYP21A2 resequencing from haplotypic products were excluded, resulting in 72 study subjects (A summary of experimental design and work flow can be found in supplementary fig. S1, Supplementary Material online). At the second stage, RCCX structure was investigated (see later) in 244 unrelated Hungarian subjects with European ancestry, and 40 unrelated subjects with two copies of CYP21A2 were included in such a way as to represent a sufficient amount of all kinds of known RCCX structure (Blanchong et al. 2000) and to be suitable for the molecular haplotyping of CYP21A2. This sorting strategy enabled us to maximize the coverage of CYP21A2 haplotype space and the discovery of rare haplotypes (these are miscalled when a statistical inference approach is applied [Tishkoff et al. 2000]) related to rare RCCX structures. However, the utilization of allele frequencies had to be rejected (the allele frequencies were not used in the subsequent bioinformatic analyses) because the subjects were sorted. Overall, 112 unrelated Hungarian subjects were included and fully investigated. The subjects gave informed consent, the study was approved by the Hungarian National Ethical Committee, and was executed according to the principles of the Declaration of Helsinki.

Molecular Haplotyping and Determination of RCCX Structures

Haplotypic RCCX structures and the suitable diploid RCCX structure combinations for the molecular haplotyping of CYP21A2 (one C4A and one C4B gene next to TNXB gene, or one C4 with the insertion allele of HERV-K CNV and one C4 with the deletion allele next to TNXB) were determined using a set of ASLR-PCRs (fig. 1) and the copy number analyses of C4 genes and HERV-K CNV.

The ASLR-PCRs principally relied on C4A and C4B allele-specific forward and reverse primers (C4A_F, C4B_F, C4A_R, and C4B_R) complementary to the discriminating nucleotide substitutions of C4 genes in exon 26. In addition to the C4 type-specific primers, STK19 and TNXB gene-specific primers (STK19_F and TNXB_R), that matched only the active genes, and the STK19P and TNXA primers (STK19P_R and TNXA_F), which adhered to both active and pseudogene of STK19 and TNX, but they were able to generate PCR products only from the pseudogenes with the C4 allele-specific primers, were applied. Finally, HERV-K-specific primers (HERV-K_F and HERV-K_R) were also used, fitting only to C4 genes with the insertion allele of HERV-K CNV (fig. 1 and supplementary table S1, Supplementary Material online). The following 10 primer pairs were applied from the possible combinations of these ASLR-PCR primers: STK19_F or TNXA_F with C4A_R or C4B_R (four pairs), C4A_F or C4B_F with STK19P_R or TNXB_R (four pairs), HERV-K_F with TNXB_R (one pair), and TNXA_F with HERV-K_R (one pair). An additional primer pair, STK19_F-HERV-K_R, was set up and described, but it was not needed in this study. The ASLR-PCRs were performed (supplementary table S2, Supplementary Material online) using LongAmp Taq DNA polymerase (New England Biolabs) according to the manufacturer's protocol with some modifications. PCRs were carried out in 10 μl total volume containing 1 U LongAmp Taq DNA polymerase, 1× LongAmp Taq reaction buffer, 300 μM of each dNTP, 0.4 μM of each primer, and 10–100 ng genomic DNA depending on DNA quality.

The copy numbers of the C4A and C4B genes, as well as the number of C4 genes with the insertion and deletion alleles of HERV-K CNV, were determined by quantitative PCR (qPCR) as described previously (Szilagyi et al. 2006; Wu et al. 2007) with some modifications. C4-specific Taqman probes (Applied Biosystems) were labeled with the fluorescent dye 6-FAM, whereas RPPH1, used as an endogenous reference (RNase P reference assay, Applied Biosystems) in multiplex reactions, was labeled with the dye VIC.

Amplification and Sequencing of the CYP21A2 Gene

To study the CYP21A2 gene, the CYP21A1P pseudogene and their possible chimeric forms, two allele-specific nested PCRs corresponding to 5′- and 3′-parts of the CYP21 genes (supplementary fig. S2, Supplementary Material online) were performed from the genotypic or haplotypic products of ASLR-PCRs generated by C4A_F or C4B_F with STK19P_R or TNXB_R, and HERV-K_F with TNXB_R primers. The subjects whose CYP21A2-specific nested PCR product was amplified from ASLR-PCR products by C4A_F or C4B_F with STK19P_R primers were considered as subjects with three CYP21A2 copies and were excluded. The nested PCRs of the 5′-part were achieved by primers adhered to the allele-specific nucleotide substitutions in exon 6 (CYP21A1P_R or CYP21A2_R), with nonspecific primer matched to the 5′-flanking region (CYP21_F). The nested PCRs of 3′-part were accomplished by CYP21A1P and CYP21A2 allele-specific primers complementary to the 8 bp indel difference in exon 3 (CYP21A1P_F or CYP21A2_F), with the nonspecific primer matched to the 3′-flanking region (CYP21_R). Each reaction (15 μl total volume) contained 1 U GoTaq DNA polymerase (Promega), 1× GoTaq colorless Flexi buffer, 1.5 mM MgCl₂, 200 μM of each dNTP, 133 nM of each primer, and ∼4 ng ASLR-PCR product directly from the ASLR-PCR mix. The cycle conditions were 95°C for 5 min, 15 cycles of 95°C for 10 s, 64°C for 5 s, and 72°C for 90 s (CYP21_F with CYP21A1P_R or CYP21A2_R) or 150 s (CYP21A1P_F or CYP21A2_F with CYP21_R), finishing with extension at 72°C for 5 min.

The full-length CYP21A2 was capillary sequenced following the allele-specific nested PCR. Nested PCR products were treated with exonuclease I (New England Biolabs) and rAPid alkaline phosphatase (Roche), then directly sequenced on both strands by 7–7 primers (supplementary table S1, Supplementary Material online) using the BigDye Terminator Sequencing Kit v3.1 (Applied Biosystems) and run on an ABI 3100 Genetic Analyzer (Applied Biosystems).

Bioinformatic Sequence and Haplotype Analyses

The sequences of RCCX CNV and CYP21 genes from Gen-Bank were used (supplementary table S3, Supplementary Material online). The 491 expressed sequence tags (EST) sequences of CYP21 genes and 6 MHC haplotype sequences of HLA-homozygous cell lines (Horton et al. 2008) from STK19 to TNXB were aligned using ClustalX2 v2.0.5 (Larkin et al. 2007). The start of the CYP21A2 gene was defined at 8 bp upstream from the start of the coding region (Higashi et al. 1986) by 5′-EST analysis (supplementary fig. S3, Supplementary Material online) (Nagaraj et al. 2007). The sequence calls of CYP21A2 were assembled with CLC DNA Workbench v5.7.1 (CLC bio) and inspected manually by two different operators. RCCX structures and CYP21A2 haplotypes, which could not be determined experimentally, were inferred with PHASE v2.1.1 (Stephens et al. 2001; Stephens and Donnelly 2003) (Specialized phasing tools for CNVs [Kato et al. 2008; Su et al. 2010] could not been used because of the lack of ability to handle the known phase information from individual to individual.). To input RCCX structure data into PHASE, HERV-K CNV and the type of C4 gene in each module were treated as independent loci. Because the trimodular RCCX structures are relatively prevalent, but quadrimodular structure is extremely rare, six loci represented the three modules. First four loci represented the 5′-modules, and zero allele indicated the lack of the particular 5′-module(s) in bimodular and monomodular RCCX structures. Because the deletion of the full RCCX region has not yet been observed, zero alleles could not be present on the last two loci representing the 3′-module. The known phasing information from the experiments was input with the phasing option (-k) of PHASE (an example is given in supplementary fig. S4, Supplementary Material online). For the connection of unconnected RCCX structure-CYP21A2 haplotypes, the experimental RCCX structure data were inferred together with the correctly resolved (experimentally determined or above 0.99 confidence probability threshold) CYP21A2 haplotypes coded as known phase. To check the correctness of both coding of the RCCX structure and the connecting of RCCX structures and CYP21A2 haplotypes, a simulated RCCX CNV data set was generated by the random connection of the haplotypic RCCX structures from a recent family study not relying on any bioinformatic inferences (Banlaki, Doleschall, et al. 2012) and the resolved CYP21A2 haplotypes. Median-joining networks were constructed from the experimental and inferred haplotypes with Network v4.6.1.0 (Bandelt et al. 1999). A chimpanzee (Pan troglodytes) CYP21A2 sequence was applied as an outgroup (root) for the network building.

Results

Experimental Determination of RCCX Structures and CYP21A2 Haplotypes

The organization of RCCX structures with respect to the number of RCCX modules, C4 gene types, and HERV-K CNV was investigated by a set of ASLR-PCRs (fig. 1), allele-specific nested PCRs (supplementary fig. S2, Supplementary Material online) and qPCRs. Furthermore, full-length CYP21A2 haplotypes determined by allele-specific nested PCR and sequencing were assigned to these RCCX structures in many cases. Although ASLR-PCR was capable of determining the C4 gene types in conjunction with the alleles of HERV-K CNV within a haplotypic module, the number of modules and relationship between modules on a chromosome were deduced from the results of ASLR-PCRs and qPCRs.

First, the deduction method is exemplified by the alignment of RCCX structures of six HLA-homozygous cell lines (Horton et al. 2008) (fig. 1). Two monomodular RCCX structures of COX and QBL in diploid state do not produce PCR fragments from TNXA and STK19P, but STK19_F and C4A_R primers result in a 15.8 kb fragment corresponding to QBL, and STK19_F and C4B_R primers generate a 9.4 kb fragment corresponding to COX. The TNXB_R primer with C4A_F and C4B_F primers (in two separate tubes) results in two 17.7 kb haplotypic products. Both CYP21A2 haplotypes can be amplified and sequenced from the two separated 17.7 kb ASLR-PCR products, and thus CYP21A2 haplotypes can be assigned to the corresponding RCCX structures. In addition to the PCR product of C4A_F and TNXB_R primers, the CYP21A2 haplotype related to QBL can be determined from the 23.3 kb product of HERV-K_F and TNXB_R primers, because this ASLR-PCR product is not generated from COX. Therefore, the diploid combination of monomodular COX and QBL is abbreviated to LA/SB.

Henceforth, the deduction is demonstrated by the results of ASLR-PCRs, allele-specific nested PCRs, and qPCRs in four samples (fig. 2). The organization of two identical monomodular RCCX structures such as LA/LA was also deduced as described earlier, but the ASLR-PCR products of two chromosomes by C4A_F-TNXB_R (A-T) or HERV-K_F-TNXB_R (H-T) primer pairs could not be separated (sample 1). Therefore, the CYP21A2 alleles of the two chromosomes could only be genotyped after the CYP21A2-specific nested PCR. Besides the products by TNXB_R (A-T, B-T, H-T) and by STK19_F-C4A_R (S-A) primer pair from the 5′- and 3′-ends of RCCX CNV, the products by C4A_F-STK19P_R (A-S) and TNXA_F-C4B_R (T-B) primer pairs were amplified in sample 2, verifying the presence of at least one multimodular RCCX structure. The presence of A-T product and the absence of product by TNXA_F-C4A_R (T-A) primer pair indicated that there was a monomodular LA RCCX structure on one chromosome, and the absence of T-A product also implied that there was no trimodular RCCX structure with the C4A gene in the middle module on the other chromosome. The absence of product by C4B_F-STK19P_R (B-S) and STK19_F-C4B_R (S-B) primer pairs indicated that there was a bimodular RCCX structure with C4A in the 5′-end and C4B in the 3′-end on the second chromosome. The size of S-A and T-B products verified that C4A genes were found together with the L allele of HERV-K CNV in a module, and the C4B gene was together with the S allele. Therefore, the diploid combination of haplotypic RCCX structures was LA/LASB in sample 2, which was concordant with the copy numbers of C4A and C4B genes (CN A and B) and the alleles of the HERV-K CNV (CN L and S). The CYP21A1P-specific products were amplified by 5′- and 3′-nested PCR from AS ASLR-PCR product as template, and CYP21A2-specific products were generated from the product of C4B_F-TNXB_R (B-T) primer pair. In the case of A-T and H-T products (these were amplified from the same LA RCCX structure), a CYP21A2-specific 5′-nested PCR product and a CYP21A1P-specific 3′-nested PCR product were generated, indicating that LA RCCX structure harbored a chimeric CYP21 gene.

Fig. 2.—

Results of ASLR-PCRs, allele-specific nested PCRs and qPCRs demonstrated in four samples (samples 1–4). The names of ASLR-PCRs are abbreviated by the first letter or the C4 gene type of forward and reverse primers, in alphabetical order: C4A_F-STK19P_R (A-S), C4A_F-TNXB_R (A-T), C4B_F-STK19P_R (B-S), C4B_F-TNXB_R (B-T), HERV-K_F-TNXB_R (H-T), STK19_F-C4A_R (S-A), STK19_F-C4B_R (S-B), TNXA_F-C4A_R (T-A), TNXA _F-C4B_R (T-B), and TNXA_F-HERV-K_R (T-H). The names of CYP21A1P- and CYP21A2-specific nested PCRs are abbreviated by the specific tag of CYP21 genes (A1P and A2) and the corresponding half of the gene from where the products can be amplified (5′ and 3′). The copy numbers (CN) of C4 genes and the alleles of HERV-K CNV determined by qPCRs are abbreviated by the types of C4 (A or B) and the long and short CNV alleles of HERV-K (L or S). Haplotypic RCCX module is abbreviated with two letters, the first represents the alleles of HERV-K CNV (L or S) and the second symbolizes the types of C4 gene (A or B). The multiplication of the two letters in a structure indicates the module number. For CYP21A1P- and CYP21A2-specific nested PCRs, a portion of ASLR-PCR mix containing ∼4 ng ASLR-PCR product was used as template. Genomic control confirms that the nested PCR products could not be amplified from the same amount of genomic DNA of the particular sample as the amount being included in a mix with ASLR-PCR product. PCR-negative control (nc) signifies the traditional control of PCR (complete PCR mix without DNA). HindIII digested lambda DNA (New England Biolabs) and 1 kb DNA ladder (New England Biolabs) were used as markers.

The LALB/LASB RCCX structures of sample 3 were unambiguously determined similarly to that of sample 2, but only a short (9.7 kb) T-B fragment could be detected in spite of the presence of H-T product. The shorter fragment may be preferred to the longer one in PCR, thus sample 3 was checked by a TNXA_F-HERV-K_R (T-H) primer pair to avoid this potential error. The presence of T-H products verified that the long (16 kb) T-B fragment became undetectable (usually, the long products of S-A, S-B, T-A, and T-B could be detected in our hands). The A-S and B-T products were derived from both chromosomes, hence the haplotypic fragments could not be separated from each other, but the H-T product was haplotypic. Therefore, only a genotypic CYP21A2-specific nested-PCR product could be acquired from the B-T ASLR-PCR product and a haplotypic product from the H-T product. Intriguingly, both CYP21A1P- and CYP21A2-specific nested PCR products were amplified from the A-S product, indicating that one of the two chromosomes harbored a CYP21A2 gene in the 5′-module. Those subjects whose CYP21A2-specific nested PCR product was amplified from the A-S or B-S product were considered as subjects with three CYP21A2 copies and were therefore excluded from the study because three gene copies from a diploid subject would have severely complicated the subsequent bioinformatic haplotype reconstruction. However, it should be noted that CYP21 haplotypes in 5′-modules can also be examined using the ASLR- and nested PCRs, which may be proven helpful for a recent research area (Tsai et al. 2011).

The diploid RCCX combinations of the first three samples could be unambiguously determined only from the pattern of ASLR-PCRs. In many subjects, RCCX combinations were unambiguous merely based on the set of ASLR-PCRs, taking all conceivable RCCX haplotypes into account. In the cases when the determination of the copy numbers by ASLR-PCR and qPCR was redundant, a perfect concordance was observed between the data of the two assays, demonstrating the reliability of both. In addition, the sequencing of haplotypic and genotypic PCR products also confirmed the reliability and accuracy of the methods. In spite of the deduction from the redundant results of different type of assays, the deduction was made unambiguously only in a proportion of diploid combinations of haplotypic RCCX structures. Some of the combinations showed the same ASLR-PCR pattern, but copy numbers could distinguish them from each other. Some combinations could not be deciphered from experimental results, hence we were only able to narrow down the number of possible combinations. For example, the presence of S-A, S-B, and T-B products and the absence of A-T products in sample 4 indicated that there were an LA and an SB module in the 5′-end of RCCX structures and two C4B genes in the 3′-end modules. However, the B-S product might be derived from a bimodular RCCX structure with a 5′-end C4B gene or from a trimodular RCCX structure with a C4B gene in the middle module. Therefore, it was not possible to determine whether SB/LASBSB or LASB/SBSB was the real RCCX combination.

Overall, haplotypic RCCX structures were experimentally determined on 110 (49%) of 224 chromosomes of the 112 subjects and full-length CYP21A2 haplotypes on 64 (29%) chromosomes. Molecular haplotyping (the experimental determination) revealed 8 different kinds of haplotypic RCCX structure variant and 35 different CYP21A2 haplotype variants (GenBank: JN034382–JN034411 and JQ993310–JQ993314). Moreover, 23 of these CYP21A2 haplotype variants were unambiguously assigned to haplotypic RCCX structures (supplementary table S4, Supplementary Material online). In addition, one chimeric CYP21A1P-CYP21A2 gene harbored by LA RCCX structure also occurred (sample 2).

Bioinformatic Reconstruction of Haplotypic RCCX Structures and CYP21A2 Haplotypes

The RCCX structures and CYP21A2 alleles undetermined by molecular haplotyping were inferred using bioinformatic haplotype reconstruction by PHASE software. The RCCX polymorphism (module copy number, C4 gene type, and HERV-K CNV in each module) data set and the CYP21A2 polymorphism data set were separately analyzed, taking account of the experimentally determined haplotypic structures or haplotypes (known phases). Haplotypic RCCX structures and CYP21A2 haplotypes were considered as resolved above the confidence probability threshold of 0.99 (this value is much stricter than those of most published works [Garrick et al. 2010]). From the 224 chromosomes, 148 (66%) haplotypic RCCX structures belonging to 10 kinds of RCCX structure variant were above the 0.99 confidence threshold, and 213 (95%) CYP21A2 haplotypes were above the 0.99 threshold (supplementary table S5, Supplementary Material online). When the 5′-parts of resolved CYP21A2 haplotypes were compared with the 377 filtered CYP21A2 5′-ESTs of ADRGL2 data set from dbEST, CYP21A2 sequences proved to be highly concordant (supplementary table S6, Supplementary Material online). The CYP21A2 haplotypes below the 0.99 confidence limit and the chimeric CYP21 haplotype were excluded from the subsequent analyses to remove ambiguous structures and haplotypes. Furthermore, the RCCX structures were assigned to the resolved CYP21A2 haplotypes by PHASE. From the 213 resolved CYP21A2 haplotypes, 199 (93%) CYP21A2 haplotypes were connected to RCCX structures above the 0.99 threshold, and only 14 (7%) could not be unambiguously assigned (supplementary table S4, Supplementary Material online). In addition, both RCCX and CYP21A2 polymorphism data sets were analyzed without known phases (treated as genotype data) to evaluate the contribution of the haplotypic information to the efficiency of haplotype reconstruction (supplementary table S5, Supplementary Material online), and a simulated RCCX CNV data set was also analyzed to check the correctness of both coding of the RCCX structures and the connecting of RCCX structures and CYP21A2 haplotypes (supplementary table S7, Supplementary Material online).

Altogether, the 213 experimental and inferred CYP21A2 haplotypes represented 61 different variants (supplementary table S4, Supplementary Material online) containing 51 segregating sites in total (supplementary fig. S2 and table S8, Supplementary Material online). The number of singleton haplotype variants was 26 (43%), implying that a considerable degree of mutation events occurred in the recent past. The combined molecular and inferred haplotyping approach finally resulted in 71 different RCCX structure-CYP21A2 haplotype variants (fig. 3), and only 2 (3%) of the CYP21A2 haplotype variants could not be assigned to a haplotypic RCCX structure. Given the genetic linkages between the haplotypic RCCX structures and the harbored CYP21A2 haplotypes, one CYP21A2 haplotype variant was related to only one RCCX structure in 48 (79%) cases, but one haplotype variant related to more RCCX structures was also observed in 11 (18%) cases. To confirm the RCCX structure-CYP21A2 haplotype variants, they were compared with the RCCX structures and harbored CYP21A2 sequences of HLA-homozygous cell lines: The structures and sequences were highly concordant (supplementary table S4, Supplementary Material online).

Fig. 3.—

Graphic representation of resolved 71 haplotypic RCCX structure-CYP21A2 haplotype variants. Haplotypic RCCX structures on the left side are abbreviated with the multiplication of the two letters of a module. In a module, the first represents the alleles of HERV-K CNV (L or S) and the second symbolizes the types of C4 gene (A or B). CYP21A2 haplotype variants are on the right side, and the names of the genes at the bottom. Eleven CYP21A2 haplotype variants are connected to more than one RCCX structure. The segregating sites of CYP21A2 haplotype variants with the related haplotypic RCCX structures can be also found in supplementary table S4, Supplementary Material online (In the figure, the CYP21A2 haplotype variants are grouped by the haplotypic RCCX structure, and in supplementary table S4, haplotypic RCCX structures are grouped by the CYP21A2 haplotype variant.).

Genealogical Network of CYP21A2 Haplotype Variants

To construct a haplotype network that allows for the unique characteristics of intraspecific level such as persistent ancestral nodes, multifurcations, and reticulations (Posada and Crandall 2001), the median-joining method was applied. CYP21A2 haplotype variants were free from the marks of crossover, presumably owing to the shortness of the gene and the low rate of meiotic (equal) crossover in the RCCX region (Cullen et al. 2002), and thus the prerequisite for the applicability of the median-joining algorithm was realized. To give an evolutionary direction to the network, it was rooted by a chimpanzee CYP21A2 orthologue. The network showed a tree-like structure with some reticulations and intensive multifurcations (fig. 4A and supplementary fig. S5, Supplementary Material online). The h58 haplotype variant was connected to the root, implying that the h58 haplotype variant was the most ancestral haplotype in the network. When haplotypic RCCX structures were projected onto the corresponding CYP21A2 haplotype variants, tight grouping related to RCCX structures was evident (fig. 4B). The haplotypic RCCX structures were not taken into consideration for the construction of the network, therefore, the CYP21A2 haplotype variants as "complex genetic markers" (Hoehe 2003) independently reflected the genealogy of the entire RCCX CNV. To test the stability of the network and the effect of inferred CYP21A2 haplotype variants on the network, an additional network was reconstructed using only the 35 experimentally determined CYP21A2 haplotype variants (supplementary fig. S6, Supplementary Material online). The position of the root was identical (connected to h58 haplotype variant), and the CYP21A2 haplotype variants remained in the original groups.

Fig. 4.—

Genealogical haplotype networks of CYP21A2 haplotype variants (the root is abridged.). (A) Haplotype network constructed from CYP21A2 haplotype variants. Red circles indicate the (sampled) CYP21A2 haplotype variants, light gray circles show the missing intermediates, and light gray arrows symbolize the allele-state changes (character-state changes) with the positions of segregating sites and the arisen allele. Two or more allele changes belonging to adjacent segregating sites have been considered as unambiguous gene conversion events: These allele-state changes are indicated together. (B) Haplotype network with projected RCCX structures constructed from CYP21A2 haplotype variants. Light gray circles indicate the CYP21A2 haplotype variants with their names. Monomodular CNV alleles (copy number on a chromosome) are indicated by yellow, bimodular by green, and trimodular by blue. Haplotypic RCCX module is abbreviated with two letters, the first represents the alleles of HERV-K CNV (L—long allele or S—short allele), and the second symbolizes the types of C4 gene (A or B). The multiplication of the two letters in a structure indicates the module number.

The h58 haplotype variant harbored by LALA or LALB RCCX structure was directly connected to the root, but there were no haplotype variants connected directly to the h58 variant. There were eight variants (h22, h24, h38, h48, h49, h50, h54, and h56) with quite different allele composition related indirectly to h58 variant through six missing intermediates (median vectors). Noticeably, none of these eight haplotype variants were carried by LALA or LALB structure. The deviations in the allele composition of h58-related haplotype-variants, the missing intermediates, and the different RCCX structures harboring these variants suggested larger evolutionary distances among h58 and its indirectly connected haplotype variants than inside the well-separated groups of directly connected haplotypes. These well-separated groups of directly connected haplotypes with branch-like junctions and with reticulation were found toward the tips of the network. With respect to CNV alleles (copy number on a chromosome), the RCCX structure groups of each allele (mono-, bi-, and trimodular) were widely scattered in the network, supporting their polyphyletic origin. Many groups with identical RCCX structure also occurred several times in the network. For instance, LA structure could be found in five distinct locations, which also indicated that they were polyphyletic groups. In contrast, some RCCX groups such as SB and LASALB occurred only once in the network, implying the monophyletic origin of these RCCX structures.

Unequal Crossovers in RCCX CNV

It is commonly understood that the mono- and trimodular RCCX structures are generated from bimodular structures by unequal crossover (fig. 5A) in humans (Yang et al. 1999; Blanchong et al. 2000). We therefore attempted to trace the origin of mono- and trimodular structures by unequal crossover based on the haplotype network. A mono- or trimodular RCCX structure group was considered as the product of an unequal crossover event if it was embedded in a group of a bimodular RCCX structure containing an identical CYP21A2 haplotype variant or if it had a direct connection to an adjacent bimodular group. Thus one parental bimodular and one resultant mono- or trimodular (recombinant) structure could be examined by means of the haplotype network from the two parental and the two resultant chromosomes of an unequal crossover (The unequal crossover of monomodular structures cannot lead to the copy number change, and the trimodular groups of the network were in tip positions, or were embedded, and consequently these structures were not regarded as parental structures).

Fig. 5.—

Unequal crossover events in RCCX CNV. (A) Scale and schematic representations of a hypothetical unequal crossover event in RCCX CNV (although both haplotypic RCCX structures with corresponding CYP21A2 haplotypes exist, it is improbable that they have been generated exactly by this unequal crossover event). The name of genes in the RCCX CNV and the insertion allele of HERV-K CNV are indicated at the top. Haplotypic RCCX module is abbreviated with two letters, the first represents the alleles of HERV-K CNV (L—long allele or S—short allele), and the second symbolizes the types of C4 gene (A or B). The multiplication of the two letters in a structure indicates the module number. (B) Unequal crossover of haplotypic RCCX structures on CYP21A2 haplotype networks. Red circles indicate the unequal crossover events, orange arrows indicate the unequal crossover or consecutive mutational events, and the numbers preceded by h represent the CYP21A2 haplotypes. Numbers are assigned to particular unequal crossover events, the detailed explanations of which can be seen at the bottom. The question marks indicate the unknown RCCX polymorphisms of the parental structures. If two possibilities are present at the generation of a particular structure, the two possible breakpoints of unequal crossover are on the border of the range where the breakpoint can be located.

Corresponding to the aforementioned definition, eight unequal crossover events were observed in the network (fig. 5B), and six monomodular and two trimodular structures arose from them. Moreover, four events resulted in independent LA structures, and two events led to independent LALALB structures, and hence, these unequal crossover events were recurrent with respect to the particular RCCX structure. In the case of some monomodular structures such as the h44 haplotype variant harbored by LA, the breakpoint in front of the 5′-end of CYP21A2 on the parental bimodular structure and at the back of the 5’-LA part on the other parental structure and the breakpoint between the module boundary of parental bimodular structure and in front of the 5′-C4 gene on the other chromosome could create the monomodular structure (fig. 5B). In effect, other breakpoints between these two breakpoints can be also envisioned, and therefore, LA-h44 structure could arise by a breakpoint located from the 3′-end of CYP21A1P to the 5′-end of CYP21A2 of the parental bimodular structure harboring h44. For the generation of SB structure, the contribution of a breakpoint between the 5′-end of the HERV-K CNV and the CYP21A2 gene could be excluded because the 5′-end of the C4B gene with an S allele of the HERV-K CNV (5′-SB part) did not occur except in itself, the arising SB structure. Therefore, the haplotype variants of SB structure presumably arose from a breakpoint around the module boundary of LASB-h22 and three subsequent, consecutive allele-state changes. In this scenario, the SB structure did not change during the allele-state changes. This is further supported by the fact that the SB structure is totally absent in a European CAH population of a previous study, because SB structure probably reduces the unequal crossover events due to a greater degree of dissimilarity compared to other RCCX structures (Blanchong et al. 2000). Similar to the LA-h44 structure, LALALB structures could be generated by breakpoints with different locations and by different RCCX structures along with the parental LALB structure, including bi- and trimodular structures.

It should be noted that the other two trimodular groups excluded by virtue of their connection to the missing intermediates were probably generated by unequal crossover as well. LASBSB might be created by two LASB structures, and LASALB might be born from a LASB and a LALB structure, as described previously (Chung et al. 2002a). In addition to unequal crossover, some events of (equal) crossover or conversions affecting C4 type-specific nucleotides (Braun et al. 1990; Jaatinen et al. 2002) were also observed in the network.

Gene Conversions in the CYP21A2 Gene

Two or more nonconsecutive allele changes belonging to adjacent segregating sites are considered as unambiguous gene conversion events (Chen et al. 2007). In addition to this criterion, an event was regarded as a gene conversion only if the allele combination of change was present in the CYP21A2 haplotype variants or haplotypic CYP21A1P sequences of the utilized external sequences (supplementary table S3, Supplementary Material online). A conversion event was considered as a nonallelic event if its allele combination occurred only in one part (branch) of the network and if it contained a CYP21A1P-specific allele or alleles (if an event meets the definition, it will be nonallelic with high probability). Overall, 11 such conversion events could be observed in the CYP21A2 network (fig. 4A and table 1). From these conversions, seven events could be regarded as nonallelic conversions. Altogether nine conversions occurred only once, but the 1109 A and 1116 C allele changes appeared twice and were located in different parts of the network, indicating that the same conversion could be recurrent, and its allele changes might belong to haplotypes with different intraspecific origins. The minimum tract lengths ranged from 4 to 264 bp (mean: 48.36 bp, median: 12 bp), and the maximum tract lengths as defined according to a previous article (Chen et al. 2007) spanned from 37 to 658 bp (mean: 377.7 bp, median: 403 bp). These values matched with the values in both the human genome and CYP21A2 gene (Chen et al. 2007).

Table 1.

Gene conversions in the CYP21A2 haplotype network

Position of Segregating Sites	Arisen Allele	Origin	Tract Length
			Min	Max
505–516	GG	CYP21A1P	12	37
624–633	GTGG	CYP21A1P	30	107
864–865	CA	CYP21A1P	2	658
1109–1116	AC		8
1109–1116	AC		8
1109–1126	ACG	CYP21A1P	18	658
1116–1126	CG		11
1425–1688	GTCGGT	CYP21A1P	264	377
2697–2764	AAGT	CYP21A1P	68	429
3080–3186	CGTCCT	CYP21A1P	107	1,433+
3152–3155	TC		4	1,433+

Note.—Segregating sites are denoted by their position numbered from the start of the CYP21A2 coding region on the PGF sequence. The minimum tract length was measured between the first and last nucleotide of allele combination of the conversion. The maximum tract length could be examined only at nonallelic conversions because it was measured from the first nucleotide difference between all CYP21A2 sequences and all CYP21A1P sequences in the 5′-direction of allele combination to the first nucleotide difference between all CYP21A2 sequences and all CYP21A1P sequences in the 3′-direction of allele combination. "+" indicates that the 3′-end of conversion tract was open (the tract was not closed where the sequenced section of the CYP21A2 gene was finished), and the maximum tract length was larger than the value in the table.

Discussion

In this study, the intraspecific evolution of the complex, multiallelic, tandem RCCX CNV has been traced by whole-gene CYP21A2 haplotype variants, which were applied as complex genetic markers. To summarize the theoretical significance, the known genetic phenomena of human RCCX CNV such as the frequent variations in copy number and in the content of C4 gene and HERV-K CNV (Yang et al. 1999; Blanchong et al. 2000), the transfer of sequence tracts by gene conversion (Tusie-Luna and White 1995; Concolino et al. 2010), and the generation of monomodular variants and trimodular variants by unequal crossover (Tusie-Luna and White 1995; Chung et al. 2002a) have been encompassed by one evolutionary framework. The studied subjects were Hungarians. The genome-wide polymorphisms of Europeans from Hungary deviate negligibly from those of the European reference population (CEU) and other European populations (Semino et al. 2000; Tomory et al. 2007; Heath et al. 2008), and the same applies to the MHC region of these populations (de Bakker et al. 2006; Szilagyi et al. 2010), therefore, the results of this study can be extrapolated for other European populations.

The delineation of intraspecific evolution of RCCX CNV mainly relied on haplotypic information obtained by a set of ASLR-PCRs, which enabled us to haplotype RCCX CNV alleles and structures in many cases that genome-wide platforms for CNV discovery have not yet resolved (Alkan et al. 2011). In contrast to genome-wide, high-throughput methods, ASLR-PCR is only feasible for particular genomic regions because of their specific primers. For a future perspective, our approach can be extended, because the DNA products of ASLR-PCR can serve as templates for high-throughput methods such as next-generation sequencing (Mamanova et al. 2010). Therefore, the advantages (haplotypic and high-throughput) of the two approaches can be merged, eliciting the full-length haplotypic sequences of large, complex, and multiallelic CNVs.

To unravel the complex structures of a duplicated region in diploid subjects, not only must the information of homologous chromosomes be separated from each other but also the duplicated modules on a chromosome. ASLR-PCR can span the module-specific parts of a CNV on a chromosome, enabling the separation of the duplicated modules. Because the centromeric modules of RCCX CNV regularly contain CYP21A2 genes on both chromosomes, the allele-specific nested CYP21A2 PCR can inherit the allele specificity only from ASLR-PCR, and its own allele specificity may seem to be unnecessary. Actually, the allele specificity of nested PCRs can prove to be rewarding, as the orthologous modules may comprise different CYP21 genes, as seen in the case of the chimeric CYP21 gene. Besides the traditional Southern-based restriction fragment length polymorphism (RFLP) (Yang et al. 1999), many methods such as pulsed-field gel electrophoresis (Chung et al. 2002b), long-range PCR (Kristjansdottir and Steinsson 2004), long-range PCR with C4 type-specific RFLP (Chung et al. 2002a), ASLR-PCR (Lee et al. 2006), qPCR (Szilagyi et al. 2006; Wu et al. 2007), multiplex ligation-dependent probe amplification (Concolino et al. 2009; Wouters et al. 2009), paralog ratio test (Fernando et al. 2010), and long-range PCR coupled to nested PCR and sequencing (Tsai et al. 2011) have been developed for the investigation of particular features of RCCX, but the lack of allele specificity and/or module separation often limit their performance. However, ASLR-PCR also has a limitation, because the separation of haplotypic modules cannot be achieved in all diploid combinations of RCCX structures.

The combined molecular and inferred haplotyping approach enabled us to construct a dense genealogical CYP21A2 haplotype network, shedding new light on the evolution of structure in a complex, multiallelic, tandem CNV. The history of RCCX dates from one primigenial duplication event, which exhibits the sign of prevalent breakpoint microhomology of genome-wide structural variations (Kawaguchi and Klein 1992; Horiuchi et al. 1993; Mills et al. 2011). This duplication occurred in early mammals or the ancestor of mammals, and it must have already existed for at least 90 million years (Hedges 2002). The common genetic features of closely related species, such as the HERV-K CNV in great apes and the 8 bp deletion of exon 3 of CYP21A1P in chimpanzee and human, are also considered to originate from one event (Kawaguchi and Klein 1992; Dangel et al. 1995). The CNV status of RCCX has been reliably proven in chimpanzee (Perry et al. 2008b), and in all probability, this status in human and chimpanzee has continuously existed since the common ancestor (Marques-Bonet et al. 2009). Although extensive RCCX polymorphism or CYP21A2 haplotype data from chimpanzee or other great ape populations has not yet been made available, some signatures have been presented by the haplotype network for the coalescence of RCCX structures. The two RCCX structures of the h58 haplotype variant were not identical to any of the eight indirectly connected RCCX structures, which were located far from each other and represented four different structures. It is hard to imagine that all the eight indirectly connected RCCX structures arose from several different recombination events. Therefore, the h58 haplotype variant and its RCCX structures should not considered as the one and only ancestral haplotype and structure, but rather one of several ancestral RCCX structure-CYP21A2 haplotype variants still extant.

Diverse and sometimes contradictory selection forces keeping the balance of various RCCX structures may underlie the continuance of the RCCX CNV. For instance, the SB RCCX structure is advantageous with a view to deleterious nonallelic conversion and unequal crossover (Chung et al. 2002a; Lee et al. 2006), but disadvantageous in terms of the retention of the C4A gene (Kawaguchi et al. 1992), and the two forces therefore attenuate the effects of each other. The haplotypes harboring the 1688 T allele (h34, h35, and h36) are also intriguing from this viewpoint, because this allele causes CAH (Rumsby et al. 1998), and should be under the effect of purifying selection. Contrary to this, more related haplotypes with the 1668 T allele were observed, implying that these haplotypes have already existed in the long term. The contradiction may be resolved by a presumed positive selection force that compensates the deleterious effects of the allele and may be generated by a genetic feature of the common LASBSB structure of these haplotypes, such as the increased C4 copy number (Yang et al. 2007). In addition to C4 copy number, an elevated cortisol response has been actually found in the heterozygous carriers of CYP21A2 CAH mutations, which may also provide greater fitness (Witchel et al. 1997). Moreover, the elevated cortisol response and the changes of other hormone levels in association with C4B copy number have recently been described (Banlaki et al. 2012), hence an advantageous phenotype determined by a subset of particular RCCX structures or CYP21A2 haplotypes may be realistic.

The cumulative effects of potential selection forces are hard to assess by virtue of the difficulty in the quantitative analysis of selection forces, and the picture is further complicated by the fact that a particular CNV allele is not, of necessity, balanced by selection. If the cumulative effect of selection forces is quite small and negative on a particular CNV allele, then the CNV allele will have existed for a while but not in the long term (Innan and Kondrashov 2010). If this CNV allele is generated (again by unequal crossover) as frequently as it is removed by selection, then it will be continuously present among CNV alleles. We speculate that the recurrent unequal crossover of a particular CNV allele, which was observed in the case of the polyphyletic LA structure, may lead to the repeated generation and removal of the particular CNV allele. Therefore, recurrent unequal crossover events may maintain the polymorphic state of a complex, multiallelic, tandem CNV.

The haplotype network has also provided some further insight into the NAHR events shaping the RCCX CNV. Besides the observed gene conversion events, recurrent unequal crossovers generating the same RCCX structure occurred several times. Although the haplotypic frequencies were not followed in this study, the consequences of unequal crossover were apparent enough for clarifying the relationship of a CNV allele and the SNPs inside the CNV. The recurrent unequal crossover events could result in the same RCCX structures: Therefore, more CYP21A2 haplotypes with rather different SNP allele patterns belonged to the same RCCX structure. However, only one recurrent gene conversion event was observed in the study, but the same can apply to the effect of gene conversions on the genetic linkage. Even if the CNV alleles were correctly inferred, the strong linkage could be hampered between CNV alleles and their harbored SNP alleles by recurrent NAHR. Therefore, the recurrent NAHR may be one of the causes for the poor tagging of multiallelic CNV alleles by SNPs (Conrad et al. 2010; Campbell et al. 2011).

Supplementary Material

Supplementary figures S1–S6 and tables S1–S8 are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/).