Decoding NF1 Intragenic Copy-Number Variations

Abstract

Genomic rearrangements can cause both Mendelian and complex disorders. Currently, several major mechanisms causing genomic rearrangements, such as non-allelic homologous recombination (NAHR), non-homologous end joining (NHEJ), fork stalling and template switching (FoSTeS), and microhomology-mediated break-induced replication (MMBIR), have been proposed. However, to what extent these mechanisms contribute to gene-specific pathogenic copy-number variations (CNVs) remains understudied. Furthermore, few studies have resolved these pathogenic alterations at the nucleotide-level. Accordingly, our aim was to explore which mechanisms contribute to a large, unique set of locus-specific non-recurrent genomic rearrangements causing the genetic neurocutaneous disorder neurofibromatosis type 1 (NF1). Through breakpoint-spanning PCR as well as array comparative genomic hybridization, we have identified the breakpoints in 85 unrelated individuals carrying an NF1 intragenic CNV. Furthermore, we characterized the likely rearrangement mechanisms of these 85 CNVs, along with those of two additional previously published NF1 intragenic CNVs. Unlike the most typical recurrent rearrangements mediated by flanking low-copy repeats (LCRs), NF1 intragenic rearrangements vary in size, location, and rearrangement mechanisms. We propose the DNA-replication-based mechanisms comprising both FoSTeS and/or MMBIR and serial replication stalling to be the predominant mechanisms leading to NF1 intragenic CNVs. In addition to the loop within a 197-bp palindrome located in intron 40, four Alu elements located in introns 1, 2, 3, and 50 were also identified as intragenic-rearrangement hotspots within NF1.

Introduction

Genomic rearrangements, including deletions, duplications, mobile-element insertions, copy-number neutral inversions, and translocations, contribute to various disorders, as well as to normal phenotypic variation.^1–6 These rearrangements are mediated by mutational mechanisms that are not completely understood. Currently, several mechanisms have been proposed, including non-allelic homologous recombination (NAHR), non-homologous end joining (NHEJ), fork stalling and template switching (FoSTeS), and microhomology-mediated break-induced replication (MMBIR).^7–10 NAHR accounts for most of the recurrent rearrangements sharing a common size and genomic interval and having clustered breakpoints.¹¹ NAHR-mediated rearrangements have their breakpoints within low-copy repeats (LCRs), also known as segmental duplications (SDs), that lead to rearrangement by misalignment.¹² In contrast, non-recurrent rearrangements have variable sizes and unique breakpoints mediated by NHEJ or FoSTeS and/or MMBIR (FoSTeS/MMBIR). NHEJ bridges, modifies, and connects two broken ends with minimal or no microhomology.^13,14 FoSTeS is a replication-based mechanism resulting from a stalled or collapsed replication fork; the stalled strand disengages and invades a nearby replication fork by microhomology.^15,16 This process may occur multiple times and may lead to complex rearrangements.^15,16 FoSTeS has been further generalized as MMBIR, which provides molecular mechanistic details based on experimental studies in multiple model organisms.^15,17 Recently, several studies have shown that the majority of the non-recurrent, locus-specific, pathogenic rearrangements are mediated by a diversity of mutational mechanisms.^18–34 However, these mutational mechanisms are not yet well understood because most of these studies are based on only a limited number of sequenced junctions. Accordingly, our aim was to explore the mechanisms contributing to a large, unique set of locus-specific non-recurrent genomic rearrangements causing the genetic neurocutaneous disorder neurofibromatosis type 1 (NF1 [MIM: 162200]). NF1 is a common autosomal-dominant genetic disorder affecting 1 in 3,000 individuals worldwide and is caused by defects in the tumor suppressor NF1 (MIM: 613113).³⁵ The primary clinical features are multiple café-au-lait spots, skinfold freckling, multiple neurofibromas, Lisch nodules, and other less common manifestations such as optic glioma, tibial dysplasia, and specific tumors.³⁶ Through comprehensive mutation analysis, we have found that ∼2% of mutation-positive unrelated individuals carry intragenic deletions or duplications of one or more exons,³⁷ but the contribution of the above-stated mechanisms to intragenic copy-number variations (CNVs) within NF1 has not yet been studied. Here, genomic-DNA-based breakpoint-spanning PCR and array comparative genomic hybridization (aCGH) were used to clone the breakpoints at the nucleotide level of the NF1 intragenic CNVs in 85 unrelated subjects. We have characterized the likely rearrangement mechanism of these 85 CNVs along with two additional previously published NF1 CNVs.^38,39 Unlike recurrent rearrangements, mediated by flanking LCRs, NF1 intragenic CNVs originate by different rearrangement mechanisms and have diverse sizes and breakpoint positions. Furthermore, in addition to the loop within a 197-bp palindrome located in intron 40, four Alu elements located in introns 1, 2, 3, and 50 were identified as intragenic-rearrangement hotspots within NF1.

Subjects and Methods

Human Subjects and Clinical NF1 Mutational Analysis

Samples from subjects with NF1 intragenic CNVs (UAB-1 to UAB-84) were ascertained from the University of Alabama at Birmingham Medical Genomics Laboratory. In addition, breakpoints were identified in one Medical University of Vienna Clinical Institute for Medical and Laboratory Diagnostics subject (MUW-1) who was shown by multiplex ligation-dependent probe amplification (MLPA) to carry a deletion of exon 2. We also included two subjects with previously published NF1 CNVs with known breakpoints (IRO-1 and UNIMI-1) to investigate the nature and origin of their rearrangement mechanisms.^38,39 In total, 87 subjects with NF1 intragenic CNVs were investigated in this study. Subjects were originally referred for NF1 diagnostic testing. Through comprehensive mutation analysis with an RNA-core assay complemented with dosage analysis by MLPA and fluorescence in situ hybridization,⁴⁰ 134 individuals with an intragenic rearrangement have been identified out of a total ∼7,200 unrelated NF1-mutation-positive individuals. Sufficient material was available for further studies in 85/134 individuals with such intragenic CNVs, including 75 deletions and 10 duplications (Figure 1). All of the DNA samples were obtained from EDTA blood. An overview of the clinical features of the individuals is presented in Table S1. This study was approved by the University of Alabama at Birmingham institutional review board. In addition, a waiver of informed consent and patient authorization has been obtained for this study.

Figure 1.

Global View of 87 Identified NF1 Intragenic CNVs Including 77 Deletions and 10 Duplications

The genomic structure of NF1 is presented in UCSC Genome Browser hg19, and custom tracks show NF1 intragenic CNVs.

High-Resolution aCGH

The DNA of test samples and normal reference samples were labeled with Cy3 and Cy5 fluorophores, respectively, and hybridized to a custom-designed microarray (Agilent design ID number: 026000, GEO platform definition number: GSE64869). Alternatively, individual DNA was hybridized and compared to unrelated control DNA. We have designed an array encompassing the coding and non-coding repeat masked regions of NF1 and the regions both upstream and downstream of the NF1 locus. Potential transcription-factor binding sites and evolutionarily conserved elements within the NF1 locus and its vicinity were also included in the design. Probes associated with highly redundant sequences were excluded. The custom array design was submitted for production via the Agilent Technologies eArray website to be synthesized on 8 × 15 K array slides. Comparative genomic hybridization was carried out, according to the manufacturer’s recommendations, on a set of positive and negative control DNA samples. The arrays were scanned at 5-μm resolution with an Axon 4000B microarray scanner. Data extraction and normalization were carried out with Feature Extraction software v.9.5. Normalized data were then imported into DNA Analytics software and the ADM-2 algorithm was used to detect CNVs; this was followed by manual examination of the covered regions.

Breakpoint-Spanning PCR

NF1 (17q11.2) spans 282 kb comprising 60 exons interspersed with large introns ranging from 0.1 to 60.7 kb. When one or both of the breakpoints of the CNV in an individual were located in large introns such as intron 1 or intron 36 (over 60 kb), aCGH was performed to narrow down the breakpoint interval. For all other individuals, long-range PCR was used to clone the breakpoints directly. Primers for breakpoint-spanning amplification and sequencing were designed with Primer3. Primer information and PCR conditions are shown in Table S2. To assess the specificity of a rearrangement product, a normal control sample from a healthy individual was always analyzed along with the mutant sample. Breakpoint-spanning PCR was carried out with the Expand Long Template PCR System (Roche) and TaKaRa LA Taq Polymerase (TaKaRa Bio). PCR bands were visualized on a 2% agarose gel. PCR products were sequenced with an automated capillary sequencer (ABI 3730) and analyzed with Sequencing Analysis v.6.0 software (ABI; Life Technologies). PCR and sequencing results were confirmed independently by repeated experiments. Mutations were named according to the recommendations of the Human Genome Variation Society (HGVS) and with reference sequence GenBank: NM_000267.3.

In Silico Analysis of Junction Fragments

For every mapped breakpoint, we performed in silico analyses of the 300 bp of sequences flanking the breakpoints. Sequences were analyzed with BLAT via the UCSC Genome Browser. The breakpoint-junction sequences were aligned and microhomology was annotated. Reference sequences were obtained via the UCSC Genome Browser (hg19). Palindromic sequences were identified by Palindrome search. Quadruplex-forming G-rich sequences (QGRS) were identified by QGRS Mapper. Repetitive elements were uncovered by RepeatMasker. The sequence similarities between repetitive elements were evaluated by Blast2. Fisher’s exact test was performed to verify whether the genomic-rearrangement frequency within AluY (chr17: 29,484,696–29,484,993), AluSx (chr17: 29,477,768–29,478,080), AluSq2 (chr17: 29,489,213–29,489,460), and AluSx (chr17: 29,678,421–29,678,733) differed significantly in comparison with the frequency within whole NF1. The mutation effect on coding sequence was evaluated by Alamut Visual mutation analysis software, v.2.6.1 (Interactive Biosoftware).

Results

NF1 Intragenic CNVs Originate by Different Rearrangement Mechanisms and Have Diverse Sizes and Breakpoint Positions

To decode NF1 intragenic CNV rearrangement mechanisms, we sequenced and aligned all 75 deletion and 10 duplication breakpoints to the human genome (UCSC Genome Browser hg19; 5 deletions [from individuals UAB-59 to UAB-63] have been published previously).⁴¹ All of these 85 CNVs were submitted to LOVD2_NF1 (curated by Rick van Minkelen). Furthermore, we included two additional published NF1 CNVs (from IRO-1 and UNIMI-1).^38,39 We analyzed all 87 junction fragments for microhomology, potential to adopt non-B DNA structures, repetitive elements, and effect on coding sequence. We further inferred whether the CNV was present in mosaic form (versus constitutional) by evaluation of the cDNA and MLPA results and rearrangement mechanisms (Table 1, Tables S1 and S3, and Figure S1). Specifically, (1) the presence of normalized MLPA copy-number values for multiple consecutive probes of 0.70–0.85 (deletion) or 1.15–1.30 (duplication), (2) the presence of abnormal cDNA fragments, and (3) sequencing and breakpoint-identification results confirming the presence of deletion or duplication of the regions implied by MLPA were considered indicative of a mosaic CNV. No further quantification of the level of mosaicism was undertaken. Breakpoints were distributed from intron 1 to intron 57 and rearrangement size ranged from 0.3 kb to 198.1 kb with a mean of 35.7 kb, a median of 12.2 kb, and a distribution biased toward smaller rearrangements (Figure 1 and Figure S2). Unlike most recurrent rearrangements mediated by flanking LCRs, the NF1 intragenic rearrangements varied in size, location and rearrangement mechanisms.

Table 1.

The Rearrangement Mechanisms that Lead to NF1 Intragenic CNVs

	Recurrent (NAHR)	Simple Non-recurrent (FoSTeS × 1, NHEJ, and Serial Replication Stalling)	Complex (Multiple NHEJ and Multiple FoSTeS/MMBIR)	Other (Alu Insertion)	Total
Deletions	16 (21%)a	55 (71%)	5 (6%)	1 (1%)	77
Duplications	3 (30%)	7 (70%)	0	0	10

^aTwo instances of Alu-Alu-mediated NAHR were most likely followed by an AluY insertion.

Junction Sequences Reveal that the Predominant Mechanism Leading to NF1 CNVs Is Based on DNA Replication

All of the junction sequences were aligned to the reference sequence (hg19) by BLAT for evaluation of their nature and origin. Simple non-recurrent rearrangements with microhomology at the breakpoint junctions were found in 43/87 (49%) individuals, suggesting the predominant mechanism may be induced by DNA double-strand break followed by microhomology-mediated end joining. Furthermore, in 22/87 (25%) rearrangements, an insertion ranging from 1–810 bp was found between the breakpoints (Table S3). Four of these insertions were most likely mediated by multiple FoSTeS/MMBIR (2 FoSTeS/MMBIR × 2 [UAB-21 and UAB-48], 1 FoSTeS/MMBIR × 3 [UAB-50] and 1 FoSTeS/MMBIR × 5 [UAB-66]; Figure 2) and one by multiple NHEJ (UAB-49). Seven insertions consisted of a short stretch of nucleotides ranging from 3 to 31 bp; this stretch could be mapped to the reference sequence close to the breakpoint, suggesting serial replication stalling and re-replication process (Figure 3). Three insertions of ∼300 bp were mediated by AluY insertion (in UAB-9, UAB-47, and UAB-51; Table S3). Only five breakpoint junctions were connected with two blunt ends without any inserted sequence (in UAB-13, UAB-25, UAB-26, UAB-42, and UAB-74); another seven CNVs had short insertion with random nucleotides ($\le$5 bp) at the breakpoints (in UAB-17, UAB-27, UAB-33, UAB-38, UAB-39, UAB-43, and UAB-46), suggesting that the replication-independent NHEJ mechanism was used to ligate these broken ends (Table S3).

Figure 2.

Example of a Complicated Rearrangement Mediated by FoSTeS/MMBIR × 5 in UAB-66

The model (not to scale) illustrating the underlying FoSTeS/MMBIR × 5 process is shown at the top. The sequence context in each breakpoint junction is shown at the bottom. The sequence color corresponds to the genomic loci at the top. The red boxes indicate microhomology at the breakpoint junctions. The genomic coordinates were annotated next to the sequence content. During DNA replication, the stalled strand in intron 39 most likely invaded backward to intron 50, stalled again and invaded forward into intron 46, stalled again and invaded backward to intron 49, stalled again and invaded backward to intron 29, and stalled again and invaded forward into intron 46. This multiple FoSTeS/MMBIR process contributed to a complicated genomic rearrangement within NF1.

Figure 3.

Illustration of Likely Serial Replication Stalling and Re-replication Processes on NF1 Intragenic CNVs

Seven NF1 CNVs were most likely mediated by serial replication stalling (UAB-14, UAB-41, UAB-56, UAB-65, UAB-68, UAB-73, and IRO-1). Arrowheads represent the replication-strand direction. The nucleotides within the green arrowheads represent the sequence flanking the breakpoint, and the nucleotides within the gray arrowheads and the light-blue boxes represent the inserted nucleotides. Genomic coordinates (UCSC Genome Browser hg19) were annotated next to the arrowheads.

Repetitive Elements Contribute to Various Genomic Rearrangements

Repetitive elements present at the breakpoint junctions were identified with the RepeatMasker track in the UCSC Genome Browser. So far, the exact minimum length of sequence homology required for NAHR is still uncertain.¹⁰ Given that the shortest Alu element contributing to Alu-Alu recombination in this study is 135 bp (AluSp, UAB-11), we used 135 bp as a cutoff value for the following analysis of repetitive-elements recombination. Overall, 152 Alu and 89 partial L1 elements of $\ge$ 135 bp are present throughout the NF1 non-coding region, and these highly homologous repetitive elements in direct orientation might mediate NAHR in various combinations (Figure 4). Blast2 analysis was performed to determine the sequence similarity between repetitive elements in identical orientation and thus to infer likely NAHR combinations (Tables S4 and S5). Regarding L1-mediated NAHR, 89 L1 elements (23 in positive orientation and 66 in negative orientation) can theoretically generate 2,398 likely combinations. However, these L1 elements are truncated, and individual copies represent only a small portion of the full L1 ∼6-kb consensus sequence (mean of 520 bp, median of 382 bp). Furthermore, individual copies of the truncated L1 elements correspond to different segments of the native L1 consensus sequence. Therefore, the majority of these truncated elements are not viable partners for NAHR. Currently, the exact minimum length of sequence homology required for NAHR is still uncertain.¹⁰ Given that the shortest length of sequence homology between repetitive-elements-mediated NAHR is 105 bp in this study (UAB-6), we used 105 bp as the minimum length of sequence homology to evaluate likely NAHR events. Only 26/2,398 combinations with sequence homology $\ge$ 105 bp between partner L1 elements are present and are likely to mediate NAHR. Furthermore, only one of these 26 likely combinations has been identified as L1-mediated NAHR in this study (UAB-5). This L1-mediated NAHR has the longest identity length (1,038 bp) and highest sequence similarity (89%) of all of the likely L1-L1 combinations. On the other hand, there are 152 Alu elements of at least 135 bp (72 in positive direction and 80 in negative direction), and they could theoretically generate 5,716 different combinations. Unlike L1 elements however, 133/152 Alu elements comprise the nearly complete consensus sequence (∼300 bp). In total, 4,470/5,716 combinations with sequence similarity greater than 105 bp between partner Alu elements could theoretically mediate NAHR. In this context, Alu elements are more likely NAHR partners than L1 elements, which is consistent with the identification of 18 instances of Alu-mediated NAHR (UAB-1 to UAB-3, UAB-6, UAB-7, UAB-10 to UAB-12, UAB-16, UAB-30, UAB-44, UAB-47, UAB-51, UAB-72, UAB-75, UAB-76, UAB-83, and MUW-1; two instances of Alu-Alu-mediated NAHR were most likely followed by an Alu insertion [UAB-47 and UAB-51]) but only one of L1-mediated NAHR (UAB-5) in this study. In addition to being involved in NAHR, Alu elements were also involved in two NHEJ events (UAB-13 and UAB-17), seven FoSTeS/MMBIR × 1 or NHEJ rearrangements (UAB-4, UAB-15, UAB-18, UAB-70, UAB-77, UAB-84, and UNIMI-1), one FoSTeS/MMBIR × 2 rearrangement (UAB-48), one multiple NHEJ event (UAB-49), and three serial replication stalling events (UAB-14, UAB-41, and IRO-1), as well as one Alu insertion (UAB-9), suggesting their important role in NF1 CNVs. Therefore, in as many as 33/87 individuals, the origin for the CNV involved Alu elements. Although Alu insertions resulting from active de novo transposition have been reported in NF1,⁴² no Alu insertion together with NAHR has been reported. In this study, UAB-51 has an AluY inserted into intron 36, which thereafter most likely misaligned with an AluSq in intron 35, resulting in deletion of exon 36 (Figure 5). Similar to UAB-51, UAB-47 also has an AluY inserted into intron 30, which thereafter most likely misaligned with an AluSp in intron 29, leading to deletion of exon 30.

Figure 4.

The Repetitive-Element Distribution within NF1

Overall, 152 Alu and 89 L1 elements ≥ 135 bp in size are present throughout the NF1 non-coding region, and these highly homologous repetitive elements might mediate NAHR in various combinations. Exon numbers are shown in squares. The direction of the arrowheads denotes the orientation of the repeats. The asterisks represent the likely NAHR events mediated by these repetitive elements and observed in this study. These Alu elements do not demonstrate equal recombination ability. For example, 130/152 Alu elements did not contribute to any recombination in this study, whereas AluY (chr17: 29,484,696–29,484,993), AluSx (chr17: 29,477,768–29,478,080), AluSq2 (chr17: 29,489,213–29,489,460), AluSx (chr17: 29,678,421–29,678,733), AluSx (chr17: 29,484,027–29,484,306), AluSz (chr17: 29,488,772–29,489,078), and AluYe5 (chr17: 29,511,386–29,511,695) have contributed multiple Alu-Alu recombinations.

Figure 5.

Example of Alu-Alu-Mediated Recombination Following an Alu Insertion Event in UAB-51

The purple circles represent the exons involved in this rearrangement, and the genomic coordinates were annotated next to the Alu elements. The AluY element inserted into NF1 intron 36 can be perfectly mapped to 34 different locations in the human genome (BLAT, UCSC Genome Browser hg19), and therefore its original location cannot be determined. Subsequently, this inserted AluY might have misaligned and connected with AluSq (intron 35) and contributed to deletion of exon 36 (∼2.5 kb).

NF1 Intragenic Rearrangement Hotspots Are Located within a 197-bp Palindrome and Four Alu Elements

Non-B DNA structures are considered to be associated with DNA double-strand breaks.^43–45 Particularly, within intron 40 of NF1, a palindromic AT-rich repeat (PATRR17; 197 bp), especially the loop of the palindrome (7 bp), has been identified as an intragenic rearrangement hotspot.⁴¹ In addition to non-B DNA structures, Alu elements were also shown in this study to be involved in recurrent genomic rearrangements. AluY (chr17: 29,484,696–29,484,993) located in intron 2 of NF1, has mediated four NAHRs (UAB-1, UAB-2, UAB-3, and MUW-1). AluSx (chr17: 29,477,768–29,478,080) within intron 1 and AluSq2 (chr17: 29,489,213–29,489,460) within intron 3, as well as AluSx (chr17: 29,678,421–29,678,733) within intron 50, have each mediated three NAHRs (UAB-1 and 2, UAB-7, UAB-10, UAB-11, UAB-30, UAB-72, UAB-82, and MUW-1). Using Fisher’s exact test, we identified that these four Alu elements also constitute a significant intragenic rearrangement hotspot within NF1 among 87 intragenic CNVs (p value < 0.0001).

PATRR17 Has Demonstrated Diverse Rearrangement Mechanisms Leading to Deletions and Translocations

Regarding five previously published PATRR17-mediated deletions (UAB-59 to UAB-63), FoSTeS/MMBIR is the most likely rearrangement mechanism because all of the partner breakpoints are located within 7.1 kb upstream of PATRR17, and microhomology is present at all breakpoint junctions investigated.⁴¹ However, in this study we identified a PATRR17-mediated deletion (UAB-68) that has no microhomology at the breakpoint-junction site, and the partner breakpoint is located in the 6.7 kb downstream of PATRR17, suggesting NHEJ, rather than FoSTeS/MMBIR, was used to ligate the broken ends. Interestingly, UAB-68’s breakpoint is also located in the loop of PATRR17, which is consistent with previously published PATRR17-mediated deletions and translocation. Besides, the sequence (AT)₆AAT (15 bp) has been inserted between breakpoints that can partially be mapped to the PATRR17 loop or perfectly mapped to the PATRR17 arm downstream of the breakpoint, supporting a serial replication event (Figure 3). Given the sequence context within PARR17, the breakpoint cannot be perfectly mapped to reference sequence hg19 because of PATRR17 sequence variations; therefore, the genomic coordinates associated with PATRR17 are annotated as 296623xx in this study (Table S3, Figure 3, and Figure S1). In addition to playing a role in PATRR17-mediated deletion, a serial replication event is also involved in a PATRR17-mediated chromosomal translocation.⁴¹ Therefore, as a rearrangement hotspot within NF1, PATRR17 has demonstrated diverse rearrangement mechanisms leading to deletions and translocations.^41,46–49

Discussion

Alu Elements Do Not Demonstrate Equal NAHR Ability

Unlike L1 elements, the majority of the 152 Alu elements within NF1 are intact and could theoretically generate 5,716 different NAHR combinations. However, only 18 NAHR combinations mediated by 22 Alu elements have been found in this study. Among these 22 Alu elements, AluY (chr17: 29,484,696–29,484,993) was involved in four NAHRs; AluSx (chr17: 29,477,768–29,478,080), AluSq2 (chr17: 29,489,213–29,489,460), and AluSx (chr17: 29,678,421–29,678,733) were each involved in three NAHRs; AluSx (chr17: 29,484,027–29,484,306), AluSz (chr17: 29,488,772–29,489,078), and AluYe5 (chr17: 29,511,386–29,511,695) each mediated two NAHRs. Overall, these seven Alu elements have demonstrated a much stronger ability to mediate NAHR than any other Alu elements within NF1. Particularly, AluSx (chr17: 29,477,768–29,478,080) and AluY (chr17: 29,484,696–29,484,993) have mediated a recurrent NAHR, resulting in exon-2 deletion in three unrelated individuals (MUW-1, UAB-1, and UAB-2) with exactly the same breakpoints (chr17: 29,477,816–29,484,743), suggesting a strong recurrent Alu-mediated NAHR. The deletion in UAB-1 was proven to be absent in both unaffected parents; MUW-1 is from Austria and UAB-2 is from the US; therefore, we consider it unlikely that a founder effect could explain the recurrence of this deletion. Interestingly, palindromic sequences 5′-CCTGAGGTCAGG-3′ (chr17: 29,477,831–29,477,842) and 5′-TCCCAGCACTTTGGGA-3′ (chr17: 29,484,724–29,484,739) have been found close to the breakpoints, suggesting that NAHR was most likely used to repair the potential palindromic-sequence-mediated double-strand breaks. In addition, although Alu elements are present throughout NF1 (282 kb), 10/18 instances of Alu-mediated NAHR cluster in a 40.7-kb interval (chr17: 29,470,913–29,511-646) between intron 1 and intron 8, suggesting an Alu-mediated-NAHR hot zone within NF1. We propose that these Alu elements within this hot zone might be susceptible to DNA double-strand break, and their partner Alu elements could be in close 3D proximity for NAHR. This finding is consistent with LCR-mediated NAHR;⁵⁰ the frequency of Alu-mediated recombination is positively associated with the flanking Alu length and inversely influenced by the inter-Alu distance. In addition to NAHR, other non-recurrent CNVs were also more likely to be smaller rearrangements (Figure S2), suggesting that broken ends tend to choose partner ends in close proximity. Given that the exact minimum length of sequence homology required for NAHR is uncertain,¹⁰ recent studies have also proposed that Alu-mediated recombination could be the result of microhomology-mediated replication errors rather than NAHR.^21,23,34 However, given that the frequency of Alu-Alu recombination is positively associated with the sequence similarity between Alu elements, NAHR might be the more likely rearrangement mechanism accounting for Alu-Alu recombination. However, further functional studies may be needed to evaluate the exact rearrangement mechanism.

Replication-Based Mechanism Contributes to the Majority of NF1 Intragenic CNVs

It was previously thought that most non-recurrent genomic rearrangements are mediated by NHEJ.^29,51,52 However, we found that replication-based mechanisms such as FoSTeS/MMBIR, as well as serial replication stalling, might be the major mechanisms leading to NF1 intragenic CNVs. NHEJ is known to ligate two broken ends with limited or no microhomology and leaves an “information scar” including cleavage or insertion of several random nucleotides at the joining site.^10,14 In contrast, FoSTeS/MMBIR takes microhomology as an essential element for stalled-strand annealing to a nearby replication fork.^15,16 Here, we identified five breakpoint junctions connected through blunt ends and seven breakpoints with random nucleotides inserted ($\le$5 bp) at the breakpoints, consistent with an “information scar,” suggesting a replication-independent mechanism, NHEJ, was used to ligate broken ends. However, 43 simple non-recurrent rearrangements showed a microhomology at the breakpoint-junction site, suggestive of NHEJ or FoSTeS/MMBIR × 1 rearrangement mechanism. Alternatively, inserted sequences found between breakpoints might be the result of serial replication stalling rather than of NHEJ. In the current study, seven likely events of serial replication stalling and four multiple FoSTeS/MMBIR rearrangements were identified, further supporting that replication-based mechanisms contribute to the majority of the NF1 intragenic CNVs. The occurrence of such replication-based mechanisms might be underestimated. For example, Lazaro et al. identified an NF1 exon 40–48 deletion with an insertion between breakpoints and proposed that homologous or nonhomologous recombination contributed to this deletion (in IRO-1).³⁹ Interestingly, this insertion can be perfectly mapped to the reference sequence flanking the breakpoint, strongly indicating that serial replication stalling and re-replication could have led to this deletion instead (Figure 3). In addition to causing deletions, serial replication stalling also leads to insertions and translocations, further supporting a significant contribution of the replication-based mechanism to genomic rearrangements.^{20,41,53–55} Other studies also indicate that the high level of microhomology observed at the breakpoint junctions suggests presence of a replication-based mechanism and a mitotic origin.^20,23,30,56 Given that microhomologous sequences are widely distributed in the genome, in addition to investigating sequence context, exploring the DNA spatial proximity and 3D genomic architecture close to the junction site is likely to reveal which DNA interval is susceptible to double-strand break or replication-fork collapse, as well as the accessibility of microhomologous sequences to enzymes associated with FoSTeS/MMBIR.⁵⁶

The Rearrangement Mechanisms Leading to Disease Differ Greatly Among Genes

It was previously thought that most genomic rearrangements were formed randomly,⁵⁷ but more recent studies indicate that this is not the case. For example, as many as 70%–80% of DMD (MIM: 310200) mutations are intragenic CNVs.^58–61 Ankala et al. proposed that DMD (2.4 Mb long) has several active replication origins and termination junctions (as expected given its size), which may explain the high intragenic CNV frequency, greater than in any other known disease-associated genes, which are usually much smaller and with less or no intragenic replication origins.²⁰ Given that mammalian replicons on average span 75–150 kb,⁶² and the average size of a human gene is 27 kb,⁶³ failure of any single replication origin in such small genes could lead to multigenic rearrangements, rather than intragenic ones. However, the frequency of the individuals’ carrying CNVs in NF2 (95 kb long [MIM: 101000]) and VHL (10 kb long [MIM: 193300]) is ∼30%, whereas in NF1 (282 kb long) and PKHD1 (472 kb long [MIM: 263200]), the frequency is only ∼2% and ∼3%–4%, respectively.^37,64 Therefore, the gene size cannot serve as the sole predictor of the intragenic CNV frequency. Other factors, including genomic architecture, spatial proximity, and cellular stress, might also contribute to the genomic-rearrangement spectrum, given that the rearrangement mechanisms leading to disease differ greatly among genes (Table S6).

Intragenic Deletions Are More Prevalent than Duplications within NF1

A striking observation of the current study is that intragenic deletions are far more prevalent than intragenic duplications. Although both intragenic deletions and duplications are expected to be pathogenic and lead to NF1, we identified 119 intragenic NF1 deletions and only 15 intragenic duplications in this study. It is unlikely that this difference is due to bias of detection from the technique used in the course of the study, given that MLPA analysis was applied to all samples. MLPA performs equally well for detection of deletions as well as duplications.⁶⁵ In addition, our findings are in line with the results reported by Imbard et al. showing duplication only in 2 out of 22 NF1 CNVs, versus 20 deletions.⁶⁶ As we found for NF1, Quemener et al. also found intragenic deletions are far more prevalent than duplications in CFTR (MIM: 219700), BRCA1 (MIM: 604370), LDLR (MIM: 143890), MLH1 (MIM: 609310), MSH2 (MIM: 120435), SERPING1 (MIM: 106100), and TSC2 (MIM: 613254) and further proposed a likely deletion/duplication mutation ratio of between 2 and 3 in the human genome.³² Additionally, Girirajan et al. also found that the prevalence of the deletion is higher than the prevalence of duplications in 15,767 individuals assessed for developmental delay and associated phenotypes.⁴ This discrepancy between deletions and duplications could be associated with the genomic-rearrangement mechanisms. Interchromosomal and interchromatid NAHR between direct copies of LCRs contribute both to reciprocal deletion and duplication, while intrachromatid NAHR only leads to deletion.^6,10 Furthermore, non-recurrent rearrangements, such as NHEJ in an intrachromatid event, also only generate deletions.³² In this context, the frequency of deletions is likely to be higher than that of duplications.

Regarding duplications, Newman et al. has found that the majority of the duplications are tandem in direct orientation.⁶⁷ Consistent with Newman et al., our cDNA and genomic DNA results both indicate that all of these ten duplications are tandem in direct orientation at the original locus. These ten duplications comprise three NAHR (UAB-75, UAB-76, and UAB-83) and seven NHEJ or FoSTeS/MMBIR × 1 (UAB-77 to UAB-82 and UAB-84) events. Interchromosomal and interchromatid NAHR between direct copies of Alu elements most likely contributed to three duplications. The remaining seven duplications could be mediated by NHEJ. When a single double-strand break occurs on one strand, one of the broken ends can thereafter invade and replicate from the sister chromatid, resulting in duplication. Another scenario could be FoSTeS/MMBIR × 1. During DNA replication, the stalled strand disengages from the original replication fork and invades to another fork located upstream (backward invasion), which also would result in a duplication. We hypothesize that NHEJ and/or FoSTeS/MMBIR × 1 account for these seven duplications.

Detection of Low-Level Mosaic CNVs

We have used MLPA to detect or confirm (after cDNA-based fragment analysis and sequencing) all of the NF1 intragenic CNVs. Four CNVs, originally identified through cDNA-based fragment analysis and sequencing, could not be detected based on MLPA (UAB-14, UAB-16, UAB-19, and UAB-73). Although MLPA is commonly used to detect CNVs, this method is not suitable for detecting mosaicism present in fewer than 30%–40% of the cells.⁶⁸ However, detection of a low-level mosaicism has important implications for genetic counseling. Failure to detect an existing pathogenic NF1 mosaic variant in parental DNA is associated with a recurrence risk of up to 50% in the offspring. A sperm donor (UAB-37’s biological father), presenting with only four café-au-lait macules, has passed NF1 to multiple offspring, prior to being genetically evaluated for gonosomal mosaicism.⁶⁹ After comprehensive NF1 molecular evaluation with RT-PCR, MLPA, and aCGH, a deletion of the NF1 exons 15–29 was found in ∼20% of sperm cells.⁷⁰ This deletion was below the detection threshold of MLPA and aCGH in the blood cells, but was detected by RT-PCR.⁷⁰ In this study, this deletion was confirmed at the nucleotide level by breakpoint identification, showing the exact same breakpoints in the blood and sperm cells of the donor compared to the affected offspring, therefore confirming an NF1 gonosomal mosaic deletion in the donor.

Mosaic NF1 CNVs are not rare and were previously estimated at 9.6% on the basis of a group of 146 individuals carrying an NF1 total-gene deletion.⁷¹ NF1 intragenic CNVs were found in 21 non-founder (familial) individuals in this study, family history was unknown in 14/87 individuals, and mosaicism was observed in 15/52 of the sporadic (founder) individuals (Table S1). Therefore, 28.8% of the individuals with sporadic NF1 due to an intragenic CNV were found to have mosaicism. Diagnostic laboratories should realize the shortcomings of MLPA and/or aCGH as an approach to identifying mosaic CNVs in NF1 diagnostic testing. cDNA-based testing has a proven high sensitivity for detecting low-level mosaic intragenic deletions because primers spanning the deletion preferentially amplify the shorter RT-PCR fragments.

Conclusions

We demonstrate that the majority of the NF1 intragnenic CNVs are non-recurrent and mediated by DNA-replication-based mechanisms. Furthermore, in addition to the loop of a 197-bp palindrome located in intron 40, four Alu elements located in introns 1, 2, 3, and 50 have now been identified as intragenic-rearrangement hotspots within NF1. This locus-centered study based on a large set of breakpoint identifications provides a mechanistic perspective for NF1 molecular etiology and might also serve as a paradigm for other genetic disorders involving genomic rearrangements.

Published: July 16, 2015

Web Resources

The URLs for data presented herein are as follows:

• Blast2, http://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE_TYPE=BlastSearch&PROG_DEF=blastn&BLAST_PROG_DEF=megaBlast&BLAST_SPEC=blast2seq

• Leiden Open Variation Database, https://grenada.lumc.nl/LOVD2/mendelian_genes/home.php?select_db=NF1

• OMIM, http://www.omim.org

• Palindrome search, http://bioinfo.cs.technion.ac.il/projects/Engel-Freund/new.html

• Primer3 primer design program, http://primer3.ut.ee

• QGRS Mapper, http://bioinformatics.ramapo.edu/QGRS/index.php

• RepeatMasker, http://www.repeatmasker.org

• UCSC Genome Browser, http://genome.ucsc.edu