The Alu-Rich Genomic Architecture of SPAST Predisposes to Diverse and Functionally Distinct Disease-Associated CNV Alleles

Abstract

Intragenic copy-number variants (CNVs) contribute to the allelic spectrum of both Mendelian and complex disorders. Although pathogenic deletions and duplications in SPAST (mutations in which cause autosomal-dominant spastic paraplegia 4 [SPG4]) have been described, their origins and molecular consequences remain obscure. We mapped breakpoint junctions of 54 SPAST CNVs at nucleotide resolution. Diverse combinations of exons are deleted or duplicated, highlighting the importance of particular exons for spastin function. Of the 54 CNVs, 38 (70%) appear to be mediated by an Alu-based mechanism, suggesting that the Alu-rich genomic architecture of SPAST renders this locus susceptible to various genome rearrangements. Analysis of breakpoint Alus further informs a model of Alu-mediated CNV formation characterized by small CNV size and potential involvement of mechanisms other than homologous recombination. Twelve deletions (22%) overlap part of SPAST and a portion of a nearby, directly oriented gene, predicting novel chimeric genes in these subjects’ genomes. cDNA from a subject with a SPAST final exon deletion contained multiple SPAST:SLC30A6 fusion transcripts, indicating that SPAST CNVs can have transcriptional effects beyond the gene itself. SLC30A6 has been implicated in Alzheimer disease, so these fusion gene data could explain a report of spastic paraplegia and dementia cosegregating in a family with deletion of the final exon of SPAST. Our findings provide evidence that the Alu genomic architecture of SPAST predisposes to diverse CNV alleles with distinct transcriptional—and possibly phenotypic—consequences. Moreover, we provide further mechanistic insights into Alu-mediated copy-number change that are extendable to other loci.

Introduction

DNA copy-number variants (CNVs) deleting or duplicating parts of genes contribute to exon shuffling and gene evolution, thus playing an important role in both Mendelian and common/complex traits.^{1, 2, 3, 4} CNVs in SPAST (2p22.3 [MIM 604277]) cause 8%–41% of autosomal-dominant spastic paraplegia 4 (SPG4 [MIM 182601]),^{5, 6, 7} the predominant form of hereditary spastic paraplegia (HSP).^{8, 9} Heterozygous CNVs deleting (n > 70 cases) and duplicating (n = 2 cases) various SPAST exons or the entire SPAST gene have been described.^{10, 11} However, it is not known how these CNVs form, whether and/or why SPAST is particularly susceptible to these rearrangements, and what the consequences of SPAST CNVs, either molecular or phenotypic, are in comparison to other types of mutations in the same gene.

Multiple molecular mechanisms have been described or proposed to generate small CNVs^{12, 13} commensurate with the genic/exonic scale of many or most rearrangements described in SPAST.¹⁴ Insights into the mechanism(s) of SPAST CNV formation, however, are largely absent because of methodological limitations: CNVs in SPAST are usually studied with multiplex ligation-dependent probe amplification (MLPA),⁶ a method that reveals the copy number of each of the 17 SPAST exons but leaves the exact location, sequence, and characteristics of CNV boundaries (i.e., breakpoint junctions) unknown.

We previously ascertained the nucleotide sequence of breakpoint junctions for three SPAST CNVs deleting the final exon of the gene.¹⁴ Although the size and genomic position of each deletion differed, all deletion breakpoints localized to pairs of directly oriented Alu elements, a repetitive DNA element of ∼300 base pairs (bp) present in >10⁶ copies in the human genome.¹⁵ This Alu-Alu-mediated rearrangement phenomenon has been reported in studies of small, germline CNVs at other disease loci,^{16, 17, 18, 19, 20} somatic rearrangements found in malignancies,^{20, 21} and rearrangements arising during human evolution.^{22, 23, 24} Such variation has generally been attributed to Alu-Alu recombination, a proposed molecular mechanism in which homologous recombination (HR) utilizes nonallelic Alu pairs as substrates to generate CNVs.^{20, 25} Intriguingly, the CNVs in our small cohort exhibited features potentially inconsistent with formation by ectopic homologous recombination: one CNV was potentially complex, and breakpoint Alu pairs shared <90% identity. The remaining four SPAST CNVs with known breakpoints do not localize to Alus.^{11, 26, 27, 28} Therefore, we sought to further investigate the mechanism(s) generating SPAST CNVs, including the way in which the Alu-rich architecture of SPAST¹⁴ might render this gene or particular exons within it susceptible to CNV mutations.

SPG4, the clinical phenotype associated with SPAST mutation, is predominately characterized by progressive lower-limb weakness and spasticity caused by degeneration of upper motor neurons in the corticospinal tract.²⁹ Occasional reports of additional phenotypes suggestive of CNS pathology exist,^{30, 31, 32} and subtle cognitive impairment has been suggested to be a moderately penetrant feature of the disease (reviewed by Dürr et al. in GeneReviews). Furthermore, severity and age of onset can vary. No consistent genotype-phenotype correlation exists for alleles in SPAST.³³ Most disease-causing variants have been thought to represent loss-of-function alleles. However, we previously hypothesized that a subset of SPAST CNVs—those overlapping the 3′ end of the gene—might also affect expression of SLC30A6 (MIM 611148), the nearest gene to SPAST, located 8 kb downstream.¹⁴ SLC30A6 encodes a zinc transporter, ZnT6. Although SLC30A6 variants have not been found in association with any disease, zinc transporter (including ZnT6) dynamics are altered in the brains of some individuals with Alzheimer disease.^{34, 35} Thus, we investigated the possibility of SPAST CNVs dysregulating SLC30A6 expression, particularly given reports of individuals with comorbid spastic paraplegia and dementia.^{31, 36}

We studied SPAST CNVs in an additional 51 subjects and now report diverse CNVs for which Alu-mediated formation predominates. These CNVs rearrange a variety of exons including 24 (47%) affecting the final (3′) exon of the gene and potentially forming fusion transcripts with or altering expression of SLC30A6 and other adjacent genes. In cultured lymphoblastoid cells from one subject with a SPAST exon 17 deletion, we identify multiple SPAST:SLC30A6 fusion transcripts. Finally, we propose alternative models to homologous recombination for Alu-mediated CNV formation based upon the observed breakpoint features of CNVs in SPAST; these are extendable to other loci, potentially enabling genome-wide predictions of Alu-mediated genome instability.

Subjects and Methods

Subjects

Subjects consisted primarily of individuals with a clinical diagnosis of spastic paraplegia and a previous finding of a SPAST CNV identified during clinical laboratory screening or a family-based research study (Table S1 available online). This study was approved by the Institutional Review Board for Human Subjects Research of Baylor College of Medicine (H-28128, H-25466, H-9170, H-29697). Remaining DNA samples were deidentified for our analyses. Subjects are further described in Tables 1 and S1. Owing to the anonymized nature of sample acquisitions, phenotypic details beyond a clinical diagnosis of spastic paraplegia are unknown in most cases. Samples from Athena Diagnostics were intentionally enriched for exon 17 CNVs. All other sources contributed all available CNVs in their collections.

Table 1.

Subjects and SPAST CNVs

Subject	Del versus Dup	SPAST Exon(s) Affected (of 17)	Other Genes Deleted or Duplicated	CNV Size (kb)	Spastic Paraplegia	Origin	Novel Rearrangementa	Notes
A1	del	17	–	5.5	y	USA	–
A2	del	17	–	14.4	y	USA	–	complex
A3	del	17	SLC30A6 ex 1–5 (of 15)	24.6	y	USA	–
A4	del	17 (partial)	–	2.7	y	USA	y
A5	del	17	SLC30A6 ex 1–2 (of 15)	18.4	y	USA	–
A6	del	17	SLC30A6 ex 1–2 (of 15)	25.7	y	USA	–
A8	del	17 (partial)	–	2.7	y	USA	y
A9	del	17 (partial)	–	2.7	y	USA	y
A10	dup	13–15	–	6.2	y	USA	y	tandem, directly oriented duplication
BAB 3166	del	13	–	1.3	n (at age 16)	USA	NA	reported in Boone et al.26
BAB 5112	del	5–17	SLC30A6, NLRC4, YIPF4, BIRC6, MIR558, MIR4765, and TTC27 ex 1–5 (of 20)	559.3	y	USA	y
BAB 3200	del	1–7	DPY30 ex 1–3 (of 5)	87.2	y	USA	NA	complex; reported in de Ligt et al.27
BAB 3327	del	17	–	5.5	y	USA	–
SPAST_1	del	17 (partial)	–	2.7	?b	USA	y
SPAST_3	dup	2–17	SLC30A6, NLRC4, YIPF4, and BIRC6 ex 1–46 (of 74)	432.2	?b	USA	y	tandem, directly oriented duplication
Spain 1	del	5–16	–	43.4	y	Spain	y
Spain 2	del	13–17	SLC30A6, NLRC4, YIPF4, MIR558, and BIRC6 ex 1–66 (of 74)	455.0	y	Spain	y
Spain 3	del	3–17	SLC30A6 ex 1–5 (of 15)	95.7	y	Spain	–
Spain 4	del	8–17	SLC30A6 ex 1–5 (of 15)	60.7	y	Spain	–
Spain 5	del	1	–	12.6	y	Spain	–
Spain 6	del	2–17	SLC30A6 and NLRC4 ex 5–9 (of 9)	170.1	y	Spain	–
Spain 7	del	10–16	–	16.0	y	Spain	–
Spain 8	del	2–7	–	48.9	y	Spain	y
Spain 9	del	6–7	–	5.0	y	Spain	–
JKF B	del	16	–	2.0	y	USA	–
JKF C	dup	9	–	6.7	y	USA	y	tandem, directly oriented duplication
Ruhr 3	del	2–9	–	65.4	y	Germany	–
Ruhr 4	del	1	–	20.5	y	Germany	–
Ruhr 5	del	2–9	–	65.4	y	Germany	–
Ruhr 6	del	5–6	–	8.3	y	Germany	–
Ruhr 7	del	17 (partial)	–	5.1	y	Germany	y
Ruhr 9	del	1	–	4.0	y	Germany	–
Ruhr 10	del	1	–	4.0	y	Germany	–
Ruhr 11	del	2–7	–	33.5	y	Germany	y
Ruhr 12	del	1	–	16.3	y	Germany	–
Ruhr 13	del	8–17	SLC30A6 ex 1 (of 15)	44.8	y	Germany	–
Ruhr 14	del	16–17	SLC30A6 ex 1–4 (of 15)	31.7	y	Germany	–
Ruhr 15	del	2–12	–	72.9	y	Germany	y
Ruhr 16	del	5	–	1.9	y	Germany	–
Ruhr 17	del	9–17 (partial)	–	27.5	y	Germany	y
Ruhr 18	del	8–17	SLC30A6 ex 1 (of 15)	44.8	y	Germany	–
Ruhr 21	del	10–17 (partial)	–	20.9	y	Germany	y
Ruhr 23	del	17	–	17.6	y	Germany	–
Ruhr 26	del	13–16	–	7.3c	y	Germany	–
Ruhr 27	del	17	SLC30A6 ex 1–2 (of 15)	24.1	y	Germany	–	complex
Ruhr 28	del	1 (partial)	–	16.5	y	Germany	–
Ruhr 29	del	8–17	SLC30A6, NLRC4, YIPF4, BIRC6, MIR558, MIR4765, TTC27, LINC00486, LOC100271832, LTBP1, 5S_rRNA	1,283.9	y	Germany	–
Ruhr 31	del	8–9	–	9.3	y	Germany	–
Ruhr 32	del	2–13	–	59.4c	y	Germany	y
Park 1	del	5–7	–	11.7	y	Korea	–
Park 2	del	5 (partial)–7	–	8.1	y	Korea	y
Park 3	del	8–9	–	8.9	y	Korea	–
Park 4	del	4–17	SLC30A6 ex 1–9 (of 15)	100.5	y	Korea	–
A37	del	17	–	10.8	y	USA	NA	reported in Boone et al.14
A38	del	17	–	16.8	y	USA	NA	complex; reported in Boone et al.14
A39	del	17	–	5.5	y	USA	NA	reported in Boone et al.14
Iwanaga	del	1 (partial)	–	2.3	y	Japan	NA	reported in Iwanaga et al.28
Mitne-Neto	dup	10–12	–	4.0	y	Brazil	NA	reported in Mitne-Neto et al.11; tandem, directly oriented duplication
Miura	del	1–4	DPY30 ex 1–3 (of 5)	69.8	y	Japan	NA	reported in Miura et al.40

Abbreviations are as follows: ex, exon; y, yes; n, no; NA, not applicable; –, no or none.

^aBased on SPAST exon or combination of exons rearranged and ploidy (del versus dup). Rearrangement of part of an exon is considered to be distinct from rearrangement of the whole exon. Previously published CNVs derived from Loureiro et al.,⁵ Álvarez et al.,⁹ Boone et al.^14,26 Sulek et al.,⁴¹ and Myers et al.⁴²

^bPhenotype unknown.

^cCNV breakpoint could not be sequenced. CNV length is an average of the minimum and maximum CNV interval identified by aCGH.

Mapping CNV Breakpoints

DNA was extracted from whole blood using the Gentra Puregene Blood Kit (QIAGEN) or obtained from previously archived samples. A total of 50 CNVs in SPAST were confirmed and mapped using a custom comparative genomic hybridization (CGH) array targeted to SPAST and flanking regions (approximately one megabase 5′ and 3′). This Agilent 8 × 15 k probe oligonucleotide array (format = G4427A; design ID = 037541) was custom designed using Agilent eArray and contains approximately one probe per 200 base pairs (bp) throughout the targeted region. Array CGH procedures have been described.³⁷ Hybridization controls were gender matched unless an individual’s gender was unknown, in which cases female DNA (HapMap individual NA15510) was used. Array image files were processed using Agilent Feature Extraction software (v.10) and CNVs called with Agilent Genomic Workbench software (v.7). One CNV (Ruhr 29) was mapped using a genome-wide Agilent 1M probe oligonucleotide CGH array (format = G4824A; design ID = 02159) because the CNV extended beyond the boundaries of our targeted array.

CNVs were further confirmed and mapped by long-range PCR with primers (Table S2) spanning their predicted breakpoints. PCR reagents and concentrations have been described previously.³⁸ The thermal cycler was programmed as follows: 94°C × 1 min; 30 cycles of 94°C × 30 s followed by 68°C × 7 min; 72°C × 10 min. Breakpoint PCR products were treated with ExoStar (GE Healthcare) according to the manufacturer’s instructions or extracted from an agarose gel using the Zymoclean gel DNA recovery kit (Zymo Research), also according to the manufacturer’s instructions, then sequenced by Sanger di-deoxynucleotide sequencing (Lone Star Labs; SeqWright; Baylor College of Medicine Sequencing Core, Houston, TX, USA).

All genomic coordinates are based on the February 2009 genome build (GRCh37/hg19) unless otherwise specified. SPAST exon numbering is based on UCSC transcript variant uc002roc.3 (17 exons). SLC30A6 exon numbering is based on UCSC transcript variant uc002rof.2 (15 exons).

STR Analysis

Alleles at 15 short tandem repeats (STRs) and subject gender were assessed using the Applied Biosystems AmpFISTR Identifiler Direct PCR Amplification Kit (Life Technologies) according to the manufacturer’s instructions.

Haplotype Analysis

For each deletion found in multiple individuals, primer pairs (Table S2) were designed to amplify 5–10 kb flanking each side of the deletion. The deleted allele was selectively amplified or was isolated based on size by gel extraction as described above. Common SNPs (allele frequency ≥ 1% in dbSNP 138) within these amplicons were genotyped by Sanger sequencing using a custom set of sequencing primers (Table S2). To assess the frequency of the observed haplotypes in the population, we obtained 1,984 phased haplotypes from the 1000 Genomes Project.³⁹

Cell Lines

Lymphoblasts were isolated from whole blood collected in acid citrate dextrose (ACD) tubes and transformed with Epstein-Barr virus to make lymphoblastoid cell lines (LCLs) by the Baylor College of Medicine Intellectual and Developmental Disabilities Research Center Tissue Culture Core. LCLs were cultured in GIBCO RPMI 1640 media with 20% FBS, then 10% FBS, and 1% antimycotic/antibiotic (all from Life Technologies) in 5% CO₂.

RNA Extraction and cDNA Synthesis

Total RNA was extracted from confluent, transformed lymphoblastoid cells using TRIzol (Invitrogen). RNA was cleaned using the RNeasy kit (QIAGEN) and quantitated with a NanoDrop ND-1000 spectrophotometer (NanoDrop). Complementary DNA (cDNA) was synthesized using qScript cDNA SuperMix (Quanta Biosciences) according to the manufacturer’s instructions (input = 1 μg RNA). cDNA quality was assessed via gel electrophoresis (not shown). A mock cDNA synthesis reaction performed without the reverse transcriptase mixture served as a negative (“No RT”) control during cDNA PCR.

PCR and Sequencing of cDNA

cDNA (2 μl) was used as PCR substrate in a 25 μl reaction mix containing 0.5 μM of each primer (Table S2) and 1× Promega PCR Master Mix (Promega). Thermocycler conditions were as follows: 95°C × 2 min; 40 cycles of 95°C × 30 s, 57°C × 30 s, 72°C × 3 min; 72°C × 5 min. PCR products were sequenced by Sanger di-deoxynucleotide sequencing (Lone Star labs; Baylor College of Medicine Sequencing Core, Houston, TX, USA) using the above cDNA PCR primers.

Quantitative RT-PCR

Quantitative RT-PCR (qPCR) assessing the expression level of SLC30A6 was performed using the TaqMan assay (Life Technologies). The Hs01071782_m1 probe set was used for SLC30A6 (spans the boundary of exon 9 and exon 10). The Hs02758991_g1 probe set was used for GAPDH (control gene [MIM 138400]; probe set spans the boundary of exon 6 and exon 7). The PCR amplification step was performed with 1 μl cDNA utilizing TaqMan Fast Advanced Master Mix (Life Technologies) and the 7900HT Fast Real-Time PCR System (Applied Biosystems). Three technical replicates were performed for each sample (subject or control).

The comparative Ct method was used to generate the relative quantification (ΔCt) of the expression level of SLC30A6 over GAPDH in an individual. The distribution of the ΔCt for SLC30A6 was established by the control samples. The SLC30A6 ΔCt for the subject was compared to this distribution using the empirical cumulative distribution function.

Results

SPAST CNVs Exhibit Diverse Sizes, Genomic Positions, and Exons Rearranged

We analyzed 51 heterozygous CNVs in SPAST (48 deletions and 3 duplications), predominately identified in individuals with a diagnosis of spastic paraplegia (Table 1). CNVs were most commonly discovered by multiplex ligation-dependent probe amplification (MLPA; data not shown). MLPA indicates only the exon(s) deleted or duplicated by a CNV, not its boundaries (breakpoints), so we fine-mapped each rearrangement using array comparative genomic hybridization (aCGH). Array CGH confirmed each CNV (Figures 1 and S1) and determined the approximate size, extent, and genomic content of each rearranged interval, enabling junction-spanning (breakpoint) PCR and sequencing reactions to be performed. PCR amplified 50 of 51 (98%) CNV breakpoints (Figure S2), and sequencing delineated the breakpoint sequence and exact coordinates of 49 (98%) of these rearrangements (Figure S3; 46 deletions and 3 duplications).

Figure 1.

Targeted aCGH Confirms and Maps CNVs in SPAST

Array CGH log₂ plots are aligned with models of SPAST and nearby genes. Very small CNVs are indicated by arrows. Coordinates refer to chromosome 2. No SPAST CNV was identified in either blood- or cell-culture-derived DNA from the control individual, BAB 3099.

A diversity of CNV alleles were identified (Tables 1 and S1, Figure 2A). To date, 40 different combinations of SPAST exons have been reported to be deleted in individuals with spastic paraplegia, and 2 duplicated;^{5, 9, 14, 40, 41} we now contribute 12 novel rearrangements of SPAST exons (Table 1), including 2 novel duplications.

Figure 2.

CNVs in SPAST Rearrange Diverse Exons and Have Breakpoints Predominately Localizing to Pairs of Alus in Direct Orientation

(A) CNVs are plotted by exact coordinates obtained by sequencing (black bars) and include both duplications (denoted “Dup”) and deletions (all others). Repetitive elements at CNV breakpoints are indicated by colored shapes and aligned with the RepeatMasker track (bottom, via UCSC Genome Browser) by colored lines. The genomic orientation of Alus is denoted by a right-facing (+) or left-facing (−) pentagon. Of 59 breakpoints, 39 (66%) localize to directly oriented Alu pairs. Note that BAB 3166 was identified in a young individual with no current spastic paraplegia²⁶ and that the phenotypes of SPAST_1 and SPAST_3 are unknown; thus, these three CNVs might or might not be pathogenic. ^∗Complex CNV (these CNVs are plotted as simple CNVs with only the first and final breakpoint repeat indicated). The breakpoints of Ruhr 26 and Ruhr 32 were not found, so these CNVs are not included.

(B) A moving average of Alu density (calculated in overlapping blocks of ten Alu elements) is aligned with (A), demonstrating that SPAST is overall an Alu-rich gene with “neighborhoods” of even higher Alu density. Alu data were downloaded from RepeatMasker.

(C) Zoomed-out plot of Alu density over a ∼1.1 Mb section of chromosome 2.

Of the CNVs we report, 36 (71%) occur in a single subject, and five groups of individuals had identical (shared regions of breakpoint microhomology) SPAST CNVs: Ruhr 9 and Ruhr 10 (del exon 1); Ruhr 3 and Ruhr 5 (del exons 2–9); Ruhr 13 and Ruhr 18 (del exons 8–17 and exon 1 of SLC30A6); A4, A8, A9, and SPAST_1 (partial del exon 17); and A1 and BAB 3327 (del exon 17; also identical to A39 from Boone et al.¹⁴). Identical CNVs in multiple subjects suggests either recurrent CNV formation or that individuals within a group are related or are the same individual. To investigate these possibilities, we analyzed 15 short tandem repeat alleles (STRs) and gender in these individuals (Table S3). No subjects sharing an identical CNV were monozygotic twins or the same individual, and in only one case (A1 and BAB 3327) were two such individuals consistent with a parent-child pair. To identify whether identical CNVs are observed on the same haplotype (i.e., potentially identical by descent) or on different haplotypes (i.e., identical by state, suggesting recurrent events), common SNPs flanking the CNVs were genotyped to establish a local haplotype. No flanking SNPs differ among individuals sharing the same CNV, thus failing to provide specific evidence for recurrence (Figure S6). For CNVs shared between Ruhr 13 and Ruhr 18, the haplotype surrounding the CNV is uncommon (5.1%), making recurrent events particularly unlikely. In other groups, the CNVs occurred on the most common haplotype.

We combined data from the 49 CNV breakpoint junctions delineated above with data from five SPAST CNVs we reported previously^{14, 26, 27} and three additional published SPAST CNVs,^{11, 28, 40} totaling 57 rearrangements and representing every SPAST CNV >30 bp for which exact breakpoint coordinates have been identified (Tables 1 and S1; Figure 2A). These CNVs range from 1,340 bp to 1,283,930 bp in size, with a mean of 73.1 kb, a median of 16.5 kb, and a distribution skewed toward smaller rearrangements (Figure 3). Thus, the majority of SPAST CNVs are of modest size when compared with the gene itself (94 kb).

Figure 3.

CNV Size: Skew toward Small Rearrangements and Association with Breakpoint Repeat Architecture

SPAST CNVs, plotted in ascending size order. Most SPAST CNVs are small (median size = 16.5 kb). CNVs flanked by Alu pairs (gray) are significantly smaller than remaining CNVs (green) (Kruskal-Wallis rank sum test, p = 0.00856; identical CNVs in multiple individuals [e.g., Ruhr 3 and 5] are collapsed to single entries, and CNVs published by other authors [which may be biased toward small, non-Alu-Alu-mediated events] are removed). ^∗Complex CNV. BAB 5112 is LINE-LINE mediated. CNVs <25 kb are magnified in the inset to show detail. The size of complex CNVs is rounded to the nearest 0.1 kb.

All known SPAST duplications (Table 1, Figure 2) are tandem and directly oriented. We previously noted that SPAST duplications are significantly underrepresented when compared with deletions among subjects with SPG4,¹⁴ a trend that continues in the data presented here (Table 1). The pathogenic duplications in subjects A10 (exons 13–15) and Mitne-Neto (exons 10–12) are both expected to shift the reading frame; interestingly, the pathogenic duplication in JKF C (exon 9) is predicted to be in-frame.

CNV Breakpoints Predominately Localize to Pairs of Alu Elements in Direct Orientation

CNV boundaries, or breakpoints, occasionally preferentially localize to repetitive genomic elements, which can play a role in mediating CNV formation.^{13, 18} Repetitive elements present at the breakpoints of SPAST CNVs are shown in Figure 2 and listed in Table 2. Of 57 sequenced SPAST CNV breakpoints, 39 (68%) localize to directly oriented pairs of Alu repeats. Such CNVs demonstrate breakpoint “signatures” consistent with Alu-mediated CNV formation (i.e., the rearranged genome contains a complete hybrid Alu, each “half” of which is derived from one of the two breakpoint Alus that flanked the deleted or duplicated region prior to its loss or gain, respectively;²² Figure S3).

Table 2.

Repeats Present at CNV Breakpoints^a

Subject	Left Breakpoint Repeat	Right Breakpoint Repeat	Microhomology (bp)	% Identity of Repeatsb
A1	AluSxc	AluY	26	81.7
A2c	AluSx	AluSx1	complex	complex
A3	AluSx1c	AluY	7	86.9
A4	AluSx1c	AluSp	15	84.8
A5	AluSx1c	AluSp	9	88.6
A6	AluY	AluSx3c	21	84.6
A8	AluSx1c	AluSp	15	84.8
A9	AluSx1c	AluSp	15	84.8
A10	AluSz6	AluSg	31	80.1
BAB 3166	–	–	6	–
BAB 5112	L1PA5	L1PA7	21	90.6
BAB 3200	AluSc	–	complex	–
BAB 3327	AluSxc	AluY	26	81.7
SPAST_1	AluSx1c	AluSp	15	84.8
SPAST_3	–	–	4	–
Spain 1	L1PA5	AluJb	4	–
Spain 2	–	AluSp	2	–
Spain 3	AluSz6	–	2	–
Spain 4	AluSq	L1M4	8	–
Spain 5	AluSz	AluSz	26	80.6
Spain 6	AluSg4	–	1	–
Spain 7	AluSc	AluY	15	84.6
Spain 8	AluSg4	AluYc	9	83.7
Spain 9	–	–	6	–
JKF B	AluSg	AluY	14	82.1
JKF C	AluSz	AluSxc	35	82.8
Ruhr 3	AluJo	AluJb	19	77.6
Ruhr 4	AluSz	AluY	25	75.8
Ruhr 5	AluJo	AluJb	19	77.6
Ruhr 6	Tigger4a	AT_rich	1 bp ins	–
Ruhr 7	AluSp	AluSp	8	87.9
Ruhr 9	AluSx3	AluSzc	5	82.7
Ruhr 10	AluSx3	AluSzc	5	82.7
Ruhr 11	AluSx1	AluSz	38	82.7
Ruhr 12	AluSx	AluSx3	9	77.4
Ruhr 13	AluSz6	AluSp	24	84.4
Ruhr 14	AluJb	AluSzc	18	77.6
Ruhr 15	AluJo	AluY	22	79.9
Ruhr 16	AluSx	AluSp	20	79.1
Ruhr 17	–	–	3	–
Ruhr 18	AluSz6	AluSp	24	84.4
Ruhr 21	AluY	AluSp	33	83.2
Ruhr 23	AluY	AluSz	11	85.6
Ruhr 26d	NA	NA	NA	NA
Ruhr 27	AluY	AluSp	complex	complex
Ruhr 28	–	L1M4	4	–
Ruhr 29	L1MC2	L1M5	0	–
Ruhr 31	AluYc	AluY	5	90.7
Ruhr 32d	NA	NA	NA	NA
Park 1	AluSx1	AluSp	7	80.7
Park 2	–	FLAM_C	3	–
Park 3	AluYc	AluSxc	8	81.1
Park 4	–	AluSq	4	–
A37	AluY	AluY	8	89.4
A38	AluY	AluSx3	complex	complex
A39	AluSxc	AluY	26	81.7
Iwanaga	–	AluSx3	1	–
Mitne-Neto	–	L1M7	0	–
Miura	AluSx1	AluJb	9	77.1

Left, telomeric; right, centromeric; –, not applicable or no repeat present.

^aAll Alu-Alu and LINE-LINE pairs are in matching genomic orientation (both negative or both positive) (Table S1, Figure 2A), with the exception of Ruhr 29.

^bNumber of identical bases divided by smaller repeat length. Comparison by Needleman-Wunsch Global Alignment algorithm.

^cRepeat contains the consensus 13-nt sequence associated with meiotic recombination hotspots, some NAHR breakpoint regions, and bound in vitro by PRDM9.⁴², ⁴³, ⁴⁴ The Alu from which the inserted sequence in the complex A2 rearrangement derives contains this sequence.

^dBreakpoint unable to be sequenced (Ruhr 26) or failed breakpoint PCR (Ruhr 32).

Breakpoint Alu pairs exhibit variable sequence identity (range = 75.8%–90.7%), possess differing numbers of bases of microhomology at the hybrid junction (range = 5–38 bp), and comprise distinct Alu families (Table 2). Of the 39 breakpoint Alu pairings, 33 (85%) consist of Alus from different families. Several Alus are present at the breakpoints of multiple CNVs. These data are consistent with analyses of multiple Alu-mediated CNVs at additional disease loci (Table 4). The repeat present at a breakpoint of the greatest number of different CNVs is an AluY (Spain 7, A38, Ruhr 23, Ruhr 27, A6), as observed previously.¹⁶ Breakpoint crossover/microhomology locations occupy various positions within the Alu substrate pairs (Figure S4).

Table 4.

Properties of Alu-Mediated CNVs Derived from Large Studies of CNVs at Specific Loci

Study	Gene	Syndrome	CNVs Studied (Unique)	Dups	Alu-Mediated CNVs	Average Breakpoint Alu Pair % Identity (Range)	Average Breakpoint Microhomology (Range)	Complex Alu-Mediated CNVs	Alus in Pair from Same Family	CNV size Avg. (Range)	Some Repeats Used in More than One CNV?
Present	SPAST	Spastic paraplegia 4	57 (44)	4	39/57 (68%)	82.7% (75.8%–90.7%)	17 bp (5–38 bp)	3/39 (8%)	6/39 (15%)	19.0 kb (1.9–73 kb)	yes
Frank et al.16	VHL (MIM 608537)	Von Hippel-Lindau syndrome	33 (32)	0	29/33 (88%)	not reported	16 bp (2–45 bp)	2/29 (7%)	15/29 (52%)	38.2 kb (0.8–249 kb)	yes
Kuiper et al.68	EPCAM (TACSTD1 [MIM 185535]; upstream of MSH2 [MIM 609309])	Lynch syndrome	45 (19)	NAa	45/45 (100%)	81% (75%–91%)	15 bp (3–32; one 0 bp)	2/19 (11%)	5/19 (26%)	11.1 kbb (2.4–23.8 kb)	yes

^aOnly deletions were studied.

^bDeletions affecting MSH2 were specifically excluded, limiting the potential size of CNVs.

The eight largest CNVs are not Alu mediated (Figure 3). To determine statistically whether CNV size correlates with the presence or absence of Alus at CNV breakpoints, we collapsed CNVs found in multiple individuals (e.g., Ruhr 3 and 5) into single entries and removed the three SPAST CNVs described by other authors, which may be biased toward small, non-Alu-mediated events because such CNVs are more readily mapped and sequenced. Among this data set, CNVs flanked by Alu pairs are significantly smaller than non-Alu-mediated CNVs (Figure 3; Kruskal-Wallis rank sum test, p = 0.00856).

Three CNVs with proximal and distal breakpoints in Alus are potentially complex (A38, A2, and Ruhr 27). A38 has been described¹⁴ and is consistent with formation by two or three template switches during DNA replication or by NAHR and a novel, familial Alu polymorphism. The origin of A2 is consistent with two template switches or a small (11 bp) insertion. The origin of Ruhr 27 is consistent with two template switches, two distinct NAHR events, or an NAHR event and a novel, familial Alu polymorphism. It is unknown whether any of these CNVs is de novo, and parental DNA is unavailable.

In addition to the 39 CNVs in our cohort with pairs of Alus at breakpoints, other signatures are present in 32% of junctional sequences. These include 13 instances of microhomology not localizing to repeat pairs, two apparently blunt end “clean breaks,” one LINE-LINE recombination (BAB 5112), one 1 bp insertion, and one potentially complex CNV (BAB 3200) (Figure S3).

Some Breakpoint Alus Contain a 13-nt Consensus Motif Enriched at Meiotic Recombination Hotspots

A 13-nucleotide motif (consensus = CCNCCNTNNCCNC) has been delineated and found in proximity to and in association with >40% of meiotic recombination crossover hotspots, at NAHR crossover hotspots in low-copy repeats flanking some genomic disorder regions, and is bound in vitro by PRDM9, a protein with a role in meiotic homologous recombination.^{42, 43, 44} Alus are enriched for this motif. We analyzed sequence within or flanking SPAST (chr2: 32,280,000–32,412,000) and found 24 of these motifs, 23 of which are in or overlap Alus (Figure 4B, Table 3). This motif is present in one breakpoint Alu of ten Alu-mediated CNVs, at both breakpoint Alus of one CNV (Park 3), and in the Alu from which inserted breakpoint sequence is derived in one apparently complex rearrangement (A2) (Figure 4).

Figure 4.

The Consensus 13-mer, CCNCCNTNNCCNC, Associated with Meiotic Recombination/NAHR Hotspots Is Present at Breakpoint Alus of Some Alu-Mediated CNVs

(A) CNVs (black bars) displayed with breakpoint Alus (pentagons; filled and intersected by a colored line if they contain the consensus 13-mer [CCNCCNTNNCCNC] associated with meiotic recombination/ NAHR hotspots⁴², ⁴³). This motif is present at a single breakpoint Alu of ten Alu-mediated CNVs, at both breakpoint Alus of one CNV (Park 3), and in the Alu from which inserted breakpoint sequence is derived in one apparently complex rearrangement (A2). Abbreviation: Dup, duplication.

(B) All instances of the 13-mer (black arrowheads).

(C) SINE elements (RepeatMasker track via UCSC Genome Browser).

Table 3.

13-mer CCNCCNTNNCCNC Associated with Meiotic Recombination/NAHR Hotspots within and Flanking SPAST^a

Coordinate	Orientation	Within Alu?	Alu Used in CNV Breakpoint?
chr2: 32,290,968	negative	–	–
chr2: 32,292,376	positive	AluSz	Ruhr 9, Ruhr 10
chr2: 32,294,873	positive	AluY	–
chr2: 32,300,868	positive	AluSz	–
chr2: 32,303,490	positive	AluSz6	A2 (complex)
chr2: 32,311,592	negative	AluSx	–
chr2: 32,317,837	negative	AluSx	–
chr2: 32,323,295	negative	AluJr	–
chr2: 32,337,660	positive	AluSg	–
chr2: 32,339,234	negative	AluJo	–
chr2: 32,346,229	negative	AluSx1 (overlaps)	–
chr2: 32,351,109	positive	AluY	Park 3, Ruhr 31, Spain 8
chr2: 32,355,085	negative	AluJr (overlaps)	–
chr2: 32,360,010	positive	AluSx	Park 3, JKF C
chr2: 32,371,388	negative	AluJo	–
chr2: 32,375,404	positive	AluSc	–
chr2: 32,378,630	positive	AluSx1	A3, A4, A5, A8, A9, SPAST_1
chr2: 32,378,906	negative	AluSx	A1, A39, BAB 3327
chr2: 32,394,608	negative	AluY	–
chr2: 32,398,678	positive	AluSx3	A6
chr2: 32,399,507	positive	AluJo	–
chr2: 32,402,087	positive	AluSz	Ruhr 14
chr2: 32,406,792	negative	AluSx	–
chr2: 32,411,515	negative	AluSx1	–

^aChr2: 32,280,000–32,412,000.

Twenty-eight CNVs are Alu mediated, unique, and noncomplex. Of these CNVs’ 56 breakpoint Alus, 12 (21%) contain the 13-mer consensus motif. To investigate whether this represents an enrichment of the 13-mer motif in breakpoint Alus of Alu-mediated CNVs (and thus may be involved in their formation), we made the following comparisons to other groups of Alus. First, seven non-Alu-mediated CNVs have a single breakpoint in an Alu (Figure 2, Table 2); of these, none contain the 13-mer (0%); this paucity was not significant (p = 0.329, two-tailed Fisher exact test). Second, there are 188 Alus within and flanking SPAST (chr2: 32,280,000–32,412,000; Figure 4C); 23 of these contain the 13-mer (12%); this decrease is not significant (p = 0.126, two-tailed Fisher exact test).

Deletions Resulting in Chimeric Genes

Although 28 CNVs are purely intragenic, seven CNVs extend 5′ of SPAST, and for 22 CNVs a breakpoint maps beyond the 3′ end of the gene; of these CNVs overlapping SPAST, 17 affect additional genes (Figure 2A, Table 1). Two deletions extending beyond the 5′ end of SPAST (BAB 3200 and Miura) have telomeric breakpoints in another gene, DPY30 (MIM 612032). DPY30 is in reverse genomic orientation with SPAST; thus, a chimeric SPAST-DPY30 gene is less likely to result from these deletions. The effect of heterozygous loss of function of DPY30 is unclear, as an association between DPY30 and any human disease has not been described, although its ortholog appears to be developmentally important in C. elegans (reviewed by Miura et al.⁴⁰).

Thirteen deletions (and one duplication) that extend beyond the 3′ end of SPAST have centromeric breakpoints in other genes (SLC30A6 = 10 deletions; NLRC4 [MIM 606831] and TTC27 = 1 deletion each; BIRC6 [MIM 605638] = 1 deletion and 1 duplication) (Figure 5). SLC30A6, BIRC6, and TTC27 are in direct orientation with SPAST; thus, the 12 deletions of the 3′ end of SPAST and the 5′ end of these genes form chimeric genes in these subjects’ genomes (Figure 5A). Multiple transcript isoforms have been identified for SLC30A6, BIRC6, and TTC27; thus, multiple fusion transcripts incorporating 5′ SPAST exons and 3′ exons of these genes are possible in subjects with such deletions (Figure 5B). Eight of twelve deletions have at least one predicted in-frame transcript. The SPAST_3 duplication is not anticipated to disrupt SPAST and thus is not likely to convey pathogenic changes to the spastin protein (consistent with this subject’s unknown status for spastic paraplegia). However, it creates an additional BIRC6:SPAST chimeric gene, the predicted fusion transcript of which is in-frame (Figure 5C). RNA was not available from any of these deletion or duplication subjects to examine experimentally for fusion transcripts.

Figure 5.

SPAST CNVs Extending into 3′ Genes Create Chimeric Genes, and Deletions Involving the Final Exon of SPAST Might Create SPAST:SLC30A6 Fusion Transcripts or Alter SLC30A6 Expression

(A–C) Deletions extending into 3′ genes.

(A) Fourteen CNVs (black bars) extend into genes 3′ of SPAST (13 deletions, 1 duplication). Three of the downstream genes (SLC30A6, BIRC6, TTC27) are in direct orientation with SPAST; thus, SPAST deletions intersecting these 3′ genes potentially yield chimeric genes.

(B) Potential fusion transcripts resulting from the deletions shown in (A) for which the 3′ intersected gene is directly oriented with SPAST. Potential fusion transcripts are based only on known splice isoforms of the 3′ genes and include all SPAST exons present, although other splice variants are possible. Predicted in-frame splicing is shown in black, out-of-frame in red. The deletion intersecting NLRC4 (Spain 6) is not shown, because this gene is oriented oppositely to SPAST.

(C) Potential fusion transcript resulting from the duplication in subject SPAST_3. This tandem, directly oriented duplication results in duplication of three genes (SLC30A6, NLRC4, YIPF4) and a potential in-frame transcript joining exons 1–46 of BIRC6 with exons 2–17 of SPAST; the normal, diploid copy number of full-length SPAST and BIRC6 are left intact.

(D and E) Deletions involving the final exon of SPAST.

(D) Deletions with 3′ breakpoints in the intergenic space between SPAST and SLC30A6 (top, “span exon 17”) or within the final exon of SPAST (exon 17) (bottom, “interrupt exon 17”). The former group of deletions may alter SLC30A6 expression by eliminating upstream SLC30A6 regulatory sequence (potential regulatory sequences shown by H3K27Ac track from UCSC), and both groups of CNVs may result in fusion transcripts with SLC30A6 by eliminating splicing to SPAST exon 17.

(E) Potential fusion transcripts resulting from the deletions shown in (D). Potential fusion transcripts are based only on known splice isoforms of SLC30A6 and include all SPAST exons present. Black, predicted in-frame; red, predicted out of frame.

None of the genes potentially involved in a fusion transcript (SLC30A6, BIRC6, and TTC27), nor any of the other genes affected by deletions spanning beyond the 3′ end of SPAST (NLRC4, YIPF4, MIR558, MIR4765, LINC00486, LOC100271832, LTBP1 [MIM 150390], and 5S_rRNA), is associated with Mendelian disease. Thus, the potential consequences to gene regulation or function of CNVs affecting these loci are unknown.

Deletion of SPAST Exon 17 Can Generate Fusion Transcripts with SLC30A6

Because SLC30A6 is the nearest gene downstream of SPAST, SPAST deletions with 3′ breakpoints in the 8 kb region between the two genes could potentially affect one or more SLC30A6 regulatory regions and/or result in splicing between the last remaining SPAST exon and SLC30A6. Seven such SPAST deletions exist among our data set, their 3′ breakpoints mapping within 6.5 kb to <1 kb of SLC30A6 (Figure 5). An additional seven deletions have 3′ breakpoints within the final exon (exon 17) of SPAST (Figure 5).

Nervous system RNA was not available from these subjects; however, a lymphoblastoid cell line from one subject with a complete exon 17 deletion (BAB 3327) was obtained. cDNA was prepared from this cell line and from a healthy control individual, BAB 3099;⁴⁵ both blood-derived and lymphoblastoid-cell-line-derived DNA from the control individual is free of SPAST CNVs as determined by aCGH (Figure 1). Two PCRs of cDNA were performed, one with primers closely flanking the deletion (forward primer in exon 16 of SPAST and reverse primer in SLC30A6 exon 2) and one with primers flanking a larger area (forward primer in SPAST exon 12 and reverse primer in SLC30A6 exon 7, the first constitutive exon of this gene) (Figure 6). A few faint fusion transcript bands were present in the control individual. However, abundant, robust fusion transcript bands were present in the SPAST deletion CNV subject (BAB 3327) (Figures 6A–6C). Sequencing of all PCR products confirmed splicing from SPAST to SLC30A6 (Figure 6E), further indicating that deletion of SPAST exon 17 can alter the expression/transcript structure of a downstream gene.

Figure 6.

A Deletion of SPAST Exon 17 Yields Multiple SPAST:SLC30A6 Fusion Transcripts

(A–D) PCR primers (arrows above and below gene models in D) were designed to detect SPAST:SLC30A6 fusion transcripts in cDNA from subject BAB 3327, who has a deletion encompassing SPAST exon 17 (black bar in D). One primer set (F1 and R1) closely flanks the deletion, and the other (F2 and R2) spans a greater distance to account for alternative splicing known to exist in SLC30A6 (exon 7 is the first constitutive exon in this gene). PCR of cDNA (A) reveals multiple transcripts in BAB 3327 not present in the control individual (BAB 3099). The majority of bands (white arrows) were extracted and sequenced. One extracted band (blue arrow) failed sequencing. Four bands (gray arrows) required a second preparation/electrophoresis to be visualized, extracted, and sequenced (B and C, white arrows).

(E) Sequencing of PCR products in (A)–(C) demonstrates a variety of SPAST:SLC30A6 fusion transcript isoforms in BAB 3327, only some of which are present in the control individual (BAB 3099). Dots indicate predicted in-frame (green dots) or out-of-frame (red dots) transcripts. Reading frame is unclear for transcripts with no dot. PCR products were sequenced in both forward (F) and reverse (R) genomic orientation, although only one orientation may be shown.

(F) SLC30A6 qPCR. In lymphoblastoid cell lines, overall level of expression of SLC30A6 in BAB 3327 does not differ significantly from that of eight healthy controls (50^th centile versus controls, empirical cumulative distribution function). Data plotted as mean ± standard deviation.

Effect of a SPAST Exon 17 Deletion on SLC30A6 Expression Level

The possibility exists that, by creating SPAST:SLC30A6 fusion transcripts or disrupting SLC30A6 upstream regulatory sequences, the SPAST exon 17 deletion in BAB 3327 may affect SLC30A6 expression level.⁴⁶ qPCR indicated that the level of expression of SLC30A6 in a BAB3327 lymphoblastoid cell line does deviate significantly from that of eight pooled control lymphoblastoid cell lines (50^th centile versus controls, empirical cumulative distribution function) (Figure 6).

Discussion

SPAST CNVs: Diverse Sizes, Boundaries, and Exons Rearranged

We delineated breakpoint junctions for 49 SPAST CNVs and document extensive diversity of disease-associated rearrangements in this gene, even for CNVs affecting a given exon. By increasing the number of published, disease-causing SPAST CNVs with known breakpoints from 7 to 54, we provide a resource for evaluating the pathogenicity of CNVs in SPAST identified during the course of genomic testing. Predicted in-frame deletions of exons 4, 8, and 9 in individuals with SPG4 suggest that these exons are indispensable, and a pathogenic, predicted in-frame duplication of exon 9 (24 amino acids) reinforces the sensitivity of the spastin AAA (ATPase Associated with various cellular Activities) domain to mutation⁴⁷ (Figure S5). No SPAST exons have been demonstrated to be dispensable. However, exon 2 or 3 deletions have not been reported in persons with SPG4; thus, these in-frame exons may be candidates for exon skipping⁴⁸ for individuals with pathogenic point mutations in them.

The CNV allele size versus frequency spectrum is skewed toward rearrangements of modest size relative to SPAST. A skewing toward small CNVs is reminiscent of the overall distribution of CNVs in personal genomes.⁴⁹ It may also reflect the mechanism(s) of formation of SPAST CNVs, discussed below.

Alu Elements Destabilize SPAST, Leading to CNVs with Unique Properties

We found that more than two-thirds of SPAST CNVs have breakpoints in pairs of directly oriented Alus. Thus, Alu-mediated CNV formation is the predominant, but not exclusive, mechanism generating rearrangements in SPAST. The observed predominance of Alu-mediated CNVs suggests that the Alu-rich genomic architecture of SPAST (36% of the gene with segments of even higher Alu density;¹⁴ Figure 2B) renders this locus susceptible to a variety of genome rearrangements and may explain why an appreciable percentage of mutations causing SPG4 are CNVs.

Alu-mediated CNVs in our cohort share properties with Alu-mediated CNVs at other disease loci, including the presence of breakpoint microhomology, rearrangement between different Alu family members, and small size (Table 4). We found that Alu-mediated CNVs in SPAST are smaller than non-Alu-mediated CNVs (which include the eight largest rearrangements in our cohort). This could reflect a “boundary” created by a zone of low Alu density near SPAST. Alternatively, it suggests a potential size preference or limit as a general feature of Alu-mediated CNV formation. Human-specific Alu-mediated CNVs²² are even smaller (median 468 bp, mean 806 bp; range 101–7,255 bp) than Alu-mediated CNVs in disease loci, potentially a contributing factor to their becoming fixed in a species.

Proposed Mechanisms Generating Alu-Mediated CNVs

The hypothesized mechanism for Alu-Alu-mediated CNV formation has been homologous recombination between nonallelic Alus, i.e., Alu-Alu recombination.^{20, 50} The SPAST CNVs we report are generally consistent with this model: pairs of Alus are directly oriented and render a complete, hybrid Alu in the recombined genome; complex rearrangements are infrequent.⁴³ However, the short length (<300 bp) and low percent identity (<91%) shared by Alus within each pair may be insufficient to meet the presumed requirements of ectopic homologous recombination derived from observational studies of NAHR at LCRs.^{51, 52} Furthermore, the three complex Alu-mediated CNVs are inconsistent with formation by NAHR unless a familial Alu polymorphism existed prior to the rearrangement. Some of the Alus at breakpoints of Alu-mediated CNVs contained the 13-nucleotide consensus motif enriched at meiotic recombination hotspots and at some LCRs involved in NAHR.^{42, 43} Although intriguing, any enrichment of this consensus motif in breakpoint Alus of Alu-mediated CNVs was not significant when compared to either Alus present at one breakpoint of non-Alu-mediated CNVs or to all Alus within and near SPAST. Finally, an analysis of previous studies at other loci (Table 4) demonstrates that moderate identity breakpoint Alu pairs and occasional complex events are general features of Alu-mediated CNVs. These findings may implicate an alternative non-homologous-recombination mechanism for Alu-mediated CNV formation.¹⁴

One alternative rearrangement mechanism may be microhomology-mediated DNA repair,⁵³ for example microhomology-mediated end joining (MMEJ) or fork stalling and template switching (FoSTeS)/microhomology-mediated break-induced replication (MMBIR). These mechanisms are consistent with the microhomology present at the breakpoints of Alu-mediated CNVs among our and others’ data (Table 4). MMEJ may explain the predominance of deletions over duplications. FoSTeS/MMBIR may explain the occasional apparently complex Alu-mediated CNV.¹⁴

Another alternative mechanism may be single-strand annealing (SSA).^{54, 55} This double-strand break (DSB) repair pathway involves the annealing of directly oriented repeats flanking the DSB after they are made single-stranded by 5′ to 3′ end resection. The removal of the 3′ flaps and filling of gaps by DNA synthesis then results in a simple, small deletion. In a study of translocation mechanisms, Elliott et al.⁵⁶ induced double-stranded breaks adjacent to a human Alu element on each of two nonhomologous chromosomes in cultured mouse cells; when the Alu elements were identical, SSA was the predominant mechanism generating translocations between these chromosomes. However, when the Alus were only 80% identical, the frequency of SSA diminished such that NHEJ became the predominant translocation-generating mechanism. Similarly, in yeast, a decrease of identity between two 205 bp repeats flanking a DSB site from 100% to 97% decreased the rate of SSA by 6-fold.⁵⁷

Another potential contributing mechanism could be homeologous recombination, reduced-frequency recombination between substrates that are somewhat but not perfectly identical.^{58, 59, 60} This model is consistent with the low percent identity (75.8%–90.7%) shared by breakpoint Alu pairs in our study.⁶¹ In addition, in 33 of 39 (85%) Alu-Alu events in this study, the substrate pairs were from different Alu families. Whereas NAHR generates recurrent CNVs with high formation rates during meiosis,⁶² our data and others’ (Table 4) demonstrate that Alu-mediated CNVs are mostly or entirely nonrecurrent; thus, the formation rate of any particular rearrangement is low, again consistent with homeologous recombination.⁶³ Although the probability of any one Alu-mediated CNV occurring is miniscule, with more than one million Alus in the genome, the cumulative ability of Alus to act as a recombination substrate and generate genomic rearrangements may be substantial.

SPAST deletions predominate over duplications. This suggests that, if formed by recombination, these events are most often intrachromatidal. Because Alus in close linear proximity along the same chromatid are generally expected to be in close three-dimensional proximity,⁶⁴ and because three-dimensional proximity has been suggested to be important for generation of rearrangements,⁶⁵ then perhaps small size is a general feature of Alu-mediated CNVs that is facilitated by chromosomal conformation itself. Long-range interactions, chromatin compartments,⁶⁴ tethering of Alus, and genome reorganization during the cell cycle⁶⁶ could also shape the spatial relationships of Alus and thus their probability of acting as rearrangement substrates. If these CNVs are formed by replicative template switching, small rearrangement size would be consistent with the known tendency of this process to occur more frequently over short distances.

Deletions Can Result in Chimeric Genes and Fusion Transcripts

Almost half of the CNVs we report extend beyond the boundaries of SPAST, many of which interrupt other genes. Although none of these other genes is known to be associated with Mendelian disease, their disruption could have phenotypic consequences. Moreover, 12 deletions extending 3′ of SPAST form chimeric genes that may yield fusion transcripts, also of unknown phenotypic consequence. Of note are the deletions forming chimeric genes between SPAST and SLC30A6 (solute carrier family 30, member 6) encoding a zinc transporter, ZnT6. Zinc and zinc transporter (including ZnT6) dynamics have been described to be altered in the brains of humans with Alzheimer disease and rodent models of dementia.^{34, 35, 67} Thus, the possibility of SPAST CNVs dysregulating SLC30A6 expression could potentially inform the clinical observation of comorbid spastic paraplegia and dementia.³¹ Neither RNA nor cells were available from any subjects with combined SPG4 and dementia.

Seven CNVs have 3′ breakpoints within the intergenic region between SPAST and SLC30A6, and an additional seven have 3′ breakpoints within the final exon of SPAST. The former group could affect SLC30A6 expression and all could potentially form fusion transcripts. We investigated this possibility experimentally using RNA from an available nonneuronal cell type (lymphoblastoid cell line) from one subject with a deletion encompassing the final exon of SPAST (exon 17). A few low-abundance fusion transcripts were present in a control individual, the significance of which is unknown. However, abundant and additional fusion transcripts were present in the SPAST deletion subject, indicating that a deletion of only an exon of SPAST can affect gene expression beyond the gene itself. The phenotypic consequence of this, if any, requires additional investigation.

The expression level of SLC30A6 in lymphoblastoid cell line RNA from this individual does not deviate significantly from that of eight control lymphoblastoid cell lines. Whether or not SLC30A6 expression is affected in this individual in neurons is unknown.

Summary

Molecular characterization of novel SPAST CNV alleles contributes to knowledge of pathogenic variants in this gene and provides insights into the importance of particular exons for spastin function. Our data suggest that the Alu-rich genomic architecture of SPAST predisposes to CNVs in this gene, contributing appreciably to its mutability and to SPG4 disease and potentially by a mechanism other than homologous recombination. Furthermore, these data allow the generation of hypotheses concerning Alus as a genome-destabilizing entity. We provide evidence that Alu-mediated rearrangements shape gene and genome structure and that the CNVs that are generated affect both SPAST and adjacent genes and extragenic regions. Thus, such CNVs can affect gene expression beyond SPAST itself, with potential phenotypic consequences.

Published: July 24, 2014

Web Resources

The URLs for data presented herein are as follows:

• 1000 Genomes, http://browser.1000genomes.org

• Agilent eArray, https://earray.chem.agilent.com

• dbSNP, http://www.ncbi.nlm.nih.gov/projects/SNP/

• GeneReviews, Dürr, A., Tallaksen, C., and Depienne, C. (2012). Spastic paraplegia 4, http://www.ncbi.nlm.nih.gov/books/NBK1122

• International HapMap Project, http://hapmap.ncbi.nlm.nih.gov/

• Online Mendelian Inheritance in Man (OMIM), http://www.omim.org/

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

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