Structural Variation in Genomes

A structural variant is conventionally defined as a genomic alteration affecting a segment of DNA 50 base pairs or larger — the size threshold that separates SV from ordinary small indels and single-nucleotide variants (SNVs), though the boundary is a convention, not a biological law. SVs fall into two broad functional classes. Copy-number variants (CNVs) — deletions and duplications — change how many copies of a DNA segment a genome carries, and therefore can change gene dosage directly. Balanced rearrangements — inversions, and reciprocal or Robertsonian translocations — reorder or relocate DNA without net gain or loss of sequence, so they act instead by disrupting genes at breakpoints, separating genes from their regulatory elements, or creating novel fusion genes. Insertions deserve separate mention: mobile element insertions (Alu, L1/LINE-1, SVA elements) are a major, ongoing source of human SV, with new L1 insertions arising roughly once every 100–200 births. Complex SVs combine several of these — a duplication nested inside an inversion, or a chain of translocations known as a chromothripsis event, in which a chromosome is shattered and stitched back together in a single catastrophic episode, often seen in cancer genomes. The comparative-genomics baseline matters here: aligning any two human genomes, or a human genome to the reference assembly, reveals millions of SNVs but only a few thousand SVs per individual — yet those few thousand SVs collectively affect more base pairs of the genome than SNVs do, because each event can span kilobases to megabases. This is why SV is sometimes called the genome's "structural" or "coarse" layer of variation, sitting above the fine-grained substitution layer studied in classical Comparative Genomics.

How structural variants arise. Three broad mutational mechanisms generate SV, and they leave distinct molecular fingerprints at breakpoints. 1. Non-allelic homologous recombination (NAHR): recombination machinery mispairs two similar-but-nonallelic repeat sequences (commonly low-copy repeats/segmental duplications >1 kb, >90% identical) and crosses over between them, deleting or duplicating the intervening DNA, or inverting it if the repeats are in opposite orientation. NAHR breakpoints cluster precisely within the repeats. 2. Non-homologous end joining (NHEJ) and microhomology-mediated end joining (MMEJ): a double-strand break is repaired using little or no sequence homology, producing variable, imprecise breakpoints often with small insertions or deletions ("microhomology" of a few bases) at the junction. 3. Replication-based mechanisms (fork stalling and template switching, FoSTeS/MMBIR): a stalled replication fork's nascent strand invades a nearby, non-allelic template, producing complex, sometimes duplicated or triplicated segments with scrambled order — the mechanism behind many complex genomic disorders and much of chromothripsis.

How SV is detected from sequencing data. Because most SVs are larger than a single sequencing read, detection relies on indirect signals rather than direct observation of the rearranged sequence in one read. Four signals are combined by modern SV callers:

• Read depth (RD): the number of reads aligning to a genomic window is proportional to copy number. A single-copy deletion in a heterozygous individual halves expected depth over that interval; a duplication increases it by ~50%. This signal detects CNVs but is blind to balanced events (inversions, translocations) since copy number does not change.
• Paired-end / discordant read pairs: sequencing produces read pairs from a known DNA fragment length (e.g., ~400–500 bp for standard libraries). If the two mapped reads are farther apart than expected, a deletion lies between them; if closer than expected, an insertion; if the pair maps in the wrong relative orientation, an inversion; if the mates map to different chromosomes, a translocation.
• Split reads / soft-clipping: a single read that spans a breakpoint aligns partly to one location and partly to another. Aligners "clip" the unmapped portion; SV callers re-align the clipped segment elsewhere in the genome to pinpoint the breakpoint to single-base resolution.
• De novo assembly / long reads: rather than aligning reads to a reference and inferring rearrangement indirectly, long-read platforms (PacBio HiFi, Oxford Nanopore) generate reads tens of kilobases long that can span entire SVs directly, or are assembled de novo into contigs that are then compared to the reference — revealing SV as an assembly-vs-assembly difference, the most direct and complete method.

Complementary technologies add resolution: array comparative genomic hybridization (array-CGH) and SNP microarrays infer CNV from fluorescence-intensity or genotype log-ratios across probes, without any sequencing; optical mapping (e.g., Bionano) images fluorescently labeled long DNA molecules (100 kb–1 Mb) to detect large SVs by direct length and pattern comparison; Hi-C (chromosome conformation capture) detects translocations and large inversions by revealing anomalous long-range contact patterns between DNA regions that should not normally interact.

How SV causes phenotype. Beyond simple dosage, SV disrupts genome function through several distinct routes: gene disruption (a breakpoint falls inside a coding exon); dosage effect (deletion → haploinsufficiency; duplication → triplosensitivity/overexpression); gene fusion (a translocation joins two genes into a novel chimeric protein, e.g., BCR-ABL1); position effect (a gene is moved next to a new regulatory element or into heterochromatin and its expression changes without the gene itself being altered); and topological domain disruption — inversions or deletions can break a topologically associating domain (TAD) boundary, allowing an enhancer to now contact the wrong gene, a mechanism established for limb-malformation syndromes affecting the SHH limb enhancer region.

Discovery that CNVs are pervasive in normal human genomes. For decades, structural variation was studied almost exclusively in the context of visible chromosomal abnormalities (karyotype-scale, detectable by light microscopy, >5 Mb) or specific disease-linked microdeletions (e.g., DiGeorge syndrome, described cytogenetically in the 1990s). Whether normal, phenotypically unremarkable people carried extensive submicroscopic CNV was unknown until the early 2000s. Two independent 2004 studies changed this. Iafrate, Feuk, Rivera, Lee, and colleagues (Scherer lab, 2004, Nature Genetics) used array-CGH — comparing genomic DNA from many unrelated, healthy individuals against a common reference on a microarray of large-insert BAC clones — and found that ~255 loci in the genome varied in copy number among healthy people, collectively covering many megabases. Independently, Sebat, Lakshmi, Troge, and colleagues (Wigler lab, 2004, Science) used a technique called representational oligonucleotide microarray analysis (ROMA) on ten unrelated individuals and detected an average of ~11 large CNVs per person (>100 kb), some spanning millions of base pairs and overlapping known genes, in people with no disease phenotype at all. This was the first systematic, genome-wide demonstration that copy-number differences are a normal, common feature of human genetic variation — not just a rare cause of syndromic disease. The Redon et al. 2006 "CNV map" study (part of the International HapMap Project's structural-variation arm, Nature) scaled this up dramatically: using both array-CGH and SNP genotyping arrays across 270 HapMap individuals from four populations, the team catalogued 1,447 copy-number variable regions, covering roughly 12% of the human genome — an order of magnitude more sequence affected than had been appreciated, reframing CNV as a major axis of human genetic diversity comparable in impact (if not in site count) to SNVs.

Establishing sequencing-based detection. Once whole-genome sequencing became routine, the 1000 Genomes Project (2010–2015) systematically applied paired-end mapping, read-depth, and split-read algorithms across thousands of genomes, producing the first population-scale SV catalogues directly from sequence rather than array intensity, and confirming that a typical human genome carries roughly 2,100–2,500 structural variants relative to the reference, most being small insertions/deletions of mobile elements or short tandem repeats, with several thousand base pairs affected by larger CNVs.

Long-read resolution of previously invisible SV. A key later demonstration came from the Human Genome Structural Variation Consortium and related long-read studies (e.g., Chaisson et al. 2019; the pangenome / Telomere-to-Telomere efforts of the 2020s): sequencing the same individuals with both short reads and long reads (PacBio, Oxford Nanopore) showed that short-read pipelines systematically missed 60–80% of SVs detectable by long-read assembly, especially insertions and SVs in repetitive or segmentally duplicated regions — direct proof that repeat-rich, hard-to-align genomic regions harbor a large hidden reservoir of structural variation, and that assembly-based methods are needed to see it.

A defining disease case: BCR-ABL1 and the Philadelphia chromosome. The clearest proof that a balanced translocation can cause disease by creating a fusion gene came from chronic myeloid leukemia (CML). Nowell and Hungerford first described an abnormally small chromosome ("Philadelphia chromosome") in CML patients in 1960; Janet Rowley demonstrated in 1973, using new chromosome-banding techniques, that this was not a simple deletion but a reciprocal translocation between chromosomes 9 and 22, t(9;22)(q34;q11). Molecular cloning in the 1980s showed the breakpoint fuses the BCR gene (chromosome 22) to ABL1 (chromosome 9), producing a constitutively active BCR-ABL1 tyrosine kinase. This causal chain — translocation → fusion protein → constitutive kinase activity → uncontrolled proliferation — was confirmed when the targeted kinase inhibitor imatinib was shown to induce remission in most CML patients, directly validating the fusion protein as the disease's molecular driver.

Example 1 — Reading a CNV from array log-ratio and sequencing depth data. A clinical microarray reports a log2 ratio of −1.0 across a 1.8 Mb interval on chromosome 22q11.2 in a patient with congenital heart disease and hypocalcemia. Interpret this and connect it to sequencing-based depth.

• Log2 ratio compares patient signal intensity to a diploid (copy number 2) reference: log2(observed/expected).
• log2 ratio = −1.0 → observed/expected = 2⁻¹ = 0.5 → the patient has half the expected DNA dosage over that interval.
• Starting copy number 2 (diploid) × 0.5 = copy number 1 → a heterozygous deletion of one copy across the 1.8 Mb region.
• If this had instead been sequenced by whole-genome sequencing at 30× average genome-wide depth, read depth over the deleted interval should fall to ~15× (half of 30×), consistent with a heterozygous deletion call from a read-depth-based SV caller.
• Clinically: a 1.5–3 Mb heterozygous deletion at 22q11.2 is the hallmark of DiGeorge syndrome / 22q11.2 deletion syndrome, explaining the cardiac defect (outflow tract malformation) and hypocalcemia (parathyroid hypoplasia) through haploinsufficiency of the ~30–40 genes in the region, most importantly TBX1, a transcription factor whose single remaining copy cannot produce enough protein for normal pharyngeal arch development.
• This is a direct dosage effect: no point mutation, no altered protein sequence — simply not enough of a needed gene product.

Example 2 — A duplication/deletion reciprocal pair from NAHR, and predicted phenotypic severity. The PMP22 gene on chromosome 17p12 is flanked by two nearly identical low-copy repeats (~24 kb, >98% identity) roughly 1.4 Mb apart. NAHR between these repeats, if it occurs on mismatched (non-allelic) chromatids during meiosis, produces two reciprocal outcomes with equal probability in principle: a duplication on one product and a deletion on the other.

• Duplication product: three copies of PMP22 (2 normal + 1 extra) → ~1.5× normal dosage → PMP22 overexpression → this is Charcot-Marie-Tooth disease type 1A (CMT1A), a demyelinating peripheral neuropathy — too much of a dosage-sensitive myelin protein disrupts normal Schwann cell function.
• Deletion product: one copy of PMP22 only → ~0.5× normal dosage → Hereditary Neuropathy with liability to Pressure Palsies (HNPP), a milder, episodic neuropathy triggered by nerve compression — here haploinsufficiency, not overexpression, is the driver.
• Quantitative check: if normal expression from two alleles is defined as 1.0 dosage units, three copies give 3/2 = 1.5 units, and one copy gives 1/2 = 0.5 units — a 3-fold difference in gene dosage between the two reciprocal NAHR products, and correspondingly the two syndromes have very different severity and inheritance presentation (CMT1A is more disabling and progressive; HNPP is episodic and often mild).
• This pair is the textbook demonstration that CNV breakpoints recur at the same genomic coordinates across unrelated patients whenever a shared repeat architecture drives NAHR — a mechanistic signature, not coincidence.

Janet Rowley trained as a physician and hematologist, and turned to cytogenetics relatively late in her career, at a time when chromosome banding techniques (quinacrine and Giemsa staining, developed in the early 1970s) first made it possible to distinguish chromosomes by banding pattern rather than size and shape alone. Working at the University of Chicago, she applied these new staining methods to leukemia cells and, in 1972–1973, showed that the small "Philadelphia chromosome" seen in chronic myeloid leukemia patients since 1960 was not simply a deletion, as had been assumed, but the product of a reciprocal exchange of material between chromosomes 9 and 22. She went on to identify recurrent translocations in acute myeloid leukemia and acute promyelocytic leukemia, establishing — against considerable initial skepticism from a field that assumed cancer chromosome changes were random consequences rather than causes — that specific, recurrent chromosomal rearrangements are causal drivers of specific cancers. This insight underlies the entire modern framework connecting structural variation to oncogenesis and to targeted cancer therapy.

Clinical genetics and prenatal diagnosis. Chromosomal microarray analysis (array-CGH or SNP array) is now first-tier testing for children with unexplained developmental delay, intellectual disability, or congenital anomalies, precisely because CNVs such as 22q11.2 deletion, 1p36 deletion, and 16p11.2 duplication/deletion are found in a meaningful fraction of such cases and are invisible to standard karyotyping (which resolves down to only ~5–10 Mb) or single-gene sequencing.

Cancer genomics. Tumor genomes are riddled with somatic SV — focal amplifications of oncogenes, deletions of tumor suppressors, and driver fusion genes (BCR-ABL1 in CML; EML4-ALK in lung adenocarcinoma; TMPRSS2-ERG in prostate cancer). Detecting these fusions from tumor sequencing directly guides targeted therapy selection — the same logic that made imatinib a landmark precision-medicine drug now extends to dozens of fusion-directed inhibitors.

Pharmacogenomics. Copy-number variation at drug-metabolizing loci — most classically CYP2D6, which metabolizes many common drugs — directly determines dosing: patients with CNV-driven CYP2D6 duplications are "ultrarapid metabolizers" who may need higher doses (or who convert prodrugs like codeine into active morphine dangerously fast), while deletion carriers are "poor metabolizers" needing dose reduction. This is one of the clearest current bridges from SV detection to individualized prescribing, foreshadowing the broader integration of structural variants into Pharmacogenomics and Personalized Medicine.

Population and evolutionary genomics. SVs, especially large inversions, can suppress recombination across the rearranged region in heterozygotes, allowing sets of alleles to be inherited together as a block — a mechanism implicated in adaptive "supergenes" controlling complex traits like mimicry in butterflies and social behavior in some birds, extending the logic of structural variation into Comparative Genomics and evolutionary biology.

Agriculture and conservation. Copy-number differences at loci controlling traits like amylase copy number (starch digestion) or coat color have been selected during domestication of crops and livestock; conservation genomics increasingly screens endangered populations for structural variants that erode fitness under inbreeding.

Genome-wide association studies (GWAS). Most GWAS to date have been built on SNP genotyping arrays that tag common SNPs well but capture CNV, inversion, and complex SV poorly. As sequencing-based and long-read-based SV catalogues mature, incorporating structural variants directly into association testing is expected to recover a portion of the "missing heritability" for complex traits — a natural extension into Genome-Wide Association Studies that this article's mechanisms and detection methods directly set up.

Ask the tutor: How would you design a sequencing and analysis strategy to detect a suspected balanced translocation in a patient whose standard microarray came back completely normal?