Structural Variant

A structural variant (SV) is a genomic rearrangement typically larger than 50 base pairs, including deletions, duplications, inversions, insertions, and translocations that alter chromosome architecture. This guide is for bioinformaticians, clinical geneticists, and molecular biologists who need a practical, source bounded framework for understanding, detecting, and interpreting structural variants using high throughput sequencing data.

Structural variants are responsible for a large fraction of human genetic diversity and are associated with both rare and common diseases. Unlike single nucleotide variants (SNVs), SVs are harder to detect and classify due to their size and the repetitive nature of breakpoint regions. This guide will help you navigate the core concepts, decision points, workflows, quality checks, common mistakes, and limits of interpretation. For a foundational overview of genomic variation detection, refer to the NCBI Bookshelf for authoritative biomedical references.

The first two sections below will establish a clear definition of structural variants and then provide a decision framework for choosing detection methods. You will learn how to implement a practical SV calling workflow, apply quality checks, avoid common pitfalls, and understand the inherent uncertainties in SV analysis. Additional resources from EMBL EBI Training offer complementary hands on modules for variant analysis.

At a Glance

Core Concepts and Definitions

A structural variant (SV) is a rearrangement of the genome that involves at least 50 base pairs of DNA. The most common types are deletions (loss of a sequence), duplications (copy gain), inversions (reverse orientation), insertions (novel sequence), and translocations (interchromosomal rearrangements). SVs can be balanced (no net gain or loss of DNA) or unbalanced (leading to copy number variation). The Galaxy Training Network provides excellent tutorials on SV discovery using both short and long read data.

Why focus on the 50 bp threshold? This cutoff is a community standard that distinguishes SVs from intermediate sized indels (which are usually 1 49 bp). In practice, many SVs span thousands to millions of base pairs and can disrupt genes, regulatory elements, or chromosome structure. A practical example: a 150 kb deletion on chromosome 15q11.2 is a known risk factor for developmental disorders. Understanding the type and size of an SV guides clinical interpretation.

SVs are identified by analyzing sequencing reads that span breakpoints. In paired end sequencing, a read pair that maps further apart than expected (discordant) or with incorrect orientation can indicate a deletion or inversion. Split reads, where a single read maps to two distinct genomic loci, directly pinpoint breakpoints. De novo assembly based methods realign the whole genome and detect SVs as differences from a reference. For a detailed technical explanation, see the Bioconductor documentation on structural variant packages like VariantTools and gmoviz.

Decision Criteria for SV Detection

Choosing the right SV detection method depends on your sequencing platform, read length, coverage, and biological question. Here are the main decision points:

1. Platform and read length: Short read sequencing (Illumina) is widely available and cost effective but has limited ability to detect SVs in repetitive or segmentally duplicated regions. Long read sequencing (PacBio, Oxford Nanopore) reads can span entire SVs and resolve breakpoints with high accuracy. Use long reads when you need comprehensive SV detection in complex genomes.
2. Coverage depth: Low coverage WGS (<15x) reduces sensitivity for SVs, especially heterozygous deletions and duplications. Higher coverage (>30x) improves breakpoint detection but increases cost. For population studies, median coverage of 20x is a common trade off Navigating tradeoffs in variant filtering for population genetic and demographic inferences to inform management, conservation, and domestication in non model marine bivalves: a case study in scallop..
3. Method type: Paired end methods (e.g., read pair analysis) are fast and suitable for deletions and tandem duplications. Split read methods (e.g., Pindel) provide base pair resolution of breakpoints but require high quality alignments. Assembly based methods (e.g., DISCOVAR) discover novel insertions but are computationally intensive. Often a combined approach using multiple tools increases recall.
4. Genome complexity: In repetitive genomes (e.g., plants, some pathogens), short read SV calling is unreliable. Long reads or optical mapping are preferred. For human clinical applications, a combination of Illumina and long read sequencing on targeted regions (like known disease genes) is a common strategy.

Your choice should also consider downstream goals. If you need to validate SVs for clinical reporting, you will likely require orthogonal confirmation (e.g., PCR, qPCR, or a second sequencing technology). For research discovery, a higher false discovery rate may be acceptable.

Practical Workflow for SV Calling

A robust SV calling workflow follows these steps. Each step should include appropriate quality checks.

Step 1: Data acquisition and quality control Download raw sequencing data from repositories like the NCBI Sequence Read Archive. Run FastQC and MultiQC to assess read quality. Trim adapters and low quality bases using cutadapt or Trimmomatic. For long reads, use NanoPlot or pycoQC.

Step 2: Alignment and preprocessing Align reads to a reference genome. For short reads, use BWA MEM and sort with SAMtools. For long reads, use minimap2. Mark duplicate reads with Picard. Ensure alignment files (BAM) are indexed. Check depth of coverage across chromosomes using mosdepth or sambamba.

Step 3: SV discovery Run one or more SV callers. A common combination: Manta (deletions/duplications), DELLY (balanced SVs), and Lumpy (discordant read pairs). For long reads, use Sniffles or cuteSV. Always provide a BAM with sorted reads and a reference genome. Example command for Manta: configManta.py --bam sample.bam --reference ref.fa --runDir manta_output. Then execute the workflow.

Step 4: Merge and filter Combine calls from multiple tools using SURVIVOR or svdb. Filter low confidence calls: require at least two supporting read pairs or split reads, minimum breakpoint quality score (e.g., QUAL >= 30), and exclude SVs in centromeres or telomeres (genomic blacklist regions). For population studies, Navigating tradeoffs in variant filtering recommends balancing recall and precision based on minor allele frequency and Hardy Weinberg equilibrium.

Step 5: Annotation and prioritization Annotate SVs with gene overlaps, regulatory elements, and population frequency databases (e.g., gnomAD SV). Use Ensembl VEP with an SV annotation plugin. Prioritize SVs that disrupt coding regions, splice sites, or known disease genes. Visualize candidate SVs in IGV to confirm breakpoint patterns.

Step 6: Validation Select a subset of high priority SVs for orthogonal validation. For deletions, design PCR primers flanking the predicted breakpoint and run gel electrophoresis. For duplications, use qPCR with a reference gene. Long read sequencing of the same sample is the most definitive validation method.

Quality Checks and Common Mistakes

Even with a solid workflow, errors can creep in. Here are common mistakes and how to avoid them:

Mistake 1: Calling SVs in repetitive or low complexity regions. Repeat masked regions produce false positives. Always exclude SVs overlapping with simple repeats, segmental duplications, or centromeres unless you have orthogonal evidence. Use a repeat masker track available from UCSC or Ensembl.

Mistake 2: Over filtering by read support. While requiring several supporting reads reduces false positives, it also removes true heterozygous SVs in low coverage regions. Balance filter stringency with expected heterozygosity. For rare variant discovery, consider keeping SVs with at least two split reads or three discordant pairs.

Mistake 3: Ignoring GC bias and coverage depth. Deletions are often called where coverage drops due to GC bias. Normalize coverage with a GC correction tool (e.g., HMMCopy) before calling copy number variants. Visualize coverage in the region to distinguish technical artifacts from true deletions.

Mistake 4: Not merging calls across callers. No single SV caller perfectly captures all types. Combining tools improves recall but introduces duplicate calls. Use merging software (e.g., SURVIVOR) with a 1 kb distance threshold to collapse overlapping calls into a single event.

Quality check: Breakpoint precision. Precise breakpoints are essential for clinical interpretation. For deletions, check that both breakpoints are supported by split reads. For inversions, require read pairs that map in the opposite orientation. If breakpoints are only defined within a 500 bp window, note the uncertainty.

Limits of Interpretation

SV interpretation carries inherent uncertainties that you must acknowledge.

Breakpoint resolution: With short reads, breakpoints are often imprecise, sometimes by hundreds of base pairs. This can misclassify an SV type (e.g., an inversion called as a deletion) or affect gene disruption status. Long reads improve precision but are not error free.

Mappability: SVs in highly repetitive regions (e.g., centromeres, ribosomal DNA arrays) are invisible to most short read methods. Even long reads struggle to map uniquely. Reported SVs in these regions should be treated with skepticism.

Population frequency filtering: Databases like gnomAD SV provide allele frequencies, but they are biased toward healthy individuals of European descent. A rare SV in one population may be common in another. Use these frequencies as guides, not absolute cutoffs.

Functional impact prediction: Predicting whether an SV disrupts a gene is not straightforward. An intronic deletion may remove a regulatory element without affecting the coding sequence. SVs that invert a gene may produce a functional fusion or a truncated protein. Experimental evidence (e.g., RNA sequencing) is often needed to confirm functional effects.

Clinical validation requirement: For diagnostic use, any SV detected by sequencing must be confirmed by an independent method (e.g., array CGH, MLPA, or long read sequencing). The uncertainty in bioinformatic calls is too high for clinical action without orthogonal validation.

Frequently Asked Questions

What is the minimum coverage needed to detect a structural variant? For heterozygous deletions, at least 15x coverage is recommended for short read data. Lower coverage reduces the chance of observing discordant read pairs. For homozygous deletions, 10x may suffice, but breakpoint resolution suffers. See NCBI Sequence Read Archive for coverage guidelines in public data.

Can structural variants be detected from RNA sequencing data? Yes, but with limitations. RNA seq can reveal fusion transcripts and exon skipping caused by SVs, but it cannot detect balanced inversions or deletions that do not affect splicing. The EMBL EBI Training portal includes a module on RNA seq variant detection.

How do I distinguish a deletion from low coverage artifact? Check discordant read pairs: deletions produce read pairs that map farther apart than the insert size. Low coverage artifacts from GC bias usually show a smooth drop without discordant reads. Also check split reads that directly cross the breakpoint.

Why do different SV callers give different results for the same sample? Each caller uses different algorithms (read pair versus split read versus assembly) and sensitivity thresholds. Merging calls from two or more callers increases confidence, but you must be aware of systematic biases. Use the same reference genome version and read alignment pipeline to reduce variability.

References and Further Reading

1. NCBI Bookshelf. Structural Variation in the Human Genome.
2. EMBL EBI Training. Structural Variant Analysis.
3. Galaxy Training Network. SV Discovery with WGS.
4. Bioconductor. Structural Variant Tools.
5. NCBI Sequence Read Archive. Public SV Datasets.
6. Navigating tradeoffs in variant filtering for population genetic and demographic inferences to inform management, conservation, and domestication in non model marine bivalves: a case study in scallop. BMC Genomics.
7. Novel variants in LINC and TTM complexes of meiotic chromosome dynamics are associated with meiotic arrest and non obstructive azoospermia. J Hum Genet.
8. RBM20 variants disrupt Ca(2+) handling and metabolism in dilated and non compaction cardiomyopathy stem cell models. Signal Transduct Target Ther.
9. Machine Learning Models for Local Optimization of Red Fluorescent Protein Variants in a Low Data Setting. J Chem Inf Model.
10. Update on Adult Acquired Flat Foot Deformity Using MR Imaging. Magn Reson Imaging Clin N Am.

Related Articles

What Is Monomeric Protein
Translation Gene
Cell Signal Booster
Gene Expression
Dna Sequencing