Figure 1. Long-read AAV genome QC separates expected full-length molecules from structural variants and non-vector packaged DNA.

An AAV preparation can contain the intended vector genome together with partial genomes, snapback forms, rearranged molecules, vector–plasmid chimeras, and host- or helper-derived DNA. Long reads can connect distant sequence elements on individual molecules, making them especially useful for structural quality control. The result is only as reliable as the reference package, sample preparation, strand-conversion strategy, read-classification rules, and orthogonal checks. This guide explains how to plan and interpret long-read AAV genome QC without treating every mapped read as an intact vector genome.

CD Genomics provides these sequencing and bioinformatics services for research use.

Key Takeaways

• Define the molecular population before sequencing. Capsid-associated, nuclease-resistant, extracted, and library-compatible DNA are related but not identical populations.
• Use complete production references. The intended ITR-to-ITR genome alone cannot identify plasmid backbone, rep/cap, helper, or host-derived impurities reliably.
• Classify read structure, not only mapping percentage. Full-length coverage can coexist with inversions, duplications, snapback forms, and chimeric segments.
• Measure workflow bias. Second-strand synthesis, annealing, ligation, size selection, and base calling can change the observed distribution.
• Keep orthogonal assays in the evidence package. Size, mass, copy number, and sequence structure answer different QC questions.

Define What the QC Study Must Measure

"Genome integrity" can refer to identity at individual bases, continuity from one ITR to the other, the proportion of full-length molecules, the distribution of structural variants, or the amount and origin of packaged non-vector DNA. These endpoints require different analysis rules. A project should state which are primary before choosing a platform. Analytical characterization of full, intermediate, and empty capsids also shows why particle class and sequence architecture should be treated as related but distinct evidence dimensions [4].

The AAV Sequencing service supports vector-focused sequence analysis, but the project brief still needs an explicit claim. If the claim concerns the fraction of molecules matching an expected ITR-to-ITR architecture, the denominator should be all analyzable capsid-derived reads. If it concerns a particular impurity, the denominator and reference competition must be specified separately.

The AAV genome integrity metrics guide provides broader metric context. A long-read study should add molecule-level definitions so that "full length," "partial," and "unexpected" remain reproducible.

Build a Complete Reference Package

The expected vector genome is necessary but insufficient. Long reads frequently span boundaries between intended sequence and production-related sequence, so all plausible sources should be represented in the mapping or tiling database. An incomplete database can force a chimeric read into the closest vector or host category.

At minimum, prepare:

• the exact ITR-to-ITR vector genome in FASTA format;
• the complete transfer plasmid, including bacterial backbone;
• rep/cap and helper plasmid sequences used for production;
• producer-cell genome build and relevant engineered elements;
• known adapters, barcodes, linkers, and spike-in sequences;
• construct version, serotype, production platform, and sample identifiers;
• a feature table defining ITRs, promoter, coding region, regulatory elements, and polyadenylation signal.

Reference orientation and circular-plasmid boundaries need special attention. A read that crosses the first and last base of a linearized plasmid reference can appear split even when the original molecule is continuous. Including rotated reference representations or a circular-aware aligner prevents false structural calls.

If several vector designs share long identical regions, record construct-specific sequence blocks before analysis. Otherwise, index misassignment or sample carryover can be difficult to distinguish from genuine cross-construct sequence. The broader Viral Vector Development Solutions page provides context for placing AAV genome QC within a research-stage vector characterization program.

Figure 2. Competitive classification requires the intended vector and every plausible production-related DNA source.

Preserve the Packaged Genome Population

The analytical target is usually DNA protected within or closely associated with the vector preparation, but sample handling determines what reaches the sequencer. Nuclease treatment can reduce external DNA, while capsid disruption and extraction recover the remaining nucleic acid. Inefficient treatment, incomplete lysis, and size-dependent extraction alter the apparent impurity profile.

Strand conversion can reshape the result

Single-stranded AAV DNA often needs annealing or enzymatic conversion before a double-strand ligation workflow. The conversion method may favor certain lengths or structures, create extension products, or fail on stable secondary structures. Direct ITR-to-ITR Nanopore research showed that intact vector structures can be recovered, while also documenting base-calling and read-length limitations [1]. More recent comparisons found that second-strand preparation influenced observed read distributions and impurity assignments [5,6].

Run process controls that expose those effects. A linearized transfer plasmid can test mapping and length accuracy. A defined mixture of vector and impurity fragments can challenge competitive classification. Replicate libraries prepared with the same input test repeatability, while an alternative conversion method can reveal method-specific bias during development.

Record molecule counts, not only gigabases

Structural precision depends on the number of independent vector molecules observed in each class. Total yield can be high while rare categories remain supported by very few molecules. Report usable reads, unique molecules when deduplication is applicable, read-length distribution, quality distribution, and the count supporting every major structural class.

Sample pooling requires caution. Multiplexing is efficient, but barcode leakage and unequal loading can create low-level cross-sample signals. Include unique barcodes, a negative library, and per-sample contamination checks when several constructs share a sequencing run.

Classify Expected and Unexpected Genome Forms

A useful classifier breaks each read into ordered segments and compares the segment path with the submitted vector map. The expected class should require both correct content and correct arrangement. A molecule containing every expected feature in a different order is not full length in the structural sense.

Structural analysis at single-molecule resolution has shown how read tiling can separate expected, truncated, snapback, and other forms [7]. The classification grammar, minimum aligned fraction, breakpoint tolerance, and handling of low-quality ITR sequence must be versioned. Without these definitions, proportions from two reports may not be comparable.

Do not equate read length with integrity

A read near the designed genome length can contain an internal deletion plus a duplicated segment, or vector sequence joined to backbone of similar total length. Conversely, a shorter observed read may reflect pore termination, shearing, or base-calling compression rather than a packaged truncation. Length distributions are useful screening plots, but sequence architecture supplies the claim.

Treat ITRs as a special region

ITRs are repetitive and highly structured. Base-level calls, terminal completeness, and orientation can be less reliable than in the payload. Platform-specific error profiles and library preparation should be considered before reporting small ITR variants. The AAV ITR sequencing guide discusses these challenges. When ITR sequence identity is a primary endpoint, an orthogonal high-accuracy assay may be needed.

Figure 3. Structural classes are defined by ordered sequence segments on individual reads rather than read length alone.

Quantify Packaged DNA Impurities Carefully

Impurity analysis should use competitive alignment to vector, plasmid, helper, and host references. Mapping the vector first and sending only unmapped reads to an impurity database can miss chimeras, because the vector portion may capture the entire read during the first step. Segment-level classification preserves mixed-origin molecules.

Report both read fraction and base fraction. A few long impurity reads and many short impurity fragments can produce different values for these denominators. For host-cell DNA, also report uniquely mapped bases or reads and exclude low-complexity or telomeric assignments that lack discriminating sequence.

Long-read batch profiling has identified transfer-plasmid backbone, helper, and host-derived reads and demonstrated that analysis of chimeric assignments affects the result [5]. Studies combining long-read sequencing with capsid fractionation have also shown that intermediate particles can be enriched for noncanonical genome content [3]. These observations make impurity results method-specific unless extraction, library preparation, references, and counting rules are harmonized.

Quantitative PCR or digital PCR remains useful for predefined residual sequences. It can provide sensitive locus-specific measurement, while long reads reveal sequence context and mixed-origin structures. The methods should be reconciled by target and denominator rather than expected to return identical numbers.

Match Platform and Assay to the Claim

PacBio consensus sequencing can provide high per-molecule accuracy when enough passes are obtained. Nanopore sequencing offers flexible long-read acquisition and direct structural observation, but chemistry and base-calling version affect accuracy. Both platforms are influenced by the upstream conversion and size distribution.

The short-read versus long-read AAV sequencing guide explains broad platform tradeoffs. For genome QC, the practical selection criteria are the required base accuracy, the importance of ITR calls, expected molecule length, available input, rare-class sensitivity, and the need for rapid iteration.

Orthogonal evidence may include:

• capillary electrophoresis or analytical ultracentrifugation for particle or size populations;
• qPCR or digital PCR for known vector, backbone, helper, or host targets;
• short-read sequencing for local depth and small-variant confirmation;
• restriction mapping or gel-based analysis for gross size patterns;
• targeted PCR and Sanger sequencing for specific junctions;
• independent library preparation to confirm recurrent structural classes.

Plan sensitivity by structural class

Rare-structure sensitivity is governed by the number of informative molecules, not by a nominal sequencing depth alone. Before data generation, define the smallest fraction that would alter a research decision and calculate how many independent molecules are needed to observe that class with a stated level of confidence. Apply the calculation after anticipated filtering, because raw reads that are too short, low quality, or ambiguously assigned do not contribute equally.

Detection and quantification should be separated. Observing one or two supporting molecules may justify targeted follow-up, but it is rarely enough for a stable fraction estimate. Require reciprocal or repeated breakpoints, support in an independent library, or a junction-specific assay for low-frequency findings that matter to interpretation. Negative controls set the background for artificial ligation and reference misassignment, while a characterized positive material tests whether the workflow can recover the structural forms it claims to measure.

Report a per-class detection limit only when the underlying assumptions are defensible. If molecule recovery varies strongly with length or structure, describe the result as an observed library fraction and show the bias evidence instead of converting it into an absolute packaged-genome fraction.

Targeted Region Sequencing can support focused follow-up when a specific breakpoint or impurity junction is already known. Long-read QC and host integration analysis should remain separate: packaged-vector samples describe the vector preparation, whereas AAV Integration Site Analysis examines vector–host junctions in exposed biological material.

Compare Batches With Locked Rules

Batch comparisons require more than running the same sequencer. Use the same vector reference version, sample preparation, strand-conversion method, library kit, base caller, alignment settings, classification rules, and denominators. Randomize preparation order when possible and include a stable control material across runs.

The AAV sequencing batch-variability guide covers sample-to-bioinformatics sources of drift. For long-read structural QC, add class-level control charts for full-length, truncated, rearranged, chimeric, and impurity-derived molecules. A shift in one class should be reviewed against read length, quality, input, and processing metadata before it is attributed to production.

Do not silently retrain classification thresholds between batches. If the pipeline changes, reprocess representative prior data or run a bridging set. Versioned outputs allow process development teams to separate biological improvement from analytical change.

Specify a Reviewable Data Package

The final package should preserve read-level evidence and a concise batch summary. Include raw-data QC, read-length and quality plots, reference manifest, mapping summary, structural class counts and fractions, breakpoint tables, impurity-source tables, representative read maps, and a list of ambiguous reads.

The AAV confirmation playbook can help plan follow-up evidence. The report should avoid a single "percent intact" headline unless the definition, analyzable population, uncertainty, and excluded reads are displayed beside it.

Figure 4. A reviewable QC package connects summary proportions to references, processing metadata, and read-level evidence.

FAQ

For research use only. Not for use in diagnostic procedures or individual treatment decisions.