Author: Suzanne Snellenberg, PhD – VP Cell Line Development
Over the past two years, the U.S. Food and Drug Administration (FDA) has signaled, through two separate draft guidances, that whole genome sequencing (WGS) is moving from a nice-to-have to an expected part of the safety package for genome-edited and allogeneic cell therapy products. Although both guidances remain in draft, the technical questions they raise can already be addressed experimentally. Here we present a study with our induced pluripotent stem cell (iPSC) cell line development (CLD) workflow to evaluate what a genome-wide, dual-platform sequencing approach reveals about the genomic integrity of a gene-edited iPSC line, and what is required to implement such an approach in practice. We describe the study design, the principal findings, and their implications for genomic safety assessment.
The regulatory signal
Two FDA draft guidances now point in the same direction. The April 2024 draft, Safety Testing of Human Allogeneic Cells Expanded for Use in Cell-Based Medical Products¹, recommends that WGS be performed on cell banks of continuous cell lines, at a minimum read depth of 50X, with results compared against cancer-associated mutation databases. For highly expanded genome-edited clones, it further calls for sequencing that identifies off-target editing, on-target editing outcomes, and vector integration events.
The April 2026 draft, Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing², goes further on the editing side specifically. It recommends multiple, complementary approaches, cell-based assays, biochemical assays, and in silico prediction, to nominate candidate off-target sites, confirmatory NGS to verify them, and genome-wide NGS-based assessment for chromosomal integrity wherever an editor creates double-strand breaks. It also draws a distinction sponsors will need to plan around: short-read sequencing for small edits, long-read sequencing where larger insertions, deletions, or structural rearrangements are possible.
Read together, the direction is clear: FDA is asking sponsors to think about genomic safety with data, generated across the product's development history, rather than relying solely on assessments performed at the final stage. For cell therapy developers, this changes what belongs in a safety package. Genome-wide sequencing evidence seems to be moving from a discretionary addition to an expected component. Not addressing the topic now may create significant risks in the future, such as review queries that cannot be answered, clinical holds, or, in the worst case, a conclusion that the cell line is not fit for purpose, which could cost millions and multi-year delays.
What we set out to test
We had two questions: could our workflow generate the kind of data these draft guidances describe, and where would the practical difficulties arise?
We sequenced three samples spanning a single cell line's development, all derived from the same donor lineage:
The rationale for sequencing the same lineage at all three stages was to enable any genomic changes to be traced back to the point at which they first arose, whether during reprogramming or during gene editing. Sequencing only the final drug substance would have left that history to inference.
We ran each sample through both short-read (Illumina) and long-read (PacBio HiFi) whole genome sequencing. The two platforms cover different ground. Illumina provides deep, highly accurate calls for small variants (SNVs and indels). PacBio HiFi, based on circular consensus sequencing (10-12 passes per molecule), can resolve structural variants, duplications, and repetitive regions that short reads cannot. Used together, each compensates for the other's limitations.
Finally, an independent clinical genomics partner interpreted the data, classifying variants against American College of Medical Genetics and Genomics (ACMG) criteria, ClinVar, gnomAD population frequency, and CADD pathogenicity scores, without pre-filtering the call set.
What the data showed
All three samples aligned at >99.9%, with 55-76x coverage across both platforms. Three autosomal recessive variants (FECH, CPAMD8, HKDC1), each a single heterozygous allele (VAF 45–58%, consistent with clean heterozygous calls (~50%)) with low disease risk on its own, were detected in all three samples. These donor-inherited variants confirm a consistent lineage across the workflow (Figure 1). A lineage check of this kind rules out cross-contamination or sample mix-ups between process stages, thereby confirming the identity of the generated cell lines.
Figure 1. Variant Allele Frequency (VAF) Across Samples. Blue line is PacBio WGS and orange line is Illlumina WGS data. *) low-confidence call; PacBio calls variant cleanly.
On-target editing was confirmed by three independent methods. Both the B2M and CIITA knockouts in the Allo Chassis™ edited cell line were called at the same position, with the same alleles and genotype, by amplicon sequencing, PacBio, and Illumina: heterozygous for two distinct repair outcomes at B2M and homozygous at CIITA (Figure 2).
Exact agreement across three orthogonal methods is strong evidence that the edits were made as intended.
Figure 2. On-Target Edit Confirmation: Three-Way Agreement. At B2M, the −5 bp and −11 bp alleles were detected at comparable frequencies by amplicon sequencing (34,171 and 30,811 reads, respectively), PacBio WGS (38/64 and 26/64 reads), and Illumina WGS (34/73 and 39/75 reads), confirming the presence of both edited alleles. At CIITA, the homozygous −2 bp allele was supported by 53,078 amplicon-seq reads, 46 PacBio reads, and 59 Illumina reads.
Genome-wide analysis identified one acquired variant that warranted further investigation: a pathogenic frameshift variant in SEC63, detected only in the edited line and absent from both the donor cells and the parental iPSCs (Figure 3). Unlike the recessive variants above, SEC63 follows a dominant inheritance pattern, so a single allele could be clinically relevant. Because the variant appeared at the editing step, it also had to be evaluated as a possible CRISPR off-target effect. In silico off-target prediction (CatsOFFinder) and guide-sequence homology analysis ruled this out: the variant sits on a different chromosome from both edit targets, with no meaningful sequence homology to either guide.
The remaining question was whether it was a genuine low-frequency variant acquired during culture or a sequencing artifact.
Figure 3. SEC63 finding. Illumina and amplicon-seq, two independent methods, both show no edited-specific signal at this locus, while only PacBio does. That pattern points to a PacBio-specific sequencing or alignment effect isolated to the edited sample, not a confirmed genomic variant.
The variant sits inside a 10 bp homopolymer run, a classic error-prone context for indel calling, with a call quality well below our other clinical variants and sub-heterozygous allele frequency. PacBio called it at VAF 28% (16/58 reads); independently sequenced Illumina data did not call it at the same position. To investigate the discrepancy, we performed targeted amplicon sequencing of the locus, generating approximately 83,000 reads per sample. At this depth, a genuine variant present at 28% VAF would have been detected with high confidence, substantially exceeding the 58 PacBio reads covering the same region. However, amplicon sequencing revealed essentially identical background signals in both the edited and unedited cell lines. Taken together, the discordant findings between sequencing platforms and the absence of support from a substantially deeper orthogonal assay strongly suggest that the PacBio call represents a platform-specific consensus-calling artifact associated with a homopolymer sequence, rather than a true variant introduced during gene editing.
We found this instructive on its own. A single sequencing platform, viewed in isolation, would have reported this as a confirmed pathogenic variant acquired during editing. It took two, independent methods to catch what turned out to be a false signal. In practice, a false positive of this kind, accepted at face value, could trigger an unnecessary root-cause investigation, delay batch release, or, in a worst case, lead to discarding an otherwise sound cell bank - outcomes with real cost and timeline consequences.
The reverse case also occurred. The independent clinical genomics partner reported a likely pathogenic stop-gain variant in VWF in all three samples, including the donor. The call came from short reads that could not be placed uniquely in the genome: VWF has a closely related pseudogene (VWFP1) on chromosome 22, the Illumina variant caller had itself flagged the call as a likely false positive, and the long PacBio reads, which span the region unambiguously, showed only the reference sequence. Each platform therefore produced one artifact, and each was resolved by the other.
One variant, unrelated to editing, was traced to an earlier process step. A pathogenic nonsense variant in VPS13B, an autosomal recessive gene, was absent in the donor material but present in both the parental and edited iPSC lines as a single heterozygous allele, indicating that it arose during reprogramming, before gene editing. Given its recessive inheritance, a single allele carries low disease risk. Tracing it to that step, rather than discovering it only in the final edited product, is exactly the value of sequencing at multiple points in the CLD workflow rather than only at release.
Beyond these findings, six further heterozygous variants were first detected in the derived lines: four variants of uncertain significance and a 73 bp intronic repeat expansion after reprogramming, and one variant of uncertain significance after gene editing. None lies at a predicted off-target site, in a gene recurrently mutated in cultured pluripotent stem cells, or among approximately 2,900 tumor genomes in the PCAWG dataset, and none showed evidence of clinical relevance. Such variants may also have been present below the detection limit before a single-cell cloning step fixed them in the line. Copy-number and structural variant analysis showed no aneuploidy, no 20q11.21 gain, and no translocation or rearrangement at either cut site.
Lessons learned
Three conclusions from this study are relevant to developers planning a genomic characterization strategy in line with the FDA draft guidances:
These results support including WGS at multiple process checkpoints, not only on the final bank, as a standard, built-in element of genomic safety assessment on our Pulse™ platform for iPSC-derived allogeneic cell therapy development. In effect, the draft guidances ask for evidence that shows sponsors and regulators what is happening across the whole genome, rather than only at the sites predicted to be at risk. Establishing this capability early, whether in-house or through a partner with a validated workflow, leaves time to investigate unexpected findings before a submission deadline.
¹ U.S. Food and Drug Administration. Safety Testing of Human Allogeneic Cells Expanded for Use in Cell-Based Medical Products: Draft Guidance for Industry. April 2024. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/safety-testing-human-allogeneic-cells-expanded-use-cell-based-medical-products
² U.S. Food and Drug Administration. Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing: Draft Guidance for Industry. April 2026. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/safety-assessment-genome-editing-human-gene-therapy-products-using-next-generation-sequencing