Ready for your all-in-one single cell sequencing solution?

In Vivo Base and Prime Editing Insights through Single Cell Sequencing

Single cell sequencing with targeted variant detection makes it possible to identify which individual cells in a tissue carry a gene edit. Bulk assays fall short here, because they return only an average across the whole population. The level of resolution during analysis particularly matters during preclinical and translational development, when dose, delivery vehicle and target cell type are still being chosen. The first patient received an in vivo base editing therapy in February 2025, and in October 2026 the first patient was treated with in vivo prime editing. Most in vivo gene editing therapies in clinical development switch off the expression of a gene whose product drives disease. A newer approach repairs the pathogenic mutation itself, so the cell produces a wildtype protein again. The two readouts primarily used today, plasma protein assays and bulk sequencing of homogenized tissue, report whether editing occurred and at what frequency across a given tissue. Neither can report in which cells editing took place, nor what fraction of the cells the therapy was aimed at were reached. Read more about that below.

The first in vivo base editing treatment reached a patient in February 2025, when a bespoke base editing therapy was designed for an infant known as KJ to correct a CPS1 mutation causing a fatal urea cycle disorder. In March 2025, Beam Therapeutics reported the first clinical genetic correction of a pathogenic mutation in a trial setting. BEAM-302 is a lipid nanoparticle (LNP) formulation that delivers a base editor to hepatocytes, where it corrects the E342K mutation in SERPINA1 that causes AATD.

By March 2026, corrected M-AAT, the wildtype form of alpha-1 antitrypsin, made up 94 % of all circulating AAT in the 60 mg cohort, while mutant Z-AAT decreased by 84 %. In every patient in that cohort, total AAT stayed above the protective threshold of 11 µM through up to 12 months of follow-up.

Base editing was not the only approach being tested on patients. On 5 October 2026, prime editing was used to treat a patient for the first time. Prime Medicine announced that the first participant had been treated with PM577a, an in vivo prime editor for Wilson disease.

As of 2026, all ongoing in vivo base editing clinical trials are in Phase 1 or Phase 1/2, and the most advanced is BEAM-302, whose pivotal cohort was first treated in July 2026. Most of these trials rely on gene silencing. For that purpose, the bulk editing rate across a tissue is a reasonable proxy for success.

All of these in vivo base editing programs target the liver. That is no coincidence, because lipid nanoparticles naturally accumulate in hepatocytes. Most of them switch off gene expression: PCSK9 and ANGPTL3 to lower LDL cholesterol, TTR to reduce amyloid-forming protein, APOC3 to lower triglycerides. All of these genes encode proteins that hepatocytes secrete into the bloodstream, so their levels can be measured via a simple plasma assay.  

Most proteins, however, are not secreted.

Take G6PC, the target of Beam’s BEAM-301 candidate in glycogen storage disease type Ia. It encodes an enzyme anchored in the membrane of the endoplasmic reticulum. There is no circulating G6Pase to measure, so downstream metabolites such as glucose and lactate serve as indicators instead.

Two separate factors affect whether a simple readout is sufficient. The first is whether the protein is secreted, since a secreted protein can be measured easily in plasma, while a membrane-bound or intracellular one cannot. The second is precision. Silencing only needs to deplete the protein, so the exact edit is less important. Correction has to restore one specific base, and any edit other than the intended one could result in a molecule that is neither the disease variant nor the wildtype. Questions that need to be answered in such a correction trial include: which cells carry the gene edit, what fraction of the therapeutically relevant population was reached, and whether those cells remained healthy afterwards. To address these questions, it is necessary to obtain single cell resolution data.

Base editing needs three components:

  • A Cas protein, either catalytically dead or converted into a nickase, fused to a deaminase that performs the chemical conversion
  • A guide RNA that delivers the complex to one specific locus
  • A target base that falls inside the editing window, which the Cas protein defines

Each class of editor works through deamination, but on a different base and with a different intermediate:

BaseEditing methodConverted baseBase read after replication
CCytidine deaminationUracil (U)Thymine (T)
AAdenosine deaminationInosine (I)Guanine (G)

On the opposite strand, these changes appear as G to A and T to C respectively. Between them, the two classes therefore cover all four transitions, which account for roughly two thirds of known pathogenic point mutations.

Cytosine base editors need two further safeguards. Left alone, the cell would simply undo the edit, because base excision repair recognizes uracil in DNA and removes it. These editors therefore carry a uracil glycosylase inhibitor (UGI) to block that repair, and they use a nickase rather than a dead Cas9. The nick is placed on the opposite, unedited strand, so the cell treats the edited strand as the template and keeps the change instead of reverting it.

Adenine base editors are cleaner in one specific respect: inosine is rarely excised, so adenine editing yields the intended product more consistently. Crucially, neither class leads to a double-strand break, so the cell never employs error-prone repair pathways that produce indels.

The editing window commonly spans positions four to eight of the protospacer, but it varies between editors. Engineered variants narrow it to one or two nucleotides, while others place it at different positions relative to the PAM. Window width can vary depending on the design, and is the main lever for limiting unintended edits at neighboring bases.

Prime editing works on a different principle and needs three components of its own:

  • A Cas nickase fused to a reverse transcriptase
  • A prime editing guide RNA (pegRNA) that targets the site and carries a template encoding the intended edit
  • A primer binding site at the 3′ end of that pegRNA, which anneals to the nicked DNA strand and primes synthesis

The template is written directly into the nicked strand. The reach is wider than the four transitions base editing offers: all twelve possible base substitutions, including transversions, plus small insertions and deletions. Unlike base editing, the edit position is set by the template rather than by its distance to the PAM, so it is not restricted to a fixed window.

Both approaches result in a mosaic population of edited and unedited cells.

But what decides which cells get edited? The process is largely stochastic, and is influenced by four factors:

  • LNP delivery
  • Chromatin accessibility at the target locus
  • Cell cycle state (several steps in repair mechanisms depend on replication)
  • Local repair environment determining whether an edit is kept or reversed

As a result, some cell types are systematically more likely to be edited than others, even within the same organ.

As described above, a gene silencing approach can lean on a surrogate endpoint. A correction program has no such shortcut, because the restored protein must carry the precise sequence change that makes it work.

Base editing harbors a higher risk for bystander edits than prime editing. Even though the guide is highly specific, the deaminase acts on any suitable base inside the editing window, not only on the intended one. If several cytosines are located within those few positions, different cells end up with different combinations of conversions. Therefore, careful target design is necessary to avoid these bystander edits. Nevertheless, distinguishing a cell that received the intended edit from one that received a bystander product requires reading the specific position. Prime editing avoids bystander effects, but it frequently yields incomplete or imprecise products that pose the same analytical challenge.

Early cytosine base editors produced a minority of C to G and C to A conversions alongside the wanted C to T, and successive generations were engineered specifically to improve this product purity. Dual editors that convert cytosines and adenines in the same reaction widen the range of outcomes further. Confirming that editing occurred therefore is insufficient. Precise analysis of the exact base edits is required to reveal which of several possible products a given cell actually carries.

As correction programs move to the mainstream of the field, the readout has to answer these questions of high regulatory importance to ensure safety and efficacy of the therapies.

In vivo editing depends on delivery, which can be uneven across different cell types. For example, an LNP formulation designed for hepatocytes will also reach other liver-resident populations, even though with lower probability. Additionally, cellular uptake does not guarantee editing. A bulk measurement from homogenized tissue reports an average across every cell type present. It therefore cannot distinguish a therapy that edited half of the intended population from one that edited a quarter of a much larger bystander population.

The number that predicts clinical benefit most accurately, however, is not the editing rate across the organ but the fraction of edited, therapeutically relevant cells.

While an edit can succeed at the sequence level, side effects of the delivery and of installing the edit need to be taken into consideration too. These could include DNA damage responses, interferon signatures triggered by the delivery vehicle or the editor mRNA, and shifts in differentiation state, but none of them is captured by measuring bulk sequencing output.

Standard single cell RNA sequencing captures transcripts at their poly(A) tail. Sequencing coverage is highest near the capture site and decreases with distance, so the 3′ end of a transcript is covered reliably while positions further upstream are not. Edit sites, however, are located where the pathogenic mutation sits, which is frequently far from the 3′ end and therefore often in a region with little coverage. Low-abundance transcripts compound this issue, because a gene expressed at low levels is rarely covered by  enough reads for confident variant calling.

FocuSCOPE® addresses this by adding transcript-specific probes alongside the standard poly(dT) oligos on the capture bead. The probes bind close to the region of interest, bringing it into the high-sensitivity zone next to the capture site, wherever it sits within the transcript. Enrichment therefore happens at capture rather than after reverse transcription, and the native sequence at the edit position is read directly. Because poly(A) capture runs in parallel, the full transcriptome of the same cell is recovered alongside the genotype. Up to ten target regions can be captured per run without losing resolution on the whole transcriptome.

The resulting dataset includes cell type, cell state, and edit status of the same cell. Tropism therefore becomes a question of which annotated cell types carry the edit, efficiency becomes a per-population figure instead of a tissue average. This therefore allows for direct comparison between edited and unedited cells of the same cell type.

Targeted RNA-based variant detection does not replace single cell DNA sequencing. Single cell DNA approaches characterize the genome itself: zygosity of the edit, co-occurrence of multiple edits, predicted off-target sites, translocations and copy number changes. They are well established in ex vivo cell therapy, where the edited cells can be sampled and analyzed before being transfused back to the patient.

For In vivo editing, the sample is a biopsy or a piece of animal tissue. With a heterogeneous population, the central question goes beyond just the genome: which cells in this tissue were edited, and how did they respond? That question needs genotype and phenotype from the same cell. Single cell genomic and transcriptomic sequencing approaches combine both information, and cover the full picture.

In vivo base and prime editing have moved from proof of concept to the clinic faster than most of the field expected. Current Phase 1 and Phase 1/2 trials are generating the data that will decide which approaches, which delivery systems, and which indications advance. Those decisions rest on knowing not just whether editing occurred, but where, in which cells, and with what side effects.

Bulk sequencing investigates the frequency of cells that were edited across a tissue sample. Single cell sequencing with targeted variant detection determines which cells carry the intended edit, how that fraction compares with the cells the therapy was meant to reach, and how the transcriptome of those cells changed as a result.

Precision editing needs precision measurement. A therapy that changes a single base in a single cell type deserves a readout at the same resolution.

The table below lists the candidates referenced in this article. All remain in early clinical development.

ProgramCompanyTargetIndicationApproachPhase
BEAM-302Beam TherapeuticsSERPINA1Alpha-1 antitrypsin deficiencyBase editing — correctionPhase 1/2 (pivotal cohort dosing)
BEAM-301Beam TherapeuticsG6PCGlycogen storage disease type IaBase editing — correctionPhase 1/2
VERVE-102Verve TherapeuticsPCSK9HypercholesterolemiaBase editing — knockoutPhase 1b
YOLT-101YolTech TherapeuticsPCSK9Familial hypercholesterolemiaBase editing — knockoutPhase 1
PM577aPrime MedicineATP7BWilson diseasePrime editing — correctionPhase 1/2

Frequently asked questions

Common questions about in vivo base editing, prime editing and how editing outcomes are measured.

Base editing uses a Cas protein fused to a deaminase to chemically convert one base into another, which covers the four transition mutations: C to T, G to A, A to G and T to C. Prime editing fuses a Cas nickase to a reverse transcriptase and writes a templated sequence into the genome, so it can install all twelve possible substitutions plus small insertions and deletions.

A bulk measurement reports one average across every cell in a homogenized tissue. It therefore cannot distinguish a therapy that edited half of the intended cell population from one that edited a quarter of a much larger bystander population. For a correction therapy, the number that predicts clinical benefit is the fraction of therapeutically relevant cells carrying the edit, not the editing rate across the organ.

The guide RNA directs a base editor to one specific site, but the deaminase acts on any suitable base inside the editing window, which commonly spans positions four to eight of the protospacer. When several cytosines sit within those positions, different cells end up with different combinations of conversions. Confirming that editing occurred therefore says little; only reading the exact position reveals which product a given cell carries.

Often not reliably. Standard scRNA-seq captures transcripts at the poly(A) tail, and coverage falls off with distance from the capture site, so edit sites in coding exons frequently sit where coverage is thin. Low-abundance transcripts compound the problem, because a gene expressed at a few copies per cell rarely accumulates enough reads for confident variant calling.

It adds transcript-specific probes alongside the standard poly(dT) oligos on the capture bead, so the region of interest is enriched at the point of capture rather than after reverse transcription. The native sequence at the edit position is read directly, while poly(A) capture runs in parallel and recovers the full transcriptome of the same cell. Genotype and phenotype therefore come from one cell rather than from two separate experiments.

As of 2026, every in vivo base editing program in the clinic remains in Phase 1 or Phase 1/2, and all of them target the liver because lipid nanoparticles naturally accumulate in hepatocytes. Examples include BEAM-302 for alpha-1 antitrypsin deficiency, BEAM-301 for glycogen storage disease type Ia, and VERVE-102 and YOLT-101 targeting PCSK9. PM577a for Wilson disease became the first in vivo prime editing therapy to dose a patient, in October 2026.

No, the two answer different questions. Single cell DNA approaches characterize the genome itself, including zygosity, co-occurring edits, predicted off-target sites and translocations. Targeted single cell RNA sequencing shows which cells in a mixed tissue carry the intended edit and how those cells responded, which is what in vivo studies on biopsy material usually need.

  • Beam Therapeutics. Positive initial data for BEAM-302 in the Phase 1/2 trial in alpha-1 antitrypsin deficiency, demonstrating first ever clinical genetic correction of a disease-causing mutation. Press release, 10 March 2025. [link]
  • Beam Therapeutics. Compelling updated clinical data from the ongoing Phase 1/2 trial of BEAM-302 in AATD to support advancement to pivotal development. Press release, 25 March 2026. [link]
  • Beam Therapeutics. BEAM-302 clinical data presented at the European Respiratory Society Congress. 8 September 2026. [link]
  • Beam Therapeutics. Second quarter 2026 financial results and first patient dosed in the global pivotal cohort of the BEAM-302 trial. Press release, 4 August 2026. [link]
  • Vafai SB, Täubel J, Ashdown T, et al. In vivo base editing of PCSK9 with VERVE-102 for hypercholesterolemia. N Engl J Med. 2026;395(7):648–659. [link]
  • Prime Medicine. First patient dosed in global Phase 1/2 clinical trial of PM577a for H1069Q-mutated Wilson disease. Press release, 5 October 2026. [link]
  • Wan, P., Tang, S., Lin, D. et al. In vivo base editing gene therapy for heterozygous familial hypercholesterolemia: a phase 1 trial. Nat Med 32, 1045–1051 (2026). [link]
  • Musunuru K, Ahrens-Nicklas RC, et al. Patient-specific in vivo gene editing to treat a rare genetic disease. N Engl J Med. 2025;392:2235-2243. [link]