🧬 Aging Wiki

in-vivo-base-editing

Properties1
aliasesbase editing, adenine base editing, cytosine base editing, ABE, CBE, in vivo base editing, somatic base editing, LNP base editing, in-vivo gene editing

10 min read

In Vivo Base Editing (ABE / CBE; LNP-mRNA somatic editing)

Base editing installs a precise single-nucleotide change in genomic DNA without creating a double-strand break (DSB), using a deaminase enzyme tethered to a catalytically-impaired CRISPR nuclease (a “nickase”) that finds the target via a guide RNA. In vivo base editing delivers this machinery transiently — as mRNA-encoded editor + guide RNA in a lipid nanoparticle (LNP) — directly into a patient so that the permanent edit is written in the patient’s own cells, most tractably the liver. It is the platform behind crispr-base-editing-pcsk9 (VERVE-102), the subject of the first published Phase 1 in-vivo base-editing trial 1.

This page exists because the method’s ramifications are at least as important as any single target: a transient dose that writes a permanent, DNA-level edit is a categorically different therapeutic shape from a drug taken for life — with distinctive durability advantages and distinctive, irreversible failure modes.

Principle

Classical CRISPR-Cas9 cuts both DNA strands; the cell repairs the DSB by error-prone non-homologous end-joining, producing indels (gene knockout) but also a risk of large deletions, translocations, chromothripsis, and p53 activation. Base editors avoid the DSB entirely 2 3:

  • Adenine base editor (ABE): an engineered adenosine deaminase (evolved from E. coli TadA) fused to a Cas9 nickase. It deaminates adenine (A→inosine, read as G), and the nick on the opposite strand biases repair to install the complementary change — net A·T → G·C 3. VERVE-102 uses ABE version 8.8 1.
  • Cytosine base editor (CBE): a cytidine deaminase + uracil-glycosylase inhibitor + Cas9 nickase; net C·G → T·A 2. CBEs generally have higher transcriptome-wide RNA off-target and Cas-independent DNA off-target rates than ABEs.

The deaminase acts on a small “editing window” (~4–8 nt) within the protospacer; any same-type base in that window can also be converted (bystander editing). Two on-target editing strategies are used: directly installing a premature stop codon, or — as in VERVE-102 — disrupting a splice site so the transcript mis-splices and read-through reaches a stop codon, abolishing protein expression 1.

Prime editing (a reverse-transcriptase + nickase + extended guide) is a related DSB-free method that can write arbitrary small edits/indels, not just transition substitutions; it is earlier in in-vivo development and not covered in depth here.

Why “no DSB” is the load-bearing safety argument

Avoiding the DSB removes the dominant source of large-scale genomic rearrangement risk. This matters most for an aging/preventive indication delivered to older patients (more pre-existing chromosomal fragility) and for any future germline-adjacent or high-cell-number application. It is the central reason base editing, rather than nuclease knockout, is the lead modality for in-vivo somatic editing of metabolic targets.

Workflow (LNP-mRNA in-vivo editing, after VERVE-102)

  1. Cargo design — an mRNA encoding the base-editor protein + a chemically-stabilized guide RNA targeting the locus. Guide and editing-strategy choice (stop-codon vs splice-site) set the on-target window and bystander profile.
  2. Formulation — co-encapsulate mRNA + gRNA in an ionizable lipid nanoparticle (ionizable amino lipid, helper phospholipid, cholesterol, PEG-lipid). VERVE-102 adds a GalNAc ligand to the LNP surface.
  3. Single IV infusion — the LNP enters the bloodstream; PEG sheds; the particle binds ApoE and is taken up by hepatocytes via the LDL receptor, and (with GalNAc) via the asialoglycoprotein receptor (ASGPR) — two convergent hepatotropic uptake routes 1.
  4. Endosomal escape & translation — endosomal acidification destabilizes the LNP; mRNA + gRNA reach the cytosol; ribosomes translate the editor, which assembles with the guide.
  5. Edit — the editor–guide complex enters the nucleus, finds the target, forms an R-loop, the deaminase converts the target base, the nickase nicks the opposite strand, and DNA repair fixes the change. The mRNA and editor are then degraded (transient exposure; LNP terminal half-life <20 h in the VERVE-102 trial) — but the genomic edit is permanent.
  6. Read-out — because target-tissue biopsy is impractical clinically, the pharmacodynamic read-out is a downstream proxy: circulating PCSK9 protein and LDL cholesterol. Allele-level editing efficiency is measured directly only in nonclinical models (primary hepatocytes, NHP liver).

Output format

  • On-target editing efficiency — % of target alleles edited in the tissue (nonclinical); reported as protein/biomarker change in humans.
  • Bystander-edit profile — frequency of unintended same-type conversions within the editing window.
  • Off-target catalog — DNA off-targets (Cas-dependent, nominated by guide-similarity / GUIDE-seq / CIRCLE-seq; and Cas-independent, from deaminase overexpression) plus transcriptome-wide RNA A-to-I off-targets.
  • Durability curve — biomarker change vs time from the single dose (months → years).
  • Biodistribution — tissue-by-tissue editing (e.g. VERVE-102 NHP work: high in liver, low in adrenal glands and spleen across 26 tissues) and germline assessment (sperm/egg, offspring) 1.

Key parameters

ParameterTypical value (VERVE-102 context)Effect
EditorABE v8.8 (engineered TadA deaminase + Cas9 nickase)Determines edit type (A·T→G·C), window, off-target spectrum
Editing window~4–8 nt within protospacerSets bystander-edit risk
DeliveryGalNAc-LNP, single IVHepatotropic via ApoE/LDLR + ASGPR; sets tissue reach
Weight-based dose0.3–1.0 mg/kg total RNADose-dependent editing; absolute total RNA administered (mg/kg × body weight) correlated with LDL reduction (Pearson r=−0.68) 1
Editor exposureLNP terminal t½ <20 hTransient editor, permanent edit
On-target PD effectup to −88% PCSK9, −62% LDL-C at 1.0 mg/kgDownstream proxy for hepatic editing extent
Durability (human)stable ≥12 mo (max 18 mo follow-up)Persistence through hepatocyte turnover

Validation and QC

  • Direct allele-level editing in primary human hepatocytes and NHP liver (the strongest editing measure; not obtainable in trial participants).
  • Off-target nomination + confirmation: unbiased genome-wide nomination (GUIDE-seq, CIRCLE-seq, Digenome-seq) followed by deep targeted or whole-genome sequencing of nominated sites; error-corrected sequencing (duplex-sequencing) is the appropriate sensitivity tier for rare somatic off-target events, since standard NGS cannot reliably call sub-1% editing.
  • RNA off-target: transcriptome-wide RNA-seq to detect deaminase-driven A-to-I (ABE) editing during the transient expression window.
  • Germline safety: sequencing of sperm/oocytes and offspring in animal models to confirm absence of heritable edits (VERVE-102’s orthologous mouse editor showed no germline transmission) 1.
  • Biodistribution: multi-tissue editing quantification to bound extrahepatic exposure.
  • Target validation upstream of all of this: mendelian-randomization — choosing a target whose lifelong loss-of-function is human-genetics-validated as protective (e.g. PCSK9; pcsk9) is what makes a permanent edit a defensible bet.

Limitations and failure modes

  1. Liver-centric reach. ApoE/LDLR + GalNAc/ASGPR make the hepatocyte the privileged target; durable, efficient, safe in-vivo editing of muscle, brain, heart, or hematopoietic cells remains the field’s central unsolved delivery problem. Most viable near-term targets are therefore secreted hepatic factors (PCSK9, ANGPTL3, TTR, LPA, FGB/KLKB1).
  2. Off-target permanence. Every off-target or bystander edit is as permanent as the intended one. There is no “stop the drug” remediation; safety must be established before dosing, against a long latency for any oncogenic consequence. gap/long-term-unknown
  3. Editing measured by proxy in humans. Circulating-protein/biomarker change conflates editing extent with downstream physiology and inter-individual pharmacodynamics; it is weaker evidence than allele-level editing of the target tissue.
  4. Durability is assumed, not proven across decades. DNA-level edits should persist through cell turnover, but the cumulative-exposure rationale that motivates early editing requires multi-decade persistence; human data extend only a few years (the longest-followed in-vivo editing program, TTR-targeting nexiguran ziclumeran, reports stable effect through ~3 years) 1. gap/long-term-unknown
  5. LNP innate-immune activation. The lipid nanoparticle — not the editor — appears to be the principal driver of acute adverse events (infusion reactions, transient transaminase rise, thrombocytopenia in earlier programs); reformulation + GalNAc reduced this for VERVE-102, but anti-PEG immunity may constrain re-dosing. gap/long-term-unknown
  6. Donor/population diversity of nonclinical off-target work. Off-target landscapes are partly genotype-dependent; off-target screening done mainly in one ancestral background under-samples human genomic diversity. gap/needs-replication

Evidence-weight implications for this wiki

When a study reports an in-vivo (base-)editing result, weight it by:

  1. How editing was measured. Direct allele-level editing in the target tissue ≫ circulating-protein proxy ≫ a clinical biomarker alone.
  2. Off-target surveillance depth. Genome-wide unbiased nomination + error-corrected deep sequencing of nominated and genome-wide sites is the gold tier; “no predicted off-targets edited by standard WGS” is a weaker claim (standard NGS misses rare edits — see duplex-sequencing).
  3. Durability follow-up vs target-cell turnover. A 6-month read-out under-tests a therapy whose entire selling point is decades of permanence.
  4. Species and donor diversity of the nonclinical package.
  5. Reversibility framing. Because the edit is permanent, weight safety evidence more conservatively than for a discontinuable drug achieving the same biomarker change (e.g. a PCSK9 monoclonal antibody).

For the specific aging relevance — gene editing as the durable end of the intervention-durability spectrum, and as a “Tier 3” longevity strategy — see interventions-by-modality and longevity-escape-velocity.

MethodRelationshipNotes
duplex-sequencingOff-target surveillanceError-corrected sequencing is the appropriate sensitivity tier to detect rare somatic off-target edits below standard-NGS limits
mendelian-randomizationTarget validationHuman-genetics validation that a target’s lifelong LOF is safe + protective is the precondition for a permanent edit
Nuclease knockout (CRISPR-Cas9 DSB)Predecessor / alternativeCreates indels via DSB; higher rearrangement risk; base editing’s DSB-free chemistry is the safety improvement
Prime editingSibling DSB-free editorRT-based; writes arbitrary small edits beyond transition substitutions; earlier in in-vivo development
Epigenome editing (dCas9-DNMT/TET)Reversible analogueModifies expression without changing sequence; reversible — the opposite durability tradeoff (see dna-methylation)
mtDNA base editors (DdCBE / mitoBEs)Organelle-genome variantTALE- or Cas-free deaminases for mitochondrial DNA; relevant to mtDNA-mutation SENS strategies (see sens-damage-categories)

Pages citing this method

Limitations and gaps

  • #gap/needs-current-protocols-anchor — no Current Protocols chapter is cited; in-vivo base-editing protocols are program-proprietary. Update if a methods compendium chapter appears.
  • #gap/long-term-unknown — human durability beyond ~3 years (TTR program) and genome-wide off-target consequences over a latency long enough for oncogenesis are unestablished for any in-vivo base-editing program.
  • #gap/needs-replication — VERVE-102 is one program; a second in-vivo base-editing PCSK9 Phase 1 trial (reported in Nat Med 2026, cited within 1) is an independent replication thread to ingest when resolved to a primary source. DOIs for the secondary in-vivo-editing programs (nexiguran ziclumeran, the Nat Med base-editing trial) are not yet captured here — add on a verification pass.
  • #stub — GUIDE-seq / CIRCLE-seq off-target-nomination assays are referenced but have no dedicated method pages yet.

Footnotes

Footnotes

  1. vafai-2026-verve-102-pcsk9 · doi:10.1056/NEJMoa2601283 · Vafai SB, Täubel J, …, Kathiresan S · N Engl J Med 2026 · PMID 42187087 · n=35 · in-vivo (human Phase 1, single-ascending-dose) · single IV VERVE-102 (ABE8.8 mRNA + PCSK9 gRNA, GalNAc-LNP), 0.3–1.0 mg/kg; up to −88% PCSK9 / −62% LDL-C; durable ≥12 mo; LNP t½ <20 h; NHP biodistribution high-liver/low-adrenal-spleen across 26 tissues; no germline transmission of the orthologous mouse edit. Closed-access (NEJM) — verified against article PDF. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9

  2. doi:10.1038/nature17946 · Komor AC, Kim YB, Packer MS, Zuris JA, Liu DR · Nature 2016 · 533(7603):420-424 · in-vitro/methods · founding cytosine base editor (BE3); programmable C·G→T·A without DSB in mammalian cells; cross-checked against the verified crispr-base-editing-pcsk9 footnote. ↩ ↩2

  3. doi:10.1038/nature24644 · Gaudelli NM, Komor AC, Rees HA, Packer MS, Badran AH, Bryson DI, Liu DR · Nature 2017 · 551(7681):464-471 · in-vitro/methods · founding adenine base editor (ABE); evolved E. coli TadA fused to Cas9 nickase; A·T→G·C without DSB; ~50% efficiency, ≥99.9% product purity, ≤0.1% indels in the 7th-gen editor; cross-checked against the verified crispr-base-editing-pcsk9 footnote. ↩ ↩2

Graph View

Backlinks