CRISPR Epigenetic Editing: How Scientists Silence and Reactivate Genes Without Cutting DNA

Last Updated: October 2026

Standard CRISPR works by breaking DNA, and the cell’s repair crew decides what the edit finally looks like. CRISPR epigenetic editing drops that bargain. It uses a Cas9 protein that has lost its cutting ability, steers it to a gene’s control region, and then writes or erases chemical marks that decide whether the gene is read. The sequence is untouched. What changes is the cell’s instruction to use it.

For years this was a laboratory curiosity with a durability problem: silencing faded within days or weeks. That has changed. In 2026, three different epigenetic editors are in human trials, and the first efficacy-style readouts are public. A liver-delivered silencer reported hepatitis B marker suppression that persisted for 17 months in some participants. A muscle-directed editor reported measurable lean-muscle gains in a rare dystrophy. Both are small, early studies, and neither is an approved therapy.

This post explains the mechanism from first principles, the main tool families, the delivery constraint that shapes every program, and what the clinical data does and does not show.

What this covers: how methylation silences genes, how dCas9-KRAB, CRISPRoff and zinc-finger editors differ, why delivery decides which diseases are reachable, the October 2026 clinical scoreboard, honest limits such as durability and off-target effects, and a checklist for reading the next press release.

What Changed for October 2026

The April 2026 version of this post was a conceptual overview with several claims I could not tie to primary sources. This rewrite replaces it. Treat the earlier numbers as withdrawn.

• Human data now exists. Tune Therapeutics presented Phase 1b/2a data for TUNE-401 in chronic hepatitis B at the EASL Congress on May 30, 2026. Epicrispr reported six-month data for EPI-321 in facioscapulohumeral muscular dystrophy (FSHD) on June 26, 2026.
• A second hepatitis B program started dosing. nChroma Bio dosed the first participant with CRMA-1001 on January 28, 2026, after clearance in Hong Kong in December 2025.
• Funding followed the data. Epicrispr announced a $90 million Series C on August 11, 2026, and said six-month follow-up results were expected in early October 2026.
• The press noticed. Nature published a news feature on September 21, 2026 titled “Epigenetic editing is here,” about hepatitis B silencing with chemical tags.
• The field is still pre-pivotal. No epigenetic editor has completed a Phase 3 study or won approval. Every efficacy number in this article comes from small, open-label, company-reported studies unless I say otherwise.

Context and Background

Gene editing grew up around cutting. The original CRISPR-Cas9 system makes a double-strand break at a guide-chosen site, and the cell repairs it, usually by non-homologous end joining, which scrambles a few letters and knocks a gene out. That is powerful for diseases where removing a gene helps, and it underlies approved therapies. It also brings known costs: unintended cuts, large deletions and rearrangements at the target, and an edit that is permanent whether or not it turns out to be wise.

The next generation softened the cut. Base editors and prime editors rewrite single letters or short stretches without a full double-strand break, which our overview of base editing and single-base CRISPR therapeutics covers in detail. They still change the sequence, and they still depend on the sequence being the problem.

Many diseases are not sequence problems. The gene is intact but it is read too much, as in FSHD, where a normally silent gene called DUX4 switches on in muscle. Or a viral genome sits in the nucleus doing damage. Or the goal is to turn a gene down in a tissue indefinitely, such as lowering PCSK9 to reduce LDL cholesterol. For these, the useful intervention is control of expression, not rewriting of letters.

Epigenetic editing targets exactly that layer. Cells already use chemistry to remember which genes are off: methyl groups on cytosines and modifications on histone proteins. Epigenetic editors borrow the cell’s own machinery and aim it. The scientific foundations are described in the 2021 Cell paper from the Weissman and Gilbert labs introducing CRISPRoff, which reported programmable silencing that was heritable through cell division. Reviews such as the Nature Reviews Cardiology summary of a PCSK9 epigenome-editing study show how quickly the idea moved to animal therapeutics.

A caution on vocabulary. “Epigenetic” is used loosely in media. Here it means two precise things: DNA methylation at CpG dinucleotides, and histone tail modifications such as H3K9 trimethylation (H3K9me3) and acetylation. The editors described below write or erase these marks at chosen loci. They do not reprogram a cell’s global identity, and they do not reverse ageing, whatever a headline implies.

How CRISPR Epigenetic Editing Works: The Reference Architecture

CRISPR epigenetic editing fuses a catalytically dead Cas9 (dCas9), which binds DNA but cannot cut it, to an enzyme or repressor domain that changes chromatin marks. A guide RNA selects the promoter or enhancer. The fusion then adds or removes methylation or histone marks, switching transcription down or up without altering the underlying DNA sequence.

The biology being hijacked

In mammals, roughly speaking, a methyl group added to the fifth carbon of cytosine, mostly at CpG sites, is a durable “off” signal when it sits in a promoter. DNA methyltransferases write it: DNMT3A and DNMT3B lay down new marks, and DNMT1 copies the pattern onto the new strand after each cell division. That copying step is why methylation can be inherited by daughter cells and why it is the central bet behind long-lasting editors.

Methylation does not block transcription by itself in most cases. It recruits readers. Methyl-CpG-binding domain proteins such as MeCP2 and MBD2 bind the marks and bring histone deacetylases (HDACs) and histone methyltransferases. The result is H3K9me3 and compacted chromatin, which keeps transcription factors out. Removal runs the other way: TET enzymes oxidize 5-methylcytosine, and the oxidized forms are eventually replaced with unmethylated cytosine. Figure 1 shows the loop.

Figure 1: The methylation cascade that CRISPR epigenetic editing hijacks. Writers add marks, readers compact chromatin, and TET enzymes provide the reverse path.

The diagram reads left to right. An unchanged DNA sequence acquires CpG methylation, readers bind it, repressive histone writers follow, and the promoter becomes inaccessible. The dashed path shows that the state is reversible in principle, which is what makes both silencing and reactivation tools possible.

The anatomy of an editor

Every epigenetic editor has three parts. The first is a DNA-binding module, which is either a dCas9 steered by a guide RNA or an engineered zinc-finger protein (ZFP) whose fingers recognize a DNA triplet each. The second is an effector, the part that does the chemistry. The third is a delivery format, such as mRNA in a lipid nanoparticle or a gene in a viral vector.

The binding module sets specificity. Cas9 is easy to retarget by changing a roughly 20-nucleotide guide, and its need for a PAM sequence next to the target limits where it can bind. Zinc fingers are harder to design but are small, human-derived, and fit inside size-limited vectors. That tradeoff recurs throughout the clinical programs.

The effector sets durability. Repressors such as KRAB (Krüppel-associated box) recruit KAP1/TRIM28 and, through it, H3K9me3 writers and HDACs. KRAB works quickly and is potent, but on its own it is typically transient: when the editor protein disappears, many sites drift back. Catalytic DNA methyltransferase domains write 5-methylcytosine that DNMT1 can propagate, which is the route to memory.

dCas9-KRAB and CRISPRi: the transient baseline

The best-known silencer is dCas9-KRAB, the core of CRISPR interference (CRISPRi). It is a workhorse for genome-wide screens because it represses a gene while the protein is present, and it is reversible. Its weakness for medicine is exactly that reversibility: if the effector is delivered as a transient mRNA, the effect tends to fade. For a screen that is a feature. For a therapy that must last years, it is a gap.

CRISPRoff and CRISPRon: adding memory

The 2021 CRISPRoff design fused dCas9 to the methyltransferase DNMT3A, its partner DNMT3L, and a KRAB domain, so the editor both recruits repressive histone marks and writes CpG methylation. The authors reported that a single transient expression could silence many genes and that the silencing persisted through cell divisions and, in their experiments, through differentiation of stem cells into neurons. A companion CRISPRon variant used TET1 to remove methylation and reactivate genes. I am summarizing the published abstract-level claims here; the paper is the place to check the specific gene list and percentages, and the later work below is the stronger evidence for durability in living animals.

Figure 2: Anatomy of a CRISPRoff-style epigenetic editor and how its two effector arms converge on a silenced promoter.

The figure shows two arms from one binding module. The KRAB arm produces fast, histone-based silencing. The DNMT3A/DNMT3L arm writes CpG methylation that DNMT1 maintains after division. Combining them gives both speed and memory, which is why multi-domain editors dominate current clinical designs.

Zinc-finger editors and hit-and-run silencing

Not every clinical editor is Cas-based. A 2024 Nature paper from Angelo Lombardo’s group at San Raffaele in Milan, described in the Nature article on hit-and-run epigenome editing, built evolved engineered transcriptional repressors that combine a zinc-finger DNA-binding domain with KRAB and DNMT3A/DNMT3L activity in a single molecule. Delivered as mRNA in lipid nanoparticles, one dose lowered circulating mouse Pcsk9 and kept it lower for close to a year. The silencing survived forced liver regeneration after partial hepatectomy, which is a stringent test, because dividing cells dilute any mark that cannot be copied.

That result matters for the thesis of this post. “Hit-and-run” means the editor is expressed briefly and then degraded, while the mark it left behind persists. If that holds in humans, the safety profile can be better than a permanent genetic edit, because the foreign protein is gone within days.

A 2025 Nature Medicine paper, a potent epigenetic editor targeting human PCSK9, tested a dCas9 fused to DNMT3A, DNMT3L and KRAB delivered as mRNA in lipid nanoparticles. The reported results were roughly 98% PCSK9 reduction in transgenic mice (with durability to at least one year) and roughly 90% PCSK9 reduction with about 70% LDL-cholesterol reduction in cynomolgus monkeys at the optimal dose, followed for at least three months. Monkeys showed transient liver enzyme elevations that resolved by day 14. These are preclinical figures from the paper’s summary and should not be read as human expectations.

Reactivation: the other half of the title

Silencing gets the attention, but the same architecture can turn genes up. Options include dCas9 fused to transcriptional activators such as VP64 or VPR (CRISPRa), TET1 catalytic domains to erase methylation, or acetyltransferases such as p300 to open chromatin. Reactivation is attractive for haploinsufficiency, where one working copy of a gene makes too little protein and boosting the healthy copy could compensate. It is also less mature clinically: activation usually needs sustained editor presence, because nothing in the cell naturally “remembers” an on state as robustly as it remembers an off one. Treat durable activation as an open research problem rather than a solved capability.

Delivery Decides the Disease: From Lab Tool to Clinical Program

An epigenetic editor is a large protein plus a guide. Getting it into the right cells at a dose the body tolerates is the real engineering problem, and it dictates which diseases are reachable first. Figure 3 gives the path every program follows, and Figure 4 shows the delivery fork.

Figure 3: The development path for CRISPR epigenetic editing programs, from construct design to first-in-human studies. Pivotal trials have not started.

The path is conventional drug development with one twist: durability must be shown at every rung. A mouse that stays silenced for a year is encouraging, but liver cells in mice and primates divide differently from human ones, and the question of whether marks survive years of turnover in people can only be answered by long follow-up.

Lipid nanoparticles and the liver

Lipid nanoparticles (LNPs) carrying mRNA are the most practical route for liver targets. After intravenous infusion, LNPs accumulate in hepatocytes, the cells translate the mRNA into editor protein, and the protein and mRNA are cleared within days. That is the hit-and-run profile. It suits diseases that live in the liver: PCSK9 and LDL cholesterol, and chronic hepatitis B, where the virus keeps its genome in liver cells.

The cost is the narrow tissue range. LNPs reach the liver readily and most other organs poorly. Repeat dosing also becomes a question when an immune response to the lipid or the payload develops. TUNE-401’s multiple-dose cohort, discussed below, is one of the first human tests of repeat dosing for an epigenetic silencer.

AAV vectors and everything else

For muscle and the nervous system, programs generally turn to adeno-associated virus (AAV) vectors. AAV can be targeted to tissues by capsid choice and can deliver a gene that keeps producing editor for a long time. That longevity is an advantage for activation and a risk for silencing editors, since continuous expression raises exposure to off-target binding and to immune responses against a bacterial protein. Dose is also a limit: systemic AAV doses are measured in vector genomes per kilogram, and high doses have been linked in other gene therapies to serious liver and other toxicities. Epicrispr’s EPI-321 is dosed at 2 x 10^13 and 4 x 10^13 vector genomes per kilogram in its first two cohorts.

Size matters here too. The commonly used Streptococcus pyogenes Cas9 is too large to fit with several effector domains inside the roughly 4.7 kb AAV packaging limit, which is why smaller Cas variants, split designs, or compact zinc-finger repressors appear in AAV programs. Sangamo’s ST-503, for example, delivers an engineered zinc-finger repressor targeting SCN9A (the gene for the Nav1.7 sodium channel) by a single intrathecal AAV injection in a Phase 1/2 study in small fiber neuropathy, according to the company’s clinical trial page. I could not verify clinical efficacy data for ST-503 at the time of writing.

Figure 4: The delivery fork. Transient LNP-mRNA suits liver targets; AAV suits muscle and nerve but gives longer, harder-to-control exposure.

Both branches converge on one question, durability: does the mark outlive the editor? Everything in the next section is, at root, an answer to that question for one tissue and one gene.

The October 2026 clinical scoreboard

The table lists the programs where I could verify public information. “Company-reported” means the numbers come from press releases or conference summaries rather than a peer-reviewed trial paper.

Program	Developer	Disease and target	Delivery	Status (verified)
TUNE-401	Tune Therapeutics	Chronic hepatitis B, HBV cccDNA and integrated DNA	LNP-RNA, intravenous	Phase 1b/2a data at EASL, May 30, 2026; Phase 2 expected to start late 2026
CRMA-1001	nChroma Bio	Chronic hepatitis B, HBV cccDNA and integrated DNA	dCas9 with epigenetic effectors (formulation not stated in sources I read)	First participant dosed January 28, 2026; Phase 1/2 NCT07200193; data expected late 2026
EPI-321	Epicrispr Biotechnologies	FSHD1, silencing DUX4	Intravenous AAV (serotype not disclosed in sources I read)	Phase 1/2 NCT06907875; six-month data on first patients June 2026; enrollment complete
ST-503	Sangamo Therapeutics	Small fiber neuropathy pain, SCN9A / Nav1.7	Intrathecal AAV, zinc-finger repressor	Phase 1/2 STAND study recruiting; no efficacy data verified

Preclinical programs, such as PCSK9 editors in primates, sit one rung below. They are the evidence that the platform can lower a validated target in a large animal, and they are the likely next wave if the liver safety data hold.

Deeper Analysis: What the Human Data Actually Shows

TUNE-401 in hepatitis B: the strongest dataset so far

Chronic hepatitis B affects more than 250 million people worldwide, according to the Nature news feature published on September 21, 2026, and current drugs rarely eliminate the virus. The virus persists in two forms: covalently closed circular DNA (cccDNA), a mini-chromosome in the nucleus, and fragments of viral DNA integrated into the host genome. A therapy that silences both with methylation would reduce viral protein output without needing to excise anything, which is the hypothesis behind TUNE-401.

According to Tune’s EASL 2026 release, the study ran at sites in New Zealand, Hong Kong and Moldova. A single-ascending-dose part tested 0.2, 0.45, 0.65 and 0.85 mg/kg by intravenous infusion, and a multiple-dose part gave up to three infusions at least four weeks apart at 0.65 mg/kg. The company reported that all participants at dose levels two to four showed repression of HBV biomarkers, including surface antigen (HBsAg), pregenomic RNA and core-related antigens. It also reported loss of pgRNA in 4 of 7 HBeAg-negative patients and HBeAg loss in 3 of 5 HBeAg-positive patients at the higher doses, with suppression observed through 17 months after dosing.

Safety was described as mild-to-moderate infusion reactions plus transient, asymptomatic transaminase elevations that returned to normal within days, with no cumulative toxicity reported from repeat doses.

How to read this. The strengths are real: a human-translatable LNP route, dose-ordered responses, and durability at more than a year. The caveats are equally real. The groups are small (single digits for the responder subsets), there is no placebo arm, and the endpoints are biomarker reductions, not a clinical cure. A “functional cure” in hepatitis B is conventionally defined as sustained HBsAg loss off treatment, and the company’s own wording refers to repression and proof of concept, not cure. I could not locate a peer-reviewed full paper for the dataset at the time of writing.

nChroma’s CRMA-1001: same disease, different construct

nChroma Bio, which I understand to descend from Chroma Medicine (corporate lineage not re-verified), is developing CRMA-1001, a dCas9 fused to proprietary epigenetic effectors targeting both cccDNA and integrated HBV DNA. According to the CRISPR Medicine News report on regulatory clearance, preclinical work in HBV mouse models showed more than a three-log reduction in HBsAg, with undetectable HBsAg and HBV DNA in 90% of mice at the highest dose and durability beyond six months. The same report says that non-human primate studies showed gene silencing lasting more than a year. A Hong Kong clinical trial certificate was approved on December 16, 2025, and the first participant was dosed on January 28, 2026.

No human efficacy data had been published when I checked. That is useful context for the claim that “epigenetic editing has been proven in hepatitis B”: the proof comes from one program, and a second, structurally different program is still pre-data. If both converge, confidence rises. If they diverge, the construct matters more than the platform label implies.

EPI-321 in FSHD: a different kind of evidence

FSHD1 is caused by failure to keep the DUX4 gene silent in skeletal muscle. DUX4 is a developmental gene that is toxic to adult muscle when expressed. EPI-321 aims to restore methylation at the DUX4 regulatory region, per the FSHD Society trial page.

Epicrispr’s June 26, 2026 release reports an open-label Phase 1/2 study with nine patients dosed across two cohorts: six at 2 x 10^13 vg/kg and three at 4 x 10^13 vg/kg, each as a single intravenous infusion. Six-month data were available for the first three patients, all of whom gained lean muscle volume, averaging about 370 mL, with some muscles showing gains near 15%. The company also reported favorable reductions in a circulating cell-free DNA biomarker linked to DUX4 pathway activity, and no serious adverse events to date. The data cutoff was May 12, 2026, and primary study completion is listed as mid-2027.

Two things deserve attention. First, a lean-muscle-volume gain measured by imaging in three people is a signal, not a functional benefit; FSHD trials will ultimately need to show strength or function endpoints versus natural history, and the company has said three-month data showed improvements across several strength and functional measures without giving numbers I could verify. Second, a mismatch exists between the 12 participants listed as the planned enrollment on the FSHD Society trial page and the nine treated patients in the release; I treat the release as the more current figure but flag the discrepancy.

Why the preclinical PCSK9 work still matters

PCSK9 is the cleanest proof-of-principle target because the biomarker is simple: LDL cholesterol in blood. The primate result described earlier, roughly 70% LDL-C reduction in monkeys with transient liver enzyme elevations, shows that an epigenetic editor can move a physiological readout in a large animal. It also shows the likely safety watch-item for any LNP-delivered editor, which is liver enzymes in the first two weeks. If the hepatitis B programs show clean repeat-dose safety, the case for taking PCSK9 or similar cardiovascular targets into humans strengthens, though none of that is yet announced in the sources I read.

Where AI and computation fit

Choosing a guide and effector is a prediction problem: which CpG islands control a given gene, which guides bind uniquely, and which edits will have off-target consequences in a given cell type. Genomic foundation models are starting to help with exactly that regulatory-sequence question. Our explainers on AlphaGenome variant effect prediction and DNA language models such as Evo cover the model side. I would not claim that any of the clinical programs above used such models for guide design; I found no source that says so.

Trade-offs, Gotchas, and What Goes Wrong

The headline advantage of epigenetic editing is that it avoids double-strand breaks. The honest list of problems is long, and several of them are the mirror image of that advantage.

Durability is unproven in humans, in both directions. Methylation is copied by DNMT1 during division, but it can also be erased by TET activity, passive dilution in rapidly dividing tissue, or reprogramming events. A 17-month hepatitis B signal is encouraging; it is not a decade. Conversely, if an effect proves too durable, the supposed reversibility is partial, because there is no clean “undo” button once the editor is gone. Reversal would require a second editor (an “on” tool such as a TET fusion), and that is far less mature than silencing.

Off-target effects change shape, but do not vanish. A dead Cas9 cannot make an unplanned cut, which removes one failure class. It can still bind unintended sites and drop methylation or KRAB-driven heterochromatin there. Because KRAB can spread H3K9me3 across neighboring chromatin, a silencer aimed at one promoter can in principle dampen nearby genes. The San Raffaele group reported that the single-molecule zinc-finger editor had fewer transcriptional off-target effects than a triple-component combination, which is evidence that design choices matter, not that the problem is solved.

Context dependence cuts both ways. A promoter that silences well in hepatocytes may behave differently in another cell type because baseline chromatin and the enzyme complement differ. The same CRISPRoff-style editor produces different durability depending on the gene, as discussed in the original work, and a mark that holds in dividing cell lines may behave differently in post-mitotic neurons or slowly turning-over muscle fibers.

Delivery imposes its own risks. LNPs bring infusion reactions and transient liver enzyme rises; both appeared in the hepatitis B and primate data above. AAV brings dose-dependent toxicity concerns and anti-capsid immunity, which usually rules out a second dose. Editors expressed for long periods from AAV may also generate immune responses against the editor protein itself.

Biomarker endpoints are not benefits. HBsAg repression, DUX4-linked cell-free DNA and lean muscle volume are plausible surrogates. None has yet been tied, in these programs, to the outcomes patients and regulators ultimately care about, such as sustained off-treatment viral control or preserved function. Anyone reading “positive proof of concept” should translate it as “the biomarker moved in a small, open-label group.”

Integrated viral DNA is a double-edged target. Silencing integrated HBV DNA could reduce HBsAg production from fragments that antivirals cannot touch. But a silenced genome is not a cleared genome, and loss of the silencing mark could allow re-expression. That is a reason to track durability and relapse after treatment stops, not a reason to dismiss the approach.

Germline and ethical boundaries. All current clinical programs are somatic, meaning the changes do not pass to children. The ethical debate around heritable editing, which flared again in news this month around embryo-editing startups, concerns cutting-based editors and sequence changes; epigenetic marks are mostly reset in early development, but this is a research question I would not oversimplify. For broader context on where gene-edited biology is moving clinically, see our piece on gene-edited pig organs for xenotransplantation.

Practical Recommendations

If you are a researcher choosing a tool, match the effector to the time horizon. Use dCas9-KRAB (CRISPRi) for reversible perturbation and screens. Use a DNMT3A/DNMT3L plus KRAB design when you need inheritance through cell division, and test durability through at least several passages and, where possible, a regeneration or differentiation challenge, as the hit-and-run work did. Choose zinc-finger binders over Cas when vector size is binding. Always measure methylation directly at the target and at neighboring loci, and measure RNA genome-wide, because phenotype alone will hide off-target spreading.

If you are an investor, clinician or analyst reading company news, discount heavily for small open-label cohorts, and ask where the numbers come from. A press release with biomarker data is a hypothesis-confirming event, not a validation. The decisive datasets will be randomized or natural-history-controlled trials with functional or virologic endpoints, which, on the public timelines I found, are not due before 2027.

If you are a patient or caregiver, the verified fact is that these are experimental studies. Recruiting trials are listed on ClinicalTrials.gov, including NCT06907875 for EPI-321 (enrollment reported complete) and NCT07200193 for CRMA-1001. Your physician, not a press release, is the right route to decide eligibility.

A short checklist for evaluating the next epigenetic editing announcement:

• Is the data peer-reviewed, a conference abstract, or a company release?
• How many participants, at which dose, and is there a control?
• Is the endpoint a biomarker or a clinical outcome?
• How long has the effect lasted, and in how many people at that time point?
• What was the delivery system, and what was the safety signal (liver enzymes, immune reactions)?
• Were off-target methylation and transcription measured genome-wide?

Frequently Asked Questions

What is CRISPR epigenetic editing?

It is the use of a catalytically dead Cas9 or a zinc-finger protein, fused to an enzyme or repressor, to change chemical marks on DNA or histones at a chosen gene. Examples are adding CpG methylation to silence a promoter or using TET enzymes to remove it. The DNA sequence does not change, so the edit is in principle more reversible than a cut-based edit, though human durability is still being measured.

How is epigenetic editing different from CRISPR-Cas9 gene editing?

Cas9 gene editing cuts DNA and relies on repair to change the sequence permanently. Epigenetic editing leaves the sequence intact and alters regulation. That avoids double-strand breaks and their associated risks, but shifts the challenges to durability, specificity of binding, and delivery. Cutting suits diseases where removing a gene helps; epigenetic editing suits diseases where a gene is read too much or too little.

What is CRISPRoff?

CRISPRoff is a dCas9 fusion described in a 2021 Cell paper that combines DNMT3A, DNMT3L and a KRAB domain to silence genes and to hold that silence through cell division. A companion tool, CRISPRon, uses TET1 to remove methylation and reactivate genes. It is a research platform; clinical programs use their own proprietary or optimized descendants.

Is epigenetic editing in human trials yet?

Yes. As of October 2026, TUNE-401 (hepatitis B) reported Phase 1b/2a data, nChroma’s CRMA-1001 (hepatitis B) has dosed its first participant, Epicrispr’s EPI-321 (FSHD) reported six-month data on first patients, and Sangamo’s ST-503 (neuropathic pain) is in a Phase 1/2 study. None is approved, and all efficacy figures are early, small and company-reported.

Are epigenetic edits permanent?

Not necessarily, and that is both the promise and the uncertainty. Methylation can be copied through cell division, which gives memory, but it can also be lost. Animal studies show silencing for about a year in mice, and Tune reports hepatitis B marker suppression through 17 months in some participants. Whether it lasts for decades in people is unknown.

Can epigenetic editing cure hepatitis B?

That has not been shown. Silencing viral genes can reduce surface antigen and other markers, and a functional cure means sustained loss of HBsAg off treatment. Current human data report biomarker repression in small groups. Larger, controlled trials with long follow-up are needed before anyone can claim a cure. This article explains the science and clinical status only and is not medical advice.

Further Reading

• Base editing and single-base CRISPR therapeutics, the sequence-changing alternative to cutting.
• AlphaGenome variant effect prediction explained, AI for regulatory DNA.
• DNA language models and Evo, genomic foundation models.
• Gene-edited pig organs for xenotransplantation, another clinical gene-editing frontier.
• Nuñez et al., Genome-wide programmable transcriptional memory by CRISPR-based epigenome editing, Cell, 2021.
• Cappelluti et al., Durable and efficient gene silencing in vivo by hit-and-run epigenome editing, Nature, 2024.
• Tune Therapeutics, TUNE-401 Phase 1b/2a data at EASL 2026.

By Riju — about