Clinical Data Support PepGen’s Potentially Disruptive Myotonic Dystrophy Type 1 Approach

Yesterday, PepGen showed first clinical data for PGN-EDODM1 in Myotonic Dystrophy Type 1 (DM1).   Although it is years behind the DMPK knockdown competition by Avidity Biosciences and Dyne Therapeutics, the single-dose data hint that its differentiated mechanism of action may prove superior in this big orphan disease indication.

DMPK1 knockdown versus CUG structural disruption

Myotonic dystrophy is an autosomal dominant condition caused by the CUG triplett expansion in the 3’ UTR non-coding region of the DMPK1 gene.  It is thought that this sequesters the MBNL1 protein and the subsequent inability of MBNL1 to carry out its RNA processing functions (alternative splicing).  As DMPK1 is largely expressed in muscle (including heart) and the CNS, the disease symptoms (most noticeably myotonia, but also muscle weakness, heart arrythmia and cognitive impairment etc) relate to muscle and to some degree the nervous system. 

Avidity’s AOC-1001 (phase 3 enrolment to be complete in mid-2025) and Dyne’s DYNE-101 address this by reducing DMPK1 transcript levels thereby liberating bound MBNL1.  Although Avidity uses RNAi triggers conjugated to an antibody and Dyne RNaseH antisense oligos attached to Fab fragments, these 2 Tfr1-targeted molecules both achieve ~-40% DMPK1 RNA knockdown of the nuclear retained transcript.

PepGen’s PGN-EDODM1, however, approaches the problem by targeting oligonucleotides to the CUG repeat itself thereby disrupting the repeat CUG helical structures that act as MBNL1 sponges.

The reason why I quite like this approach is because it does not affect DMPK1 expression.  Mouse models show that DMPK deficiency causes deficits like myotonia and cardiac conduction problems.  In DM1, due to the nuclear retention of the CUG-expanded RNA, DMPK levels are already reduced by 50% and reducing it by another 40% (as AOC-1001 and DYNE-101 do…on average) raises on-target safety questions not posed by the PepGen approach. 

Early splice data support validity of CUG disruption

PepGen has now evaluated about a handful patients each after 28 days following a single dose of 5mg/kg and 10mg/kg for the CASI22 score.  CASI22 is a measurement of MBNL1-related alternative splicing and should be the first measurable change along the mechanistic therapeutic trajectory.  At this early timepoint, there is a nice dose-dependent response of -12.3 and -29.1 for the 5 and 10mg/kg cohort, respectively.

Such clean dose dependency has not been seen in the Avidity and Dyne programs.  Moreover, -29.1 is clearly superior that of -12 (2mg/kg) and -9 (4mg/kg) seen with not 1, but 2 doses of Avidity’s molecule and roughly on par with Dyne’s -25 after 2 doses of 6.8mg/kg.

Whether this translates to the clear functional improvements in myotonia and other symptoms specific to DM1 as for AOC-1001 and DYNE-101 remains to be seen, but logic would say ‘yes’.

A concern with the PepGen approach is that it uses oligonucleotides attached to a cationic cell-penetrating peptide for delivery.  Not only is that less specific than the Tfr1-targeting of Avidity and Dyne, similar approaches, for example by Sarepta in their next-gen DMD exon skipping program, have run into safety issues.  This certainly needs to be watched as PepGen escalates doses further and repeat doses.

Increasing the Reach of ADAR Therapeutics by Protein Structure Prediction

'Only' being able to convert an ‘A’ to an ‘I’ when genome editing can seemingly re-write the code at will had been seen as a major limitation of the technology.  Examples such as the large and underserved market in correcting the piZZ genotype in alpha-1-antitrypsin disease using A-->I RNA editing were the exception.  With the help of artificial intelligence-enhanced structure prediction tools, however, the financial incentive to pursue genetic disease should increase again.  Indeed, ProQR’s first Rett Syndrome candidate already points in this direction.

Addressing the p.R270X mutation

Rett Syndrome is an X-linked dominant genetic haploinsufficiency neurological disorder caused by mutations in the MeCP2 protein leading to decreased functional activity as a master regulator of gene expression and neuronal development.  There are many ways to cause the loss of activity of a protein, so mutations can typically be found throughout an affected haploinsufficiency gene.

ProQR’s 1st candidate for Rett Syndrome addresses the p.R270X mutation where the arginine codon CGA at position 270 in the protein is mutated to the stop codon UGA, thus leading to a truncated protein that is also destabilized at the mRNA level via the NMD pathway.  Luckily, the CGA codon contains an actionable ‘A’ and although editing it to a G-like Inosine would not restore the wild-type protein, it happens that mice with tryptophan-coding UGG at position 270 behave like rescued wild-type mice.

Beyond p.R270X

It is well established that a second mutation can modify, if not rescue a disease caused by a first mutation.  At a protein level, this may be due to functional restoration by coaxing the protein back to its original functional structure.  

Rett Syndrome affects around 50,000 females in the US and EU of which ~3,500 would be due to p.R270X.  Point mutations across the MeCP2 gene overall account for 60% of cases. Such numbers are too low for one to expect to come across compensatory mutations in the MeCP2 from population genetics.  So instead of relying on serendipity one may systematically ask whether the structural pathological change resulting from a given point mutation or group of point mutations (characterized for example by destabilization of the DNA-binding domain of MeCP2) could be reverted back towards wild-type if one ADRA edited one of the ~350-400 Adenosines in the MeCP2 mRNA based on AI-powered structure prediction tools like AlphaFold.

This could then be verified by a cellular functional assay of MeCP2 as a master epigenetic regulator.  Regarding approval and clinical trial population, a label may just say that a patient with a very rare mutation that could not be confirmed in clinical trials due to low subject numbers may still qualify for the drug as long as such an assay supported it.  This path has already been trodden, for example with Vertex’ Cystic Fibrosis drugs.

AI increases IP value around platform drug modalities

Artificial intelligence is an amazing development.  The above strategy is just one example of how it may decrease future drug development cost and efficiency, thus increasing the attractiveness of going after rare diseases again.  I would not be surprised therefore if structure prediction may help to address the most common missense mutation in Rett Syndrome, p.R106W, which affects 3x the patient number of p.R270X.  

There are, of course, many more applications also to RNA editing such as chemistry based on structure prediction around the A to be edited when paired to the editing oligo. 

AI, however, is also a scary development for people like me.  When I came across this concept, it seemed like Gemini had it all figured out already.  I fully expect that in a year’s time, I will have to re-think the point of blogging full-stop when AI can tell for itself which are the more or less valuable concepts.  One thing is sure, AI will open the flood-gates for therapeutic strategies leveraging genetic platform technologies like ADAR, RNAi, antisense, CRISPR and gene therapy.  Investing in companies with fresh, gate-keeping IP in these technologies could therefore more valuable than ever.  It is therefore possible that sooner than later I will be watching all this from the beach as a passive investor in these platforms.  

Why I Am Taking My HAE Cure Idea to My Grave

Last year, as I was researching the hereditary angioedema (HAE) therapeutics landscape, I came across data in the C1 inhibitor gene regulation literature that gave me a proverbial heureka moment: if I could do ‘X’, then I would have a cure for HAE. Dazed by the excitement of being in possession of such an idea, I verified that the enabling technologies (genome editing and LNP delivery) were fit for purpose.

It basically seemed that all I had to do was to verify my approach in tissue culture and the rest would be ‘simple’ business development. In the vast majority of cases, HAE is caused by mutations in the C1 Inhibitor gene. It is a rare disease affecting close to 10000 patients in the US and EU. Nevertheless, it is a lucrative, currently ~5B market with therapeutics mostly replacing missing C1 inhibitor in the form of recombinant or purified C1 inhibitor or agents interfering with proteins involved in bradykinin generation or the bradykinin B2 receptor. 

It is getting a bit crowded, but there still is room for improvement to achieve the goal set by the HAE community: a life free from disease. Intellia’s NTLA-2002, a one-time CRISPR Cas9 therapeutic now in phase 3, achieves a 80-90% attack rate reduction by destroying prekallikrein gene activity. The efficacy is in line with competing therapeutics that need to be taken indefinitely and cost around half a million USD- annually. 

A good estimate for the price of NTLA-2002 would be $3M, meaning that after 6 years this convenient treatment would start saving the healthcare system serious money for a disease that gets diagnosed typically in the late teens or early 20s. Nevertheless, the financial market could not be less excited by NTLA-2002 as indicated by the lackluster response to its clinical data. In theory, NTLA-2002 could treat every HAE patient and rake in $30B in the process. Compare that to the annual $5B and growing market for the traditional forever treatment options which will cost the system $60B plus over the next 10 years alone…without even treating every patient and leaving treatment gaps due to insurance that like to drag their feet. 

We are talking here about cost alone, not even how a one-time treatment would benefit a HAE patient: less stress, both financial and emotional which in itself is a major risk factor for HAE attacks. Considering the lack of investor interest in NTLA-2002, I concluded that going from 80-90% attack rate reduction to 100% attack rate reduction following a one-time treatment, is not worth the time and financial risk. Disclosing the idea also does not make sense since it would destroy any IP value if the idea was picked by somebody else or conceived independently, further reducing chances it would ever get to patients. 

Just recently, Pfizer, a major player in hemophilia therapeutics, declined to bring a reasonably effective and safe hemophilia A one-time gene therapy discovered by Sangamo to market. Again, I ask myself, why would Pfizer ever want to destroy a lucrative market it participates in? I can only see how an outside company would want to disrupt such a rare disease market (hemophilia is bigger than HAE, so that situation could actually happen e.g. in the form of Regeneron). 

What about patients? I am slowly coming to accept that the best possible treatment for a disease will often not be the end-game in terms of therapeutic development. Sometimes it is pharma’s profit model, sometimes society that does not feel that a patient deserves better. Take for example a rich country like Singapore where patients do not have access to modern HAE medicines and where even women are still treated with the androgen sledgehammer. 

It seems we have arrived at a turning point where it has become hard to invest in best possible therapeutics in a resource-constrained environment where money gets diverted to the military and AI. Making matters worse for rare disease drug development, as international ties continue to deteriorate, the target population becomes even smaller.

The Lp(a)-Lowering Drug Race

Cardiovascular disease (CVD) remains a leading cause of death and morbidity worldwide.  According to the WHO, 18 million of the 68 million deaths annually can be attributed to it.  That's more than 1 in 4. This does not include those whose quality of life is severely impacted by it.  Despite the positive impact of LDLc-lowering drugs (statins, PCSK9 inhibitors), blood-lowering, anti-diabetic, and weight-loss medications, other risk factors remain to be addressed.

Among them is lipoprotein little a [Lp(a)], an independent and causal risk factor for cardiovascular disease.  Except for a few with extremely high Lp(a) who were lucky enough to be diagnosed, often after family members encountered cardiovascular events early in life, it has been largely met with a shrug in general medicine.  The dynamic, however, is quickly changing now as the PCSK9 outcomes trials are strongly supportive that lowering Lp(a) should be able to lower CVD risk (O'Donoghue et al 2019) and new strategies mostly centered around sequence-targeted therapeutics allow for robust Lp(a) reductions.

This week, Novartis disclosed that the eagerly anticipated results from the pivotal trial of the most advanced of these clinical candidates, the RNaseH antisense compound pelacarsen has been delayed.  It was originally projected to deliver results in 2025, but is now guided to read out in the first half of 2026.  As this has raised a few eyebrows, I will take the opportunity to provide an overview of the Lp(a) competitive landscape and the pharmaceutical companies involved.

    

Lp(a) is a genetic determinant of CVD risk independent of LDLc and ApoB100

Lp(a) is quite similar to LDL-cholesterol lipoprotein particle, but in addition to carrying one molecule of apolipoprotein B 100, cholesterol and phospholipids in its outer shell, there is a apolipoprotein (a) molecule attached to ApoB100 via a disulfide bond.  If LDL cholesterol is referred to as the ‘bad’ cholesterol, Lp(a) on a per particle basis is the lipoprotein supervillain.  It not only gets more easily through the blood vessel endothelial cell layer so that it becomes a substrate for plaque build up, it is quite inflammatory as a result of being a good carrier of oxidized phospholipids.  It is also thought that there may be a prothrombotic effect partly due to its similarity and evolutionary relationship with plasminogen.   

Lp(a) distribution in general population (from Varvel et al 2016)

Unlike LDLc, Lp(a) levels are largely genetically determined and cannot be modified by nutrition and exercise.  They stay more or less constant after 5 years of age.  Levels range widely within a population.  Whereas many have Lp(a) levels of 10mg/dL or less (~22nM), the distribution has a long tail with ~10% of the population, especially black people, who have more than 90mg/dL.  This puts them at ~1.4x and higher risk of suffering CVD events like CVD death, myocardial infarction or of developing calcific aortic valve and peripheral artery disease (Erqou et al 2009).  The increased risk is maintained after having suffered a first event with almost 2x the risk for persons in the top decile versus those below 10mg/dL (Madsen et al 2020).  This is relevant when thinking about using Lp(a) lowering as a strategy for primary or secondary CVD prevention.

Evidence that intervention can reduce CVD risk

Currently the best advice one can give to somebody with highly elevated familial Lp(a) is to address all the other modifiable CVD risk factors.  In terms of medication, this typically means LDLc-lowering by statins and, even better so, the newer PCSK9 inhibitors.

This is because PCSK9 inhibition reduces Lp(a) by 20-30% (Sabatine et al 2017; Schwartz et al 2018; Ray et al 2017) .  The high variability of the Lp(a) reduction between subjects allowed for the determination that in the FOURIER trial of Amgen’s evolocumab there was a 15% relative risk reduction for certain cardiovascular events for each 25nM Lp(a) lowering (~11.6mg/dL; O'Donoghue et al 2019).  Remarkably, those above the median Lp(a) at baseline (37nM) not only had the highest absolute Lp(a) reduction, they had a 23% lower CVD hazard ratio versus only 9% for those below the median.  Importantly, these findings were independent of the magnitude of LDL-c lowering in the same subjects  which was the same across the Lp(a) ranges.

This in my mind is the strongest evidence that not only life-long differences in Lp(a) have a meaningful impact on CVD risk, but even just 5 years of pharmaceutically lowering Lp(a) may have a big impact.

The Lp(a) competitive landscape

It is therefore not surprising that half a dozen or so notable Lp(a)-lowering efforts are in clinical development.  At the same time, medical societies like the EAS, EHA and AHa and KOLs are busy raising awareness around Lp(a).  Still, a lot of work remains to be done: less than 1% in the western medical systems have been tested, despite the fact that guidelines are increasingly recommending getting tested at least once in a lifetime. While universal numbers are hard to come by, reports from various health systems suggest that testing is finally gaining traction (e.g. Bhatia et al 2023).  The pump is thus getting primed for the successful roll-out of Lp(a)-lowering agents.

RNaseH pelacarsen (Ionis, Novartis)

Leading the pack is pelacarsen, originally discovered and developed by Ionis Pharmaceuticals and subsequently licensed to Novartis in 2019.  Pelacarsen is a subcutaneously administered GalNAc-ASO working by RNaseH-mediated knockdown of apo(a) mRNA.   It is currently in a global pivotal phase 3, Lp(a)HORIZON outcomes study evaluating its ability to reduce major cardiovascular events (MACE) in a secondary prevention setting in 8000+ patients with Lp(a) >70mg/dL.

The 80mg once monthly should achieve 80% Lp(a) lowering based on a phase IIb study in a similar population (Tsimikas et al 2020).  Safety and tolerability have been good.  This means an absolute reduction of ~55-60mg/dL for a patient at the lower end of the range, a number that has been estimated (Madsen et al 2020) to be enough for ~20% relative risk reduction.  

Given the uncertainties about the exact relationship between pharmaceutical Lp(a) lowering and risk reduction, Novartis conservatively added a second cut analyzing the effect of pelacarsen in the population with >90mg/dL of baseline Lp(a) for efficacy.  If the study hits in either cohort, it can be considered successful.

The study is an event-driven one, meaning that in this case 993 MACE need to accumulate.  The fact that the data monitoring committee (DMC) has apparently communicated to Novartis that the projected results are now due in 2026 instead of 2025 suggests that insufficient MACE events have occurred since study start in 2019 and completion of enrolment in mid-2022 (2 ½ years minimal follow-up).  

It is plausible that the slower-than-anticipated event rate has to do with a population that is highly controlled for other risk factors, especially LDLc, as I expect most to be on maximally tolerated statins and a good proportion on PCSK9 therapeutics, too.  More optimistically, slow event accrual could also be due to an extraordinary efficacy of pelacarsen.  I suspect, however, that the trial protocol provided for interim looks not only for futility (early looks), but also for efficacy at this late stage of the trial.  The most likely scenario therefore is that the DMC saw maybe 800 events and concluded that the study would benefit from running its full course.  At the very least, the study was not stopped for futility.  

Personally, 70mg/dL is too low a baseline Lp(a) to expect a strong MACE response.  I see Lp(a)-lowering to have most initial utility in the top Lp(a) decile (>90-100mg/dL) and it would therfore not surprise me if HORIZON only hit in that bracket.

RNAi olpasiran (Arrowhead, Amgen)

Hot on the heels of pelacarsen is olpasiran, an RNAi agent discovered by Arrowhead Pharmaceuticals, licensed to and now in development by Amgen.  This liver-targeted RNAi trigger commenced a pivotal phase 3 trial in another secondary prevention study for atherosclerotic cardiovascular disease in 2022 and has finished enrolling 7000+ subjects in just 1 ½ years.  This means that results from OCEAN(a) Outcomes could become available within a year of Lp(a)HORIZON in late 2026. 

The apparent enthusiasm by Amgen is encouraging since they, as the developer of PCSK9 antibody Repatha, are best positioned to understand the likely impact of modulating Lp(a) on CVD outcomes.  In addition, the efficacy of olpasiran trounces that of pelacarsen with a time-averaged Lp(a) reduction of more than 95% (vs -80%) despite of using quarterly instead of monthly subcutaneous dosing (O'Donoghue et al 2022).  The marked difference in efficacy will also allow for insights into whether, as is widely assumed now and as is the case for LDLc, it is the absolute Lp(a) lowering that matters or the relative percent lowering. 

Finally, I like the study because it focuses on a population in the top decile of baseline Lp(a) (>200nMol, ca >90mg/dL) where genetically CVD risk increases exponentially.  I also would not be surprised if lowering Lp(a) well below the genetically defined 50mg/dL risk threshold brings further benefit in an interventional setting.  Doing everything to keep these streaky macrophages from bursting by depriving them of their oxidized fatty meals could be quite beneficial.

RNAi lepodisiran (Dicerna, Eli Lilly)

Two years behind olpasiran in terms of phase 3 initiation, but every bit as potent and even more long-lived in its Lp(a)-lowering activity is RNAi rival lepodisiran.  Lepodisiran is a Dicer-substrate RNAi molecule discovered by Dicerna (now Novo Nordisk) and developed by Eli Lilly.

Lp(a)-lowering by single dose of lepodisiran

In a single dose study, a large dose of 608mg, achieved time-averaged ~95% reduction in Lp(a) over a year.  A lower dose of 304mg effectively lowered Lp(a) by a peak ~98% in the first months before leveling off at around 90% at the half year mark (Nissen et al 2023).  

The ACCLAIM-Lp(a) ASCVD phase 3 study in 12500 subjects in secondary prevention and for those at high risk will test 3 semi-annual doses before moving on to annual dosing.  Though the dose has not been disclosed, I consider both 608mg and 304mg dosages possible with this dosing regime.  The choice will have probably been strongly informed by whether Eli Lilly believes the clinical benefit of going from 10nM to 5nM Lp(a) will balance out formulation, drug administration, and potentially safety issues coming with the high dose.

Besides Eli Lilly, Novo Nordisk, Amgen, and pretty much everybody with a stake in the lipid-related CVD market believe in the advantages of infrequent drug administration as evidenced by their investments in this area.  Novartis this quarter saw the PCSK9-targeting RNAi drug Leqvio (licensed from Alnylam) cross the $1B run-rate for the first time.  It is 284mg of canonical siRNA oligonucleotide administered basically semi-annually.  This demonstrates the shift from daily oral pills that most people will have stopped using after a year despite their need for life-long treatment.  This is understandable since most of them do not suffer acute symptoms. 

2-dose study of zerlasiran

This brings us to the final notable RNAi contender, zerlasiran by Silence Therapeutics.  In early trials, 300mg showed clear potential for maintaining Lp(a) reduction of more than 95% using quarterly dosing.  Being phase 3 ready, Silence Therapeutics stood to benefit the most from a timely Lp(a)HORIZON reveal by Novartis.  Not only may have they been able to adjust their Lp(a) target levels for their outcomes trial, a successful readout may also have been the trigger for a Big Pharma like Pfizer or Merck with no stake in Lp(a) yet to acquire zerlasiran, if not acquire Silence Therapeutics (market cap $300M) altogether.  

Life-long Lp(a)-lowering using CRISPR genome editing

If you like infrequent dosing, why stop at semi-annual or annual when there are life-long options going after the same apolipoprotein(a) target?  This includes base editor VERVE-301 by cardiovascular specialty company Verve Therapeutics which is partnered with Eli Lilly, but still in preclinical development.  Already in phase I is competitor CTX-320 by CRISPR Therapeutics using more traditional CRISPR Cas9 cleavage of apo(a).  Both use standard intravenous LNP delivery to liver hepatocytes.

Finally, in addition to RNAi and CRISPR, Eli Lilly is also developing an oral small molecule drug. Muvalaplin is designed to inhibit the interaction between ApoB100 and Apo(a) and thus the biogenesis of the Lp(a) particle.  Daily doses barely lower Lp(a) by 60% and as a small molecule there are concerns about hemodynamic effects given the similarities of apo(a) and plasminogen (Nicholls et al 2023).

Disclosure: I have been and am still positioning myself for HORIZON by investing across the Lp(a) landscape, including Silence Therapeutics, Verve Therapeutics and Ionis.  Focused Lp(a) testing companies would also fall into my interest area, so please pitch me ideas if you have some in this area.

Huntington’s Disease Therapeutics Breaking Through (Finally!)

Huntington’s is one of the large devastating monogenic diseases for which disease-modifying therapies have yet to materialize.  This is about to change as a combination of big data sciences (single cell sequencing and GWAS) together with good old molecular biology is unraveling the detailed mechanisms of this neurodegenerative disorder.  This provides an understanding for when and how to intervene.  New therapeutic modalities, in particular sequence-specific therapeutics like oligonucleotides and genome editing that have emerged in recent years will play an important role in this endeavor.

In this blog entry, I will try to summarize the research findings and how individual steps along the pathogenic pathway may be targeted.  Hold on, it will be an exciting one.

Somatic repeat expansion

Huntington’s Disease is linked to the expansion of the CAG triplet repeats found in exon 1 of the huntingtin gene.  If you inherit a copy of huntingtin from either parent that carries 40 repeats of CAG or more you will almost certainly develop symptoms of HD (manifest HD).  Symptoms intially involve emotional instability and cognitive deficits, progressing to include the famous movement abnormalities, ultimately culminating in life-shortening problems with basic physiological processes such as feeding and breathing.  These follow neurodegenerative processes which can be observed decades before symptom onset (Scahill et al, 2025), most notably severe striatal atrophy.  If you start with a number that is closer to 40 at birth, symptom onset may be in the early 60s, but if it is 50 or so, it will be more like at age 30.

Newer findings (Handsaker et al, 2025), show that having a huntingtin copy with 40 CAG repeats per se is not toxic to a given cell.  But having 36 (incomplete disease penetrance), and especially 40 repeats or more is essential for stage 1 of pathogenesis: somatic repeat expansion.  Genome-wide association studies (GWAS) have highlighted the critical role of the mismatch DNA repair (MMR) in facilitating such repeat expansion, also in post-mitotic cells in the CNS (Gem-HD, 2015).  It is possible that as the transcription bubble in the triplet repeat DNA is formed during huntingtin mRNA transcription, chances of the DNA bubble collapsing out of register increases exponentially with repeat length.  The consequently misaligned DNA strands would then be recognized by MMR and haphazardly resolved, in most cases by adding a CAG triplet on both strands. 

The reason why symptom onset can differ by decades depending on starting CAG numbers is because repeat expansion itself is length-dependent and the initial expansion is quite slow until it picks up steam later (Handsaker et al, 2025).  It is when repeat numbers reach ~150 in a given striatal projection neuron (aka medium spiny neuron), toxic substance X (see discussion below) gets produced and the cell starts to suffer.

The exact timing of manifest HD for a given CAG repeat number, however, also varies substantially and depends on additional factors, many of them genetic.  Genome-wide association studies (GWAS) looking at those differences have clearly shown that proteins in the mismatch DNA repair pathway mediate repeat expansion (MSH3, FAN1, MLH1 etc) and have critically contributed to the mechanistic model of somatic repeat expansion outlined above (Gem-HD, 2015; Gem-HD, 2019).  

An obvious therapeutic strategy therefore would be to inhibit repeat expansion by targeting DNA repair or by cutting down its size, if not get it below the 36 risk threshold using CRISPR genome editing (e.g. by twin prime editing).  A number of companies among them Ionis, Biogen, Rgenta, Atalanta, LoQus23, Prime Medicine are attempting just that.  Just binding DNA with zinc fingers has also been shown to inhibit expansion (Mathews et al, 2024).  This strategy would also be able to simultaneously reduce the generation of toxic factor X. 

DNA repair could in theory be targeted by small molecules.  There is, however, perhaps unsurprisingly ample evidence that doing so indiscriminately throughout the body may be too risky a strategy.  For example, early research of somatic repeat expansion in HD showed that knockout mice of the MMR protein MSH2 develop cancer (Wheeler et al, 2003), and while having heterzygous mutations in MMR genes slows HTT repeat expansion, it causes Lynch Syndrome, the most common genetic predisposition for cancer (Skarping et al, 2024).

More targeted approaches are therefore called for.  These may include delivery specifically to the cells most affected by repeat expansion (i.e. the striatal projection neurons) or at least limit delivery to the post-mitotic and therefore less cancer-prone CNS using oligonucleotides, mRNA and/or DNA (by AAV, LNP).  Alternatively, one could identify and copy rare missense or exon-skipping genetic modifier mutations in the human population that are protective against repeat expansion without predisposing to cancer.  This could involve CRISPR prime or base editing, or non genome-editing ADAR RNA editing.   

Toxic Factor X

Somatically expanded CAG repeats in the huntingtin gene DNA per se, however, is not what causes cellular toxicity.  This is illustrated by GWAS hits that disentangle somatic expansion from age of disease onset and disease duration (GEM-HD, 2019).  Furthermore, while CAG repeat length at birth strongly correlates with age at onset of disease and death, other genetic factors determine the duration between onset of disease and death (Keum et al, 2016).  Finally, while the MSH2 knockout mice mentioned earlier (Wheeler et al, 2003) did not exhibit CAG expansion, they still showed some striatal-specific huntingtin inclusions characteristic of HD, albeit at much reduced levels compared to wild-type mice.

Historically, HD was thought of as a 'polyglutamine' disease, meaning that it is extended polyglutamine stretch that is toxic to a cell, either by forming inclusions causing cellular stress or by inappropriately interacting with other proteins.  In fact, polyglutamine inclusions are characteristic of disease pathology as they are found in the cells most affected by HTT. Interestingly, these inclusions are best detected by antibodies directed at the N-terminus of HTT encoded by exon 1.

However, this polyglutamine-only model is seriously challenged by the observation that other CAG expansion diseases carry extensions in untranslated UTR elements and by the fact that mutations in HTT exon 1 itself that extend the stretch of glutamine by non-CAG synonymous codons do not increase inherent toxicity and may in fact delay disease onset and death.

In addition to protein, differences in HTT expansion should affect RNA structure and RNA polymerase II transcription.  Toxic RNA is thought to be at the root of other repeat expansion diseases such as type I myotonic dystrophy as they by sequester important RNA-binding proteins away from their normal functions.  Intriguingly, HTT RNA clusters in a repeat length-dependent manner around the huntingtin gene locus and these clusters appear to be seeded by exon 1 RNA (Ly et al, 2022).

Exon 1 mRNA

In the 90s it was found that merely expressing a fragment of the human huntingtin in mice comprising exon 1 makes for a good mouse model of HD (Mangiarini et al, 1996).  The R6/2 line is widely used to this day, also because it models highly aggressive disease pathogenesis.  Little did the authors of that paper probably suspect that the highly toxic exon 1-only huntingtin is not only generated in a length-dependent manner in mice (Sathasivam et al, 2013)...it is also expressed in patients (Neueder et al, 2017).  In fact, there is (yet to be published) talk (Huntington's Study Group 2024, Wexler Symposium 2024) that exon 1 mRNA production is tied to somatic repeat expansion. 

If you now consider that exon 1 is extremely toxic on its own, gets generated in a repeat number-dependent manner, and that the characteristic N-terminal huntingtin inclusions may not represent cleavage fragments of full-length huntingtin and that exon 1 mRNA may seed huntingtin RNA clusters, a compelling model emerges whereby once CAG somatic repeat expansion has reached a certain length, highly toxic exon 1 mRNA/protein gets produced and triggers manifest HD.  This would also explain why the duration of manifest HD is independent of the original repeat length.

Mechanistically, it is possible that the expanded CAGs can impact the dynamics at which RNA polymerase II moves along the huntingtin gene which may allow for cryptic polyadenylation sites in huntingtin intron 1 to be recognized.  The formation of complex R-loop structures between nascent RNA with extended CAGs and equally CAG-extended DNA may play a role here.   

Therapeutic Strategies Aimed at Toxic RNA/Protein

There are a number of obvious modalities you could deploy to target the toxic agent in HD.  These include RNaseH antisense, RNAi and genome editing for HTT knockdown, steric blocking antisense oligonucleotides to inhibit intron 1 polyadenylation site usage.  Genome editing and ADAR RNA editing may also be used to interrupt the stretch of CAGs.  Disrupting even just 10% or so of 150 CAG repeats in a given RNA or protein could have a huge impact on its toxic property.

While many of such agents are in preclinical development, some have reached the clinic already.  Unfortunately, straightforward lowering of total, full-length  huntingtin by RNaseH antisense (ASO) by Ionis and Roche failed to show benefit in a pivotal clinical trial (McColgan NEJM, 2023).  In fact, the trial with tominersen was stopped early for toxicity as it may have done more harm than good as the therapy seemed to drive ventricular expansion along with striatal atrophy.  While some argue that this shows the perils of lowering wild-type along with mutant huntingtin, this not only is inconsistent with human genetics showing no such effect in people with heterozygous null mutations in the huntingtin gene.  If HD was caused by huntingtin insufficiency, there would be many more mutations aside from CAG repeat expansions that associated with HD.  More likely, toxicity stemming from the heavily phosphorothioated oligonucleotide backbone led to increased death of already stressed striatal cells.  

But sure, there should be no harm in targeting exclusively or predominantly mutant huntingtin as does Wave Life Sciences with their RNaseH antisense oligos targeting naturally occurring variations (SNPs) in the HTT gene correlating with the mutation.   

Both strategies, however, fail to address highly toxic exon 1-only mRNA/protein as they are targeted downstream of exon 1.  This is where UniQure’s DNA-directed RNAi and Alnylam’s synthetic oligonucleotide RNAi that target huntingtin in exon 1 itself come in.  

In fact, the clinically more advanced of the two (Alnylam's ALN-HTT02 has just entered phase I), UniQure's AMT-130 has generated data suggesting almost disease stabilization as measured on the widely used cUHDRS scale along with stabilization, if not lowering of neurofilament, a biomarker for CNS injury over time.  The data was so compelling when compared to external control in a propensity weighted manner that the FDA and UniQure are now working on expeditiously bringing the intrastriatally administered AAV therapy to the market possibly in 2026 already.   

At this stage, I should disclose that I have started to invest in UniQure.  Being first-to-market with a disease-modifying drug for HD would obviously be enormously lucrative for a well-financed company with an $800M market cap.  While their intrastriatal approach makes it difficult to tell how much huntingtin knockdown is achieved, also on a single cellular level, and how much is therapeutically impactful, the increasingly promising clinical evidence and them probably pursuing the most impactful target increases my confidence in their success.  Alignment with the FDA on the statistical analysis of their ongoing study which would form the basis for an accelerated approval along with upcoming 3-year follow-up data in the next 6 months will be key.

Please let me know in the comments below, if you come across evidence disproving the exon 1 hypothesis.  I will keep looking, but have yet to find it.

Longer-term, I am excited by the prospect of exon 1-targeting therapeutics with agents slowing somatic repeat expansion, an area seeing significant investment.  Since somatic expansion is a stochastic, asynchronous process and overlap generation of toxic RNA/protein, combining the two mechanisms will likely be most impactful.