My notes from the RNA Therapeutics Symposium (RNATx) at UMass Medical School June 26-28, 2024.

Morgan Maeder | Therapeutic development of epigenetic editors

Dr. Maeder is VP of platform technologies at Chroma Medicine. Her talk is about epigenome editing. Chromatin conformation determines whether a gene is active or inactive. Inactive genes have promoters whose chromatin is closed, compacted, inaccessible due to DNA methylation and/or histone marks. Chroma works on editors to methylate DNA to repress targets, or demethylate DNA to activate them. Editors typically include a DNA-binding domain fused to a DNA methylation or de-methylation domain, and sometimes a transcription effector domain such as KRAB.

They have a program to silence PCKS9 through DNA methylation [Tremblay 2023]. The strongest hits in a cell line were all within 100-200 bp of the TSS. In primary hepatocytes, CRISPRi was transient, bouncing back by day 14, while the epi-editors provided durable repression, reducing expression >90% for 14 days. In vivo in human PCSK9 mice they have stable >98% silencing out to >300 days. They used whole genome bisulfite sequencing to look at methylation. Absent treatment, PCSK9 promoter CpGs are heavily methylated in brain but not liver; with treatment, liver acquires methylation just like brain has. They did not detect any significant off-targets. They did 70% partial hepatectomy to force the hepatocytes to divide and re-grow. Even in the context of this elevated cell division, methylation was maintained.

To test their epi-activators, they dosed mice that had previously received the PCSK9 silencer. This used a construct with a TET as the DNA demethylase. Within 7 days post-dose PCSK9 expression was restored to 100% of untreated. Whereas the silenced mice had CpG methylation extending well beyond the promoter, the re-activated mice had the demethylation concentrated pretty close to the promoter, not spreading further beyond, yet this was enough to fully restore expression.

In their Hepatitus B (HBV) program they are silencing the viral DNA regardless of whether it is an episomal loop, called cccDNA, or integrated into host genome. This is in contrast to nuclease-based cutting approaches, which would risk translocation and even could increase host genome integration.

They are also doing ex vivo cell therapy. They show that by multiplex epigenetic silencing they can make allogeneic CAR-T cells without risking chromosomal rearrangement which you would if you used nucleases to cut.

Q&A

Q. How is it that binding at one site gives you methylation for hundreds of base pairs?

A. DNMT3A/3L polymerizes and extends up and downstream. What we do is just provide the first copy, and then polymerization receruits endogenous machinery that extends the methylation.

Q. For those of us who are sadly ignorant about virology, can you explain both for natural HBV and for the AAV model system you used, does that viral DNA acquire histones, and do you know if the DNA methylation is silencing it by causing compaction around nucleosomes, versus direct repulsion of transcription factors?

A. Yes, it does appear that cccDNA acquires some degree of “chromatinization” and that our editors may work in the same way as they do on host DNA.

Q. Is this so specific that you could do allele-specific too?

A. Yes, this is specifically possible. Biology has examples (imprinted genes) where only one allele is methylated. The key is just whether your DNA-binding domain is sufficiently allele-specific.

Kiran Musunuru | Therapeutic gene editing for cardiovascular and metabolic diseases: from the leading cause of death to N-of-1 disorders

Dr. Musunuru began with the example of a 10-year-old child who has an LDL of 800 due to familial hypercholesterolemia, and has to undergo plasmapheresis weekly, and has already had 3 open heart surgeries. This is just the most extreme example; ischemic heart disease is the world’s leading killer, including in low and middle income countries, and has grown even more common since 2000. All approved therapies targeting PCSK9 require lifelong re-dosing. Dr. Musunuru’s work with Verve Therapeutics aims to use an LNP encapsulating Cas9 mRNA and a PCSK9-targeting guide RNA to do a “one and done” therapy. Verve dosed the first patient in New Zealand in 2022. Initial trial results just announced in Nov 2023 show that in the single patient at the 0.6 mg/kg dose they achieved 55% LDL lowering, which is close to the theoretical maximum achievable by PCSK9 knockout, suggesting saturation editing in hepatocytes.

In his academic lab at Penn, Dr. Musunuru is trying to use a similar approach to do corrective editing for rare disorders of metabolism [Brooks 2023, Brooks 2024]. A first target is phenylketonuria (PKU), caused by recessive variants in PAH. It only takes 10% wild-type expression of PAH to rescue plasma phenylalanine levels and disease phenotype. They are trying to correct two of the most common variants. All the mouse studies are done with human-specific reagents in humanized mice (his lab made 6 lines modeling different mutations). The mice have impaired growth and hypopigmentation. For the pathogenic variant P281L, within 24 hours of treatment, blood Phe levels go down by half, and by 48 hours, they are down to the wild-type level. A lifelong cure within 48 hours of treatment. They did a very similar approach with just a different editor (an adenine base editor / ABE) and a different guide for R408W, which is 22% of disease alleles in patients, and got similar results. Even though the time to effect is very fast, the whole pipeline to do the in vivo studies takes 1 year.

How can we take advantage of the inherently programmable nature of this therapy to do best by our patients? The 3rd through 6th most common PAH mutations each affect <6% of patients. Do we have to do everything all over again? They are setting up cell systems to screen base editing solutions for variants 3 through 6, and they believe that they should be able to skip in vivo experiments for those entirely. They found great base editors for #3, #4, and #5; they didn’t find a good one for #6 which might require prime editing or something. The in cellulo screening only takes a week.

A big problem is inequity of access to these therapies, and he doesn’t just mean the money to pay for approved expensive rare disease drugs. He means access to even have a therapy developed. There may be common founder alleles causing PKU in Africa and Asia that are less well characterized and patients with those variants will not have effective therapies. He calls this “mutational discrimination”. Not only will industry go after the variants most common in the U.S. and Europe, academics will too because they’re viewed as the best test cases, and that’s exactly what he himself just did. To solve this problem we need a patient-first approach. Any patient with a hepatic inborn error of metabolism, should have a fair shot at a personalized editing drug.

He is now working on 4 different rare inborn errors of metabolism in 4 different genes, affecting infants, for which liver transplant is currently the only option. They hope that eventually FDA will let them go to the clinic with cell culture data alone, but certainly for the first few test cases, it’ll be critical to have in vivo studies. They are rapidly generating mouse models of each using Rosa26 safe harbor locus knock-in for each variant.

Their vision for combating mutational discrimination is to convince regulatory agencies that once we have a full set of IND-enabling preclinical studies and Phase I for one variant one disease, we should be able to add additional guides for the same gene, and eventually other genes, with only on- and off-target editing data in cellulo, in the form of IND amendments rather than new INDs.

Q&A

Q. Is FDA open to this?

A. Yes, we asked them and they are open to eventually getting towards this kind of pathway where you just amend the IND.

Q. What is the cost of all this?

A. Remember that these kids are severely ill from day 1 of life, and almost never leave the hospital. If they get a liver transplant that is $800,000 in care. We think we can get the discovery of each marginal editor/guide down to $100,000 per patient.

Omar Abudayyeh & Jonathan Gootenburg | Programmable molecular technologies for genome editing and cell control

Together these two PIs run the “Abugoot” lab at Harvard Medical School.

Most genetic diseases are highly allelically heterogeneous, and it may be tough to make so many different base editors to address every patient’s mutation. In recessive loss-of-function diseases, supplying a copy of the lost gene would be a more universal therapy across all different patients with the same gene. They developed PASTE [Yarnall 2023] to try to make DNA integration programmable to guide integrases. They showed they can use it to get >10% integration for payloads of up to 36kb. They delivered it in vivo in a 3-vector adenovirus system and were able to get 1-2% insertion in hepatocytes.

In the long run, they think triple adenovirus delivery is not the right approach. They spun out a company, Tome Biosciences, which is trying to do the same using a combination of LNPs and AAV. They presented data at ASGCT in May 2024: after iterative improvements, they have improved delivery and are now able to get 9% restoration of PAH in hepatocytes in non-human primates.

They are also working on Cas7-11, which is an RNA-directed nuclease [Ozcan 2021, Kato 2022, Strecker & Demircioglu 2022]. They found one isolate at the bottom of Tokyo Bay that was highly functional and could cleave RNA highly precisely, leaving behind fragments at clearly defined molecular weights on a gel, in contrast to Cas13 which tends to totally degrade its target. Cas13s also have a lot of collateral activity and harm cell viability, which Cas7-11 does not. They solved the structure and got rid of unnecessary domains, making it fit in an AAV. They used this to develop an analog of PASTE for RNA. It is called PRECISE [Schmitt-Ulms & Kayabolen 2024], Programmable RNA Editing & Cleavage for Insertion, Substitution, and Erasure. They can get up to 40% editing, though it is lower for very highly expressed transcripts.

What else is out there? Other than Cas7-11, almost every editing tech out there is based on Cas9. He wants to broaden that scope. Cas9 is but a small slice of the diversity of CRISPR systems. There are actually far more Class I CRISPR systems than Class II (which is what Cas9 is). They and Feng Zhang found two other RNA-programmable nucleases, the TnpB and IscB families, which are very abundant in prokaryotes [Altae-Tran & Kannan 2021, Karvelis 2021]. Can it really be that only prokaryotes have such systems? TnpB is similar to the Fanzor1 and Fanzor2 families of proteins [Bao & Jurka 2013]. They looked computationally for proteins similar to Fanzors, and found a huge number across different kingdoms [Jiang & Lim 2023]. They found versions in clams, mussels, and green algae and evaluated their suitability as mammalian gene editors. They don’t have protospacers, so instead of PAMs, they have target adjacent motifs (TAMs) which look like TTTA. They also natively have an N-terminal sequence that functions as a nuclear localization signal (NLS) - this is in contrast to Cas9, where you have to add an NLS in order to enter the nucleus in eukaryotic cells. Promisingly, these enzymes are small and do effectively introduce indels at a programmable locus in human cells.

Finally, they used ADAR to develop a programmable protein expression system using ADAR sensors [Jiang 2022]. You can use it to detect and even quantify expression of a gene by luciferease reporter, or to kill cells in a targeted way using iCaspase as the reporter. You can multiplex it to create simple logic gates, e.g. “AND”, where you only get a reporter signal if two different genes are expressed.

Steve Dowdy | How do we pull off the great endosomal escape?

Endosomal escape is the great limiting problem for all oligonucleotide therapeutics [Dowdy 2023]. The problem dates from 4 billion years ago, when lipid bilayers came to encapsulate an organism and keep out foreign nucleic acids. Whereas small molecules can permeate the membrane, oligonucleotides get in via endocytosis, but as long as they are enclosed in an endosome, they are intracellular but not in an active compartment. For PS ASOs, only 1% of dose escapes endosomes, for double-stranded siRNAs it is 0.3%, and fror neutral PMOs and PNAs it is <0.1%.

How are there are 27 FDA-approved oligonucleotide therapies [Egli & Manoharan 2023]? A few are for Duchenne and benefit from dystrophin deficiency compromising muscle cell membrane integrity, making it easier to get in. Most are for liver and rely on GalNAc. GalNAc binds ASGPR on hepatocytes. ASGPR has 500,000 copies per cell and recycles every 15 minutes, for an incredibly high throughput pathway into the endosomes. It takes 50,000 - 70,000 copies of ASO per cell to saturate response, while for siRNA which are more metabolically stable, it only takes 2,000. Fitzgerald NEJM 2016 from Alnylam on inclisiran shows that it takes 2-3 weeks to get maximal RNAi response. That delay is due to endosomal escape. Brown et al NAR 2020 show that only 0.3% of GalNAc siRNA ever escapes from the endosome. Dowdy estimates that GalNAc siRNAs escape at a rate of just 5 molecules per cell per hour. We are giving many times the dose we need to give.

The endosomal-lysosomal pathway is incredibly complex and undefinable; you see diagrams of it but the truth is there are all different compartments with different markers and pH, and we have no idea which specific compartment(s) the oligonucleotides escape from. When you try to pharmacologically or genetically perturb the system, it is like having an accident on the L.A. freeway at 5:00pm: every route shuts down, and the blockage is so bad everywhere you can’t actually tell where the accident is by looking at the traffic. If we can’t perturb the system, can we study how oligos “naturally” escape from endosomes? PS ASOs bind the lumenal membrane of their vesicle and therefore are close to the wall of the compartment. siRNAs which typically have PS linkages only at the ends and not throughout, bind less tightly. PMOs which are neutral do not bind at all. One possibility is that nanosecond-long tears in the endosomal membrane allow things to get out, and then “heal” back like oil on the surface of water. If you’re close to the membrane, you are more likely to happen to flit out during the momentary opening.

How can we enhance endosomal escape? The most obvious, and deadliest, approach is endosomal rupture - by endolytic agents such as chloroquine. This lets everything out, including antibodies, and causes a massive innate immune response. This approach is completely clinically inviable. What could be more selective for drugs? One idea is cationic CPPs, with positive guanidinium groups. The problem is that siRNA and PS ASOs are anionic, so if you conjugate them to anything cationic, they just stick to themselves like superglue. PMOs are neutral, so you can conjugate them. But the cationic CPPs stick to everything including blood cells [Ho 2001]. There are people still doing this with PPMOs for muscle, say Entrada and Sarepta. The oligos bind to the outside surface of muscle cells, and as the Duchenne boys try to move, their fragile membranes have nanoscopic ruptures which allow the oligos into cytosol. Another option is hydrophobic conjugates, which work but have very high cytotoxicity because the hydrophobic group just wants to insert into the middle of the lipid bilayer and just stay there.

If this was easy, it would be solved by now.

Enveloped viruses are professional endosomal escape artists. Flu and SARS-CoV-2 for example. They use a ligand such as hemagglutinin (HA) or spike, which has an outer hydrophilic domain that binds the cell surface receptor, concealing a hydrophobic inner domain. Once the hydrophilic domain gets the virus into the endosome, the pH change causes cleavage of the hydrophilic domain, exposing the hydrophobic domain which inserts into the endosomal membrane and causes a very localized membrane destabilization that allows the virus out into cytosol. Each virus has 10-20 copies of this on its surface, and they come in trimeric architectures because this gives you the right amount of area of disruption on the mebrane. The viruses escape from endosomes with an efficiency of 30-70%, so 100 times better than our oligonucleotides do. You can’t just add HA or spike onto the oligo because their molecular weight is 210 kDa (vs. say 14 kDa for the oligo itself) plus the’re incredibly potent antigens and everyone already has immunity to them. We need to get the size down to 2 kDa. People have tried to mimic this with peptides, but no one has figured out a way to get that kind of sophisticated function in such a short amino acid sequence.

Dowdy’s lab is working on uEED monomers, with a hydrophilic mask, endo cleavable linker, hydrophobic core, ideally all composed of moieities that are naturally occurring in our bodies or have clinical precedent. Then we can search the space of how many of these monomers of which composition we need in order to get function. 10 years ago they worked on oligo phosphotriesters [Meade 2014] and tried over 100 bioreversible versions. The problem is the kidneys are really good at filtering this out. Even for GalNAc siRNA, which get into liver cells very fast, 30% of dose is lost because the kidneys are yet faster. They had to find a way to mask this and make it look like a protein so kidneys would leave it in the blood. They tried hydrophilic glycoside masks with an endo linker and then a hydrophobic indole ring [Jadhav 2023]. They have 11 options of endo/lyso restricted glycosidases. When glucuronidase clips off the sugar, exposing the carbamate on an aromatic ring which gets decarboxylated into CO2 leaving an exposed indole ring with no charge, which should insert into the membrane. It’s very stable in serum because there are no glycosidases in serum. Addition of β-glucoronidase at pH 5 causes quantitative cleavage within 30 minutes. They are still working on optimizing this with a goal of 10x increased escape without added tox, and they have a diversity of chemical options to explore.

Q&A

Q. If we solve the problem, how often will we have to dose?

A. Of course, the gradual release from endosomal depots is why we can dose inclisiran only twice a year. We can imagine having a single drug vial with two versions of the molecule, with and without the endosomal escape conjugate. Maybe 20% of the dose would be conjugated, and would escape right away and get us immediate target engagement, and then the 80% would give us the sustained activity over months. Remember that there are many diseases we still can’t touch due to the kinetics of endosomal escape, such as COVID, where the virus replicates faster than escape (see Alnylam/Vir collaboration), or cancer, where cell division dilutes the oligos too fast. Endosomal escape conjugates would make those diseases addressable by oligonucleotides.

Ken Yamada | Extended nucleic acid (exNA): a novel platform technology development for enhancing siRNA potency in vivo

Of the 27 FDA-approved oligonucleotide drugs, all 6 siRNAs are for liver [Egli & Manoharan 2023]. We are way behind on targeting siRNA to other tissues. Luckily oligonucleotides can be thought of has having separate “dianophores” and “pharmacophores”, so if we can develop an improved scaffold, we can plug in sequences to target almost any gene we need to target in a given tissue.

Phosphorothioate (PS) is generally used to improve stability of oligonucleotides, but it can still be cleaved in vivo. Dr. Yamada set out to make an even better linkage that would be more nuclease-resistant. One attempt was iE-VP [Yamada 2021]. Inspired by nature, which uses DNA methylation and histone lysine methylation to alter local nucleic acid structure, he set out to introduce an additional methyl group into the nucleic acid backbone, thus inventing the exNA linkage [Yamada & Hariharan 2023]. At many positions in an siRNA, exNA was highly detrimental to RISC activity, but when multiple exNA were added at the far 3’ end, especially positions 18-20 or 19-20, it was not only tolerated but actually increased RISC activity. They therefore tested stability to 3’ exonucleases and found that exNA vastly improved it. Whereas 2 PS linkages at positions 19-20 on the 3’ end yielded a half-life of 1.1 hours, 2 exNA with PO gave 9 hours, and 2 exNA with PS gave 32 hours.

They tested subcutaneous (SQ) injection of DCA-conjugated [Biscans 2021] exNA siRNAs versus equivalent without exNA. Cmax was improved by 3.1-fold, but the half-life was also much longer such that the AUC was improved by 5.8-fold. At 10 mg/kg SQ at a 1 month post-dose timepoint they had 16x improved drug accumulation in liver and 2x - 8x improved drug accumulation in several other tissues. Oligonucleotides always go to kidney, of course. But siRNAs do so only for clearance and are generally not functional (do not engage target) in the kidney. But for the DCA-exNA-siRNA they saw excellent silencing in kidney.

In the CNS, they have found that exNA gives improved potency for several targets including SOD1, where a single dose of exNA siRNA in G93A mice gives much greater survival improvement than tofersen, and with chronic dosing of exNA siRNA, the animals are still alive as of when the preprint was posted last week [Weiss 2024].

Maire F. Jung | Advances in therapeutic oligonucleotide delivery

Dr. Jung is VP and Chief Scientific Officer for RNA at Eli Lilly’s Institute of Genetic Medicine.

A bit of background on Lilly’s journey and strategy in genetic medicine. 20 years ago they dabbled in ASOs, then stopped. From 2018-2021 they set out to build a platform in siRNA, starting with deals with Dicerna and Avidity. From 2022 to present they sought to expand modalities for instance adding editing, including recent deals with Scribe and Verve. From 2024 onward their focus is on improving delivery.

We all love the potential of siRNA, but it’s been a rough journey of hype and disappointment, as famously illustrated by [Khvorova & Watts 2017]. Early work (2000s) focused on formulation (e.g. LNP) for delivery, and mostly it was a bust. Later work (2010s) pivoted to chemistry (conjugates) with first approval of an siRNA drug finally in 2018. Alnylam has now had several approved liver drugs, and dosed the first CNS siRNA (for APP). Dr. Jung thinks the future is precision delivery to cell types/tissues other than the liver.

All the clinical success with siRNA has been built on chemical modifications to increase half-life of RNA. Most of those modifications are also thermally stabilizing (they increase melting temperature of duplexes), but the tradeoff is that this means tighter binding even in the presence of mismatches, so it can make off-targets worse. Just recently there have been some new modifications reported that are thermally destabilizing, and thus allow the potential to achieve a balance, improving durability without worsening off-targets.

Lilly has lepodisiran in Phase III (trial called ACCLAIM-Lp(a)), it is their flagship GalNAc siRNa program against LPA for the liver. The phase I was published [Nissen 2023] and showed that at high dose 608 mg, >95% suppression of serum LPA maintained 1 year post-dose.

How do we find “the next GalNAc” for tissues other than liver? There are improvements in many tissues, including CNS [Mummery 2023], sometimes using simple lipophilic conjugates (probably referring to Alnylam’s C16).

Lilly has achieved improvements in adipose-selective gene silencing using conjugated siRNA delivery to inguinal white adipocytes. It is completely liver- and kidney-detargeted, with no significant target engagement in either tissue. Adipocytes are about 50 μm in diameter and are basically a big lipid droplet with the actual cellular machinery in a bubble on one side. Their imaging shows that the siRNA accumulates exclusively in that cellular machinery.

Lilly is also using a transferrin receptor-specific ARC (antibody RNA conjugate) to deliver siRNA systemically and cross the blood-brain barrier. It is a TfR antibody, a linker, and an siRNA. In both mouse and monkey, even 0.25 mg/kg “effective siRNA” (that does not contain the antibody’s molecular weight) gives >50% KD, and 5 mg/kg gives about 80-85% KD. The knockdown is relatively uniform across the brain. Everyone has long struggled with the “neuraxis”, getting uneven distribution after intrathecal dosing; the systemic dosing appears to be more uniform. They have also stacked IV delivery q1w, IV, 28 days in mouse; it looked like about 90% knockdown.

Q&A

Q. Does your TfR ARC go only to brain?

A. No, it goes everywhere TfR is expressed, including liver.

Q. What is the dose level and conjugate for the adipocyte delivery?

A. I’ll tell you next time!

Joshua E. McGee

McGee’s research is on self-amplifying RNAs (saRNA) by encoding a viral RNA-dependent RNA polymerase (RdRp) together with a gene of interest [McGee 2023]. Once the initial RNA is expressed, you get replication and amplification of the RNA including the gene of interest. The kinetics look like a natural viral infection to your immune system, and allow you to get a lot of antigen expression with a very low dose. Thus, this should be a promising way to deliver mRNA vaccines. Problems are that we have innate immune responses to recognize RNA-dependent transcription, and the saRNA is long (at least 9 kb). Modified nucleotides have transformed mRNA therapeutics, but it turns out most of them, such as pseudouridine, do not work with saRNA because they prevent RdRp binding. They set out to do a screen to identify modifications that would be compatible with saRNA. They found 3 that are compatible: hm5c, 5mC, and N1mΦ. When they took the saRNA in vivo, they found that unlike using unmodified RNA, the expression did not peak until about a week post-dose, adn there was still detectable antigen at 28 days post-dose. In a SARS-CoV-2 challenge study in mice, the 5mC-modified saRNA yielded a statistically significant improvement in survival compared to WT saRNA.

Q&A

Q. What about the opposite - select RdRp mutants that can tolerate pseudouridine?

A. Yes, there was a pre-print about that just recently! (Perhaps referring to [Quintana 2024]).