Covalent drugs straddle these two worlds. They can be used to target proteins previously thought undruggable, and to create more selective/tolerable versions of existing inhibitors. Such drugs have experienced a surge in popularity in recent years. Ibrutinib and Acalabrutinib, both covalent BTK inhibitors, were the subjects of For Blood and Money1 Sotorasib and Adagrasib, covalent inhibitors used to treat KRAS G12C-mutated NSCLC and colon cancers, were the first approved drugs that managed to bind to RAS. We’ve known RAS to be a cancer driver since the 80s, so it’s fairly big deal that covalent drugs were the ones to crack the code after decades of failed efforts!

Unlike traditional small molecule drugs, which function via non-covalent interactions, covalent ones contain an electrophilic ‘warhead.’ Once in position (which, importantly, is achieved non-covalently), this warhead is able to accept electrons from its nucleophilic counterpart (often cysteine) and form a covalent bond. There are technically reversible and irreversible covalent inhibitors, but for the purposes of this note we’ll focus only on irreversible. Irreversible is a bit of a misleading term, and in reality means the protein will be inhibited by the drug until it’s resynthesized (which could take 24 hrs plus). This is, however, very different from standard small molecules, where inhibition occurs for minutes to single digit hours.

Traditional, non-covalent, small molecules present real challenges when it comes to targeting ‘undruggable’ proteins that play a role in disease. These proteins tend to be characterized by flat surfaces with no obvious binding sites (think of trying to grab onto a flat, slippery, cliff face rather than a beginner bouldering hand hold). That’s not an environment where small molecules tend to thrive: the lack of clear binding sites means an interaction requires a far more extensive number of non-covalent bonds. That can be achieved with a much larger protein, but not with a molecule that’s ~100x smaller.

Subscribe now

Covalent small molecules have an advantage here, namely that they’re not completely reliant upon binding affinity. This is unique compared to these other approaches for targeting ‘undruggable’ proteins. The thesis behind Aktis’ approach is that its miniproteins are large enough to generate a sufficient number of non-covalent interactions to bind to a flat protein. Consequently, their S1 emphasizes the sub-nanomolar/nanomolar binding affinity of its pipeline candidates. Revolution Medicines’ chief innovation is around molecular glues, which again act non-covalently!

Things are a little different when it comes to covalent compounds. These compounds operate in two steps:

1) Like traditional small molecules, the compound first binds to the protein via non-covalent bonds.

2) Next, the small molecule’s warhead forms a covalent bond with a closeby nucleophile.

One way to conceptualize these steps is to think about effectively shutting the front door on a windy day. Step one is closing the door. This puts the door in the correct position, but at some point it’ll get blown back open. So it is for regular small molecules, where the aim isn’t extended inhibition. However, if the goal is to keep the door closed overnight one needs to actually turn the lock. So it is with a covalent small molecule that’s trying to inhibit a protein for an extended period of time.

Binding affinity is still very important here. Without it, the molecule can’t get in position to form a covalent bond. However, the maximum potential rate of covalent bond formation (denoted as k_inact) is also very important. k_inact tells you the speed with which a covalent bond forms with the target protein assuming all of the target protein has formed the initial non-covalent complex.2 Taken together, k_inact/K_I describes the overall efficiency of covalent bond formation for these covalent inhibitors.

In other words, measuring binding affinity alone doesn’t give an accurate view into how well a covalent compound may be performing. This is well-illustrated in an example given by DrugHunter:3

Ritlecinitib (Litfulo) is a covalent JAK3/TEC kinase family inhibitor, developed by Pfizer, for the treatment of alopecia areata, an autoimmune condition causing hair loss. The goal of the compound was to selectively inhibit JAK3, a task it doesn’t look very good at doing based on K_i alone. When you look at k_inact and k_inact/K_i, however, things look much more promising. The general takeaway is to be wary of a covalent drug developer spending too much time discussing binding affinity alone.45

The covalent nature of these compounds additionally means one can worry less about the drug’s pharmacokinetics. For non-covalent small molecules, tracking its plasma concentration (C_p) over time is critical for determining when a next dose might be needed. The lower the plasma concentration, the less drug there is inhibiting the target protein. For covalent compounds this isn’t the case. Plasma concentration matters initially, but only until the warhead reacts with its protein target. Once that occurs, the target remains inhibited until re-synthesis. In other words, C_p stops correlating with drug effect.

Historically, development of covalent molecules was shunned given concerns about toxicity and off-target effects. Covalent drugs can lead to hapten formation, where the drug covalently binds to a protein and the resulting complex causes a massive immune response. This isn’t something one has to worry about when a molecule can’t bind covalently! Moreover, off-target effects become a bigger issue when covalent bonds are involved: inadvertently inhibiting a non-target protein for 24 hours plus is a bigger concern than inhibiting it for only a few minutes. That’s not to say we weren’t using covalent drugs: Penicillin and Aspirin act covalently, but their covalent nature wasn’t known at the time of approval. Put another way, the challenge with trying to develop a drug known to be covalent is that it would be discarded/deprioritized in favor of other compounds.6 This is no longer the case. Revolution Medicines’ lead candidate may act only non-covalently, but the company has earlier-stage candidates that throw in a covalent bond on top of the molecular glue. The 2021/2022 approval of Sotorasib/Adagrasib means covalent drugs are now seen as having real potential for hitting ‘undruggable’ proteins. There are even companies like Vividion Tx, Frontier Medicines, and Scorpion Tx, whose entire thesis is that chemoproteomics can find protein binding pockets that are promising targets for covalent compounds. Vividion was acquired by Bayer for $1.5 billion, and Lilly will spend potentially $2.5 billion acquiring only one of Scorpion’s candidates (the other assets have been spun-out into a new entity called Antares).7 Such big pharma interest in covalent first approaches would’ve been hard to imagine 15 years ago.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

Terns’ general business strategy has been the following: develop small molecule drugs targeting an already validated mechanism of action, but where there’s an opportunity to improve efficacy/safety/tolerability.1

Its lead candidate is a good illustration of how this works and why this strategy is an appealing one. CML is caused by a chromosomal translocation, where a piece of chromosome 22 breaks off, merges with chromosome 9, and vice versa. Consequently, chromosome 9 is longer than usual, and chromosome 22 is shorter. Chromosome 22 now contains a BCR-ABL1 fusion gene, and in this state is described as the Philadelphia chromosome.2

This BCR-ABL1 fusion gene encodes for an abnormal tyrosine kinase that remains stuck in the ‘on’ position. Much like Ras proteins, healthy tyrosine kinases play a vital role in regular cellular processes. Critically, these kinases are meant to switch between active and inactive. In the case of BCR-ABL1, this switch between on and off doesn’t happen, and tyrosine residues are instead phosphorylated constantly/uncontrolled cellular proliferation occurs.

The good news with tyrosine kinases is they have well-defined active binding sites, and so historically weren’t considered undruggable. Kinases are enzymes that closely match the enzymes one learns about in high school biology. Designing an inhibitor, then, is just a question of designing a molecule that can inhibit that enzyme-substrate binding. This relative straightforwardness has resulted in over 50 FDA approved tyrosine kinases inhibitors for the treatment of various cancers. In the case of CML, the first approved inhibitor was Imatinib (Gleevac) in 2001. There have been a number of ‘second-generation’ inhibitors approved since then (Dasatinib, Nilotinib, and Bosutinib) with pretty direct trade-offs of increased potency for greater side-effects. It’s worth emphasizing how much of a step forward Imatinib was: this inhibitor altered CML into a cancer with a ~30% 5-year survival rate into one with a survival rate of 70%+.3

Unfortunately, these well-defined binding sites (or orthosteric sites) come with downsides. The primary one is there’s not much difference between the BCR-ABL1 binding site and the binding sites of other, non-mutant, tyrosine kinases present in the cell. The result is that healthy tyrosine kinases are inhibited, causing unpleasant side-effects. This led to a fairly fragmented market for second-gen TKIs. They collectively comprise over half of the 1L CML market, but there wasn’t a clear winner. Dasatinib and Nilotinib both peaked at ~2 billion ish in sales, with Bosutnib at a more modest ~600mm.4

Subscribe now

This drastic improvement in survival rates, combined with non-ideal tolerability profiles, makes CML quite an interesting disease to develop new treatments for. The side-effects of these second-generation inhibitors are not insignificant, which leads to a high rate of patients switching medications. Terns’ management estimates that “40% of CML patients on these active site TKIs switch therapy within 5 years due to lack of sufficient treatment response or side effects.”5 The company also estimates that, within G7 nations, there are 17,000 new frontline CML patients every year and 13,000 CML patients who switch medications.6 If you can develop an effective medication with a real improvement in toxicity you can help patients for far longer than a new therapeutic in, say, the pancreatic cancer space might be able to.7

One solution to the tolerability issues present in 2nd gen TKIs is to alter BCR-ABL1’s activity by binding to an allosteric site rather than an orthosteric one. In the context of an enzyme, an allosteric site is a binding site that, while separate from the active site, has a direct effect on the active site. In other words, binding to the allosteric site can alter the shape of the active site, and thus increase/decrease the likelihood of substrate binding. Allosteric inhibitors can almost be thought of as putting a boot on a car. It doesn’t stop the engine from working, but the result is still a nonfunctional vehicle. Allosteric inhibitors don’t directly block the active site, but they still render the enzyme inactive.

Historically, the challenge with allosteric sites is that they’re often hard to actually locate on a protein (the reasons why are likely a subject for a different piece). By hard, I mean hard enough that D.E. Shaw’s founder has spent enormous sums of capital building supercomputers to run molecular dynamics simulations in the hope of finding these targets. Targeting allosteric sites also, however, comes with very real benefits. Most notably, allosteric sites differ between different tyrosine kinases much more so than orthosteric sites do. The result is that off-target effects should be much less, which should mean one can take dosing higher and thus more effectively attack the cancer. Targeting an allosteric site is especially helpful in cases of tumor resistance, where the BCR-ABL1 active site may have mutated so as to avoid these first-gen/second-gen inhibitors.

Terns is leveraging this allosteric approach, following in the footsteps of Novartis’ Asciminib (Scemblix), considered a third generation TKI. Asciminib, like TERN-701, is a STAMP inhibitor (specifically targeting the ABL myristoyl pocket). This ABL myristoyl pocket is an allosteric site that can lock BCR-ABL1 in its inactive position. Asciminib received accelerated approval for the treatment of third line CML in October ’21, and then received accelerated approval for first-line in October ’24. The drug has been enormously successful thus far, doing 1.3 billion in ’25 revenue with a ~4 billion peak sales estimate. In the U.S., Novartis estimates Asciminib has 23% front-line share, 57% second-line share, and 59% third-line share.8

Despite Asciminib’s success, there does seem to be a clear opportunity for improvement. Asciminib is a therapy that primarily competes with second-generation inhibitors on tolerability rather than efficacy.9 That’s valuable, but not as valuable as a therapeutic that can deliver better efficacy along with improved tolerability. Asciminib also annoyingly requires that patients fast for two hours before taking the medication and for one hour after.

The key bet with Terns is that its CML candidate can deliver on all fronts. It’s also uniquely derisked. There are a number of companies pursuing allosteric inhibitors for clinically validated targets, but very often for targets where:

1. Orthosteric inhibitors are the only validated options or

2. The target is known to be a disease driver but was previously considered undruggable.

In those cases the allosteric approach has an obvious appeal in terms of avoiding off-target effects/inhibiting a protein we previously couldn’t bind to, but there’s always the chance that binding to the allosteric pocket causes some unanticipated side effects.10 Given the approval of Asciminib, there’s no risk that the allosteric binding specifically leads to side-effects that make TERN-701 untenable. Moreover, there’s precedent that TERN-701 could move from third-line all the way to first-line, and there appears to be a decent amount of physician willingness to switch from TKIs that have been on the market for much longer to newer approaches. TERN-701 can be taken with food, and is hoping to avoid the side-effects of Asciminib (most notably pancreatic toxicity and hypertension). If that ends up being the case (and assuming other, different, side-effects don’t come up), TERN-701 could end up taking Asciminib’s 4 billion opportunity and then some.

All this said, TERN-701 is relatively early in development; the company’s most recent Phase 1 clinical data update included 38 efficacy-evaluable patients. Once this cohort is split into patients in the trial because of lack of efficacy on their last TKI (20) versus lack of tolerability on their last TKI (16) the numbers get quite small.11 The market’s reaction to this data has been extremely positive, I think largely because the existing body of knowledge on effective MoAs for treating CML is comprehensive enough that a large clinical trial isn’t seen as necessary to conclude improved efficacy/tolerability.12

There was some debate upon the merger announcement as to whether Merck had given Terns a low-ball offer. Leerink was a proponent that the price was too low, putting peak sales at ~6.2 billion and arguing the acquisition price should sit at 1.5 - 2x peak sales/another bidder should come in. The Schedule 14D-9 came out yesterday and paints a more complete picture of the acquisition process. There were six potential bidders involved in varying degrees: Party A, B, C, D, E and Merck. Party E never responded to initial outreach, Party D said it wasn’t interested, and Party A and B both said they couldn’t arrive at a valuation above Terns’ current market capitalization.13 Interestingly, Party C pulled out after seeing additional clinical data:

“At the meeting, Ms. Burroughs presented new CARDINAL trial data showing TERN-701’s consistent performance compared to the data presented at the ASH Annual Meeting. However, the MMR achievement rate was lower, potentially due to more patients being pre-treated with asciminib in the evaluable population.14 The MMR achievement rate stayed within Terns’ disclosed confidence interval after the ASH Annual Meeting, with no overlap with the asciminib interval. The data supported a similarly encouraging safety and tolerability profile for TERN-701 as the data from the ASH Annual Meeting. Ms. Burroughs speculated that some aspects of the updated data may have contributed to Party C’s decision to withdraw from discussions to potentially acquire Terns”15

Party C’s withdrawal illustrates an important point when determining Terns’ fair value. The risk with this asset is that its efficacy drops/side-effects increase as more patients are evaluated, such that it ends up trending closer to Asciniminb. Such a drop would lead to a more modest peak sales number, and could result in similar dynamic to 2^nd gen TKIs where both Asciminib and TERN-701 top out at ~2 billion. That’s a perfectly fine outcome, although not a 6.2 billion one.16

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

The importance of targeting PPIs is perhaps perfectly illustrated through RAS proteins. RAS proteins have been known to drive many types of cancers since the 1980s, and yet for decades there wasn’t anything that could be done to inhibit these proteins’ activity. There was no obvious binding site that could be leveraged to inhibit RAS function. RAS exists primarily in two states: active and inactive, or RAS(ON) and RAS(OFF). In its active state, RAS is bound to GTP; in its inactive state RAS is bound to GDP. To further complicate drug design efforts, RAS binds to GTP and GDP with picomolar affinity, which makes those proteins incredibly hard to displace with an alternative.

In healthy cells, RAS proteins are essential for cellular proliferation, differentiation, and survival. Typically, RAS exists primarily in the off state; if you looked at the ratio of inactive to active RAS proteins in the average cell, RAS(OFF) would far outweigh RAS(ON). This dynamic is often flipped in cancerous cells, where increased RAS(ON) activation allows the cancerous cells to flourish at the expense of their host. It’s worth emphasizing that these RAS mutations are known to be driving the cancerous growth; we’re not discussing a passenger mutation. RAS exists in three different isoforms: HRAS, NRAS, and KRAS. Most conversations center around KRAS, as it’s a more prevalent oncogenic driver. RAS mutations are most commonly present in lung (~30% of all cases), colon (~40% of all cases), and pancreatic cancers (>90% of all cases), so most of the discussion centers around treating those diseases:

The first approved RAS-targeting therapeutics were Sotorasib and Adagrasib, approved in 2021 and 2022, respectively. Both were initially approved for previously treated patients with KRAS G12C-mutated non-small cell lung cancer (NSCLC). Both are now also approved as part of combination therapies to treat KRAS G12C-mutated colon cancer.

Sotorasib/Adagrasib represent true innovation and are based on the work of Kevan Shokat out of UCSF. The KRAS G12C targeting piece is essential for understanding how these drugs work. For patients with this mutation, the 12^th codon of the KRAS gene is altered. This alteration turns what should be the amino acid glycine into the amino acid cysteine. This cysteine mutation, specifically, is actually very useful when it comes to drug development. It’s highly nucleophilic, or in other words eager to form covalent bonds. Sotorasib/Adagrasib takes advantage of this nucleophilicity, forming a covalent bond with the cysteine mutation when RAS is in an inactive state.1 These compounds don’t try to disrupt the actual RAS-GDP complex; such an approach would be incredibly challenging given the strong binding affinity between RAS-GDP. Instead, the covalent bond with the cysteine mutation prevents RAS from switching back into its active state, and so should positively affect the ratio of RAS(ON) to RAS(OFF) proteins in the cell.

Sotorasib/Adagrasib’s binding to the cysteine residue makes them highly selective. This cysteine mutation only occurs in mutant KRAS proteins, so there’s little risk of affecting healthy KRAS or off-target effects.2 There are downsides, however:

• Both drugs work initially, but treatment resistance does eventually occur.

• Given than RAS(ON) is what’s driving cancer growth, Sotorasib/Adagrasib’s RAS(OFF) binding isn’t ideal. Lowering the number of mutant RAS(ON) proteins in circulation is a good thing, but you’d still ideally be directly targeting RAS in its active state.

• These drugs work because of the cysteine amino acid specifically; again, cysteine is highly motivated to form covalent bonds. It’s much harder to design drugs targeting a different amino acid mutation, even if it’s still on that 12th codon. The G12D mutation is actually a more common driver of cancer, but it’s not as nucleophilic.

• The KRAS G12C mutation is only present in 14% of NSCLC cases and 3% of colon cancer cases. The G12C mutation is even less prevalent in pancreatic ductal adenocarcinoma (1-2%), the poster child of RAS-driven cancers.

This last point illustrates a drawback of the mutation-specific approach generally: it means one’s limited to developing drugs in a mutation-by-mutation manner. In an ideal world, there’d be a drug that’s just as effective treating patients with a G12C mutation as it is at treating those with G12D, G13C mutations, or others.

Subscribe now

Historically, there was good reason to think such an approach wasn’t feasible. The challenge with broadening the selectivity of a RAS inhibitor is that this increases the odds you’ll end up affecting the body’s healthy RAS cells. In other words, you’re going in the opposite direction from the general trend in cancer treatment. Pluvicto targets the PSMA antigen because it’s heavily expressed in prostate cancer cells and minimally expressed in healthy ones. RAS multi-selective inhibitors target RAS proteins because they’re cancer drivers, but with the trade-off that you risk affecting the healthy RAS proteins present throughout the body. In other words, it’s a little bit like traditional chemotherapy.

It is here that Revolution Medicines enters the picture. The company went public in February 2020, and at the time was best described as a very interesting small molecule company going after a variety of targets, one of which was RAS. Its focus has narrowed since then, and Revolution Medicines now describes itself as “a clinical stage precision oncology company developing novel targeted therapies for RAS-addicted cancers.”3 Its two most mature candidates are:

Daraxonrasib – a multi-selective RAS inhibitor currently in registrational trials for PDAC (the most common form of pancreatic cancer) and NSCLC. This pan-RAS inhibitor makes use of only non-covalent bonds.

Zoldonrasib – a G12D-selecitve RAS inhibitor in registrational trials for PDAC. Like existing RAS inhibitors, Zoldonrasib makes use of a covalent bond to increase selectivity.

Critically, Revolution’s RAS inhibitors bind to RAS(ON) rather than RAS(OFF), and so more directly attack what’s causing cancer proliferation. The company’s candidates are all small molecules, but small molecules that act in quite a unique way. Daraxonrasib and Zoldonrasib are considered molecular glues, which means they’re small molecules used to affect protein-protein interactions.4 The small molecule binds to one protein, which alters that protein’s binding surface. This altered surface enables an interaction with another protein that wasn’t previously possible. These glues have actually been used clinically for a while, even if it wasn’t clear that was the mechanism of action: take Cyclosporin A, which has been approved for use in organ transplants since the early 1980s. Cyclosporin A enters the cells and binds to Cyclophilin (most commonly cyclophilin A). This binding alters the shape of Cyclophilin, which now enables the binary complex to bind to Calcineurin. Inhibiting Calcneurin function dampens the immune system, thus preventing organ transplant rejection.

Revolution Medicines’ candidates act by blocking RAS(ON) from initiating its deleterious downstream effects. Like Cyclosporin, the inhibitor enters the cell and binds to Cyclophilin A. This binding alters the shape of Cyp(A), which enables this complex to now bind to RAS(ON) and inhibit its function5 This protein-protein interaction is, in the case of Daraxonrasib, comprised entirely of non-covalent interactions. That’s what one would expect when trying to modulate a PPI: for protein surfaces with no obvious binding site, one is reliant upon many weak intermolecular contacts.6 It’s worth emphasizing that Cyp(A) and the small molecule are both necessary for this RAS inhibition to occur; neither is sufficient by itself. Revolution refers to the RAS(ON)-Cyp(A)-small molecule block as a tri-complex.

Both Daraxonrasib and Zoldonrasib are exciting therapeutic candidates for two reasons:

1. They’re potentially able to target RAS(ON)

2. They can do so for a cancer type, PDAC, that has very few promising treatments, is often metastatic at the time of diagnosis, and comes with a RAS mutation an overwhelming amount of the time. Revolution’s decision to focus on its G12D inhibitor over its G12C candidate is smart: G12D mutations are present in ~40% of PDAC cancers.7

Said differently, there’s a very good reason why the company received a National Priority Voucher for Daraxonrasib. The uniquely fascinating part of Daraxonrasib, specifically, is that its side-effect profile wasn’t intolerably severe. This is the case even though it’s inhibiting mutant and wild-type RAS proteins. Dr. Kenneth Olive at the Olive Lab has a theory for why this is the case:

“First, RMC-7977 does not inhibit RAS continuously in tissues. Over the course of 24 hours after a dose, the RAS pathway is fully inhibited within a few hours, but it is back to normal after 12-24 hours. We think this allows normal tissues to recover and get enough of the normal RAS function to continue to be healthy. At the same time, we found that tumor cells react violently to the sudden loss of RAS signaling, with many of them dying within a few hours of the first dose (and more dying after each subsequent dose). Tumor cells seem to be addicted to RAS signaling whereas normal healthy cells can get by for a bit without it. This makes pan-RAS inhibition “tumor-selective”8

I think it’s fair to say excitement about Revolution Medicines is baked into the stock price. The company needs a lot to go right beyond treating second-line metastatic PDAC to grow into a 20 billion dollar valuation. 61,000 patients in the US are diagnosed with PDAC every year. 52% of those are diagnosed with metastatic. Of that 52%, ~56% actually get treatment. Of that 56%, ~46% receive second-line treatment. That puts you at 8200 patients every year. If you assume 100% patient penetration (a bit unlikely!), that these patients stay on Daraxonrasib for 8.5 months (the median PFS survival period in the company’s released results), and that Daraxonrasib is priced at ~20k per month (similar to Sotorasib/Adagrasib), that gets you ~1.3 billion USD in peak U.S. sales each year.9 There’s of course potential for Daraxonrasib on the NSCLC side (and on the colon cancer/breast cancer side), but in general that’s a more crowded space. Furthermore, Daraxonrasib does unfortunately seem to work only for a period of time before the cancer progresses again; this isn’t a cure.

That’s not necessarily to say the valuation doesn’t make sense. The 2L PDAC data really has been quite promising so far, and could represent a big step up from the current standard of care. Additionally, there’s plenty more in the late-stage pipeline. If Daraxonrasib is approved for 1L metastatic PDAC (whether as a combo or solo treatment), things start to get really interesting. Moreover, Revolution Medicines really is the company that’s figured out a potential way to disrupt RAS in its active form. I would guess that part of the valuation premium here is down to how innovative the approach is, even compared to other developers of RAS-targeting treatments. A pan-RAS inhibitor is also the archetypal pipeline in a product: RAS is known to be implicated across plenty of different cancers (even if, thankfully, standards of care for other cancers tend to be better than for pancreatic). Finally, Baker Bros is a large shareholder, and historically they have a bit of an eye for promising approaches to cancer! (see Seagen)

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

The company’s business model is simple: sell/rent instruments to academic/research customers, rent them out to any customers developing therapeutics, and then enjoy milestone/royalty payments on top of that lease revenue. MaxCyte counts CRISPR Therapeutics/Vertex as customers, who co-developed Casgevy, the first (and only) FDA-approved therapy leveraging CRISPR.

There are a few plausible reasons MaxCyte has been hit so hard over the past 5 months:

• The Casgevy launch hasn’t been a homerun thus far. To be sure, there’s significant logistical complexity to delivering an ex-vivo gene therapy around the world. But the anemic start has led to questions on what the appetite for these one-time therapies really is.1

• There’s been a shift in sentiment/funding away from ex-vivo cell/gene therapies and towards in-vivo. MaxCyte’s instruments aren’t needed for manufacturing in- vivo therapeutics.

• There’s probably some forced selling from funds that won’t hold stocks with a sub 100mm market cap or sub $1 price.

• There was definteily some forced selling after MaxCyte was dropped from the Nasdaq Biotechnology Index in late December.

These headwinds are on top of a 2025 that was already looking disappointing. Customers cut the number of programs in the pipeline/went under altogether, there was general instrument purchasing hesitancy, and two large customers streamlined manufacturing in such a way that reduced the number of electroporation instruments required.2

As a result, MaxCyte has gone from trading at an incredibly frothy valuation when it listed on the Nasdaq back in 2021 down to trading at less than 0.5x tangible book value.3

This seems like a bit of an overreaction for a company providing an essential instrument for the manufacturing of a therapeutic. In most cases, if there’s no electroporation then there’s no way to get the gene-editing complex (whether CRISPR or otherwise) into the cells non-virally.4 If MaxCyte’s instruments disappeared tomorrow Casgevy’s revenue would drop to zero. Vertex/CRISPR Tx would have to go through a lengthy process of validating a different electroporation instrument, in addition to a rigmarole with regulatory bodies across various countries.

These are MaxCyte’s individual revenue segments:

The ‘Core Revenue’ portion of the business has two attractive characteristics: low churn and a high percentage of recurring-ish revenue (~75% of ‘24 core revenue was licenses or consumables). That’s not to say it has the characteristics of an enterprise software business. Many of MaxCyte’s customers operate with plenty of science risk, and if ex vivo gene-editing approaches don’t take off the business will be in a challenging spot. Having said that, once a clinical company starts using a MaxCyte electroporation instrument, churn is very unlikely. Much like clinical testing companies using Illumina, switching providers creates massive workflow and regulatory headaches.

The SPL portion of revenue is frustratingly lumpy and carries with it a good amount of uncertainty. If a company fails in phase 1 clinical trials and folds its ex-vivo program(s), then the value of the related SPL agreement goes to zero. That said, the revenue that could come from these agreements is 100% gross margin, and a critical part of making the MaxCyte business model work. Management does have a reasonable amount of visibility into how many SPL agreements will be signed every year. The company is often working with these customers in the early research phase of developing a clinical asset, and so has a good sense for which of these customers are preparing to submit IND applications. MaxCyte expects to sign 3-5 such agreements in 2026.

Subscribe now

Casgevy is a decent illustration of how MaxCyte’s business should ideally work. MaxCyte and CRISPR Therapeutics signed their initial agreement in March 2017. As Casgevy progressed through clinical trials there were milestone agreements along the way (in the mid-6-figure to 7-figure range). Management is cagey about breaking out SPL revenue by customer, but the 8.5mm in SPL revenue back in Q423 was likely all related to FDA approval of Casgevy. Going forward, MaxCyte will collect royalty payments that are ~1% of Casgevy sales, on top of instrument leasing/consumables revenue.

Here are the 5 programs MaxCyte expects to enter pivotal studies in the next 6 – 18 months:

There are a few things worth flagging here. Most notably, both Caribou and Imugene need to raise additional capital to carry out their pivotal trials. That’s extra challenging when pharma companies and investors are more excited about in-vivo than ex-vivo. I think the undisclosed customer program in Phase 1/2 trials is Beam’s Risto-Cel candidate for severe sickle-cell disease. In some ways this isn’t the most attractive asset for MaxCyte to have in its portfolio. Risto-cel would be a direct competitor to CRISPR/Vertex’s Casgevy, and so doesn’t really grow the pie of potential royalty sales.5 That’s true, but the other way of looking at it is MaxCyte gets twice the milestone revenues for treating the same indication.

MaxCyte is trying to diversify beyond just electroporation instruments/consumables, as seen with the company’s acquisition of SeQure Dx last year for $4.5mm in cash. SeQure Dx provides off-target analysis services for in-vivo and ex-vivo cell/gene therapies; it’s an early-stage business, expected to bring in ~2mm in revenue for 2025. At least part of the goal here is to shift the business away from existing only as an index on the growth of ex-vivo cell/gene editing therapies. I don’t have all that substantive of an analysis to offer on this acquisition, but I would note that the acquiree’s services were used in the development of the N of 1 CRISPR therapy for baby KJ that was (rightfully!) so heavily publicized. This also means the FDA evaluated the off-target analysis done by SeQure Dx, and was comfortable enough with the results to give the go ahead. Biopharma companies tend to like using tools that the FDA is already familiar with, and probably especially so in a more nascent field where there’s already a good number of unknown unknowns.

It’s certainly true that there’s been a sentiment shift away from ex-vivo approaches towards in-vivo. However, companies like CRISPR and Beam have been quite transparent that in-vivo has always been where they’ve hoped to eventually go. That hasn’t stopped them from putting capital behind ex-vivo approaches, and remaining excited about these approaches at the most recent JPM Healthcare Conference:

“We’re looking at 4 to 6 months for the entire journey to potentially cure yourself of sickle cell disease. So clearly, we believe this would be very attractive and bode well for Ristacel’s uptake within the market. But in fact, we think we can actually grow the market because you can use an existing amount of clinical infrastructure and potentially with numbers like these, treat more patients. So we’re very excited about the potential of the program and its differentiation” – Beam CEO John Evans

“In oncology, here are the baseline characteristics, very severe patients, all high risk, several primary refractory, all with SPDs above 2,000 millimeter square. And in these patients that are very sick, that -- many of them have actually gone through T-cell engager therapies and not responded, one even went through an auto CAR T before and then they were treated with zugo-cel and they had complete responses. This is remarkable. I don’t think there’s any other allogeneic therapy out there that has shown responses after auto CAR T.” - CRISPR CEO Samarth Kulkarni

Not exactly the nail in the coffin for ex-vivo! On the big pharma side, there have been a lot of recent acquisitions on the in-vivo side of things (AbbVie bought Capstan, Gilead bought Interius, and BMS bought Orbital). Eli Lilly bought Orna a bit over a week ago, but also recently announced a collaboration with CRISPR Therapeutics evaluating zugo-cel with pirtobrutinib in aggressive B-cell lymphomas. Again, not exactly a nail in the coffin.

None of this is to say that MaxCyte hasn’t run into problems recently, and few people would argue the company’s ‘21 valuation was justified. Moreover, the business isn’t profitable, which makes current headwinds harder. I gave a rough outline of the MaxCyte/CRISPR Therapeutics agreement earlier, which illustrates how the business model works and highlights its challenges. A 1%-ish royalty agreement means the therapeutic has to perform well to be of real value to MaxCyte, and that a lot of the company’s terminal value is wrapped up in revenue far out in the future. Consequently, the slowdown/culling of early-stage ex-vivo clinical programs has implications for the business beyond just some foregone instrument revenue.

I don’t dispute any of this, but I’m not sure the appropriate response is for the company to be valued at <0.5x tangible book. Management reduced headcount and expects cash burn of 10-15mm for next year. Cash/cash equivalents/short-term investments sat at 105.7mm as of Q325, and the company has no debt, so is unlikely to run out of money/default in the short-term. Its business model involves a lot of selling consumables and licensing instruments. In other words, high-margin activities. These activities bring with them the potential for milestone/royalty payments, which are even better than high margin. Very importantly, their customer base includes companies that are real trail blazers in the cell/gene therapy space. Over the past year, ~97% of insider buying/selling has been buying. The executive team is well-paid, but over 70% of the CEO’s comp is in the form of stock/options, which is presumably fairly motivating.

There is, of course, plenty I could be wrong about. There’s always science risk, and MaxCyte having a broad portfolio of customers doesn’t mean it’s not possible that Casgevy’s the only program that pans out. Alternatively, maybe a few do pan out, but don’t hit enough of a commercial ramp to be of real value to the business. Maybe many pan out but then are beaten out quickly by emerging therapeutics with better efficacy. There are no guarantees with biotech tools suppliers!

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

The general structure of a therapeutic radiopharmaceutical is well illustrated in RayzeBio’s S1 filing:1

The linker and chelator are critical pieces of radiopharmaceuticals, but aren’t a differentiated part of what Aktis is doing. Linkers keep the radionuclide (and, for typical ADCs, whatever the cytotoxic drug is) attached to the targeting agent until it is close to or inside the malignant cells. In the context of radiopharmaceuticals, chelators are used to prevent the radionuclide from spilling out of the therapeutic construct and affecting healthy tissue.

Where Aktis does start to differ from other companies is in its choice of radionuclide. The company is working on radiopharmaceuticals that use Actinium-225, an alpha-emitting isotope. This isn’t necessarily a unique approach (there are a number of companies trying this in clinical trials), but it’s not the one taken by FDA-approved Lutathera or Pluvicto.2 Both of these use Lutetium-177, a beta-emitting isotope. There are a few differences between alpha and beta particles:

• Beta-particles are made up of an electron; alpha-particles are two protons and two neutrons (and so look quite like a Helium ion). Consequently, alpha particles are much heavier.

• Alpha particles cause double-stranded DNA breaks, beta particles cause single-stranded breaks. This could be why Lutathera and Pluvicto slow cancer progression but do not cure it. Single-stranded breaks are easier for cells to fix than double-stranded ones.

• In layman’s terms, alpha particles have a more limited ‘blast radius’ compared to beta particles. The damage they cause is more severe but less widespread.

• I won’t get into it here, but alpha particles remain effective in hypoxic tumor environments, an area where beta-emitters/traditional radiation treatments struggle. This isn’t really relevant to the purpose of this piece.

Historically, it was thought that alpha particles’ more limited radius meant they would be a fit for treating very small groups of cancerous cells but wouldn’t exert much impact on larger tumor sites. It’s turned out that these particles actually can significantly damage large tumors (see here and here), but it’s not entirely clear why.

Excitement about these alpha-emitting isotopes, and particularly Ac-225, took off after a 2016 paper detailed their effectiveness in treating two patients with metastatic castration resistant prostate cancer. These two patients either didn’t respond to beta-emitters or weren’t a good fit for them, and both responded remarkably well to treatment with peptide-conjugated Ac-225. These peptide targets aimed at PSMA, much like Pluvicto, and caused marked drops in PSMA positive lesions.

Subscribe now

Regrettably, Ac-225’s promise was limited by severely constrained supply and concern about its radioactive decay chains. Ac-225 doesn’t really show up in nature so has to be created synthetically, which is quite a complicated process! Until recently it could really only be made in any notable-ish quantities at one location. The issue with Ac-225’s radioactive decay chain requires some explaining of radionuclides in general:

An element is defined by its atomic number, or the number of protons in its nucleus. This is the number right above an element’s name in the periodic table (Carbon has 6 protons, Calcium has 20, Lutetium has 71, Actinium has 89). Isotopes of an element all have the same number of protons, but a varying number of neutrons. When this ratio of neutrons to protons is unfavorable (the threshold here depends on the mass of the element), the isotope becomes unstable. Unstable isotopes don’t enjoy existing in that state, and so try to head towards stability. The process of radionuclides searching for stability is termed radioactive decay, and it is here that radioisotopes become useful for therapeutic purposes. During radioactive decay there’s an emission of energy (in our case an alpha or a beta particle), and, as a result of that emission of energy, the radionuclide becomes (most of the time) a different radionuclide/stable nuclide.3 It’s this emission of energy that can disrupt cancerous cells.

The nuclide that a radioisotope decays to is termed a daughter nuclide, and unfortunately these can cause problems. In the case of alpha-emitting particles, daughter nuclides ‘eject’ with so much energy that they can break out of the chelator and cause damage to healthy tissue.4 For Ac-225 specifically one of its daughter nuclides is Bi-213, which causes toxicity in the kidneys.

This daughter nuclide problem increases the importance of Aktis’ delivery platform. Antibodies come with two relevant downsides when it comes to alpha-emitting radioisotopes: they have a much longer half-life than peptides (think days to weeks rather than hours) and exhibit poor tumor penetration. The longer half-life is an issue because it means the antibody-conjugate is circulating through the healthy parts of a patient for an extended period of time. In other words, a patient is exposed to toxicity with no clinical benefit. For alpha-emitting particles this extended half-life becomes a double issue, as it gives daughter nuclides, particularly Bi-213, plenty of time to cause trouble for the kidneys. In the case of beta-emitting isotopes, antibodies’ poor tumor penetration theoretically matters less; with alpha-emitting particles, however, it means that they’re really not able to reach much of the tumor.

Miniproteins solve these issues. Their half-life is similar to that of peptides, and their size means they’re able to reach a lot of the tumor. Of course, the miniprotein doesn’t get rid of the fact that Ac-225 has problematic daughter nuclides. With the miniprotein approach, however, the radionuclide payload is rapidly delivered to cancerous cells. Consequently, radioactive decay predominantly occurs within these cancerous cells rather than out in the bloodstream, and one shouldn’t have to worry about systemic toxicity.

It’s worth noting here the uniqueness of Aktis’ approach. I mentioned above the case of two patients with metastatic prostate cancer. In that instance, a peptidomimetic (small molecule designed to mimic a peptide) was used to target the PSMA protein, and so half-life/daughter nuclides are less relevant considerations. Peptides/peptide-mimics have short half-lives, effectively get into the tumor, and get in fast enough that one doesn’t have to worry as much about daughter nuclides. The real constraint, then, for alpha-targeting therapies using a linear peptide/small-molecule conjugate has been around securing enough Ac-225 supply. Again, this is far from an easy problem to solve!5 It is, however, one problem less than what Aktis is trying to figure out. The company not only has to solve the supply bottleneck, but also has to design targeting molecules that aren’t antibodies but bind like antibodies. Without this, alpha-emitters remain out of reach for targeting proteins that are highly expressed on cancerous cells, but have a flattish surface rather than a clearly defined binding site.6

The final point worth making on Aktis is the amount the company is spending: typically companies don’t say they’ll burn 140-150mm advancing an ongoing Phase 1b trial! Typically companies also don’t run a Phase 1b trial with a target enrollment of 150 participants. For reference, RayzeBio, another alpha-emitting radiopharmaceutical player, enrolled 17 patients in a Phase 1b trial for its lead candidate and probably spent a tenth of what Aktis is planning to.

This isn’t an apples-to-apples comparison for a few reasons, but most notably because Aktis believes its lead candidate has broad applicability, and so is running its Phase 1b trial across multiple tumor types:

“Our lead product candidate, [225Ac]Ac-AKY-1189, was generated using our miniprotein radioconjugate platform and is designed to deliver 225Ac, a highly potent alpha-emitting radioisotope, to Nectin-4 expressing tumors. Nectin-4 is a cell surface protein found on a wide variety of tumors and has very limited expression in normal adult tissues. Nectin-4 is also the target of enfortumab vedotin, or Padcev, an approved ADC for the treatment of locally advanced and metastatic UC……..We believe AKY-1189, when conjugated to a radioisotope, has the potential to treat UC as well as other Nectin-4 expressing tumors”7

The choice to target patients with urothelial cancer makes sense. Padcev, which is mentioned frequently in Aktis’ S1 filing, was the first antibody-drug conjugate approved for the treatment of urothelial cancer. Padcev similarly targets the Nectin-4 protein, and has estimated peak annual sales of 7 billion USD (depending on who you ask!) There’s reason to be a tad skeptical, however, of AKY-1189’s applicability in other Nectin-4 expressing tumor types. The other Nectin-4 expressing cancer that Aktis mentions by name is triple-negative breast cancer (TNBC). While cancerous cells in TNBC often do express the Nectin-4 protein (this paper estimates Nectin-4 is expressed in 62% of TNBC cases), Padcev’s Phase 2 results for TNBC were on the disappointing end. Moreover, there are already approved ADCs for TNBC, most notably Trodelvy. Importantly, Trodelvy targets the Trop-2 protein, which is expressed in over 90% of TNBC cases. Quite a bit better than 62%.

The one very brief note I’d make on valuation is that it doesn’t necessarily make sense to evaluate these radiopharmaceutical companies based on their current clinical pipelines. Their supply chain/manufacturing capabilities very much come into play, which is something Eli Lilly and Bristol Myers Squibb both made clear in their Point Biopharma/RayzeBio acquisitions. This funnily enough means that one risk to valuation is that the alpha emitting isotope supply problem is solved a little too well, such that big pharma sees much less value in expertise/capabilities there.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

The recent story of cancer treatment has really been around improving the ability to precisely target cancerous cells over healthy cells. This isn’t what traditional chemotherapy does, as explained by Endocyte pre its acquisition by Novartis:1

“Traditional cytotoxic cancer chemotherapies kill rapidly dividing cancer and normal cells in an indiscriminate manner, leading to significant toxicity in patients. The need for patients to recover from this toxicity can limit the ability to deliver effectively-dosed cancer therapy. In addition, cancer therapies for a given tumor type are generally selected based on observations of efficacy and toxicity in that patient population and not, in most cases, based on an understanding of the differences between tumors on a molecular level.”2

Antibody-drug conjugates (ADCs) try to improve upon both these limitations. Take Padcev, used to treat a subset of patients with cancers of the bladder or urinary tract. The monoclonal antibody portion of the drug binds to Nectin-4, a surface protein overexpressed in a variety of cancers. Once this binding occurs the ADC is then brought within the cell (via receptor mediated endocytosis), and the cytotoxic payload MMAE (monomethyl auristatin E) gets to work preventing further cancerous cell division. Unlike chemotherapy, MMAE is too toxic to be delivered to patients ‘naked.’ Its conjugation with an antibody enables it to very precisely target the cancerous cells, rather than just any cells that are dividing quickly.

Therapeutic radiopharmaceuticals are similar in principle to ADCs. They’re specifically targeted, and contain a payload that in most cases couldn’t be delivered to patients absent some type of conjugate.3 In the case of radiopharmaceuticals, as the name suggests, a radionuclide is used to kill the cancerous cells. There are two FDA approved targeted therapeutic radiopharmaceuticals that have generated significant investor/pharma excitement: Lutathera and Pluvicto.4 Both are owned by Novartis, both treat types of cancer, both use a peptide or peptidomimetic-conjugate, and both are expected to have peak annual sales exceeding a billion (exceeding 5 billion in the case of Pluvicto).

Aktis Oncology, which went public earlier this month (with an 100mm anchor investment from Eli Lilly), is another company playing in the targeted radiopharmaceutical space. The S1 spends a lot of time emphasizing the benefits of its miniprotein radioconjugate platform, as well as describing the limitations of using either peptides or antibodies to direct radionuclide payloads.

In an ideal world, antibodies probably wouldn’t be the conjugate of choice for guiding cancer treatment. They’re expensive/time-consuming to produce (which affects production of the final product and the ease of validating initial targets during drug discovery), challenging to modify during the drug development process, can cause an immune response in patients, and exhibit poor tissue penetration in solid tumors.

Compared to antibodies, peptides are cheaper/faster to produce, easier to modify, don’t cause the same immune response, and exhibit much better solid tumor tissue penetration (all else equal, a smaller amino acid chain is going to have an easier time penetrating tissue than something much larger). Unfortunately, in the context of treating solid-tumor cancers peptides currently come with some downsides.

Subscribe now

Peptides (and many small molecules!) work well when there’s a clearly defined binding site to interact with. This point is well demonstrated by Pluvicto and Lutathera. Pluvicto targets the prostate-specific membrane antigen (PSMA), a protein overexpressed in those with prostate cancer. PSMA is also an enzyme, so as expected has a well defined active site. Lutathera targets the somatostatin receptor 2 (SSTR2), which is both overexpressed in certain gastroenteropancreatic neuroendocrine tumors and exists as a receptor for an endogenous peptide.

Regrettably, there are plenty of proteins overexpressed in cancer that don’t have such perfectly defined binding sites. It is here that peptides (and small molecules!) struggle. In these cases, affecting cancer growth requires mimicking protein-protein interactions (PPIs).5 Protein-protein interactions are quite different from how an enzyme/substrate or peptide/receptor work together. In high school biology, one is taught to think about an enzyme and its substrate fitting together like a lock and key. Such an analogy doesn’t work when thinking about PPIs. Crudely, one can think of the difference in indoor climbing/bouldering terms. At the beginner level, holds can be so thick and made for hands that a climber can keep himself on the climbing wall with only one point of contact. Things change, however, once one moves onto more difficult routes. At that stage, the holds become so small/awkward to hold onto that multiple points of contact (and keeping your body tight into the wall rather than hanging off leisurely) and required to succeed.6 So it goes for enzyme/substrates vs PPIs. The enzyme active site is designed to make it easy for the substrate to hang on. PPIs, however, are like the harder climbing route. The surfaces of these proteins are largely flat, and held together primarily by weak hydrophobic interactions, requiring many more points of protein-protein contact.

This presents a few challenges for the peptide approach:

1) When not bound to the target protein, peptides exist in a linear, flexible state. Binding to the protein requires that the peptide take on a well-defined 3D structure, but that change from less-ordered to more-ordered is not entropically favorable! Consequently, peptides tend to have quite weak affinity for these protein targets. When there’s a well defined binding pocket for a peptide this entropic penalty can be overcome due to the strong bonds formed, but that doesn’t occur with a mostly flat surface.

2) Because peptides typically exist in that flexible state, this means they can take on different 3D structures to bind with proteins other than the one with therapeutic potential. This increases the risk of off target affects. Put differently, peptides often have poor selectivity.

3) Because peptides exist in that linear state they have a limited surface area. This limits the number of hydrophobic interactions that can occur.

A peptide that can mimic PPIs would ideally have a larger surface area and already be in the conformation required to bind to the protein of interest. The larger surface area allows for greater hydrophobic interactions, and the pre-set conformation means there’s not the same entropic penalty for the peptide binding to the protein.7

It is here where Aktis’ miniprotein platform comes in. The company’s miniproteins are 40-70 amino acids in length, and so while longer than typical peptides are far smaller than typical antibodies. Critically, this slightly longer length means these miniproteins do have a 3D conformation, and so have a larger surface area, better binding affinity, and less potential for off-target interactions. In theory, miniproteins gives Aktis all the aforementioned benefits of peptides, but also overcome their principal limitations. This is why Aktis is developing therapies for Nectin-4 and B7-H3 expressing tumors.8 Historically, these proteins have been a better fit to target with antibodies rather than peptides, and so are an ideal proving ground for the company’s thesis. In its S1 Aktis claimed to have “reproducibly generated sub-nanomolar affinity miniprotein binders to specific tumor targets.”9 This would be a big deal, putting its miniprotein binders on par with Lutathera, and indicating the company has overcome binding affinity issues typically present with peptides and PPIs .

From an investor standpoint, it’s important that the length of these miniproteins mean they’re considered biologics, not small molecules.10 In other words, the patent protection for these therapeutics would be twelve years rather than five. This isn’t the case for Lutathera or Pluvicto, which is why Lantheus Holdings is trying to bring a Lutathera generic to market. This should mean Aktis has an advantage raising additional capital when compared to any peptide-conjugate competitors - the net present value calculation is quite different when your product is protected for an extra seven years!

Correction: In a previous version of this post I spoke about PPIs as though Aktis was trying to affect protein-protein interactions in some way, as a KRAS inhibitor would. Aktis is instead designing miniproteins that can bind to these flattish protein surfaces for the purpose of getting into the cells those proteins are expressed on. The binding affinity problem that Aktis is trying to solve is the same one that developers targeting protein-protein interactions are trying to solve: how do you bind to a protein when it’s a large, flat, surface with no clear binding site.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

An increasing amount of effort, from Alnylam and others, has focused on delivery methods that reach beyond the liver. The appeal of this is straightforward: there are plenty of extra-hepatic diseases that could be treated with nucleic acid approaches; it’s just a question of reaching the relevant tissue types. Capstan is an example of this. The company’s designed an antibody fragment conjugated LNP that accumulates in the spleen rather than in the liver, and binds specifically to CD8+ T-cells.

Beyond the spleen, delivery to muscle cells is another area companies have put plenty of resources toward. Part of the rationale here has to do with existing FDA-approved treatments for DMD. Sarepta has three different FDA approved exon-skipping therapeutics, and yet it’s unclear how beneficial these treatments really are for patients. These exon-skippers do not use an antibody, antibody-fragment, or cell-penetrating peptide (CPP) to enable enhanced muscular delivery, and so it seems reasonable to think that conjugating a PMO with something that improves muscular delivery should enjoy a straightforward path to approval. There’s always science risk, but the regulatory risk is at least somewhat diminished. As a result, most of the companies working on nucleic acid therapies for muscular diseases have some sort of DMD exon-skipper in the works. All of them also have a treatment for DM1, another form of muscular dystrophy that, similar to Huntington’s Disease, is caused by an expanded number of trinucleotide repeats. In the case of DM1, patients have too many CTG repeats in the DMPK gene. This in turn causes a larger than usual DMPK pre-mRNA, which then inhibits various splicing proteins from doing their job. This inhibition of splicing proteins is why the effects of DM1 are so broad: the splicing proteins can’t do their job, which results in splicing errors for a multitude of transcripts, and in turn impacts a multitude of proteins. Said differently, the problem with DM1 isn’t necessarily with the mutated DMPK gene itself, but instead with the cellular machinery that it inhibits.1

DM1’s impact on splicing activity means that, for a therapeutic to work, it actually needs to enter a cell’s nucleus. This is an added layer of complexity. In the case of typical RNAi therapies, such as Alnylam’s, the goal is to affect the mature RNA transcript, and so therapeutics only need to get beyond the cell membrane and enter the cytoplasm. In the case of treatments for DM1 (and exon-skipping approaches for DMD), the goal is to affect the pre-mRNA. Suppressing expression of DMPK mRNA doesn’t help anything: there’d still be the same splicing errors at the pre-mRNA stage.

Subscribe now

The same delivery problem applies to exon-skipping treatments for DMD, although for different reasons. Here the issue isn’t that the mutated dystrophin gene is affecting the splicing of other proteins but that the patient isn’t producing a functional dystrophin protein. Exon-skippers trick the cellular machinery into skipping over a mutated exon as pre-mRNA is processed into mRNA, resulting in (hopefully) at least some functional protein. These DMD and DM1 treatment approaches illustrate the same point. When developing targeted RNA therapies for muscular disorders the problem is actually two-fold: figure out how to effectively reach the muscles and figure out how to enter the nucleus.

Thus far, there have been varied strategies for trying to reach the muscle cells:

Monoclonal antibodies are an obvious choice for two reasons: not only are there many existing FDA approved monoclonal antibodies, there are also plenty of FDA approved antibody-drug conjugates (although these are for treating cancer rather than genetic diseases). Consequently, the safety profile is pretty well understood.2 Historically, the issue with using mAbs to get to muscle cells has been around choosing which receptor to target. The downside of targeting TfR1 is that’s the receptor that transferrin primarily binds to. If transferrin can’t bind to TfR1, then your body doesn’t have a way to transport iron, and you’ll become anemic. Avidity engineered at antibody that shouldn’t interfere with this transferrin binding, but it’s not a total fix and they did still report some anemic patients.

Using an antibody fragment, as Dyne is doing, is also a somewhat obvious choice. While there are not yet FDA approved antibody-fragment conjugates of any kind, there are FDA approved standalone antibody fragment treatments, and plenty of antibody-fragment conjugates in clinical trials. Dyne’s decision is explained well in its annual report:

“We selected a Fab antibody over monoclonal antibodies, or mAbs, due to its potential significant advantages when targeting TfR1 to enable muscle delivery, including enhanced tissue penetration, increased tolerability due to lower protein load and reduced risk of immune system activation due to the lack of the Fc domain, the portion of an antibody that interacts with the immune system, on the Fab. To identify the proprietary Fab we use in our product candidates, we generated and screened proprietary antibodies for selectivity to TfR1 in order to enhance muscle specificity and for binding to TfR1 without interfering with the receptor’s function of transporting iron into cells. Binding to TfR1 may also enable delivery to the CNS.”3

All the above points are true. Any company that isn’t Avidity in this space is keen to point out the downsides of an antibody-based approach, particularly given the potential immune response. The magnitude of a patient’s immune response becomes very relevant in the case that Avidity’s treatment approach is correct in theory but hampered in practice by dose limitations because of the immune system.4 The point on tissue penetration is also interesting. All else equal, smaller proteins will have an easier time penetrating tissue than larger ones, and so it makes sense that an antibody fragment does better in this respect. The point I’d quibble with Dyne on is the benefits of not including an Fc domain in its fragment. While it lowers the risk of immune activation, it also negatively impacts a therapeutic’s half-life.

PepGen and Entrada are taking advantage of cell-penetrating peptides (CPPs), although in different ways. Using peptides as a delivery method has been appealing to researchers for quite some time, but first generation CPPs were found to be highly toxic. The more arginine residues a peptide had the more effectively it was able to penetrate cells, but also the more toxic it was in animal studies. As one would expect, PepGen and Entrada spend a good amount of time explaining how their peptides are different from the doomed first-generation instantiations.

Both companies emphasize their improved endosomal escape capabilities; first generation CPPs demonstrated very poor levels of endosomal escape, which is a bit of a problem. You can’t affect change in the nucleus if you can’t even get into the cytoplasm! More generally, it’s worth noting that peptides can cause kidney problems and hypomagnesemia. Until late last year Sarepta was actually developing a new version of its exon 51 skipping therapeutic that was conjugated to a peptide. The reported mean dystrophin increase was substantially better than its existing treatment (5.17%, so quite close to Dyne’s recent data, versus Sarepta’s existing <1%), but development was discontinued due to prolonged hypomagnesemia and declines in kidney function.

Arrowhead’s avenue to targeting the muscles is different from any of the above, and potentially has broader applicability. The company conjugates siRNA to a ligand that binds to the avb6 integrin receptor, which is mostly found on epithelial cells and has been used by Arrowhead in its pulmonary disease candidates. Historically, this receptor was thought to only be found on epithelial cells, but was then unexpectedly found present in skeletal muscle. Arrowhead also explored using antibodies to target muscles:

“So the reason we went with the integrin targeted approach and the advantages we see one, potential for improved knockdown, which we showed in the previous slide in 2 different species. And then additionally, we think that less drug is always better for the patients and the transferrin receptor targeted approach with an antibody oligo conjugate will require a higher total dose because siRNA is linked to this large monoclonal antibody compared with the integrin peptide targeted approach. For example, if you normalize for the siRNA dose, the amount of siRNA in a 12 mg/kg dose of ARO-DUX4 is equivalent approximately to the amount of siRNA in a 100 mg/kg of a DUX4 antibody oligo conjugate. And we see that this is an advantage, potential efficacy advantage, potential safety advantage and a potential advantage in terms of cost of goods.”5

In other words, much like PepGen and Entrada they emphasize the ability to take dosage higher. The quirk I’d mention with Arrowhead’s strategy is that using the avb6 integrin receptor means there’s ample opportunity for off-target effects – the therapeutic payload is not only ending up in the muscle cells.

When it comes to developing improved exon-skipping therapeutics for DMD, there’s not a real difference in approach between companies other than the delivery mechanism. There may be a difference in the exons being targeted, but the theory is all the same: prevent the RNA machinery from including a faulty exon in the mRNA transcript.

With DM1 treatment approaches, however, there are some interesting differences. As said above, DM1 is caused by too many CTG repeats in the DMPK gene. These repeats negatively affect a cell’s splicing ability, which in turn negatively impacts multiple proteins. In my view, Avidity, Dyne, and Arrowhead use a somewhat blunter strategy to combat DM1 by trying to reduce DMPK RNA expression overall. The less DMPK that’s expressed, the less harmful CTG repeats will be present in the nucleus, and the less a cell’s splicing machinery will be impaired. Entrada and PepGen, however, get more specific. Rather than reducing overall DMPK RNA expression, these companies focus on binding only to the expanded repeats. If PepGen/Entrada can get this to work it’s likely the better approach. The DMPK gene serves a purpose, so ideally you don’t want to be suppressing expression of its healthy, non-deleterious, portions. PepGen/Entrada’s approach isn’t totally dissimilar from what Wave Life Sciences is trying to do with Huntington’s. Rather than silencing both mutant and wild-type HTT, as companies like Alnylam and UniQure try to do, Wave instead tries to target only the mutated allele. In the case of Huntington’s, you still want the healthy HTT protein to be expressed!

It’s also worth mentioning the type of nucleic acid each company uses to affect gene expression. Specifically, it’s quite interesting that Arrowhead and Avidity both decided to use siRNA for their DM1 treatments. Typically, siRNAs are known for acting in the cytoplasm, and so you wouldn’t expect them to exhibit much activity in the nucleus. That said, both companies have reported some level of splicing correction, which indicates the siRNA is entering the nucleus somehow and at enough of a rate to display actual effectiveness. This is an interesting case of in practice being different from in theory, something that I’m sure is quite frustrating to see for anyone reading who’s a UChicago alum.

Below is a summary of where each company is in clinical trials for both DM1 and any DMD exon skippers:

There are a few points to mention. The first is that Avidity is furthest along in becoming a true platform. The company has an additional candidate in Phase 1/2 trials for treating FSHD (Facioscapulohumeral Muscular Dystrophy), and plans to submit a BLA for this candidate in 2H26. Dyne is close behind Avidity, but doesn’t yet have its FSHD candidate in clinical trials. At first glance Avidity’s exon-skipping data looks much better than Dyne’s, but that’s because exon 44 patients naturally produce more dystrophin. Both product candidates are exciting: if approved, Avidity’s exon 44 candidate will be the first to market for DMD patients with a confirmed exon 44 mutation. Ideally Dyne would be showing a dystrophin increase of greater than 10%, but 5.46% is significantly better than Sarepta’s EXONDYS 51. After halting progress on its exon 51 candidate given disappointing results, PepGen is now a single product company, but a very interesting one. For its DM1 candidate, they reported a mean splicing correction of 53.7% in phase 1 trials, a number that’s far above what either Avidity or Dyne have reported (16% and 27%, respectively). Critically for peptide-conjugated therapeutics, kidney issues were mild.6 Entrada’s work on cyclic peptides is fascinating, and management is eager to point out the theoretical advantages its approach could have against competitors. Its phase 1 exon 44 data didn’t surface any kidney issues, which is encouraging. That said, actual clinical data on efficacy is quite important here!

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

An area where RNA hasn’t met much success yet is powering one-time treatments for genetic diseases. At first glance this makes sense: any therapeutic delivered in the form of RNA is inevitably going to be degraded by the body’s cellular machinery. RNA is inherently transient; it’s not supposed to exist forever. There’s a lot of work that goes into prolonging RNA therapeutics’ persistence in the human body, but the goal is to lengthen its existence, not make it last indefinitely.

Put differently, the goal of current RNA-based therapies is different from the goal of one-time gene therapies like Zolgensma or Elevidys. In those cases, the goal is to deliver a functional gene to the body, which will remain there permanently, and continuously be transcribed and then translated into a functional protein. In principle, however, gene therapies can be used to deliver more than only genes that will be translated into proteins. Plenty of genes are transcribed into something other than mRNA, and these other types of RNA can be incredibly useful! In theory, it’s possible to deliver a gene to a patient that would permanently produce a type of non-coding RNA. These therapeutics are probably best thought of as both gene therapies and RNA-interference therapies. While they would be delivered as genes, their end state means they’re taking advantage of the same mechanisms as companies like Alnylam or Arrowhead.

Two types of naturally occurring RNA have received attention for their potential or actualized therapeutic capabilities: siRNA and miRNA. Both exist primarily to prevent certain mRNAs from being translated into proteins, both are similar lengths (~20-24 nucleotides) and both make use of cells’ RNAi (or RNA interference) pathway. There are important differences: siRNA is double-stranded, has high specificity, and primarily leads to mRNA degradation. miRNA is single-stranded, has greater potential for imperfect binding (and thus for off-target effects), and primarily inhibits mRNA translation rather than encourages degradation. There’s also a non-naturally occurring approach to modulate mRNA expression, antisense oligonucleotides (or ASOs). ASOs are short, synthetic, single-stranded DNA or RNA strands that don’t take advantage of pre-existing RNAi machinery but do bind to mRNA and either inhibit translation or precipitate mRNA degradation. Tofersen, developed by Ionis and Biogen, is an FDA approved ASO treatment for a subset of ALS patients that initiates degradation of mutant SOD1 mRNAs. Vyondys 53, developed by Sarepta, is an FDA approved ASO treatment for a subset of DMD patients that binds to mRNA and prevents translation taking place.1

Subscribe now

At a high level, the RNAi pathway that miRNA and siRNA are part of, and that therapeutics try to take advantage of, is straightforward enough. In the case of siRNA, the process typically kicks off with a double-stranded RNA (whether from a virus or otherwise) entering the cytoplasm. A dicer enzyme then cleaves that double-stranded RNA into a shorter ~20-24 length-ish fragment, which we term an siRNA. That siRNA then joins a group of proteins known as the RNA-induced silencing complex, or RISC. Next, one of the RNA strands (termed the passenger strand) is degraded, and the siRNA guide strand binds to mRNA and prevents translation. Unlike siRNAs, miRNAs don’t come from an endogenous source, and so are transcribed in a cell’s nucleus. miRNA is initially transcribed into a long, single-stranded, pri-miRNA, which is then cleaved by the Drosha enzyme into pre-miRNA (those are two different things rather than a typo!) This pre-miRNA then exits the nucleus and, like siRNA, is cleaved by the dicer enzyme into miRNA. Again like siRNA, this miRNA then joins the RISC complex, the passenger strand is degraded, and the miRNA goes onto prevent translation.

Thus far, only siRNA-based therapeutics and ASOs have received FDA approval. Alnylam has been a real powerhouse on the siRNA side. Its drug Onpattro, for the treatment of hATTR, was the first FDA-approved drug to leverage siRNA. hATTR is caused by a mutation in the gene that codes for the TTR protein, so Onpattro functions by binding to wild-type and mutant TTR mRNA to prevent subsequent protein expression. The company’s a true RNA platform, and since Onpattro’s approval has brought five more siRNA-based drugs to market. It’s worth noting that Alnylam’s products are delivered as siRNA, rather than as longer double-stranded RNA that first must be cleaved by the dicer. This is the standard approach for siRNAs used therapeutically.

miRNA-based therapeutics have not yet enjoyed the same level of success, which has been driven by a few factors. The first is its greater potential for off-target effects. This is driven by miRNA’s intrinsic tolerance for imperfect binding, as well as occasional loading of the passenger strand, rather than the guide strand, into the RISC. In both cases, this leads to suppression of the wrong transcript. The second reason is that, historically, exogenously delivered miRNAs have ended up boxing out endogenous miRNAs, which in turn really disrupts a cell’s regular functioning. In short, a cell’s RNAi machinery is busy enough working with the exogenous miRNA that it neglects to process its own miRNA. This results in abnormal levels of gene expression, which is a real problem!

UniQure has devoted a lot of effort towards solving these challenges with its miQURE platform. The company received a lot of press recently thanks to its pivotal Phase I/II study data for AMT-130 and the FDA’s flip flop on the drug’s accelerated approval pathway. AMT-130 is a treatment for Huntington’s, a severe neurodegenerative disease caused by too many CAG repeats in exon one of the HTT gene. This in turn leads to nerve cell damage in the brain.2 For the most part, those with Huntington’s will start to develop symptoms at between 30-50, and live ~15-20 years after initial signs of the disease. What begins as small involuntary movements gradually progresses to losing the ability to walk or talk, psychosis, and severe loss of cognitive function, among other things. Late-stage patients require around the clock care, and there is currently no cure to the disease nor a way to slow its progression. AMT-130 contains the genetic material for an miRNA that binds to the HTT transcript, thus preventing protein expression.

UniQure’s basic insight with its miQURE technology was to use an atypical miRNA scaffold, miR-451, that isn’t processed by the Drosha and Dicer enzymes. In other words, this miR-451 doesn’t follow the process that was laid out above, and thus should avoid the risk of overloading the cellular machinery. miR-451 also ‘is produced without the complementary passenger strand present, significantly reducing the rate of off-target effects.’3 Consequently, UniQure’s side-stepped two significant issues with miRNA-based therapeutics. Now, one still has the risk of off-target effects driven by miRNAs’ intrinsically imperfect binding to mRNAs. That requires less innovation to mitigate, and is dependent upon the specific miRNA sequence chosen to target a disease. In the case of AMT-130, UniQure tracked mRNA expression of transcripts similar enough to the mRNA transcript AMT-130 was meant to bind to, and found no marked changes in expression.4 The real innovation here is UniQure’s miQURE technology, rather than the company’s measuring of mRNA expression alterations for similar transcripts.

Now, UniQure’s technology being fascinating doesn’t mean that its AMT-130 candidate will actually work, nor that it will work better than competing approaches:

There are a few key takeaways here: UniQure’s candidate is further along in its clinical trials than competitors, its delivery method is the most complex, and PTC Therapeutics is one to watch in the short-term. UniQure being furthest along does really matter: this is a disease where there’s an enormous unmet need, and so being first to market is an advantage. It’s also a disease where progression is relatively slow, and so having data at 36 months after treatment rather than only after 24 is valuable for determining clinical effect. That said, its complex delivery method limits how quickly UniQure will be able to serve patients. There are ~68 centers in the U.S with the needed ClearPoint instrumentation, and the procedure has to be performed by a a neurosurgeon.5 That’s a contrast from intrathecal delivery, and even more of a contrast from PTC’s pill. However, investor excitement about PTC’s therapeutic has been more muted. The company did report a drop in blood HTT protein levels, but hasn’t shared the magnitude of the change in cUHDRS, other than saying there was one. cUHDRS measures actual Huntington Disease progression, and so while a drop in HTT levels could be encouraging it doesn’t mean a whole lot if there’s not a statistically significant change in cUHDRS.6 UniQure and Alnylam’s focus on targeting exon-1 fragments is worth mention: the CAG repeat expansion that causes Huntington also causes aberrant splicing that results in an exon 1 only transcript. This transcript is then translated into its own shortened protein, which accumulates in patients with the disease. There are researchers who believe you can’t have success slowing Huntington without inhibiting translation of these exon-1 fragments, so that’s something to keep in mind. I find Wave and Vico’s allele-specific approach to be fascinating: the wild-type HTT gene of course serves a purpose, so it’s reasonable to think silencing mutant and wild-type HTT translation could lead to harmful effects. Having said that, Wave Life Sciences had to nix two other allele-specific Huntington candidates targeting those with SNP 1 and SNP 2 mutations, so maybe there’s something about this approach that just doesn’t work.

Of course, how the FDA behaves is incredibly relevant to UniQure’s success, especially in the short-term. When UniQure released its Phase I/II results in late September the company planned on submitting a biologics license application (BLA) in the first quarter of ’26, with a potential US AMT-130 launch later on in the year. This schedule was based on previous discussions with the FDA on what would constitute primary evidence for an accelerated approval. Unfortunately, this timeline was completely thrown into question on Monday, when UniQure announced the FDA has now shifted the goalposts for an accelerated approval pathway. I don’t have anything to add here, other than it seems like an odd move for the FDA to make with a severe disease that doesn’t have any available treatments.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

AbbVie made headlines recently with its ~2 billion USD all cash acquisition of Capstan Therapeutics. Capstan’s lead product, CPTX2309, is a CAR-T treatment for B-cell mediated autoimmune diseases that’s in phase 1 trials. What makes Capstan interesting is not that it’s developing an autoimmune disease treatment, nor that it’s developing one that leverages a CAR-T approach, nor that its target antigen is CD19. Rather, what makes the company so fascinating is its potential therapy:

1. Is in vivo, and so doesn’t require a specialized treatment center, an initial regimen of chemotherapy, harvesting a patient’s cells, a weeks to months period of editing those cells, or an eye-popping sticker price. Critically, it also doesn’t require donor cells, and so unlike other allogeneic treatments in clinical trials one doesn’t have to worry about mitigating graft vs host disease.

2. Delivers the chimeric antigen receptor non-virally. mRNA coding for the CAR is contained inside an LNP.

3. Is part of Capstan’s broader tLNP platform. This platform is exciting because it’s an instance of using LNPs to reach something other than the liver. That requires some innovation!

There’s more to be said on the above, starting with (3). Currently, LNPs are limited in that they really can only target the liver.1 Capstan has found a way around this impediment: first, they were able to design an LNP that primarily accumulates in the spleen. The spleen makes white blood cells, so accumulation here is very helpful when trying to affect the immune system! Secondly, Capstan conjugated an antibody fragment to the LNP. This antibody fragment enables the lipid nanoparticle to bind precisely to a specific cell type. In the case of Capstan’s lead product, the fragment facilities LNP binding to CD8+ T-cells. The mRNA is then delivered into these cells, and the chimeric antigen receptor is successfully expressed.2

Capstan’s decision to target CD8+ T-cells differs from typical CAR-T approaches which target both CD8+ cells and CD4+ cells. The rationale for this is twofold: CD4+ cells are often overexpressed in patients with lupus, so you may not want to stimulate additional activity here. Moreover, CD4+ cells are responsible for driving cytokine release syndrome (CRS), an excessive immune response that’s a well known side-effect of CAR-T treatments.3

Subscribe now

That mitigation of side-effects is key, and leads to a related one on why point (2), Capstan’s non-viral approach, is so important. Treating autoimmune disease is a very different thing from treating refractory/relapsed cancers. If patients with serious cancer don’t get treatment they die; that is thankfully not the case for those with autoimmune disease. To be sure, autoimmune disease can wreak havoc. Organ damage, especially to the kidneys, is a real risk for those with lupus, and it’s estimated that 10-15% of those with this disease will die prematurely.4 That’s serious, but still quite a different prognosis than for any cancer patient eligible for CAR-T treatment. There’s also a reason that this piece by Capstan researchers mentions the total number of people affected by B-cell mediated autoimmune disease: the goal is to develop a treatment for the whole patient population, not only those with the most severe instantiations of it. For the majority of those with a B-cell mediated autoimmune disease, it wouldn’t make sense to consider CAR-T treatment when the side-effects can be life-threatening. Taking steps to minimize the odds of cytokine release syndrome is a meaningful stride towards widening the potential addressable patient population. The second, and perhaps more important, stride that Capstan has taken here comes back to its decision to deliver the CAR non-virally.

I’ve covered the potential benefits of non-viral CAR delivery previously. The key concern for viral methods centers around the semi-random integration of the lentivirus/retrovirus into the T-cell DNA. This comes with the risk of insertional mutagenesis, whereby the viral DNA integration causes a mutation within the T-cells. These mutations could theoretically cause a second cancer.

I say theoretically, but there have been cases of T-cell lymphoma in patients who received CAR-T treatments, and the FDA now requires that treatment providers list secondary malignancies as a potential side-effect. It’s so far been very difficult to discern precisely what’s causing these secondary cancers. Their incidence is thankfully low, and for a patient with already serious cancer the risk/reward makes sense. But again, that calculation changes when it’s a patient with even a severe case of autoimmune disease. Delivering the CAR via LNP, as Capstan does, avoids that insertional mutagenesis risk altogether. Furthermore, it meaningfully differentiates the company from other players in the in-vivo CAR-T space. Interius (acquired by Gilead), EsoBiotec (acquired by AstraZeneca), Umoja (recently raised 300mm), Novartis, and Kelonia are all working on in-vivo CAR-T treatments, but all of them remain constrained by lentiviral delivery.

There’s a natural question here as to whether Capstan’s approach threatens existing CAR-T cancer treatments. In the near-term I think this answer is no. For those with cancer, the persistence of CAR-T cells in a patient post-treatment is seen as critical to keeping him/her successfully in remission. For those with autoimmune disease that’s not the case, which is exactly why the CAR can be delivered in the form of mRNA. Transient expression is good enough here! The obvious solution to modifying Capstan’s approach for cancer would be repeat dosing, but then one has the added hurdle of worrying about the immune response to repeated mRNA intravenous treatments.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.

MaxCyte ($MXCT), a supplier of electroporation instruments/consumables that enable ex-vivo non-viral cell therapies, would also fit into the biotech picks and shovels category. As with quite a few other small cap life science tools businesses, its stock price has cratered 90% since going public in 2021. Even better, if you’d bought at the beginning of 2025 you’d have lost over 60% of your money.1

Ex-vivo non-viral cell therapies refer to treatments where:

• Cells are treated outside the body

• Whatever alters those cells is delivered using a method that doesn’t involve containing that alternation mechanism within a virus

The most immediate applications of this approach are for treating genetic diseases. Casgevy, a treatment for sickle-cell/beta thalassemia, takes a patient’s blood stem cells, modifies those stem cells using CRISPR-Cas9 to knockout the BCL11A gene, and then reinfuses blood cells back into the patient. This is the only FDA approved CRISPR therapy, and MaxCyte powers the electroporation process. Less immediately, companies are working on ex-vivo non-viral treatments for certain blood cancers, B-cell mediated autoimmune diseases, and solid tumors. Of these three, blood cancers have FDA approved, ex-vivo, viral treatments as a template.

FDA approved ex-vivo cell treatments for blood cancers (CAR-T therapies) share two commonalities:

• They use a patient’s own T-cells.

• They deliver the CAR (chimeric antigen receptor) to the patient’s T-cells using a virus (and so don’t use electroporation technology).

I’ve written previously about the challenges with existing CAR-T treatments and the promise of allogeneic (meaning donor-derived, rather than patient derived) approaches. Put briefly, autologous (patient-derived) treatments are costly, take weeks to manufacture, entail further weakening a cancer patient’s immune system, and require that a patient’s T-cells are sufficiently healthy. The last part isn’t a given when one’s already undergone rounds of heavy cancer treatment.

Allogeneic CAR-T treatments instead use donor cells, but this has its own set of challenges. Again put briefly, the donor T-cells are unhappy to be in a foreign environment, and so attack the patient (this dynamic is termed graft vs host disease); the patient’s T-cells and NK cells are unhappy that foreign T-cells have entered the body, and so attack the donor T-cells.23

The hope is that we can leverage gene-editing to dampen these host and donor immune responses. This dampening happens in two ways. The first is by using a double-stranded break to knock-out one or many genes in the donor T-cells, much as Casgevy does with autologous blood cells. The second has to do with where the chimeric antigen receptor DNA is actually inserted within the donor T-cell.

Subscribe now

As said above, among FDA approved CAR-T therapies the CAR is delivered to T-cells via a modified virus. The virus then integrates into the T-cell DNA, and the chimeric antigen receptor is successfully expressed. Importantly, delivering the CAR inside a virus means we have very little control over where in the T-cell DNA the virus and CAR integrate. In the case of allogeneic CAR-T therapies, there’s a preference for non-viral delivery. Moreover, gene-editing tools mean we can actually be picky about where the CAR integrates.4 The CAR is typically inserted into the T-cells’ TRAC locus, knocking out TCR expression and thus limiting graft-vs-host disease.

MaxCyte’s electroporation instruments/consumables can enable this non-viral delivery. There’s a lot of complexity here, but crudely speaking these instruments deliver an electric pulse to the cell, temporarily opening up the cell membrane and allowing the gene-editing complex to enter. Using a viral vector to deliver the gene-editing package isn’t a good option, as that requires delivering the complex in the form of DNA and so increases the risk of off-target effects.5 Consequently, developers prefer to deliver the complex as either mRNA or as an already formed protein. Both of these will be cleared out relatively quickly by the cell, thus diminishing the risk of off-target effects.6

MaxCyte’s business model is straightforward to understand. The company provides instruments as well as the consumables required to run these instruments. MaxCyte sells instruments to academic customers, but rents them out to any customers developing a therapeutic. With these customers it also negotiates a Strategic Platform License (SPL) agreement; these agreements include milestone payments and, most of the time, a royalty agreement should a therapeutic be commercialized. SPLs are a vital part of the business and represent a real strength when compared to a company like Maravai. The challenge with existing as a supplier for a one-time sickle-cell treatment is that the addressable patient population isn’t that large. This doesn’t matter as much when you’re Vertex and are collecting a hefty sum per dose, but it does matter if you’re a small Vertex supplier like MaxCyte. Royalty agreements ensure that MaxCyte can keep its own business economical even when powering one-time treatments for genetic diseases and refractory cancers. As one would expect, instrument and consumables usage grows as clinical programs progress from early-stage to (hopefully) commercial availability.

MaxCyte’s focus on leasing out instruments, in combination with its SPL agreements, explains why margins have historically averaged over 80%. While margins are strong, revenue has not grown as expected. After peaking at ~44mm in 2022, the top-end revenue estimates for this year (32.5mm) are actually modestly below where revenue was in 2021 (33.9m). This weakness has been driven by what one would expect given the biotech market generally: lower than expected consumables sales, longer than typical instrument sales cycles, and rationalization of customer clinical programs. Interestingly, the installed instrument base has still grown nicely since its Nasdaq IPO, from over 500 in 2021 to 814 as of Q22025. The challenge has really been around weaker customer usage of these instruments.

MaxCyte could be a compelling buy here, for the following reasons:

• It’s a supplier for a lot of the flagship gene editing companies, including CRISPR Therapeutics, Beam Therapeutics, and Sana Biotechnology. As mentioned above, it’s a supplier for Casgevy, the only FDA approved CRISPR therapy.

• It’s not a perfect index on the success of allogeneic CAR-T therapies taking off, but it’s a pretty good one. Its electroporation devices enable cell therapies leveraging a wide variety of gene-editing technologies (TALEN, ARCUS, etc), not only CRISPR.

• Its price to book ratio is ~0.75. The company does not carry any debt.

There are of course risks:

• Allogeneic CAR-T treatments could end up not living up to the hype. Maybe it’s just too difficult to get donor T-cell persistence in a patient. If these treatments don’t work, it’s very unlikely that any B-cell mediated autoimmune disease or solid tumor approaches would still hold any kind of promise.

• Allogeneic CAR-T treatments could end up working, but only those developed by a company like Intellia, which explicitly doesn’t use electroporation to deliver the gene-editing complex. Again, this would in turn have implications for autoimmune disease/solid tumor clinical programs.

• In vivo gene-editing therapies end up taking off but ex-vivo therapies do not. MaxCyte’s electroporation technology isn’t needed for in vivo treatments.

• Competitors in the space, such as Lonza and Thermo Fisher Scientific, start taking a lot of market share.

• The end market could continue to deteriorate as it has since 2022.

These are all real risks, and ultimately if allogeneic CAR-T treatments don’t succeed in clinical trials then MaxCyte probably isn’t going to be worth much. Intellia’s decision to not use electroporation for its CAR-T treatments is a fascinating one and may lead to good outcomes, but Casgevy’s efficacy at least indicates that using electroporation is unlikely to render a treatment unsafe or ineffective. Additionally, Intellia is in the minority with its approach; MaxCyte is a better way to bet broadly on the success of the allogeneic space. It would be a little strange if in vivo gene-editing therapies took off while ex vivo didn’t, but it’s not a totally implausible scenario. AbbVie recently bought Capstan Therapeutics, whose lead asset is an in vivo CAR-T therapy for B-cell mediated autoimmune disease. A successful in vivo approach would pose a real threat to alternative ex-vivo approaches as well as ex-vivo CAR-T treatments for blood cancer. In my view it’s difficult to be overly concerned about competitors: the quality of customer logos make it clear MaxCyte offers a compelling product, and unlike Thermo Fisher/Lonza the company’s success is almost totally dependent on winning in this space. A lot of the thesis here does come down to valuation: MaxCyte’s powering key players in an emerging space where investor hype got out over its skis too early. Now things have gone in the entirely opposite direction, and I think MaxCyte is a way to bet on things reversing.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence. There is very real science risk here!

Disclosure: I hold a small position in MaxCyte.

These life science business quirks are precisely why Maravai is so interesting. The business was started in 2014 by two life sciences executives and private equity firm GTCR. The goal was to be a kind of baby Danaher, acquiring companies in the diagnostics/life sciences space. It operates two business segments: Nucleic Acid Production (NAP) and Biologics Safety Testing (BST). The NAP business began with the acquisition of TriLink in 2016, which is best known as a provider of mRNA capping solutions but also serves the mRNA market more broadly. The BST business began with the acquisition of Cygnus Technologies in 2016. Growth for this segment has been anemic since 2020 (3.4% CAGR), but it possesses some excellent characteristics. Its products ensure there are no process-related impurities in drug bioproduction; impurities can cause an immune response in patients and reduce therapeutic effectiveness, so are fairly important to spot! Once a company picks an impurities detection provider it makes no sense to churn off:

“Because of the extensive validation required for these products, these components are frequently purchased for the life of our customers’ products and we believe they are unlikely to be substituted. In addition, our analytical tools are used in the design and development of manufacturing processes and often will be used throughout the life cycle of our customers’ manufactured products. As a result, our customer relationships may span many years.”1

Maravai is the dominant player in this safety testing space. It has a broader product portfolio than competitors and, as of the 2024 annual report, was used in all 24 FDA approved CAR-T Cell and Gene Therapies. Unfortunately, as MBI has pointed out, product usage is inversely correlated to a drug candidate’s position in clinical trials. That is to say, revenue ramps down rather than up once a product is commercially available. A further downside to this segment is the majority of sales (58%) go through distributors, so management has less end market insight than would be ideal and can’t cross-sell products from the NAP division. The perk of this outsourcing is margins look much better than they otherwise would.

The focus of this piece will be on TriLink’s CleanCap product, which is really what Maravai is known for. mRNA capping accounted for 72% of Nucleic Acid Production revenue in 2024, and was used by Pfizer/BioNTech in the development and production of their mRNA COVID-19 vaccine. The utility of the capping product is straightforward: mRNA is an inherently unstable molecule, and not the only kind of RNA that’s transcribed in the nucleus. When DNA is transcribed to RNA in regular cells, a 5 prime cap is added to each mRNA molecule. This 5’ cap protects mRNA from degradation by nucleases and aids in translation once the mRNA exits the nucleus and enters the cytosol. Capping is just as essential for therapeutics leveraging mRNA as it is for regular cellular processes. Delivering uncapped mRNA molecules to a patient results in activation of the immune system and consequent degradation of the uncapped mRNA. Put differently, it causes an immune response with no clinical benefit.

I’m actually not going to spend much time explaining how Maravai’s capping technology differs from its competitors, as I’m not convinced the strength of this product is the dominant consideration for how the company will perform in the coming years. I will note that a product’s capping efficiency, which refers to the percentage of mRNA molecules that are successfully capped, is of vital importance to Maravai’s customers. You can think of capping efficiency in the mRNA world as conceptually similar to full to empty capsid ratios in the gene therapy world. Just as a poor full to empty capsid ratio risks an unwanted immune response that ultimately puts FDA approval of your product at risk, so it is with poor capping efficiency.

The yield, or amount of mRNA produced in the capping process, is another salient consideration. The better the yield, the more final product can produce for the same amount of money and in less time. It's worth emphasizing that yield is a secondary consideration. A capping process with extremely high capping efficiency and low yield still results in a potentially efficacious therapeutic. A capping process with poor efficiency but very high yield does not.

I say that yield is a secondary consideration because it helps explain why CleanCap hasn’t entirely taken over the market. The product boasts strong metrics on both capping efficiency (95%) and yield (95%), but these stats weren’t enough to convince Moderna to use the product. Instead, the company elected to use a post-transcriptional capping method that was more time-consuming/lower yield but that had a capping efficiency of 100% rather than 95%.2

As one would expect, Maravai’s business went on a tear in 21/22, with revenue peaking at 883mm in 2022 before crashing back down to 259mm in 2024. Supplying a critical component of mRNA vaccines turns out to be a good business when there’s a global pandemic going on! Its stock price collapse (down 95% from its peak in late ‘21) stems from multiple compression as well as a steep drop in revenue. When the business initially went public it appeared to have a high level of predictability. Covid vaccines were meant to be more than a one time thing for vast portions of the world population, providing a recurring-ish revenue stream that would enable management to continue its predominantly growth by acquisition strategy. Unfortunately for Maravai, the rosy scenario of frequent Covid boosters didn’t come to fruition.

Subscribe now

The most optimistic bull case for Maravai goes something like this: It’s true that the business, and the cell/gene therapy space more generally, has struggled in recent years. It’s additionally true that mRNA-based infectious disease vaccines haven’t experienced the expected adoption beyond 2022 and suffered from some fairly severe unforeseen PR issues. However, mRNA still has real potential in the infectious disease space, as well as applications far beyond that area:

“The first clinical trial involving an mRNA therapeutic agent took place in 2016. Since then, over 1,500 clinical trials are now in the pipeline, encompassing a wide range of medical applications.

These trials include vaccine development programs targeting infectious diseases such as avian flu, Lyme disease, malaria, HIV, tuberculosis, shingles, rabies, yellow fever, respiratory syncytial virus (RSV), and Zika. Beyond infectious diseases, mRNA based programs are addressing various medical conditions, including ornithine transcarbamylase deficiency, glycogen storage disorders, alpha-1 antitrypsin deficiency, acute lymphoblastic leukemia, Hurler syndrome, ovarian cancer, cardiovascular disease, and autoimmune disorders.

Cell and gene therapy programs also leverage mRNA across multiple therapeutic modalities, such as CRISPR/Cas9, transcription activator-like effector nucleases (TALENs), enzyme replacement therapies, allogeneic CAR-T cells, and base editing. These advancements underscore the broad and growing impact of mRNA technology in revolutionizing healthcare.”3

In short, Maravai stands to benefit significantly as mRNA programs move beyond Covid vaccines and from clinical trials through to regulatory approval/commercial availability. Furthermore, the approved therapeutics will cover a broad range of diseases, resulting in a much more diversified revenue base going forward. One should underwrite the business based on the exciting future applications of mRNA rather than where Covid boosters might be trending this year.

On the infectious diseases side, I’ll briefly note that HHS cancelled its 590mm in funding for Moderna’s avian flu vaccine, and that uptake of mRNA-based RSV vaccines has been disappointing so far. On the cell/gene therapy side, I think this bull case ignores some important nuances. Firstly, mRNA is used to varying degrees within the therapeutic modalities mentioned above:

• For CRISPR/Cas9 approaches, mRNA can be used to code for the Cas9 protein that makes the double stranded break in the DNA helix. However, for those CRISPR therapeutics that don’t leverage base-editing, using mRNA for the Cas9 is neither the default approach nor viewed as the ideal one.4 Casgevy, the only currently FDA approved CRISPR treatment, delivers the Cas9 protein rather than the mRNA that codes for it.

• When it comes to CRISPR treatments leveraging base-editing, mRNA is used when the goal is to target the liver. LNPs, the delivery vehicle used in all mRNA vaccines, are very good at getting to this organ and so it makes sense to utilize the Cas9 mRNA rather than the actual protein.5 Unfortunately, CRISPR therapies attempting to target other organs don’t use mRNA or LNPs.6 This has an important consequence: it’s likely not a good idea to invest in Maravai based solely on an expected proliferation in base-editing treatments. One would need to be confident in the proliferation of base-editing treatments and (1) either that LNPs will be developed/utilized to target organs beyond the liver or (2) that the number of CRISPR applications via liver delivery is so large it doesn’t matter.

  • The second question for Maravai and base-editing is whether the company can effectively compete with Danaher. Danaher was the supplier/manufacturer for the majority of materials used in the base-editing treatment for a newborn with liver disease. Furthermore, it was recently selected to be the supplier/manufacturer for the recently announced Center for Pediatric CRISPR Cures.

• For TALENs mRNA delivery is the default approach, but that’s a gene-editing tool that’s been largely dwarfed by CRISPR and so is used in far fewer clinical trials (although that doesn’t mean it couldn’t end up being very effective!) Cellectis, Allogene, and Iovance are running clinical trials leveraging TALEN.

• The current slate of FDA-approved enzyme replacement therapies deliver the actual enzyme to patients, rather than the mRNA that codes for said enzymes. That said, there are companies (such as Moderna) working on mRNA-based approaches to enzyme replacement therapies.

• For allogeneic CAR-T therapies, it’s again untrue that mRNA is the default method to deliver the Cas9 protein, although it is the default method when TALEN is used and for the CRISPR approach developed by Intellia.

The other question to mull over is how large Maravai’s business can become even if cell/gene therapies leveraging mRNA do end up taking off. A lot of the excitement about these therapies centers around their potential to treat rare diseases. I hope that such therapies succeed, but they’re unlikely to be a boon to Maravai. Treating rare diseases works economically when you’re the actual therapeutic provider; it also works for a company like Dyno Therapeutics that develops mission critical therapeutic delivery vehicles and so can command hefty royalty fees in return. It doesn’t work when supplying an mRNA cap that costs what it costs regardless of the end therapeutic sticker price. To be sure, cell/gene therapies aren’t only targeting rare diseases. CRISPR Therapeutics and Verve are individually working on base-editing treatments that have the potential to treat broad patient populations. CRISPR’s CTX310 program aims to treat those with very high cardiovascular disease risk, and estimates the potential patient population in the U.S. and Europe to be 14mm. However, even assuming CTX310 passes clinical trials with flying colours, is given to all 14 million people, and is using Maravai’s capping technology, it’s only intended to be a one-time treatment. While 14 million people is a lot, it’s only a small fraction of the 4.6 billion Pfizer/BioNTech Covid vaccines that have been shipped since December 2020. It’s an especially small fraction when Maravai management recently admitted that the capping space has become more competitive and prices will need to come down going forward:

“Pre-pandemic, the mRNA reagent and service market was mostly TriLink, honestly. And so it definitely is a more competitive space, both on the RUO discovery side as well as the GMP side. We know that, and I think we embrace that. That's why you see continued push for technology development and enhancement. The reality is that COVID era programs were scaling processes people have been working on for 5 to 10 years, and sometimes more. And this is a period not only of reset from the pandemic and the volumes of infectious disease vaccine, but also people have had the opportunity to test new innovations -- we know and embrace that people will not use the same reagents they used 5 years ago. They will not use the same processes they used 5 years ago; and that, ultimately, the cost of mRNA needs to come down to have it take its rightful place as a ubiquitous platform in medicine.”7

My intent with this note isn’t to argue that a world in the nearish future where there’s a plethora of cell/gene therapies targeting rare and common diseases isn’t possible. My intent is to point out that a lot has to go right for Maravai specifically to profit from this success. Companies like Verve, Intellia, and Beam all stand to gain a lot if their therapeutics are shown to work; the same is not necessarily true of Maravai if it’s supplier to these companies. Maravai’s used its Covid revenues and debt to position itself for future growth, which has led to OpEx increasing YoY even as revenue goes down.8 That’s not necessarily a bad strategy, but it might be when you don’t have the cash that Danaher does and so can’t always acquire the number 1 or 2 player in a space. It also might be when there’s high executive turnover (as there is at Maravai), and so an inconsistent acquisition taste (underwriting Berkshire is a different task once Buffett steps down!) I mentioned above that Danaher’s well-positioned in the base-editing realm, and so it’s an interesting signal that its subsidiary Aldevron recently announced a non-exclusive deal with Maravai for its capping technology. The company’s capping franchise and biologics safety testing business makes it a good target to be bought out by a more diversified CDMO, but as a standalone organization it fits pretty firmly in the ‘too-hard’ pile.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

What makes CRISPR-Cas9 exciting for treating human disease is that we’re able to create synthetic gRNA, and so in theory can use the complex to affect whatever DNA sequence we like. Take sickle-cell disease, a blood disorder. We know there’s a sub-segment of those with sickle cell who never stop producing fetal hemoglobin. These patients either exhibit very mild symptoms or are entirely asymptomatic. We also know the gene responsible for repressing fetal-hemoglobin transcription. If I can design a gRNA strand complementary to this gene, the Cas9 will make a double-stranded break, and the human body’s imperfect DNA repair mechanism (non-homologous end-joining) will inactivate the gene in its efforts to rejoin the break. This is termed a gene-knockout approach, and is exactly what CRISPR Therapeutics/Vertex have done with Casgevy.

We can also modify the Crispr-Cas9 complex so that it alters a single nucleotide rather than makes a double-stranded break. An adenine base-editor turns an adenine base into a guanine, and a cytosine base-editor turns a cytosine into a thymine. This in turn changes an A-T base pair to a G-C one, and a C-G base pair to a T-A. An adenine base-editor was used in the case of newborn KJ in an effort to correct his CPS1 enzyme deficiency.

The potential applications of CRISPR don’t stop there (Intellia’s Annual Report is a fun place to look for where CRISPR could be used in the future. The company’s trying to tackle a lot!) CAR-T is an area where CRISPR may end up being especially useful. CAR-T, like other immunotherapies, is a way to harness the body’s immune system to more effectively fight off cancer. A patient’s own T-cells are removed, then modified in a lab to contain a chimeric antigen receptor (or CAR). Once transfused back into the patient, these chimeric antigen receptors will bind to antigens on cancerous cells, and so can more effectively fight off the cancer. In other words, the lab serves to train a patient’s T-cells to recognize the malignant cells and fight them off. For example, FDA CAR-T therapies often bind to CD19, a protein expressed in B-cells, and so worth targeting when treating B-cell cancers. There are currently 7 FDA approved CAR-T therapies, all of which are used for various blood cancers in cases where the cancer has returned after treatment or stopped responding to treatment.

The seven FDA-approved treatments are all autologous, meaning they use a patient’s own T-cells rather than those of a donor. Autologous CAR-T therapies have several downsides. Modifying a patient’s own T-cells is far from an easy thing to do, and the process takes weeks to months, during which time the cancer can progress. The complexity of modifying these cells results in a costly treatment. Kymriah, the first FDA approved CAR-T treatment, is priced at $475,000; there isn’t much in the way of economies of scale when delivering bespoke therapies! The total cost is greater once one accounts for the costs of leukapheresis (the process of removing a patient’s white blood cells) and lymphodepletion therapy (a dose of chemotherapy the patient receives pre-infusion). Furthermore, an autologous approach requires that a patient’s T-cells are healthy enough to undergo modification. As one would expect with those who are very sick, this unfortunately isn’t always the case.

For the reasons given above, there’s a lot of interest in developing allogeneic (donor derived), rather than autologous, CAR-T treatments. There are real benefits to an allogeneic approach: T-cells don’t need to be harvested from a sick patient, the patient doesn’t have to wait around for treatment, there are economics of scale in the manufacturing process, and patients with non-optimal T-cells can still be treated. Currently, however, this approach poses issues on both the donor and patient side:

Graft-versus-host disease – the donated T-cells see the recipient’s T-cells as a threat. This is due to a mismatch in human leukocyte antigens (HLAs), proteins on human cells that let the immune system know which cells are native/foreign. The donor T-cells detect they’re in foreign territory, and so begin attacking the patient rather than helping him/her. The donor cells’ T-cell receptors (TCR) are responsible for launching this attack.

Rejection via host T cells - The host -cells aren’t not pleased to have the donor cells there. Again, this is due to mismatches between donor and acceptor HLAs. Two types of host T-cells attack these donor cells: CD8 and CD4.

Rejection via host NK cells – NK cells, or natural killer cells, are another type of white blood cell that attacks cells it deems harmful. Host NK cells also attack the donor cells.

Subscribe now

Leveraging CRISPR-Cas9 is a potential avenue to solve these issues. Companies differ in their approaches to mitigate host T-cell and host NK cell rejection, but the CRISPR approach to prevent graft-versus-host disease has been fairly uniform. The ability to create synthetic gRNA means scientists can decide where to insert the chimeric-antigen receptor, something that isn’t possible with typical CAR-T transduction methods. With CRISPR, the CAR gene can be inserted into the TRAC locus, which is the section of the T-cell responsible for TCR expression. Placing the CAR gene here disrupts the TRAC locus, thus preventing TCR expression and graft-versus-host disease.

Avoiding rejection via host T-cells and NK-cells is more complex. As mentioned above, CD8 and CD4 host T-cells both attack the donor cells. CD8 cells are activated when they detect foreign HLA I molecules; CD4 cells are activated when they detect foreign HLA II molecules. Knowing this, the obvious solution would be to use CRISPR-Cas9 to knockout HLA I/II expression in the donor cells. Unfortunately, natural killer cells spring into action when they detect cells with little to no HLA-I expression. Knocking out HLA-I might prevent CD8 cells springing to life, but at the cost of activating NK cells!

The easiest method, at least initially, to solve for the issue of host immune response is lymphodepletion, which is a standard part of the autologous CAR-T process anyway. Lymphodepletion gets rid of a large number of patient lymphocytes (T-cells, B-cells, NK-cells), thus giving the incoming CAR-T cells plenty of room to proliferate and fight off the cancer. This in turn increases the odds of CAR-T cell persistence, which has been continually linked to better outcomes. Of course, this means the patient’s immune system is in a comprised state, which is why infections are a common side effect of CAR-T.

As said above, this works as an initial way to limit host immune response. The patient’s immune system will come back, and at that point T-cells and NK cells will get to work once they notice HLA mismatches in the donor T-cells. This response limits CAR-T persistence, and thus the effect of the therapy. One option here is to up the dose of chemotherapy and thus keep the immune system suppressed for longer, but that in turn further increases the risk of patient infection.

An effective treatment approach that leverages allogeneic CAR-T, then, is all about figuring out how to get those CAR-T cells to endure in patients. This problem is in some ways made more acute by disrupting the TRAC locus: this step prevents graft-versus-host disease but actually reduces donor T-cell persistence. CRISPR Therapeutics and Intellia Therapeutics are both working on allogeneic CAR-T therapeutics that leverage CRISPR technology, but have taken different steps to get there.

Intellia has focused predominantly on evading immune detection. They knockout HLA Class II genes, thus evading rejection by host CD4 T-cells. As mentioned above, knocking out HLA Class I genes isn’t as simple, because that results in activated host NK cells. Intellia’s workaround here is to knockout only HLA-A, which is part of HLA Class I. They then keep HLA-B, C, and E, and search for T-cells donors whose HLA-B and C alleles are matches to that of the patient. This should prevent host NK Cell and CD8 T-cell rejection. Moreover, it would significantly simplify the donor matching process: management estimates 20 donors would cover 80% of the US population.

CRISPR Therapeutics takes a blended approach of avoiding immune detection and knocking out genes that reduce T-cell persistence in other ways. They knockout the entirety of HLA Class I, and additionally knockout the genes that code for Regnase-1 and TGFBR2. Knocking out Regnase-1 should increase T-cell persistence and antitumor immunity; knocking out TGFBR2 should prevent CAR-T cell exhaustion (when T-cells stop functioning as they should). CRISPR Therapeutics’ decision to knockout the whole of HLA Class I and not touch Class II means there will be an immune response from host CD4 T-cells and host NK cells, but they’re betting that knocking out Regnase-1 and TGFBR2 will make up for that.

Developing allogeneic CAR-T treatments, whether using CRISPR or not, has been a tough place to be over the past few years. Enthusiasm for gene/cell therapies has dropped dramatically since Covid, leading to frequent layoffs and slashing of programs in development. Most of the companies in the space are in phase I trials with market caps of sub 300mm USD (Caribou, Cellectis, Celyad, Fate, Imugene). Allogene has a large B-cell lymphoma treatment in phase 2 trials, but its market cap also sits at ~264mm. Not all of them leverage CRISPR. Allogene and Cellectis both use TALEN, another gene editing tool that leverages mechanisms initially found in bacteria; Celyad doesn’t use gene editing at all given concerns about off target effects; Imugene uses yet another gene-editing tool (ARCUS) discovered by Precision Biosciences.

Sana Biotechnology only has phase I candidates, but has done a fantastic job raising money and a promising, non CAR-T, diabetes type 1 treatment in the pipeline. Gracell was acquired for a billion by AstraZeneca, but the acquisition announcement emphasized the company’s autologous CAR-T manufacturing capabilities rather than its allogeneic programs. Roche spent a billion to acquire Poseida Therapeutics in late 2024, and in that case indicated real interest in the company’s CRISPR edited allogeneic CAR-T candidates. CRISPR Therapeutics and Legend Biotech are doing fine, but both have FDA approved therapies in other areas and do much more than just allogeneic CAR-T!2 AvenCell, a private company working on both autologous and allogeneic candidates, looks quite interesting. The company licensed Intellia’s platform for its allogeneic candidates, and raised over 100mm in late 2024, a tough time to be raising capital when you’re not an AI business! Blackstone Life Sciences was its founding investor back in 2021, which is either another data point on the style drift that occurred in the Covid era (the fund typically invests in much later-stage drugs) or an indication the company’s working on something with a relatively high likelihood of success.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

It's worth understanding CRISPR and the current gene therapy landscape generally to appreciate how cool of an advancement this is. The CRISPR-Cas9 mechanism was initially discovered in bacteria, and serves to protect these bacteria against harmful foreign invaders. Like humans, bacteria have an immune system that responds to external threats. Part of the way an immune system functions (and why vaccines are helpful) is by learning from previous infections. The polio vaccine delivers an inactivated form of the polio virus, which causes my immune system to develop antibodies to fight the disease. Should I somehow then get exposed to polio, these antibodies recognize the virus, spring into action, and prevent serious illness.

In bacteria and archaea, CRISPR-Cas9 works somewhat analogously. When a bacteriophage (a virus that infects bacteria) injects its DNA into bacteria, an enzyme complex cuts out a short piece of the phage’s DNA and inserts it into a CRISPR array. This DNA segment is transcribed and forms a portion of guideRNA (gRNA), which works with the Cas9 enzyme to identify bacteriophages that have invaded in the past. The next time invasion occurs, the CRISPR-Cas9 complex recognizes the phage, and Cas9 makes a double-stranded break in its DNA, rendering the virus ineffective. In other words, the CRISPR-Cas9 complex is akin to an actually effective defense coordinator when the Eagles are on fourth and inches. Based on historical data, the DC recognizes that they’re going to use the tush push. One in existence just hasn’t yet figured out how to make the equivalent of a double-stranded break and prevent the first down.

In the case of the bacteriophage, once this double-stranded break occurs the phage is ideally further degraded and destroyed. If the effects of CRISPR-Cas9 stopped there, there wouldn’t be much to get excited about with its potential applications for treating disease. If there’s a method that definitely doesn’t work for fixing a faulty gene, it’s making a double-stranded cut and hoping the cellular repair machinery doesn’t rejoin it.

Subscribe now

What makes CRISPR-Cas9 so intriguing is threefold:

1. The innate DNA repair mechanisms that spring into action after Cas9 makes its double-stranded break.

2. The modifications that can be made to guideRNA. Researchers can modify gRNA to direct the CRISPR-Cas complex towards a precise section of human DNA. This can either be a segment of DNA with a deleterious mutation, or a segment that’s responsible for up/downregulation of a gene. The ability to design gRNA sequences means one could theoretically use CRISPR to fix any disease caused by a DNA mutation.1

3. The modifications that can be made to the Cas9 protein. For treating some diseases the double-stranded break is helpful. In other cases we want the CRISPR-Cas9’s targeting ability, but don’t want it to make a cut.

Put simply, this CRISPR-Cas9 complex that evolved to protect bacteria from foreign invaders can be leveraged to treat and (hopefully) cure many human diseases.2

In the case of humans, the repair mechanism from point (1) that will most often spring into action is non-homologous end joining.3 This rejoins the double-strand, but imperfectly. An imperfect rejoin is incredibly helpful when the aim is to switch off, or knockout, a gene. This is the approach Vertex and Crispr Therapeutics took when developing CASGEVY, the only CRISPR treatment that has (thus far!) been FDA approved for a broad patient population. Casgevy is used to treat sickle-cell disease (SCD) and transfusion-dependent beta thalassemia (TDT), both of which are inherited blood disorders. Its design was based on the following observation:

“In most patients with SCD or TDT, HbF [fetal hemoglobin] disappears in infancy, at which point the symptoms of the disease begin to manifest. However, some patients have elevated levels of HbF that persist into adulthood, a condition known as hereditary persistence of fetal hemoglobin, or HPFH. These patients are often asymptomatic or experience much milder forms of disease because elevated HbF compensates for the defective adult hemoglobin.”4

Vertex/CRISPR’s strategy is based on upregulating (producing more of) the amount of fetal hemoglobin in a patient. To achieve this, CRISPR/Cas9 makes a double-stranded break in a region of the BCL11A gene, a gene responsible for repressing fetal hemoglobin transcription as one ages. The double-stranded break is repaired through non-homologous end joining, which then means the gene is unable to act as intended and fetal hemoglobin production is increased. This gene knockout approach is especially helpful when the aim is to treat more than one disease. Increased fetal hemoglobin helps those with both SCD and TDT, so you’re both helping more people and can amortize the R&D spend across a larger patient population. This isn’t a luxury Bluebird Bio enjoyed, which meant it was developing separate therapies for sickle cell and beta-thalassemia.

The CRISPR-Cas therapy to treat KJ went a step further, and is similar in philosophy to some ongoing clinical trials leveraging CRISPR (see Beam and Verve for further examples). As mentioned above, the gene knockdown approach depends upon the broken double-stranded DNA being rejoined imperfectly. There wouldn’t be a whole lot of value to the method if the break was then perfectly corrected! This isn’t a method, however, that comes in handy when the goal is to actually fix the mutation rather than just knock a gene out of commission. Fixing the mutation requires first altering the CRISPR-Cas9 complex so that it can no longer make double-stranded DNA breaks. Next, the CRISPR complex is modified to include deaminase, a type of enzyme that can alter a single nucleotide.

It's easy at this stage to get bogged down in overly scientific jargon and have your eyes glaze over. There are a few critical pieces to keep in mind:

• DNA is a double-stranded helix composed of 4 different nucleotides: Adenine (A), Thymine (T), Guanine (G), and Cytosine (C). The double-strands of DNA are held together by hydrogen bonds between nucleotides on opposing strands. A pairs with T, and G pairs with C. In the case of KJ, he had an error in his genetic code where what should have been a cytosine was instead a thymine. Consequently, he had an A:T base pair rather than a G:C one.5 A point mutation like this is where a deaminase comes in handy.

• In its natural state, CRISPR-Cas9 does not allow precise edits to one’s DNA. Achieving specific nucleotide edits requires modifying the complex and adding a deaminase enzyme. This approach is termed “base editing.”

This base-editing method is what was used to treat KJ. The CRISPR-complex was modified to include an adenine base editor, which as the name suggests edits an adenine base and turns it into a guanine.6 Thus, an A:T base pair at the variant location becomes a G:C one. The therapy was delivered intravenously and contained within a lipid nanoparticle (LNP), an equally vital part of the equation. I’ve discussed this before, but delivery vehicles are a vital component of getting gene therapies right. The right therapy with the wrong delivery mechanism means you don’t have a functioning therapy at all (see Verve’s recent announcement for an illustration of this). The individualized nature of the treatment for KJ made LNPs an especially good choice: the challenge with gene therapies delivered via AAV vector is you can only dose the patient once. Attempting to dose the patient again will drive the immune system into total overdrive. LNPs do cause an immune response, but it’s not severe enough that treatment is limited to one dose. Doctors didn’t know at the outset how much of the treatment KJ would need, so the LNP gave them much needed flexibility.

Thus far KJ has been dosed three times and shown meaningful improvement. At the time the New England Journal of Medicine paper was released, he was eating a regular amount of protein and on a much lower dose of medication. That the treatment method has (so far) worked is an important validation for a whole basket of CRISPR companies, many of which have experienced a tough few years in the aftermath of Covid (Beam, CRISPR Therapeutics, and Intellia are down 85%, 80%, and 95%(!) from their respective 2021 highs.) It’s also a good look for Danaher, whose companies Aldevron and IDT Technologies were both used in the development of KJ’s treatment. Aldevron was acquired in June 2021 as a bet on the growth of gene and cell therapy, which then fairly promptly entered a slump. Its involvement is at least evidence that Danaher bet on the right horse. Acuitas Therapeutics, which provided the LNP for the treatment, presumably also experienced a post-Covid drop in its business. Its partnership with Pfizer/BioNTech on the mRNA vaccine, and now with the treatment for KJ, indicates the organization is still poised to benefit from the theoretical growth of nucleic acid therapies in the future.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

Location is of pretty serious importance here. Being just off a highway exit brings in a reliable stream of traffic with a greater propensity to buy more than just gas. A city/densely populated suburb storefront is more likely to bring in regulars with a Zyn or Marlboro habit. Rural locations sit in a more challenging position. While there may be less competition, it’s for good reason! Fewer potential customers and greater poverty rates weigh on the magnitude of success gas stations can have in those areas.

Casey’s General Stores has embraced this challenge, opening 72% of all its stores in areas with fewer than 20,000 people. The structural issues present with operating in these regions forced the company to direct a lot of attention toward its food offerings, and more specifically on high-margin ones. Casey’s began offering freshly made pizza in 1984, and today sells it in almost all stores. Its food menu expanded from there, now including items like hot sandwiches and croissants.To top it off, 97% of stores sell bakery items such as donuts and cookies.1 Management has continually emphasized the size of its pizza business to investors, and as of 2023 Casey’s was the 5^th largest pizza chain in the U.S., at least when measured by number of kitchens. As much as I’d like to imagine the Casey’s C-suite looking at Domino’s stock price performance since IPO and deciding to put further emphasis on pizza operations, that’s fortunately not what occurred. Its pizza business, and management’s focus on higher-margin food items more generally, has been highlighted at least since the company’s 1994 annual report, the first available on the current website:

“It is management's policy to experiment with additions to the Company's product line, especially products with higher gross profit margins. As a result of this policy, the Company has added various prepared food items to its product line over the years. In 1980, the Company initiated the installation of "snack centers" which now are in approximately 99% of the stores. The snack centers sell sandwiches, fountain drinks, and other items that have gross profit margins higher than those of general staple goods. Casey's also has introduced the sale of donuts prepared on store premises, available in approximately 99% of the stores as of April 30, 1994, as well as cinnamon rolls and cookies, and is installing donut-making facilities in all newly constructed stores……. Casey's began marketing made-from-scratch pizza in 1984, expanding its availability to 818 (93%) stores as of April 30, 1994. Management believes pizza is the Company's most popular prepared food product”

Subscribe now

There are mixed opinions on the quality of the pizza, but some people do seem to be fanatics, as evidenced by Dave Portnoy getting booed at the Iowa State Iowa football rivalry for his poor review.2 High-margin food brings two benefits. The first is that it makes rural gas stations economical. The second is that if your food is good enough people will actually come to your store because of it. A Friday night can double as a night to pick up pizza and fill up your gas tank because you’re at Casey’s anyway. Casey’s concentration on rural locations also means that half of their stores aren’t in an area with other major pizza competitors; one doesn’t have to worry about competitive pressures from Domino’s or Papa John’s if there isn’t one close by!3 This leads into a related point: it’s not a good use of Domino’s time to figure out how to make its four-wall unit economics work in areas with a limited population. Figuring this out would probably require adding something beyond its standard menu items, resulting in further complexity for a profit pool that just isn’t large enough relative to other opportunities. Funnily enough, offering pizza plus gas may be the ideal format to operate in sparsely populated regions.

Casey’s went through a management change in 2019, with Darren Rebelez coming on as CEO. Rebelez was previously President of IHOP Restaurants, and prior to that was COO of 7-Eleven. A new COO (also ex 7-Eleven) and CFO were brought on shortly after. In 2023 management announced a three-year strategic plan, that aims to grow store count, accelerate the food business, and enhance operational efficiency. The third piece of the pillar seems to mainly be about reducing employee hours The second has led to thin crust pizza, revamped sandwiches, and some experimentation with chicken wings in Des Moines stores. Chicken wings can be a riskier endeavour than it initially seems. Ideally, the goal with rolling out new food items is to capture a greater share of customer spending. The risk, however, is that chicken wings end up cannibalizing the pizza business and so drag down overall food & beverage margins. At least so far management hasn’t seen this happen.4 The most interesting, and easiest to get wrong, portion of Casey’s strategic plan is the effort to grow store count. The business added 154 stores in fiscal 2024, ~70% of which were via acquisition. In November 2024 (which is part of Casey’s 2025 fiscal year), the company acquired Fikes Wholesale, which operates CEFCO Convenience Stores. This is Casey’s largest acquisition in history and boosted store count by 198, including 148 stores in Texas.

There’s a bull and a bear case to be made around the size and pace of Casey’s recent acquisitions. The bull case is that Casey’s has a tremendous amount of room to grow its store footprint. As of 2024 it was the third largest convenience store chain in the U.S., but that figure’s a bit misleading. Casey’s is just over a third the size of Couche-Tard, which operates Circle K and sits in second place. It’s under a tenth the size of 7-Eleven, which has the number one spot. Moreover, it has yet to open stores in large portions of the country!5

Paradoxically, its rural history could be what enables the company to succeed in areas with increased competition. To succeed in low-traffic regions it had to offer good, high-margin food. That’s not something gas stations in busier areas needed to spend time on to make their unit economics work. Murphy’s USA, which sits in the fourth place spot, is a clear example of this.6 Most of Murphy’s stores are located near to a Walmart, which means the company had no problem generating traffic and didn’t see the need to build kitchens and offer pizzas made on-site. Casey’s ends up looking very compelling when competing head-to-head against stations like these. This food advantage was acquired over many decades, and it’s not an easy task for a gas-station/convenience store to quickly build a fresh-food menu that customers are genuinely excited about! It's true that Casey’s has been abnormally acquisitive, but management’s taking a smart approach to these acquisitions, and a 2-ish times debt to EBITDA ratio is hardly excessive. 125 out of the 198 acquired CEFCO stores have kitchens, which makes adding pizza both relatively straightforward and an easy to boost margins for a historically protein-heavy chain. Furthermore, there’s natural purchasing economies of scale as CEFCO transitions over to using Casey’s vendors.7

The bear case is that Casey’s is straying too far from its roots, and that this makes the future harder to underwrite for management and investors. Historically, the business section of its annual report has included a statistic about the percentage of stores located in areas with fewer than 5,000 people. In 2024 that statistic was changed to the percentage of stores located in areas with fewer than 20,000. That represents some style drift, and could hurt the competitive strength of Casey’s food offerings. Management has said that at the ~50% of locations with national pizza brand competition Casey’s comes out about a dollar or two cheaper. As store mix moves more in this direction there’s a risk that Domino’s, Papa John’s, Pizza Hut etc decide to get more promotional to maintain or take market share. Moreover, until 2021 Casey’s annual reports would brag about installing donut-making equipment in every new store. This no longer appears to be a focus for the company, and at least some stores have now transitioned away from the practice. The pessimistic take on this, especially in combination with efforts to reduce employee hours, is that management is attempting to lower the maintenance opex of the business in a way that will come back to bite them. To top things off, Casey’s paid 11x pro forma adjusted EBITDA for CEFCO, not exactly a cheap multiple! The company’s presentation on the acquisition notes that 40 CEFCO sites are located in the Florida/Alabama panhandle, with plenty of people living nearby. Normally that would be a positive, but perhaps not when your annual report continually harps on the company’s rural nature. Dollar General moving to higher-traffic areas probably wouldn’t be met with much investor enthusiasm!

There is of course also the EV adoption question. While Casey has started putting in charging stations, the company has no stores on either of the coasts. Management has noted that muted electric vehicle uptake in middle America (at least currently) keeps them somewhat insulated from EV trends. A fun question to mull over is how much EV adoption ends up hurting gas stations with strong food brands, particularly when those gas stations are installing charging stations. It’s not clear to me that Buc-ee’s will be all that affected once electric cars really take over the roads. That’s probably less true for Casey’s or Wawa, but still true to a degree.

At ~32x earnings, one wouldn’t necessarily call Casey’s cheap. The counter to that is earnings per share have compounded at ~19% per annum over the past 5 years. Not exactly nothing! The company does have plenty of room to grow geographically, and increasing energy drink/nicotine consumption is a tide that lifts all gas station boats.8 The risk is that management is no longer executing the classic Casey’s playbook, and could run into challenges in these more competitive markets.

Share

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

As a brief review, analyzing gene expression levels has historically been done using bulk-RNA sequencing. This method reads out the average gene expression levels from a sample, but doesn’t allow you to zoom in and examine individual cells. This lack of granularity presents an issue when rare cell types are driving a disease. Single-cell RNA sequencing (scRNAseq) enables researchers to look at gene expression on the individual cell level. Standard scRNAseq methods don’t, however, include spatial information on these cells, and so researchers are still left in the dark on how these cells may be interacting with one another. This can again be quite relevant when examining things like tumor resistance, which at times is driven by the microenvironment a tumor resides in rather than the tumor’s transcriptome.

In situ analysis goes a step further than both single-cell and spatial, giving researchers a sub-cellular (rather than single-cell) level of resolution with spatial context. This comes with trade-offs: it’s more time consuming/expensive, you can’t analyze anything close to the whole transcriptome, and you need to know ahead of time which genes you want to target (so you can’t use it to discover a novel transcript you didn’t know was there). Put more simply, in situ’s a great tool to zero in on a very particular area at a level of detail that regular spatial analysis doesn’t allow. For example, you might use spatial analysis to figure out where gene expression changes are taking place in lung cancer tumors, and then zoom in with an in situ lung panel to examine a particular tumor subtype. Importantly, in situ analysis is imaging-based rather than NGS-based, making it the one product in the single-cell/spatial world where Illumina and other sequencing providers don’t benefit from a growth in its usage and don’t have a front-row seat to how the technology is used.

There are four in situ providers one frequently comes across: 10x Genomics, through its Xenium instrument, NanoString Technologies (now a part of Bruker) through its CosMx SMI Instrument, Vizgen, through its MERSCOPE instrument, and Resolve Biosciences, through its Molecular Cartography Platform. Resolve Biosciences hasn’t updated its publications page since June 2023, went through a leadership change in 2023, and has no clear way on its website to contact a sales representative. Consequently, I’ve mostly excluded Resolve from the following discussion.

Beyond sensitivity and specificity, studies comparing different instruments’ performance focus on a few key factors:

Transcript counts per gene – the number of transcripts a platform picks up associated with a specific gene. This matters because so much of RNA analysis is about quantifying gene expression! All else equal, an instrument that can identify more transcripts per gene than a competing offering will give researchers a better window into the questions they’re trying to answer.

Plex – the number of distinct genes an in situ platform can target at one time. An 1000 plex panel allows researchers to pinpoint 1000 genes in a tissue sample; a 300 plex panel would only allow researchers to pinpoint 300.

Cell Segmentation Performance – the ability of a platform to accurately draw cellular boundaries. If boundaries are too wide transcripts end up getting assigned to the wrong cell.

Subscribe now

I looked at three different pre-print studies comparing in situ platforms. The first analyzed the performance of Xenium, CosMx, and Merscope when examining FFPE samples from various human tissue types; the second analyzed the performance of Xenium and CosMx when examining only FFPE prostate samples; the third analyzed the performance of Xenium, Molecular Cartography, Merscope, and three non-commercial methods when examining mouse brains. It’s of course worth noting that these studies are all just pre-prints, and that 10x, Vizgen, and NanoString have all improved their respective technologies since these papers were published.

Xenium emerged superior for the two studies examining FFPE tissues, capturing the highest number of transcript counts per gene with high sensitivity and specificity. That’s not to say Xenium won on every metric: its cell segmentation abilities were the worst of the bunch, and it and Merscope’s panels were much lower plex than what CosMx offered (CosMx was 1000 plex; Xenium and Merscope were more than half that).

Poor cell segmentation seems to be forgivable in the FFPE cases: Xenium’s default segmentation algorithm may be lacking, but it resulted in a minimal number of incorrectly assigned transcripts. Furthermore, one isn’t obligated to use an instrument provider’s algorithm, and as in situ usage increases there’s reason to think independent algorithms will get better and better.1 Put more simply, if an instrument has poor sensitivity/specificity there’s no way for a researcher to fix that; it’s inherent to the instrument. If a provider’s segmentation algorithm isn’t great, however, a researcher is free to use a different one.2

Xenium was not the winner when it came to mouse brains. In this study Merscope won out (followed by Xenium), and Xenium’s poor cell segmentation was flagged as leading to “a substantial increase of incorrect molecular assignment.” Merscope may be the optimal choice when it comes to mice! This leads into a related point, which is that the right instrument may just depend on the type of sample studied.

There are differing opinions on the importance of CosMx’s better plexity. On the one hand, plex obviously matters. It enables researchers to visualize a greater number of genes, which is clearly useful for answering scientific questions! That said, higher plexity is less useful if it comes with much lower sensitivity and specificity. In situ loses its utility if it’s identifying transcripts that aren’t there (poor sensitivity) or missing transcripts that are (poor specificity). Researchers are excited about single-cell/spatial/in situ methods because granular gene expression data that moves beyond bulk methods can be so informative. As one paper put it:

“A key observation from this analysis has been the importance of assay sensitivity. Simply, many analyses depend on the robust detection of relevant transcripts. Although CosMx data has consistently higher noise that masks lowly expressed transcripts, the challenges that emerge in accurate cell typing and data interpretation tend to come from the weak detection of highly expressed cell type markers….. it is critical to underscore that many biological questions will depend on the accurate detection of a small number of genes: tumor classification based on the presence of a single biomarker, distinguishing between subtly polarized cell states, the spatial mapping of cells expressing immunosuppressive factors, and more.”

Again, none of the above is to say that plex isn’t a relevant factor. CosMx now offers 6000-plex RNA panels, a 6-fold increase from when these studies were done. 10x offers 5000-plex panels, which interestingly come with a lower-sensitivity per gene than its lower plex methods. 10x’s decision to offer increased plexity at the cost of lower sensitivity seems to indicate that some researchers are happy to make that trade-off (although presumably only up to a point). It also means that CosMx’s plex is now only 20% better, rather than 2.5x better as it was the time of these preprints.

I’d be remiss to study the differences between these technologies without examining the companies these technologies reside in. A reasonable takeaway from looking at Vizgen, 10x Genomics, and NanoString is that they enjoy suing each other. NanoString went into Chapter 11 proceedings after a lawsuit with 10x over its GeoMx NGS-based spatial analysis tools didn’t go its way. For a period, NanoString was also unable to sell its CosMx in situ products in Europe following injunctions from the European Unified Patent Court and a court in Germany. Bruker acquired NanoString’s assets out of bankruptcy for 392.6mm, or ~2.3x 2023 revenues in May of 2024. Those 2023 figures are somewhat misleading: Other than a consumables carve-out for those who purchased an instrument before November 2023, NanoString is now barred from selling its GeoMx products in the US. Bruker management have said the NanoString business is currently running at a bit under 10mm a month, and they don’t expect revenue to return to pre-Chapter 11 levels in ’25.3 There’s a second, continuing, lawsuit between 10x and Bruker over patent infringement/anti-trust violations on the in situ side of things, with the trial set for August 4th of this year. Until recently, Vizgen and 10x Genomics were suing each other over their in situ technologies. 10x was suing Vizgen for patent infringement and Vizgen was countersuing 10x for antitrust violations. Those suits were settled in early February.

I think there’s a question as to how hard Bruker will push to see NanoString’s in situ business succeed. Bruker’s 2024 revenues were 3.37B, but operating income dropped from 437mm in 2023 to 253mm for 2024, its cash & equivalents are down from 488.3 in 2023 to 183.4mm in 2024, its stock is down 50% over the past year, and it’s been on a partially debt-fueled acquisition spree. NanoString succeeding isn’t of existential importance to the business, and spending too much time in litigation is unlikely to be an activity investors have much patience for.

Vizgen and 10x Genomics, on the other hand, have much more of an incentive to fight for success. Vizgen’s capabilities are more limited when it comes to both plexity and pre-designed gene panels. It offers 1000-plex panels as compared to Xenium’s 5k plex panels, and 3 pre-designed panels against Xenium’s 11, but that doesn’t mean it can’t succeed as a business. That’s especially the case with the ill will 10x created through initiating so many lawsuits. I’m not sure it's fair to view 10x as a totally bad actor here: patents exist to protect innovation, and so while it may be true that suing competitors can put them into bankruptcy (as it did with NanoString) or force them to restructure and do layoffs (as it did with Vizgen), it’s also true that it’s important to live in a society where IP is protected. That said, that doesn’t mean a lot of academic researchers were pleased by 10x’s tactics, which could end up negatively affecting 10x’s brand with critical buyers.

The other question to mull over when thinking through the future of in situ is whether improved NGS-based spatial analysis technologies will cannibalize some use cases. Until very recently, it was the case that NGS-based spatial analysis wasn’t able to give you information at a single-cell level. 10x’s Visium HD, the company’s most recent NGS-based spatial offering, now offers this. Given Visium’s far greater throughput (it can do the whole transcriptome rather than target 5000 genes), this enhanced resolution will likely be enough to keep a portion of researchers from moving towards in situ. That doesn’t really pose a problem for 10x, but it poses a very real one for Bruker, now barred from selling NanoString spatial products in the U.S., and Vizgen, which doesn’t offer an NGS-based spatial solution.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

1) The company spent 8 billion acquiring Grail, an acquisition it went ahead with despite objections from U.S. and European regulators. This led to an activist battle with Carl Icahn, the ouster of Illumina’s CEO and CFO, multi-billion dollar impairment charges, and Grail eventually being spun-out in mid-2024.

2) China banned imports of Illumina sequencers in March of this year.

3) Increased competitive pressures from venture-backed players (Element and Ultima), BGI Genomics (which put pressure on Illumina’s China business before the ban) and Roche (although they haven’t yet entered the space). One point to keep in mind here is that Illumina serves the whole spectrum of the short-read sequencing market, not just customers who demand the most sophisticated/highest throughput instruments. Element Biosciences may only be aiming to serve mid-throughput customers, but that still presents a real threat.

4) Illumina’s core business, sequencing, is based upon Jevons Paradox holding true. Its newer machines are more expensive upfront, but able to perform individual sequencing runs at a more affordable price point than legacy machines. In theory, this unlocks increased demand which then leads to increased revenue etc. Whether or not the paradox holds in the long-term, what it leads to in the short-term is investor uncertainty as sequencing volumes grow without a correspondingly large increase in revenue. As Edward Larkin points out, consumables revenue grew at only 7% per year from 2019-2024 as consumables volume grew 25% per year.

The bear case for Illumina is straightforward: this is a company serving a market that’s maturing and increasingly competitive. The bull case is similarly straightforward. Illumina’s struggled with poor management, exemplified by its acquisition of GRAIL; now that Grail has been divested, you’re left with a company that remains the market leader in next-generation sequencing. This market’s growth may have slowed as of late, but this is a temporary blip rather than an indication of market maturity. Growth will re-accelerate as NGS Applications like liquid biopsy takeoff and interest in multi-omics grows. Illumina sits in the best position to benefit from that growth.

Both the bull and bear cases include, whether explicitly or otherwise, a view on the growth potential of single-cell and spatial sequencing. The bear will say that single-cell/spatial analysis is really the domain of 10x Genomics, and isn’t necessarily an attractive place to be anyway. 10x has run into substantial growth headwinds as of late, has a stock that’s down more than 70% over the past year (even after its 20% jump yesterday!), and isn’t afraid to be litigious with competitors.1 The bulls will say single-cell/spatial is one of Illumina’s many potential growth avenues. I think it’s worth exploring the single-cell/spatial space more thoroughly, as well as considering why these are areas where Illumina wants to compete.

Next-generation sequencing typically refers to the sequencing of DNA. All cells in the human body have the same DNA, but vary widely in function; this difference is thanks to transcription and translation. A portion of DNA is transcribed into coding and non-coding RNA, and this coding RNA (termed mRNA) is then translated into proteins. As one might expect, there’s value to sequencing more than just DNA. Sequencing RNA can provide insight into what genes are actually expressed in a certain sample type. Analyzing cancer tissues is a good illustration of where this is useful: looking at gene expression levels in cancerous vs non-cancerous tumors can give researchers and clinicians ideas on which genes to target in fighting a cancer and which RNA biomarkers are worth tracking over the course of treatment. For example, RNA-sequencing will tell you the enzyme tyrosine kinase is over-expressed in many cancerous tissues when compared to non-cancerous tissues.2 Tyrosine kinase inhibitors serve to prevent this over-expression.

Historically, bulk-RNA sequencing (which is not a growth area Illumina is pursuing nor an area where 10x Genomics plays) has been the dominant method for analyzing gene expression levels in tissues. While this gives you average gene expression levels in a sample, it doesn’t enable you to look at individual cell transcriptomes. This can present problems when rare cell types are driving things like treatment resistance – these cells express RNA at such low levels that they won’t show up in a bulk method! This is where single-cell RNA sequencing (scRNAseq) comes in, which allows you to study individual cell transcripts within a tissue sample. Again, this can be especially helpful when looking at cancer: scRNAseq has identified rare drug-tolerant RNA variants in breast cancer and which specific immune cells are associated with better outcomes in non-small cell lung cancer.3 This granularity just isn’t possible with bulk methods.

As with all methods, single cell sequencing has downsides. Most notably for the purposes of this note, neither bulk-RNA nor single cell methods include spatial information, and so don’t allow you to analyze how cells interact and communicate with one another. In the oncology space, this spatial information can provide another form of insight into the behavior of tumors. A view into a tumor’s microenvironment (the ecosystem surrounding a tumor) can help answer questions like why different tumor subpopulations within a patient might be responding differently to treatment. In other words, within the cancer context single-cell RNA sequencing is useful when tumor resistance is due to a variant’s transcriptome. Spatial is useful when tumor resistance is due to the environment said tumor resides in.4

Subscribe now

Illumina has made forays into both the single-cell and spatial analysis spaces. It acquired Fluent Biosciences, a player in single-cell, back in July 2024, and management has said they will roll out a technologically sophisticated and competitively priced spatial platform in 2026.

The company sits in a privileged position when it comes to developing a product roadmap: It has a good sense of where sequencing prices are likely to trend over time and is embedded within its customers’ workflows. Its insights into where sequencing prices are headed is what led to the incubation of Grail in 2016: multi-cancer early detection from a blood test didn’t make economic sense under current sequencing prices, but it would a few years down the line as prices continued to fall.5 Consequently, Illumina gave Grail heavily discounted pricing based on where management believed prices would land.

Of course, Grail wasn’t the only company working on cancer detection tests, which is why the FTC and European regulators objected to Illumina’s acquisition just a few years after initially spinning out the business. The conflict between Illumina and its non-Grail early cancer detection customers was straightforward to see: the company owning Grail meant it would be incentivized to give Grail preferential treatment over other customers. Illumina was chasing a worthy potential growth avenue, but at the expense of alienating its clinical customer base. If the sequencing giant was getting into early cancer detection, what would stop it from pushing further into the diagnostics market? (As an aside, I think this is a problem Roche will run into with its newly announced sequencer. It may pose a threat to Illumina on the research customer side, but I think those developing clinical tests will be more wary of using a provider with a large pre-existing diagnostics business).

Illumina’s deep familiarity with its customers’ workflows gives it a clear view into what these customers are using sequencers for and which use-cases are growing quickly. Presumably, this is what’s driving its forays into single-cell and spatial analysis. To be clear, Illumina already benefits from the growth of these segments: single-cell and spatial analysis require sequencers as part of their respective workflows. Its single-cell and spatial solutions are not attempts to enter a space it’s not a part of, but instead attempts to handle larger portions of those workflows and grab a bigger slice of the profit pools.

The degree to which Illumina succeeds in these new areas depends upon whether single-cell/spatial adoption is constrained primarily by pricing and ease of use or by functionality. Its single-cell product portfolio is far smaller than 10x Genomics’ and much more limited in the sample types it can accept. If gaining a foothold in the market near-term is dependent on functionality, then Illumina loses. If, however, the key bottleneck for market adoption is pricing and ease of workflow, then Illumina has a real opportunity to not only penetrate the single-cell/spatial markets but also (and critically!) drive incremental demand towards its sequencers. This is especially the case in 2025, when almost every sequencing/single-cell/spatial provider has cited elongated instrument purchase timelines as a revenue headwind. Unlike 10x Genomics, Fluent Biosciences’, and hence Illumina’s, single-cell solution doesn’t require the purchase of additional expensive equipment beyond a sequencer:

Not requiring an expensive instrument purchase is a significant advantage in a period of general macro uncertainty and NIH funding uncertainty. This gives Illumina a real leg-up against the current heavyweight in the single-cell space. 10x Genomics’ CEO of course believes success depends on more than just pricing, but he is also adamant that lower pricing unlocks further demand. This is exactly why 10x released cheaper consumables and instruments in 2024 that initially serve as revenue headwinds:

“Within our single cell portfolio, we introduced a number of new products that deliver a lower price per cell and per sample. We believe there is a great potential to grow single cell revenue by reducing prices to increase volumes over time.” – Serge Saxonov, 10x Genomics CEO, Q42024 Earnings Call

The final point to make on the single-cell and spatial opportunities is why they present more interesting routes for Illumina as compared to multi-cancer early detection tests. As discussed above, the challenge with developing MCED tests is that it put Illumina in direct conflict with a set of customers, which in turn sent a worrying message to all clinical customers that their sequencing provider might eventually compete with them. This dynamic isn’t present in single-cell/spatial, where Illumina is competing with its partner vendors rather than its own customers. If anything, Illumina offering additional parts of the single-cell/spatial workflow is helpful to its customers because they have one less vendor to deal with in the procurement process.

As much as Illumina harps on its instrument differentiation and lower cost of overall sequencing workflow on earnings calls, I find it hard to believe management hasn’t been aware for quite some time that they’re facing a more competitive market going forward. This is why the company tried to acquire PacBio back in in 2018, acquired and then divested Grail in recent years, and acquired Fluent a little under a year ago: these are all attempts to create new growth vectors. At the same time, management’s incentivized to keep reasonably quiet about any excitement in the single cell/spatial arena. It’s challenging to not be in regulators’ firing lines when you have 80% market share in the market you predominantly serve. The upshot of this dynamic is that single-cell/spatial would have to contribute meaningfully to Illumina’s bottom line before coming up with any real frequency during earnings calls and investor presentations.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

Disclosure: as of 5/6/25 I have a small position in Ilmn. Somewhat embarrassingly I then sold it on 5/13. The China risk, combined with research consumables down in the mid to high single digit range in Q1, means that growth just needs to be a lot higher in other areas to compensate for those headwinds. I need to have more insight into China/the NIH/U.S. govt research budget to be comfortable holding it here.

Elevidys is currently the only FDA-approved gene therapy for treatment of DMD, an approval that wasn’t without controversy. The therapy failed to cause a statistically significant change in the primary endpoint measured (NSAA, a test to evaluate motor abilities), but did lead to improvements in other metrics such as time to rise from floor and time to climb four steps. While Elevidys presents a step forward, there’s still much be desired when it comes to treatment.

Fortunately, Sarepta’s not the only company working on therapeutics for the disease, and several others (Solid Biosciences, Regenxbio, and Genethon) are working on gene therapy treatments that are structured differently from Sarepta’s.

Solid Biosciences has a good overview of how gene therapies work in general:

“Gene transfer, a type of gene therapy, is designed to address diseases caused by mutated genes through the delivery of functional versions of those genes, called transgenes. The transgenes are then utilized by the body to produce proteins that are absent or not functional prior to treatment, potentially offering long-lasting clinical benefit. In addition to a transgene, our gene transfer candidates include a viral capsid or vector (a protein shell utilized as a vehicle to deliver a transgene to cells in the body) and a promoter (a specialized DNA sequence that directs cells to produce the protein in specific tissues).”1

The theory behind a gene therapy for Duchenne is straightforward: if you can deliver a functional version of the dystrophin gene then symptoms should markedly improve. The practice is complicated, however, by the large size of the dystrophin gene (it’s the largest in the human body), which means companies and researchers are forced to make a judgement call on which portions to include.

This judgement call is somewhat simplified by the existence of Becker’s Muscular Dystrophy (BMD), a milder form of muscular dystrophy where patients have a much longer life expectancy and progressive muscle weakness is not as severe. BMD patients do produce some dystrophin, but in a shortened form. These patients enable researchers to back into a shortened gene construct that could prove effective for those with Duchenne; getting DMD patients to a place where their symptoms present as a mild form of BMD would be an enormous step forward. Unfortunately, the size of the dystrophin gene in these BMD patients is still too large to be inserted into an AAV capsid (which is why the transgene designed by Solid Biosciences differs from the transgene designed by Sarepta differs from the transgene designed by Regenxbio), but nonetheless serves as a useful roadmap.

Subscribe now

The viral vector and promoter choice are other critical aspects of designing a gene therapy. If you get the promoter choice wrong the gene will be translated in the wrong tissue of the body; if you get the vector choice wrong you can end up delivering the gene to the wrong cells and causing a severe immune response in patients. The below slide from a Regenxbio presentation gives a good overview of how Sarepta and its competitors differ in AAV vector, promoter, and transgene construction:2

When thinking about choices on transgene construction, it’s useful to keep in mind that proteins don’t act in a silo but instead interact with each other. Dystrophin plus the proteins that help it do its job form a larger dystrophin-associated protein complex (DAPC). Getting dystrophin expression in patients with DMD is hopefully fruitful, but you also want the proteins that typically interact with the dystrophin protein to be able to do so! This is done by including certain protein-binding sites that are part of the dystrophin gene. Decisions about transgene design, then, really come down to decisions about which proteins you think it’s most vital for dystrophin to be able to interact with. [Those less scientifically inclined should feel free to skip the next two paragraphs. The bottom line is companies make different choices about which protein-protein interactions to prioritize, which in turns impacts the therapy’s efficacy]

Genethon’s transgene construction doesn’t differ from Sarepta’s, which is explained by the seven year research agreement and five year license & collaboration agreement between the two organizations for Genethon’s DMD program. Regenxbio’s management emphasizes their decision to include the CT-domain as part of its transgene. The CT-domain comes at the tail-end of the dystrophin protein, isn’t included in any competitors’ transgenes, and plays an important role in interactions with the α-dystrobrevin protein and five syntrophin proteins (all parts of the DAPC). The hope behind including the CT-domain is three-fold: in mice it increased the half-life of the therapy (the longer the capsid-transgene combo remains in the muscles the more functional protein that can be expressed!), enabled better muscle recovery, and improved restoration of the DAPC (which makes sense; it’s hard to have the entire complex if the associated proteins aren’t able to bind anywhere).

Solid Biosciences’ transgene doesn’t contain the CT-domain, but instead includes the nNOS binding domain. nNOS (neuronal nitric-oxide synthase) is an enzyme that’s believed to ensure enough blood gets to the skeletal muscles during exercise. This is something that doesn’t happen for those with DMD and is a direct contributor to muscular dystrophy rather than only a coinciding event.3 This enzyme can’t be recruited without a-syntrophin, a protein that’s part of the DAPC and can’t bind to dystrophin without spectrin-like repeats 16 and 17. Regenxbio’s construct allows for syntrophin interaction at the CT-domain, but its lack of R16/17 (which can be seen in the above slide) means its construct can’t recruit nNOS.

This can all sound quite abstract, but the reason Solid’s choice is so interesting is because the FDA specifically flagged Sarepta’s construct as not including R16/17:

“The extreme truncations in Sarepta’s microdystrophin [microdystrophin is defined as any shortened version of the dystrophin protein] protein result in absence of important functional domains. For example, Sarepta’s micro-dystrophin does not bind either neuronal nitric oxide synthase or α-syntrophin, two proteins known to play a synergistic role to protect muscle cells. Recruitment of neuronal nitric oxide synthase by wild-type dystrophin at the sarcolemma through spectrin-like repeats 16 and 17 (R16/17) helps control local blood flow by antagonizing sympathetic vasoconstriction. It is therefore unclear to what extent Sarepta’s micro-dystrophin can function similarly to wild-type dystrophin or to shortened forms of dystrophin in patients with BMD.”4

At least currently, Solid including R16/17 seems like a good bet. Its day 90 data for the first three candidates (note the small sample size!) in the Phase 1/2 trial was very positive, with microdystrophin expression levels higher than what Sarepta and Regenxbio reported at the same point in time.

There’s also a potential scenario where Solid’s therapy doesn’t end up succeeding but there’s tremendous value created through its proprietary AAV-capsid. SGT-003, Solid’s current gene therapy candidate, is the company’s second attempt at developing a gene therapy to treat DMD. Its first attempt, SGT-001, had a similar focus on nNOS and a-syntrophin but utilized a different capsid, AAV9. SGT-001 was plagued by safety issues, which led to frequent pauses in the clinical trial and eventually a more permanent program pause in 2022.5 SGT-003 uses a capsid designed in-house, AAV-SLB101, that has improved skeletal muscle tropism (ability to infect the skeletal muscle) when compared to AAV9. Improved tropism matters because an AAV capsid is a virus and so elicits an immune response. The better the tropism, the lower the dose can be, and so the less you have to worry about an immune response affecting the therapy’s efficacy/safety. Solid has licensed out this proprietary capsid to 15 academic labs and corporations, and so stands to benefit enormously if these labs/corporations bring a therapy using this capsid to market. The company’s additionally putting effort into developing capsids to tackle cardiac and neuromuscular diseases, with a similar intent to license them out if successful.6

Sarepta’s CEO has been overly dismissive of any competitors’ efforts on earnings calls, but in doing so he makes a few good points on the company’s short-term advantages.7 As exciting as Solid’s results are, so far only 6 patients have been dosed with the therapy, which means there’s very limited data on both safety and efficacy. That puts those with DMD in a challenging situation: choose a therapeutic with a better known safety-profile that has limited efficacy or choose one that doesn’t yet have a ton of safety data but could end up being a better treatment.8 Moreover, Sarepta has a real distribution advantage given its years of serving DMD patients through exon-skipping therapies. Its existing relationships with infusion centers, which are used for both exon-skipping and gene therapy treatments, means the company is just better positioned than competitors when it comes to rolling its gene therapy out at scale. Furthermore, even if Regenxbio and Solid both have clinical trials that go well, the longer they take to get to market the more existing DMD patients Sarepta has the opportunity to dose. Gene therapies are intended to be one time, in large part because a patient’s immune system will recognize the AAV virus the second time and do a much more effective job at stopping it from reaching the intended cells. Solid management argues that its capsid is differentiated enough from Sarepta’s AAVrh74 capsid that redosing opportunities may exist for the patients who have already received Elevidys, but then one runs into the question of whether insurance will cover a second treatment that costs in the low millions, particularly if it displays a questionable level of efficacy.9

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

Private equity funds at least somewhat solve the problem of investors acting in their short-term interests to their detriment of their long-term ones. No matter how poor a fund’s performance might be in any given year, a multi-year lockup is a multi-year lockup. There are, however, a few drawbacks to this model. The first is that, in addition to your track record, raising another fund is inevitably influenced by the macro/fundraising environment of that point in time. The second is that a big chunk of compensation comes from realized gains once an investment is sold or goes public. These gains take time to materialize and are inevitably lumpy in nature. That’s not necessarily ideal as the partner of a privately held investment fund, but it’s even less ideal as the partner of a publicly traded one. When a chunk of a employee compensation is tied to a company’s share price, the market’s perception of that company as lumpy/difficult to forecast is going to cause the company to look for fixes to that perception.

Blackstone, a name synonymous with private equity, is a company that’s done just that. The firm was founded in 1985 as an M&A advisory business, and from there expanded into private equity, hedge funds (including a fund of funds), real estate, and credit.1 It makes money, as all private equity firms do, by charging both a management fee that’s a percentage of assets and a performance fee for returns above a certain hurdle. (Note: Blackstone manages assets on behalf of insurers within its credit portfolio. It doesn’t itself operate an insurance business).

Increasingly, Blackstone has been raising money for what it terms “perpetual capital” vehicles. These vehicles span all of Blackstone’s business segments, and can be broadly split between offerings for retail investors and offerings for institutions. Blackstone caps redemptions for its perpetual retail products at a certain percentage of net asset value per month, and institutions can’t redeem capital from these vehicles unless new capital comes in to fill their gap. Given that these funds are perpetual, performance fees are charged on unrealized, rather than realized gains (these fees are defined as “fee related performance revenues” when Blackstone reports earnings.) The frequency of these performance fees vary (As examples, BREIT, Blackstone’s real estate product for retail investors, charges performance fees once a year, Blackstone’s Infrastructure fund for institutions charges performance fees every three years). Today, perpetual capital under management sits at 444.8 billion, or ~40% of overall assets.

It’s worth exploring the challenges that these perpetual capital vehicles do and do not solve. They do mitigate the vicissitudes of the fundraising cycle and they do mean that Blackstone has a more stable stream of income. That said, there’s still a large amount of variance in those recurring streams. As a case in point, fee related performance revenues for the company’s real estate segment were up 401% from 2020-2021, then down 37% from ’21 to ’22, down another 73% from ’22 to ’23, and then down a further 31% from ’23 to ’24. Performance fees work well when the underlying funds perform well, but less well when the funds don’t! Moreover, perpetual capital vehicles are unfortunately not able to totally prevent investors pulling money over short-term performance concerns. When BREIT withdrawals picked up in late 2022 to more than 2% of NAV per month, the company announced it was limiting withdrawals (something it is very much allowed to do!). This is turn caused increased media coverage, a drop in Blackstone’s share price, and further requests for withdrawals that exceeded the 2% NAV limit until April 2024.

Subscribe now

The life of an asset manager also follows an arc where one typically generates less alpha as fund size becomes larger. There are multiple possible explanations for this, but the explanation based purely on AUM growth rather than the portfolio manager now having enough money that he/she no longer cares as much about performance is that a larger fund size just makes deploying capital harder. This is most clearly the case with a successful microcap public equities investor: part of the reason microcaps are such an inefficient space is because most funds are too big to justify spending any time looking at companies with a sub 250mm market cap. A microcap manager whose fund grows in size must now spend his/her time looking at a basket of stocks that are much more efficiently priced, so substantial outperformance becomes a lot harder.

Given that its AUM now sits at over a trillion dollars, Blackstone has had to think about the trade-offs between fund size and performance more than most. Management argues that its scale confers a few advantages:

1. It can be one of the only seats at the table for extremely large deals.

2. There’s a lot of cross pollination that can happen across business segments. The real-estate teams can talk to the hedge fund teams can talk to the infrastructure teams, etc etc. This leads to insights that a single asset class firm wouldn’t uncover.

There’s a real question as to whether being one of the few seats at the table is always an advantage. You’re able to do deals that few investors can do, but that means there are few investors at the table when you try to exit an investment, which becomes an especially acute problem when the IPO market is tepid. Nonetheless, there do seem to be market segments, such as healthcare, where Blackstone’s size confers true benefits. Its pure life sciences business began in 2018 with the acquisition of Clarus, and is primarily focused on “financing the last mile of clinical development of pharmaceutical products.” In other words, Blackstone is funding things like drugs in phase 3 clinical trials, where the science risk has been significantly reduced but there’s a substantial funding need. There’s a considerable supply/demand gap in life sciences R&D, so it’s a space where Blackstone’s scale means it’s one of the few seats at the table and that it’s meeting a funding need that wouldn’t be met otherwise. The exit opportunities are also different. The downside of Blackstone being one of the few seats at the table for a private-equity deal is that there are then only a few seats at the table when Blackstone wants to exit this deal. On the life sciences side, however, the exit opportunities are more binary and less subject to market conditions. If the clinical trial/insurer reimbursement process is a success, Blackstone makes a lot money. If the clinical trial or reimbursement process isn’t a success, then Blackstone makes very little. The cross-pollination opportunities in life sciences are very real. Blackstone owns a significant amount of life-sciences real-estate (although that’s been a performance headwind as of late!), and has purchased two clinical trials related businesses out of its private equity funds. That’s not to say cross-pollination doesn’t have its downsides – if Blackstone gets its AI bets wrong its real-estate funds (via data-center investments), infrastructure funds (via data-center and power generation investments), and private credit funds (via its financing of CoreWeave) will all suffer.

There’s a third advantage to Blackstone’s scale that is the most relevant to its business going forward: it enables the long-term growth of its private wealth and insurance/credit segments. Historically, private wealth has been under allocated to private markets when compared to institutional money (Blackstone quotes that institutional investors are 30% allocated as compared to private wealth’s 1-2%). Some of this difference is structural: Wealthy families and individuals have greater liquidity needs than a university endowment! Even so, a portion of this difference is because there hasn’t been an easy way for individuals below a certain net worth to invest in privates. Consequently, Blackstone’s put an enormous number of resources into developing this business, an effort that began well before its competitors in 2011. There’s good reason to think the firm is exceptionally well positioned to succeed here, as management has pointed out:

“I think the one advantage, I’d say in this [the private wealth] market versus the institutional market, there you can have thousands and thousands of individual private equity firms or real estate firms, credit firms. I think when you get to private wealth, the brands are going to matter, the scale, the ability to service. And I think it’ll be a smaller number of players in that segment. It’ll grow over time, but it requires something different. And we have a pretty meaningful first mover advantage, $240 billion of total assets, and we are absolutely committed to delivering great performance and great service to the underlying customer. So, we recognize it’s going to be more competitive. Others will try to do things in the marketplace.”2

If you’re a small private equity firm, it just doesn’t make sense for you to spend time building out a private wealth segment. Not only are you unlikely to have the capital or patience or distribution know how to build relationships with a substantial number of financial advisors or navigate regulatory hurdles, but you don’t have a sufficiently recognizable brand name. Even if there were no distribution/regulatory challenges to figure out, there’s little reason for an advisor to recommend an unknown fund to its clients. “No one ever got fired for buying IBM” can easily be changed to “No one got fired for investing in Blackstone.” Moreover, Blackstone’s existing private wealth presence means it’s relatively easy to raise money for any additional funds launched in this channel: BXINFRA, its infrastructure product for private wealth, launched with over 5x the assets of its competitors’ products, and “over 90% of advisers who allocated to this strategy had previously allocated to another Blackstone perpetual vehicle and over 50% allocated to all 4 of our perpetual flagships.”3 It’s hard to imagine a scenario where a shift in private wealth allocations towards illiquids doesn’t end up seriously benefiting all the large players, but especially Blackstone given its early start. I think this holds even if fund performance dips somewhat (although there’s obviously a limit to this). The private wealth category is essentially a way for Blackstone to convert its years of generating alpha for institutions into an ‘alpha-lite’ product serving a massive TAM.

There are similar dynamics present in Blackstone’s insurance business, which now sits at 230 billion in assets. Blackstone itself is not an insurer (unlike Apollo, which owns Athene), but instead manages money on behalf of insurance companies. Importantly, this insurance money is the stickiest of any money Blackstone manages:

“Investment advisory agreements related to certain separately managed accounts in our Credit & Insurance and Hedge Fund Solutions segments, excluding our separately managed accounts in our insurance platform, may generally be terminated by an investor on 30 to 90 days’ notice. Separately managed accounts in our insurance platform can generally only be terminated for longterm underperformance, cause and certain other limited circumstances, in each case subject to Blackstone’s right to cure.”4

While this money is sticky, it’s invested in Blackstone’s credit strategies, and so is lower fee. These lower fees end up as an advantage. For managing insurer’s money to be worth the squeeze you need two things: the ability to manage truly massive amounts of capital, and the ability to originate not only a substantial volume of private credit deals but a volume of high-quality private credit deals that will actually generate outperformance. Said differently, this just isn’t a business line that an average private equity or private credit firm can decide to get into, no matter how much excitement there may be about the asset class in general! Brand name (and the firm’s longevity) is vital here, too. Insurance is not a business where one is supposed to take large amounts of risk! Like private wealth, this is a case where it’s just not worth taking the career risk to invest an insurance company’s capital in an upstart private credit firm.

The big risk that Blackstone and the other dominant players face with private credit is that success in the business is reliant upon the amount you can originate. This creates a natural incentive to let origination quality suffer in the name of growing assets, a strategy that will fare especially poorly when managing money on behalf of the insurance industry.5 This incentive only increases as the asset class gets more competitive, a dynamic that is very much present now! In the past few months, Blackrock spent 12 billion to acquire HPS, JP Morgan announced it would invest 50 billion of its capital in private credit deals, and Goldman CEO David Solomon said private credit was “one of the most important structural trends taking place in finance.” Long-term underperformance has very real implications for insurers, and the end result of a decline in private credit performance is that they get spooked and pull money once they can.

It's a shame I didn’t write this note back when in 2012, when Blackstone was trading at 10x earnings rather than the ~43x it trades at today. There’s of course much to like about the business. Schwartzman refers to Blackstone as “the reference firm in our industry”, which really is true. Jonathan Gray, the firm’s COO, described much of its strategy as a “commitment to our capital-light, brand-heavy, open architecture model.”6 In other words, an approach similar to a former portfolio company, Hilton. Blackstone’s reputation gives it a tremendous advantage in growing its private wealth, insurance, and more traditional business segments. Its move towards perpetual capital means that a significant portion of revenue is at the very least recurring-ish. That doesn’t however, mean that the firm isn’t quite dependent upon the overall macro environment for its performance, and, as seen above, that both management and performance fees don’t drop dramatically (and for a prolonged period, which inevitably impacts fundraising in those years) when portfolio performance is poor. If the macro over the next year is smooth things probably look very good: real-estate performance improves, private wealth doesn’t get spooked, defaults stay low so money keeps coming into credit, and the IPO market opens up again so realized performance revenues soar.7 This could very well happen, and I don’t doubt that private wealth and private credit are huge opportunities. But I do wonder whether 43 times earnings is the right price to pay for a business that can have large amounts of lumpiness should the macro go south.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

But long-read sequencing has been, and is, the better option for certain types of genomic analysis:

• Assembling entire genomes. Genomes are substantially longer than anything a sequencer can read in one run, and so need to be split into pieces, run through the sequencer, and then re-assembled. As one would guess, an 100 piece puzzle is easier to put together than a 10,000 piece puzzle. Assembling a genome (or transcriptome) with short-read sequencing means you have less overlap between individual strands, which makes it harder to put the pieces together accurately and causes gaps/errors in the final product.

• Reading highly-repetitive sequences. This comes in especially handy for agricultural use cases (wheat has a genome with over 85% repeat content!), and rare diseases where the mutation causing the disease is present within an often repeated sequence.

• Detecting structural variants, which are defined as genomic alterations larger than 50 base pairs. In the past this benefit was offset by long-read’s overall inaccuracy, which made it a very poor tool for detecting any non-structural variants (“Is this a single point mutation or did our machine just make a mistake?”)

• Phasing – Humans have two copies of each chromosome, one from each parent; each copy is referred to as a haplotype. Phasing refers to the separation of these inherited copies into haplotypes. It’s especially useful for the diagnosis of rare diseases like Angelman’s, which is caused by a mutation in the chromosome inherited from the mother.

• Methylation patterns. Long-read sequencing measures methylation patterns by default. Short-read sequencing requires additional library prep.

On top of this, there have been large strides made in both long-read cost and accuracy. The cost of sequencing a human genome using this method has come down from thousands of dollars to under a thousand, and accuracy has improved dramatically from an error rate of 10% to one of 0.1%.1

PacBio has long been the dominant player in long-read sequencing, a truth one wouldn’t recognize from looking at the company’s stock price, which is not only down ~76% over the past year but down 90% from when it went public in 2010. Its business model is essentially the same as its short-read sequencing peers: customers buy sequencers, which are infrequent/lower-margin purchases; these sequencers need consumables for each sequencing run, which PacBio sells at at a much higher margin. Said differently, the business is somewhat analogous to Apple, where infrequent hardware purchases enable the higher-margin, recurring revenue. Unlike Apple, however, sequencing companies are continually releasing machines that are not only better, but markedly cheaper. The bet here is that the more affordable sequencers become, the more demand for sequencing will increase. This increased demand has two components: the first is that institutions/companies that don’t currently have the balance sheet to buy a machine will do so once the cost comes down enough. The second is that existing customers have plenty of additional sequencing runs they’d currently like to do but can’t yet afford given each run’s cost. The way this bet plays out in sequencing companies’ financial statements is straightforward: margins are lower in years when a new sequencer is released, and consumables revenue for these new sequencers then takes a few years to fully ramp up.

Subscribe now

Over the past few years, there’s been some uncertainty from investors that lower sequencer prices will continue to spur increased demand, as evidenced by analyst questions on both Illumina and PacBio earnings calls. This uncertainty illustrates the challenge of new product launches coinciding with a tougher macro environment: it’s easy for management to put extended sales cycles and lower than expected annual pull-through down to macro troubles, but it’s possible that there’s genuinely less of an appetite for sequencing than PacBio/Illumina anticipated.2 In PacBio’s case, it’s difficult to make a determination either way. On the long-read side, the company released two new machines over the past two-ish years: Revio, the latest version of its traditional sequencer, and Vega, a much more affordable, benchtop, option that’s lower throughput and has a higher-cost per sequencing run.3

The argument that PacBio overestimated long-term demand for long-read sequencing rests on comments from the company’s Q1 earnings call, when Revio utilization came in lower than expected:

“In Q1, more customers than we expected utilized their Revio systems at less than 20% capacity. Many of these are newer customers and the average age of the systems in this category was less than four months, which we believe indicates that these customers are still early in their ramp up. The pace of the ramp is dependent on a number of factors including, the timing of sample availability, lab readiness, and funding among other reasons. We are putting measures in place that we believe will help these customers ramp to their full utilization as timely as possible, and help drive consumable growth going forward.”4

It could be true that these customers are simply taking longer to ramp; it could also be true that these customers won’t actually ramp up to the expected levels. The recently announced Revio price cuts and the release of the much cheaper Vega option might be evidence of management realizing the latter is true. The worst case scenario (albeit an unlikely one) here would be something like what happened to Blackberry, where price cuts fail to stimulate further demand and instead result in less money to do R&D, which is a vital component for a sequencing company!5

I think the best counterargument to the above is what happened when PacBio initially released Revio in 2022. Revio actually cannibalized sales of its Sequel II sequencers faster than expected, to the point that management had to make excess inventory adjustments on the income statement. That indicates that at least some of PacBio’s customer base has a real appetite for higher throughput machines that can support increased research activity. Company management has emphasized the work that’s been done to lower the DNA input requirements for Revio and simplify the library preparation steps for sequencing runs; these are efforts that should both widen the addressable market and increase annual pull-through per sequencer. Lowered input requirements substantially expands Revio’s use cases (it’s now easier to sequence DNA from heel pricks and saliva samples), and easier library prep should result in decreased friction and thus increased customer activity.

Historically, PacBio’s struggled to branch out beyond academic and research applications, which is a real limitation for the business. A big part of the reason Illumina could enjoy success going forward is that the company is partially indexed to the growth of liquid biopsy/MRD market. If you think we’re in the early innings of these tests, then Illumina is a good way to bet on the picks/shovels in the gold rush.6 You can’t analyze a liquid biopsy sample if you don’t have a high-throughput, short-read sequencer! Moreover, diagnostics companies are very unlikely to change sequencing providers once test development is under way, and even more so once a test receives FDA approval. Unfortunately for PacBio, long-read sequencers are not used in the liquid biopsy/MRD workflow. Management is aware of this, and so developed its own short-read capabilities by acquiring Omniome in 2021 and Apton Biosystems in 2023. Its short-read sequencing system, Onso, was launched in August of ’23 with a sales pitch focused on the system’s accuracy. Illumina’s NovaSeq X systems advertise an error rate of Q30 (1 error per 1000 bases); PacBio’s Onso advertises one of Q40 (1 error per 10,000 bases). Accuracy of course matters, and Natera expressing interest in Onso is a positive signal, even if such interest wouldn’t result in real revenue for several years.7 That said, Illumina brings more to the table than just accuracy (as this reddit post goes into). Furthermore, now is a particularly inopportune time to get into the short-read sequencing market. Everyone, including the VC world, has woken up to the success of Illumina. Element, Ultima and BGI Genomics are all going after the short-read space, something that wasn’t happening to any notable degree pre-Covid. I think the best reason to believe PacBio has an advantage over venture-backed players is that its historic work in long-read puts it in a better position to understand what customers are looking for other than throughput, accuracy, and cost (eg. sequencers break often. If you don’t have a great customer service team you’re going to struggle long-term in this business). The fact that the company’s been around for a while also helps: it’s challenging to buy more consumables if your early-stage sequencing provider goes out of business!

It's not necessarily easy to figure out a valuation that makes sense for PacBio. On an EV/’25 Revenue multiple it’s valued remarkably like Illumina, at 5.7x as compared to Illumina’s 5.61x. This doesn’t make a lot of sense (although the organizations are similar in that executive turnover has been a cause for concern at both!). PacBio has negative EBITDA (Illumina’s is positive), its full-year revenue was down 23% YoY (Illumina’s was down ~2%), its short-read platform is far from proven out, and in Q1 of ‘24 management abandoned its 2026 target of at least 500mm in revenue. It’s true that Illumina fumbled the bag with its GRAIL acquisition and is facing increasing competition, but it’s still the king of short-read and powers liquid biopsy/MRD tests for the likes of GRAIL and Natera. PacBio sits in a fundamentally different position: it primarily serves the much smaller long-read market and is years out from its short-read sequencing efforts bearing real fruit. I think the stock would likely pop if management announced plans to abandon its short-read efforts and use that money for debt paydown, but that’s a strategy that could easily be taken too far. The sequencing industry is a prime example of a space where if you cut R&D too much you’ll quickly approach obsolescence.

Correction: In the first version of this research note I wrote “I think the stock would likely pop if mgmt announced plans to abandon its short-read efforts and use some of that money for buybacks.” As two people were quick to point out on Twitter that’s far from the smartest move in the world given the company’s debt levels! The general point is that it’s difficult to pull money away from R&D for buybacks or debt given how critical innovation is in this space.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence

Sarepta Therapeutics is one of these companies, and its first three commercialized DMD treatments leverage exon skipping (considered an RNA targeted therapeutic) rather than gene therapy. Exon skipping is relatively straightforward, at least conceptually. DMD is caused by a mutation in one of the seventy-nine exons that comprise the dystrophin gene.1 In the case of Duchenne, this exon mutation prevents any dystrophin protein from being produced. As the name suggests, exon skipping skips over this mutated exon during pre-mRNA splicing, enabling the production of a functional, albeit shortened, dystrophin protein.2 Sarepta has three different exon skipping treatments that differ by the exon that’s skipped: EXONDYS 51 skips exon 51, VYONDYS 53 skips exon 53, and AMONDYS 45 skips exon 45.

Exon skipping has a few downsides, the most notable of which is that the treatment isn’t suitable for patients with a dystrophin gene mutation that doesn’t involve exon 45, 51, or 53. Unfortunately this segment of the patient population is significant, with 71% of those with DMD have a mutation that current exon skipping therapeutics aren’t suitable for. The second, and lesser, downside is that exon-skipping treatments are administered weekly.

Consequently, Sarepta has also focused its efforts on developing a gene therapy, ELEVIDYS, that can treat a much larger percentage of the patient population. Taking a gene therapy approach to DMD presents a bit of a challenge given the size of the dystrophin gene, which as far as we know is the largest in the human body. As one would expect, Sarepta used an AAV vector as the delivery mechanism; AAV vectors are favored in gene therapy because they typically don’t cause a massive immune response, but are limited by the size of the gene they can carry.3 The workaround is to use a truncated version of the coding portion of the dystrophin gene, but that comes with the downside of the treatment not being curative. Given the limited nature of existing DMD treatments, this is a trade-off worth making. ELEVIDYS comes with the added benefit of being a one-time treatment, rather than one that requires repeated dosing.

Sarepta received initial FDA approval for ELEVIDYS in June 2023, but only for those between the ages of 4 and 5. Importantly for everyone involved, this FDA approval was expanded this past June for ambulatory and non-ambulatory individuals aged 4 and above, or 80% of patients with DMD in the US. The approval wasn’t without controversy; the primary endpoint measured in the clinical trial was a statistically significant change in the NSAA (North Star Ambulatory Assessment), which is used to measure the motor abilities of those with DMD. Trial participants who received ELEVIDYS did not perform significantly better on the NSAA than those with the placebo, but they did perform meaningfully better on two secondary endpoints: time to rise from the floor and the four-stair climb. Dr. Peter Marks, Director of the Center for Biologics Evaluation and Research at the FDA, had this to say on his approval decision:

“Overall, the demonstrated benefits of ELEVIDYS in the treatment of ambulatory individuals, and the expected benefits of ELEVIDYS in non-ambulatory individuals, with DMD over 4 years of age who are eligible to receive this therapy in improving key functional endpoints such as the ability to stand, walk, or climb stairs, outweigh the risks. The benefit to risk considerations are favorable taking into account the existing uncertainties, such as the ultimate duration of response. Although it might be argued that other gene therapy products in development may prove superior to ELEVIDYS in future clinical trials, these products have yet to receive regulatory approval. During this time, the availability of this gene therapy option may help slow or prevent irreversible decline that might otherwise occur in both ambulatory and non-ambulatory individuals, particularly since the latter have few or no alternative treatments available to address their imminent further decline in function over time.”4

The key takeaway here is that while Sarepta’s gene therapy treatment is a step forward for patients, it remains very far from a cure. This makes it difficult to model the success of the business over time. In the shorter-term things look very good, because there simply isn’t much in the form of competition. Many of the competitors listed in Sarepta’s annual report are focused on exon skipping, which, as said above, is fundamentally limited in its ability to address the patient population. Vertex and CRISPR Therapeutics are exploring Crispr/Cas9 approaches to treat DMD, but neither company has anything in clinical trials yet. Pfizer nixed its DMD gene therapy candidate earlier this year after disappointing phase 3 results; Solid Biosciences and Regenxbio both have gene therapy candidates in phase I/II trials, but even if all goes well neither is expected to submit a BLA filing before 2026. While the next few years look promising for Sarepta, ELEVIDYS’ limited efficacy leaves the company very exposed should a competitor develop a superior treatment.

Subscribe now

ELEVIDYS additionally represents a business model shift for the company from primarily recurring revenue to primarily one-time. The therapy ranks among the most expensive treatments in the US, coming in at 3.2mm per patient. That’s a staggering amount of money, but exon-skipping treatments aren’t cheap either.5 As mentioned, patients receive exon-skipping treatments weekly, and at a cost of 750k-1.5mm per annum. Beyond a few years, ELEVIDYS is the more cost-effective option. I think this business model shift ends up being quite a positive thing for those with rare diseases. The move from recurring to larger one-time revenue means Sarepta is heavily incentivized to continue innovating and has a large amount of cash to do so! That doesn’t mean Sarepta should become a pure gene therapy company. Recurring treatments can just be the right option for treating disease, and it’s unlikely to be a good thing for innovation if Sarepta goes out of business after pivoting entirely to a one-time treatment that a competitor then develops a clearly superior version of. Management is taking a prudent approach here: they don’t expect the exon-skipping treatments to be cannibalized by ELEVIDYS until 2026, and recently announced a collaboration agreement with Arrowhead Pharmaceuticals. Arrowhead develops rare disease treatments that leverage siRNA technology, and so require recurring administration to be effective. This balance of RNA therapeutics with curative gene therapies enables Sarepta to have a reliable stream of income while forcing the company to continue to innovate.

Management has also taken a prudent approach to its gene therapy pipeline, which is currently focused on LGMD, a collection of muscular dystrophy diseases that vary in severity. SRP-9003, currently in phase 3 trials, is a therapy designed to treat LGMD2E, a severe form of muscular dystrophy that affects a far smaller portion of the population than DMD. Importantly, SRP-9003 uses the same delivery vector as ELEVIDYS, which meaningfully derisks the treatment. This is an effective playbook for companies to tackle rare diseases:

1. Start with a gene therapy treatment tacking a disease with a relatively high prevalence rate.

2. Use the vector from that therapy to target diseases that affect the same type of cells, but with lower prevalence rates.

3. Repeat.

When one takes a cursory look at Sarepta’s valuation the natural response would be to conclude it’s extremely expensive, at 76x FY24 estimated earnings. What that misses, however, is what’s expected to be a massive revenue ramp in 2025. ELEVIDYS revenue came in at 181mm last quarter, an increase of 49% QoQ. That revenue figure is expected to double this quarter, and management expects 2025 revenue to come in at between 2.9B – 3.1B, two thirds of which will be ELEVIDYS. Consensus estimates have earnings of $10.99 a share next year, which would mean the stock trades at ~10.7x 2025 earnings. Furthermore, management doesn’t expect peak year ELEVIDYS sales to materialize until later this decade, suggesting further top and bottom-line growth beyond 2025.6 The current valuation makes a lot more sense in light of the sharp expected increase over the next few years.

What gives me pause is how confident one can be in picture of the business beyond 2025. As said above, the FDA’s approval of ELEVIDYS wasn’t without controversy, and the treatment’s sadly far from a cure. This creates plenty of room for Solid or Regenxbio to take share should their treatments prove more effective. It’s difficult to calculate terminal value for a business when revenue for its main product could drop substantially almost overnight! The other risk to the stock (beyond side-effects and reimbursement dynamics), is that investors lose confident in management. The company’s CEO, Douglas Ingram, has expressed an extremely high level of certainty in 2025 revenue estimates.7 That level of confidence means he’ll lose a lot of investor trust if 2025 guidance turns out to be too high.

Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence