Capsida's Trailblazing Moment: What the Field Owes the Next BBB Program
By Fatma Elzahraa Eid, PhD | TheBioMLClinic
Brief disclosure: I am a named inventor on patents and author on publications related to AAV capsid engineering and CNS gene delivery, developed during my time at the Broad Institute. I now operate independently. This post does not represent any prior employer, current advisory client, or collaborator. The mechanistic analysis presented here is my own scientific interpretation of publicly available data. Full disclosures at the bottom.
A child died.
Brain swelling (cerebral edema) within days of receiving a gene therapy that had cleared every preclinical safety bar the field knows how to set. The IND had FDA clearance. The Good Laboratory Practice (GLP) toxicology study showed no adverse histopathology. The capsid achieved greater than 70% neuronal transduction across critical brain areas in nonhuman primates. By every metric the field currently uses, this program was ready.
And yet.
This post is not about assigning blame. Capsida did what the field does, what any of us would have done. They engineered a capsid to cross the blood-brain barrier, validated it rigorously in the best available preclinical models, obtained regulatory clearance, and dosed their first patient. The tragedy is not a failure of their science or their diligence. It is a failure of the field's collective toolkit: the questions we do not yet ask routinely, the assays we do not yet run as standard, the biology we do not yet examine before we commit to a receptor.
That is what I want to address here.
Capsida's program is the first to reach this frontier at clinical scale. They will not be the last. The field has a responsibility, now, before the next BBB program doses its first patient, to learn everything this moment is trying to teach us.
What happened, in plain language
CAP-002 was designed to treat STXBP1 encephalopathy, a devastating neurodevelopmental condition with no disease-modifying therapies. Wild-type AAVs cannot achieve the brain-wide neuronal coverage this disease requires. Capsida's engineered capsid solved that problem: it crossed the blood-brain barrier efficiently and transduced neurons broadly, as demonstrated in NHP studies showing greater than 70% neuronal coverage across critical brain regions.
The first dosed patient, a child, died of cerebral edema days after treatment. The autopsy confirmed brain swelling as the cause of death but did not conclusively establish the underlying mechanism. Capsida subsequently disclosed, just last week at ASGCT 2026, that their capsid uses ADAM15 as its primary receptor for crossing the BBB, and that they believe receptor biology is a key factor in what went wrong. They are now developing a next-generation capsid using a different receptor, RECK.
The root cause remains unconfirmed. What follows is my scientific interpretation: a working hypothesis built from the published biology of ADAM15, AAV immunogenicity, and CNS immune responses. I offer it not as established fact but as the analytical frame I believe the field needs to take seriously as it designs what comes next.
A note on epistemic boundaries
Before going further, I want to be explicit about what this post is and is not.
The autopsy confirmed cerebral edema as the cause of death. It did not identify the mechanism that produced it. No public evidence has established a direct causal chain from ADAM15 engagement to the edema. Multiple biological mechanisms are plausible contributors, and I will address them in turn.
What I am offering is a framework for asking better questions, informed by the receptor biology, the immunogenicity literature, and the structural gaps in current preclinical practice. The intent is not to assign causality to any single pathway but to identify which biological variables the field should be evaluating more systematically before the next program reaches the clinic.
I am a computational scientist with more than twenty years of experience working mostly on biological problems, including AAV capsid engineering. The mechanistic dimensions of this analysis should be read in that context, and alongside input from immunologists, clinicians, and CNS biologists who bring expertise I do not.
What could have caused it: the full mechanistic landscape
Before making any argument about receptor biology specifically, intellectual honesty requires addressing the full space of plausible contributing mechanisms. The underlying cause has not been established. Any of the following could have contributed, and they are not mutually exclusive.
Innate immune activation via TLR2 and TLR9. These are the two best-characterized innate immune sensors for AAV vectors, and they work through distinct mechanisms. TLR2 recognizes the capsid surface as a pathogen-associated molecular pattern (PAMP) and is expressed on endothelial cells, liver nonparenchymal cells, and monocytes; activating NF-κB and driving proinflammatory cytokine production including IL-1β and IL-6. TLR9 operates differently: it sits inside the endosome and detects the unmethylated CpG motifs in the vector genome, triggering interferon-α release from plasmacytoid dendritic cells and priming T-cell responses. In the CNS context, microglia express both TLR2 and TLR9, meaning that a high-efficiency BBB-crossing capsid delivering a large genome-containing load into brain tissue engages both sensing arms simultaneously in the most immunologically sensitive tissue compartment in the body.
Importantly, the published kinetics of TLR-mediated innate responses to AAV are rapid but transient: cytokine and interferon transcripts rise within hours of administration and largely resolve within six hours for single-stranded vectors. This timing makes TLR2/TLR9 activation an unlikely direct cause of a death that occurred days after dosing. Their role in this cascade, if they played one, is more plausibly that of an early initiating signal: priming the inflammatory environment in the hours after dosing, amplifying downstream complement activation and endothelial stress responses that then build over subsequent days. Neither TLR pathway is confirmed as causative here. Both represent well-documented biological risks for this class of program that are worth screening for at the library design stage, precisely because their early activation may set the conditions for what comes later.
Complement activation. High-dose intravenous AAV is a documented activator of both the classical and alternative complement pathways; observed clinically across multiple programs at doses above 1×10¹³ vg/kg, manifesting as thrombotic microangiopathy, acute kidney injury, and severe inflammatory responses. Critically, complement activation follows a different timescale than TLR-mediated innate responses: alternative pathway complement antigens have been detected remaining elevated through day 7 after AAV infusion in clinical monitoring data, placing complement activation squarely within the days-long window consistent with a death occurring a few days post-dosing. There is also a dose-dependent dimension: at lower doses, AAV capsids actively bind complement regulatory proteins including factor H and iC3b, partially inhibiting the cascade. At the high doses required for brain-wide neuronal coverage, that self-regulatory capacity appears to be overwhelmed, which may partly explain why lower-dose NHP studies did not predict a complement-driven adverse event.
One piece of context that is striking: at doses above 1×10¹³ vg/kg, a systemic AAV infusion delivers a capsid surface area exceeding 20 square meters, comparable to nearly half the pediatric lung surface area. The innate immune response to that physical surface alone, independent of any receptor interaction, is a potent stimulus. In the CNS specifically, activated complement contributes to cerebral edema through membrane attack complex formation, causing BBB disruption and brain cell swelling, a mechanistic link that is documented in the CNS injury literature. Whether complement activation was a contributing factor to the edema in this case is unknown. That its timing, dose-dependence, and CNS consequences make it a biologically plausible contributor and a standing risk for any high-dose pediatric CNS IV program, is established.
Pre-existing immunity and NAb-complement interaction. A mechanism I did not initially include and that expert feedback has rightly flagged as a real omission is the role of pre-existing neutralizing antibodies in amplifying complement activation. A large fraction of the human population, estimated at 40–70%, carries pre-existing anti-AAV antibodies from prior environmental exposure to wild-type AAV. When these antibodies form immune complexes with circulating capsids, they can activate the classical complement pathway directly; a mechanism that is distinct from, and additive to, the alternative pathway activation driven by capsid surface area alone. Published data show that seropositive donors with higher anti-AAV IgG1 levels activate complement significantly more than seronegative donors when exposed to AAV9 capsid, with mass spectrometry confirming increased binding of immunoglobulins and complement factors to the capsid surface.
Whether the patient in the SYNRGY trial was screened for pre-existing NAbs, and at what titer, is not publicly known. What is known, and what makes this mechanism particularly difficult to manage, is that NAb status alone is not a reliable predictor of complement risk. In a gene therapy study in DMD patients, complement activation occurred in patients who were seronegative for both NAbs and total binding antibodies at baseline, meaning no current antibody assay would have predicted the event. Complement activation at high doses appears to be an independent risk factor that operates even in the absence of pre-existing immunity. This means NAb screening is necessary but not sufficient. And it points to a genuine field-level gap: NAb assays are not standardized across the field, there is no FDA guidance on how to run them, and their predictive value for complement activation specifically (as distinct from transduction efficiency) has not been established. That gap needs to close.
A further layer of uncertainty compounds this: current NAb assays are poorly sensitive at low titers, there are no standardized definitions of seropositivity across laboratories, and low-titer antibodies below the detection limit of standard assays may still carry biological activity. This means that a "seronegative" screening result carries more ambiguity than it appears. Whether sub-threshold NAb levels contribute meaningfully to complement activation specifically has not been established, but the question cannot be answered until the assays themselves are sensitive and standardized enough to ask it properly.
ADAM15 receptor biology. This is the mechanism I will focus on most, and I will address it in the next section. I focus on it not because it is confirmed, but because it is the most design-specific variable: the one that was chosen, and the one Capsida itself has identified as a focus of their ongoing investigation. It is also the variable most actionable at the earliest stage of capsid engineering.
Transgene biology. At greater than 70% neuronal transduction brain-wide, the scale of STXBP1 transgene expression in a single dose is extraordinary. STXBP1 is a haploinsufficiency condition; the therapeutic goal is to restore expression, not to flood neurons with supraphysiological levels of protein. Whether any aspect of transgene overexpression, promoter activity, or off-target expression contributed to cellular stress or inflammatory signaling in this patient cannot be excluded from public information.
Vector genome sensing beyond TLR9. The packaged genome can trigger innate immune responses through additional pathways, including cGAS-STING sensing of DNA escaping into the cytoplasm, and RIG-I/MDA5 sensing of double-stranded RNA generated from inverted terminal repeats. These pathways are documented in the AAV immunogenicity literature and represent additional sensing arms that may not be adequately represented in standard preclinical panels.
Manufacturing and product quality variables. Empty capsid ratio, process-related impurities, and aggregation state can all influence immunogenicity. The published GLP characterization described CAP-002 as well-tolerated, but without access to tissue samples from the autopsy, manufacturing variables cannot be independently evaluated from public information.
The underlying disease. STXBP1 encephalopathy itself carries a risk of sudden unexpected death in epilepsy, and this is not a minor caveat. As one expert noted in public commentary after the death, it is possible that the gene therapy contributed to the patient's death, and it is also possible that the death was unrelated. The autopsy's inability to establish a root cause means this cannot be ruled out. Transparent disclosure of the full clinical and pathological dataset, which Capsida has been seeking from the trial site, is essential for the field to learn from this event.
Each of these mechanisms is plausible. Several may have interacted. I am not dismissing any of them. What I am arguing is that among them, receptor biology is the variable most amenable to systematic evaluation at the earliest design stage, and the one the field has historically asked the least about.
The receptor is not a neutral choice
This is the part of the story I believe the field is underweighting, not because it is confirmed as causative, but because it represents a class of question that capsid engineering does not currently ask routinely.
When we engineer a capsid to cross the BBB, we are not just solving a delivery problem. We are choosing a biological interaction: a specific receptor, with its own expression pattern, its own downstream signaling, its own role in normal physiology. That choice is part of the drug, not a footnote to it.
ADAM15, which is a disintegrin and metalloprotease, is the receptor CAP-002's capsid binds to cross the BBB endothelium via transcytosis. It is not a passive transporter. The risk is concentrated precisely there: at the endothelial cell, the site of transcytosis, not downstream in the neurons the capsid ultimately transduces.
The published literature over more than a decade is consistent: under inflammatory conditions, ADAM15 promotes endothelial barrier dysfunction. Its overexpression increases vascular permeability and promotes neutrophil transendothelial migration via Src/ERK1/2 signaling. Its deficiency attenuates barrier dysfunction in inflammatory models including sepsis and acute lung injury. It has been implicated in atherosclerosis and inflammatory bowel disease, specifically through its role in disrupting adherens junction integrity.
More recently, the mechanism has been characterized in greater detail: under inflammatory stimulation, ADAM15 mediates cleavage of CD44, disrupting the structural integrity of the endothelial glycocalyx (the protective matrix that lines blood vessels and acts as a first barrier against macromolecular leakage). Reduced glycocalyx coverage increases exposure of endothelial cells to circulating inflammatory mediators, which in turn causes adherens junction disorganization and plasma protein leakage. This is not an instantaneous process. It evolves over hours to days of sustained inflammatory signaling, which is temporally compatible with a death occurring a few days after dosing, and with a cascade in which earlier TLR-mediated innate activation and complement amplification provide the inflammatory context that ADAM15 then amplifies at the barrier level.
This is the mechanistic picture I find most plausible, framed as a hypothesis: TLR2/TLR9 activation in the hours after dosing initiates an inflammatory environment. Complement activation builds over subsequent days, overwhelming the capsid's partial self-inhibitory mechanisms at this dose. ADAM15 engagement at the BBB endothelium (a receptor whose function under inflammatory conditions is to increase barrier permeability) may have lowered the threshold at which the CNS compartment failed to contain that inflammatory load. The result, days later, was cerebral edema.
I want to be precise: this is a cascade hypothesis grounded in published biology, not a confirmed mechanism. Each link in this chain is plausible and supported by literature. The chain as a whole has not been demonstrated in this context. But it is the hypothesis that best fits the available timing, the receptor biology, and the known immunological behavior of high-dose IV AAV in a pediatric patient.
Compare this to other BBB receptors the field is working with. TfR1 (the transferrin receptor) is a transport protein whose job is moving iron. It does not participate in inflammatory signaling. It is not involved in barrier permeability regulation. ALPL (alkaline phosphatase on brain endothelium) is similarly non-inflammatory. RECK (the receptor Capsida is now pivoting to) has a distinct biology worth examining carefully before assuming it resolves the problem, but its involvement in inflammatory permeability pathways is not established the way ADAM15's is.
The contrast is not incidental. Receptor biology should be the first question asked in BBB capsid engineering. For ADAM15, a literature review available to anyone would have returned a decade of publications on endothelial hyperpermeability and inflammatory signaling. Whether that review would have been sufficient to change the development decision, given the severity of the disease and the strength of the delivery data, is a legitimate question. But the review should have happened, and its findings should have explicitly informed the preclinical safety package and clinical monitoring plan. Whether such a review happened, and what weight it was given, is not something I can know from the outside. What I am arguing is that the field has no standard requiring it, and that absence is the gap worth closing.
Why the NHP studies missed it. And why that is a structural problem
Capsida's GLP NHP toxicology study showed no adverse histopathology. That study did what GLP studies are designed to do. The problem is not that it was inadequate by current standards; it is that current standards were not built for what engineered BBB-crossing IV capsids actually require.
Three gaps compound each other.
Species biology. Human CNS immune biology differs from NHP in ways that are not fully characterized. Microglial activation thresholds, complement regulation in the CNS compartment, and species-specific expression patterns of immune receptors all introduce translation gaps that a standard NHP GLP study is not designed to resolve. This is not a criticism of the model; it is an acknowledgment that the model was not built for this question.
Age. The preclinical safety data came from adult animals. The patient was a child. Pediatric complement activation kinetics, cytokine response profiles, and CNS immune surveillance differ meaningfully from adult systems. This is not a rare edge case; it is the primary patient population for STXBP1 programs.
Dose and tissue panel. Standard preclinical immunogenicity assessment centers on peripheral blood and liver. For a CNS-targeted IV program delivering a high-efficiency capsid into brain tissue, that panel misses the most relevant immune compartment. Microglia are present throughout the brain, express TLR2 and TLR9, and respond to AAV vectors in ways that peripheral blood readouts do not predict. There is also a subtler dose-related issue: at lower effective CNS doses, AAV capsids partially inhibit their own complement activation by binding regulatory proteins including factor H. As doses increase to achieve brain-wide neuronal coverage, evidence suggests this self-inhibitory capacity is overwhelmed: a threshold effect that standard NHP safety studies, which typically assess tolerability rather than the limits of immune self-regulation at dose escalation, are not designed to detect.
The NHP studies were not wrong. They were not designed to ask the questions this biology requires. That is the field's gap to close.
What the next BBB program should do differently
These are not calls for new assays to be invented. The component tools exist. What is missing is field-level agreement that they are required for this class of program, and validated standards linking these readouts to clinical risk thresholds, which the field still needs to build.
[1] Ask the receptor question before you choose your target.
Before committing to a BBB receptor, the first question should be: what does this receptor do? Where is it expressed? Is it involved in barrier integrity, inflammatory signaling, or leukocyte trafficking? Does it have a documented role in pathological permeability? This is a literature review, not an experiment. For ADAM15, that review was available and would have returned a substantive body of evidence on endothelial barrier dysfunction. Receptor biology should be a formal checkpoint in the target selection process, not an afterthought.
[2] Counter-screen your capsid library against TLR2 and TLR9 activation in parallel with your receptor screen.
TLR2 recognizes the AAV capsid surface as a pathogen-associated molecular pattern; TLR9 detects the CpG content of the vector genome. Both are well-documented innate immune sensors for rAAV vectors, and both are expressed in microglia. Running the same capsid library against innate immune activation readouts (including TLR2-associated signaling assays and TLR9-mediated interferon responses) in parallel with receptor binding screens would allow programs to enrich candidates for lower innate immune liability from the start. The component assays exist. What the field does not yet have are validated predictive thresholds linking these readouts to clinical CNS toxicity risk, which is precisely why building and standardizing them should be a field-level priority.
[3] Assess the surface chemistry of your insertion before library construction.
The peptide inserted into the capsid loop changes the local surface properties of the vector. Hydrophobic and positively charged insertions are generally associated with greater potential for complement interaction and innate immune recognition than hydrophilic ones. This is grounded in basic peptide and surface chemistry, though the specific quantitative relationship for engineered AAV insertion peptides is not yet established at the level of validated predictive models.
Surface chemistry assessment can inform which peptide sequences are prioritized in library design, rather than serving only as a late-stage filter. Computationally, peptide biophysical properties (charge, hydrophobicity, flexibility, and alpha-helical content) are already being used to predict AAV packaging efficiency, demonstrating that these features are learnable from sequence alone. The same framework is directly applicable to immune risk scoring at the library design stage, before any capsid is synthesized. Experimentally, single-particle atomic force microscopy can directly measure surface charge and hydrophobicity of assembled capsids, providing a biophysical readout that complements computational predictions for prioritized candidates.
The component tools exist on both the computational and experimental sides. What is missing is the field-level decision to apply them to immune risk, not just to packaging and yield.
[4] Include human microglia in your CNS immunogenicity panel.
For any CNS-targeted IV program, the preclinical immunogenicity panel should include human microglia (the brain's resident innate immune cells, distributed throughout the tissue the capsid is designed to reach) expressing both TLR2 and TLR9. Their response profile is not captured by PBMC or hepatocyte panels. Whether microglial activation contributed materially in the CAP-002 case is unknown. What is clear is that for a high-efficiency BBB-crossing IV program, their response is the most relevant readout and the least commonly assessed. The cells are commercially available. The cytokine assays exist. The missing element is the field-wide agreement that this is required.
[5] Prepare for complement activation at the clinic, especially in pediatric patients.
Complement activation at high systemic AAV doses is documented across multiple clinical programs, manifesting at doses above 1×10¹³ vg/kg. For CNS IV programs in children: complement inhibitors should be accessible at the clinical site, intracranial pressure monitoring should begin early and continue frequently, and adult NHP safety data should not be assumed to fully predict pediatric immune responses. There is no public evidence establishing complement activation as the proximate cause in the CAP-002 case, but it represents a documented biological risk for this class of program that clinical teams should be prepared for, not discovering in real time.
[6] Standardize NAb screening and establish its relationship to complement risk.
Pre-existing anti-AAV antibodies are common in the population and can amplify complement activation through classical pathway engagement, a mechanism additive to the dose-dependent alternative pathway activation described above. Currently, NAb assays are not standardized across the field, there is no FDA guidance on how to conduct them, and their predictive value for complement activation is unestablished. A patient screened as seronegative by one assay may not be seronegative by another. For high-dose CNS IV programs in pediatric patients, this is not an acceptable level of ambiguity. The field needs to agree on what NAb screening looks like before dosing, and then do the work, across programs and datasets, to establish what NAb levels actually predict in terms of complement risk, not just transduction efficiency. Separately, capsid engineering for antibody evasion, already pursued for efficacy reasons, should now be evaluated explicitly for its effect on complement activation risk; these are related but not identical properties, and optimizing for one does not guarantee improvement in the other.
What the field needs to build
The five (now six) recommendations above are reactive: they make existing programs safer. What the field actually needs is to treat immune risk as a designed-in property of engineered capsids, evaluated computationally and experimentally at scale, before any single candidate is nominated.
That means two things.
A standard immunogenicity panel for engineered BBB-crossing capsids. Agreed upon across the field, the way neutralizing antibody assays are standardized. TLR2 and TLR9 innate activation readouts. Complement surface risk assessment. Microglial cytokine panel. Receptor biology audit. Standardized NAb screening with complement activation correlation. Every engineered capsid entering a CNS IV program should pass this panel, and the field should agree on what "pass" means. Which requires building the correlation datasets that link these readouts to clinical outcomes. That work has not been done. It should be.
Computational tools that make immune risk assessable during library design, before the first capsid is synthesized. The training data exists: TLR2 and TLR9 activation measurements, complement biology, capsid sequence-function datasets, and emerging NAb-complement correlation data. If 100,000 sequence variants can be scored computationally for innate immune liability and receptor biology risk before any wet lab work begins, the experimental assays become confirmation rather than discovery. The technology to build these tools is available now. The field investment to validate them is not yet happening at the scale this problem requires.
Neither of these exists at field standard today. Both are buildable. The scientific foundation for both is solid.
On calling this "Capsida moment"
I want to say something directly.
What happened in the SYNRGY trial is already being referred to as a 'Capsida moment'; I understand the instinct. It marks a before and after in BBB gene therapy. But that framing carries a weight it should not.
This is not Capsida's failure. It is the field's frontier.
Capsida did what was required by current standards. They took a program to the clinic because the disease is devastating, the unmet need is real, and the science was rigorous by every metric the field has agreed upon. If another company, or an academic group, or a larger institution, had been first to clinic with a high-efficiency BBB-crossing IV capsid for a pediatric CNS indication, they would mostly be having the same conversation today. The gaps this event revealed are not specific to one company's decisions. They are the field's collective blind spots.
I would call this the field's trailblazing moment. Capsida was first across a frontier that the whole field was racing toward. The cost of being first was catastrophic, and it was not theirs alone to bear. The knowledge now available (about receptor biology, CNS immune responses, the limits of NHP translation) is available to every program that follows. The question is whether the field acts on it before the next patient is dosed, or after.
I believe we should act now. The questions to ask are clear. The component assays exist. The computational tools are buildable. What is missing is the collective will to make them standard before the next IND is filed, not in response to the next adverse event. That is the field's responsibility. It starts now.
Full disclosures
Prior work and potential conflicts of interest. I am a named inventor on patents related to AAV capsid engineering and CNS gene delivery, including work on TfR1-targeting capsids, developed during my tenure at the Broad Institute's AAV Engineering Program. I am also a first or co-author on peer-reviewed publications in this area. This prior work directly informs my scientific perspective and may constitute a competing interest. This post does not represent the views of the Broad Institute, any prior employer, or any current advisory client or collaborator of TheBioMLClinic.
Independence. I operate as an independent computational scientist and advisor through TheBioMLClinic. I have no financial or professional relationship with Capsida Biotherapeutics. This post was not coordinated with or endorsed by Capsida or any other organization.
Disciplinary scope. I am a computational scientist with more than twenty years of experience working mostly on biological problems, including AAV capsid engineering and machine learning-guided variant design. I am not a clinical immunologist, a virologist, or a physician. The mechanistic interpretations in this post reflect my reading of the published literature and should be evaluated in that context, and in conjunction with expert input from immunologists, CNS biologists, and clinicians who bring expertise I do not.
Unconfirmed mechanism. The biological interpretation of ADAM15's role, the proposed mechanistic hypotheses, and the discussion of alternative mechanisms are scientific interpretations of publicly available data. The underlying cause of the cerebral edema in the SYNRGY trial has not been officially established. Nothing in this post should be read as a causal claim about what happened to this patient.
Educational intent. This post is intended to open a scientific discussion and identify areas where the field's current practices may warrant examination. It is not clinical, regulatory, or legal guidance for any program. It is not a confirmed blueprint that programs should follow. It is a starting point for a conversation the field needs to have.
On the patient. A child died. That is the fact that makes all of this matter. I have tried to write about this event with the seriousness it deserves. If I have fallen short of that in any way, I welcome the correction.
This piece was updated on 05.18.2026 and on 05.19.2026 to include the role of pre-existing neutralizing antibodies in complement activation, following expert feedback from the AAV gene therapy community and Heikki Turunen, PhD, an AAV expert.
Fatma Elzahraa Eid, PhD is the founder of TheBioMLClinic, an independent advisory practice at the intersection of AAV capsid engineering and machine learning.
Comments
Post a Comment