Gene Delivery Is the Real Bottleneck: Four Signals from ISSCR 2026


Boston / Montréal — a working note on what the 2026 ISSCR keynotes actually said about where cell and gene therapy is heading.
Read the ISSCR 2026 agenda and the four keynotes look unrelated. Twenty years of iPSC. Brain organoids. Chromatin and circadian control of neocortical fate. A new class of gene delivery vehicle.
Sit with all four and they collapse into one question: the science works — can you get it where it needs to go, and can anyone afford it? That is the gene delivery bottleneck, and it now shows up at four different scales at once.
Molecular — Feng Zhang, on getting a Cas9 ribonucleoprotein into a cell without AAV or LNP.
Tissue — Madeline Lancaster, on getting a drug past the blood-brain barrier.
Temporal — Yukiko Gotoh, on delivering an intervention inside the window where it still changes the outcome.
System — Shinya Yamanaka, on getting a manufactured cell into a patient at a price a health system will pay.
Four scales, one problem. That is what a field looks like when it finishes its discovery phase and walks into its engineering phase — and most of the people in it are still operating on discovery-phase instincts.
I covered each keynote separately for GBI Monthly (Chinese-language originals linked at the end). This piece is the synthesis: what changes for the people who have to build, fund, or regulate this.
Signal 1: The iPSC moat moved from biology to manufacturing
Yamanaka's talk reads like a retrospective until you notice what he spent the most time on. Not differentiation protocols. Cost.
The clinical scoreboard he presented was honest rather than triumphant: four of six Parkinson's patients showing symptomatic improvement after dopaminergic progenitor transplantation; two of four spinal cord injury patients showing improvement signals. Real, early, not yet a product.
The infrastructure numbers were the actual news. By March 2026 the CiRA Foundation had established 34 cell lines from 8 donors, supplying research-grade cells to 166 institutions and clinical-grade cells to 125 — supporting 20 Japanese clinical studies and one US trial. Three parallel cost-reduction strategies are running at once: an HLA-homozygous allogeneic bank covering roughly 40% of the Japanese population, HLA gene-edited iPSCs aimed at global coverage, and an autologous "My iPS Cells" program started in 2025. The closed manufacturing process he described produces 200–500 colonies and roughly 10 million cells within three weeks from 20 mL of whole blood.
A Nobel laureate presenting colony counts and process timelines in a keynote is itself the signal.
His summary line was that moving iPSCs from research to clinic proved roughly a hundred times harder than expected — across GMP manufacturing, quality control, batch consistency, genomic stability, immune rejection, cost, and sustained supply. Not one of those is solvable with a paper.
What actually changed: for a decade the differentiating asset was a differentiation protocol. Most of those are now published and reproducible. The scarce capability is making cells consistently, at scale, under GMP, at a unit cost a payer will tolerate. That is a CMC problem, not a stem cell biology problem — and CMC moats are unglamorous, hard to put in a deck, and very real. Expect value to migrate from platform biotechs toward CDMOs and process equipment.
If you want the underlying technical background, I've written separately on iPSC cell therapy and on Yamanaka factors and cellular reprogramming.
Signal 2: AAV's safety ceiling opened a gene delivery window — and it is narrower than it looks
The timing of Zhang's talk was almost too neat.
In June 2025 the FDA issued a safety communication on Sarepta's Elevidys following fatal acute liver failure reports. On 14 November 2025 the agency approved a boxed warning and narrowed the indication to ambulatory DMD patients aged four and up, with distribution to non-ambulatory patients already voluntarily suspended, plus a mandated post-marketing study following 200 patients for at least twelve months.
Then Zhang presented a DMD mouse model — using neither AAV nor LNP. Engineered dARC protein nanoparticles packaging Cas9 RNP directly, delivering statistically significant dystrophin editing after intramuscular injection (p = 0.0001), built on the 2021 Science work showing that endogenous retroviral-derived proteins self-assemble into virus-like particles (SEND). A parallel eCIS platform, derived from bacterial contractile injection systems with re-engineered tail fibers, is being aimed at human cells.
Both incumbent gene delivery platforms are hitting structural ceilings at the same time. AAV's problem is dose — systemic hepatotoxicity and immune response cap what you can give. LNP's problem is hepatotropism and tolerability on repeat dosing. That simultaneity is exactly what creates an opening for something new — and it reframes the gene delivery bottleneck as a dosing problem rather than a precision problem.
Here is where I'd push back on the enthusiasm, from experience.
Protein nanoparticles are elegant in a paper and have already hit a wall in a company. Aera Therapeutics — founded by Zhang, and where I worked — launched in 2023 with $193M and a story that was entirely PNP. By May 2025, Endpoints reported Aera shifting its pipeline emphasis to lipid nanoparticles for T-cell targeting while continuing PNP development.
The lesson for anyone evaluating this class of asset: the journal timeline and the pipeline timeline are not the same timeline, and the gap is measured in years. Zhang was explicit that the data shown was unpublished mouse work, with immunogenicity, long-term safety, and manufacturing scalability all open.
If I'm underwriting a delivery platform, my first question isn't editing efficiency. It's which tissue, at what dose, and does it still work on the second administration? Delivery platforms are not valued on first dose. They're valued on redosing.
For background on why AAV got here, see my primer on AAV gene therapy.
Signal 3: Organoids are crossing from model to tool
CNS programs fail at the highest rate of any therapeutic area, and a large part of the reason is that animal models don't represent the human brain.
Lancaster's 2013 Nature protocol established self-organizing brain organoids by exploiting the neural default model — tissue patterning itself without external factors. Her group has shown human neural progenitors maintain slower cell cycles and longer proliferative windows than rodents, which is precisely why rodent timing doesn't transfer.
Two pieces of this year's work matter commercially:
Mechanism, not morphology. Her team showed that mutations in KDM5B and CHD7 interact with androgens to shift neuron ratios and circuit connectivity — the first direct observation of gene–hormone interaction in an organoid system, and a concrete handle on autism's sex asymmetry.
Choroid plexus as an entry point. Rather than breaching the blood-brain barrier, the approach redesigns the drug as cargo that choroid plexus receptors actively recognize and transport into cerebrospinal fluid. The barrier becomes a loading dock.
The commercial dividing line for every organoid company is whether they can convert "looks like a brain" into "predicts what happens in a human." Pharma buys predictive power. Companies that clear correlation-to-human-data, batch variability, throughput, and cost become long-term suppliers. Those that don't are running a very expensive microscope.
Gotoh's keynote added the dimension almost everyone ignores: time is an operable variable. Clock genes — BMAL1 and CLOCK driving terminal differentiation, PER and CRY maintaining stem cell properties — oscillate in embryonic radial glia (p < 0.0001) and gate neural progenitor fate through a defined E12–E16 window. Her data suggest maternal circadian disruption may affect cortical developmental timing more than fetal mutations do. If that holds, the intervention set expands well past drugs, into monitoring, behavior, and environment.
What this means for each side of the table
Large pharma. CNS franchises are the direct beneficiary. The barrier has been the most expensive wall in neuroscience R&D, and if the choroid plexus route works, valuation logic shifts from "how good is the antibody" to "can your molecule be carried in." Expect BD attention to move toward gene delivery platforms, which are cheaper to acquire than assets.
Biotech. Companies whose differentiation is a differentiation protocol will find fundraising harder, because that is no longer differentiation. Process automation, closed systems, and rapid release testing — structurally boring, structurally in demand.
CDMOs. The most underpriced segment here. Yamanaka's numbers amount to a public stress test of the field's cost structure, and CiRA works because it is run as public infrastructure. Private markets can only replicate that scale through CDMOs. Cell therapy CDMO capacity gets scarcer, not looser, over the next five years.
Regulators and payers. Elevidys has already reset FDA's risk appetite on one-time therapies: accelerated review no longer buys you a pass on long-term safety. In parallel, CMS and commercial payers are being pushed toward outcome-based and installment structures for million-dollar one-time treatments. Yamanaka's "a price patients can afford" translates, in the US market, directly into that conversation.
Japan. CiRA's three-layer architecture — national cell bank, academic foundation, clinical network — is the only version of this that is genuinely operating at scale anywhere.
The emerging-ecosystem playbook, with Taiwan as the worked example
Most of the world's biotech ecosystems are not Boston, and the strategic question for them is not "how do we get a first-in-class asset." It's "where in this value chain can we be structurally hard to replace." Taiwan is a useful worked example because its regulatory position just changed.
Don't crowd into organoid modeling. Build the manufacturing layer. There are already too many labs making brain organoids; a late entrant has no edge. But cell therapy process automation, closed systems, rapid release testing, consumables and instrumentation map directly onto forty years of semiconductor and precision-machinery capability. The field needs people who can take yield from 60% to 95%. That is a different skill set from publishing another protocol, and it's one Taiwan actually has.
Treat conditional approval as a data-acquisition window, not a revenue window. Taiwan's two regenerative medicine acts took effect on 1 January 2026. Products that have completed Phase 2 with demonstrated safety and preliminary efficacy can obtain a conditional license of up to five years. Used to start billing early, the license lapses when the conditions aren't met and nothing remains. Used to systematically build real-world evidence, long-term registries, and a safety database that FDA or EMA will recognize, you end up holding an asset nobody can buy. Japan's conditional approval pathway ran this exact experiment; the lesson is already available.
Build the national cell bank now, or don't bother. Copy CiRA's model first and improve later: HLA-homozygous banking, defined population coverage, operated as infrastructure rather than product, supplying both research- and clinical-grade material domestically and abroad. Taiwan's relatively concentrated medical center system and population HLA distribution mean fewer lines could cover a higher share of the population than in Japan. But this is a first-mover asset — once another Asian market builds one and secures international recognition, a second bank is worth only its domestic demand.
For investors: fund delivery, not indications. Capital in smaller ecosystems reliably prefers assets with clinical data, an indication, and a visible exit. This cycle's scarcity is in platforms — specifically extrahepatic delivery. Long payback, high technical risk: precisely the profile that needs patient and government-adjacent capital, rather than waiting to in-license after someone else de-risks it.
My take
Having worked on both the drug side and the delivery-platform side here in Boston, the strongest impression from these four talks is that the field is moving collectively from discovery into engineering — and most participants' instincts haven't moved with it.
Discovery-phase logic: find a mechanism, publish, incorporate, raise, get acquired. That script worked beautifully from 2015 to 2021. Engineering-phase logic rewards reproducibility, yield, unit cost, batch consistency, and supply chain resilience. None of that gets you a Nature cover. All of it determines whether your product survives Phase 3 and a payer review.
Yamanaka's "hundred times" line reads to me as a warning rather than modesty. The seven things he named cannot be solved by insight. They're solved by years of process engineering — which is what Japan is good at, and what a lot of ecosystems believe they're good at without having proven it.
The second impression: delivery has been underrated for far too long. A decade of genetic medicine narrative focused on the editing tools — ZFN, TALEN, CRISPR, base editing, prime editing, each more precise. But precision was never what blocked the clinic. Getting in was. And giving enough to work without hurting someone was. Sarepta's boxed warning, Aera's practical pivot toward LNP, Zhang's continued bet on dARC and eCIS — three views of the same unsolved problem.
Whoever solves efficient, low-immunogenicity, repeat-dosable gene delivery to extrahepatic tissue owns a toll booth on the next decade of genetic medicine. That is worth more than any single indication.
Third, and more uncomfortable: three of the four keynotes came from Japan and the UK, and none of them was framed as "look what we discovered." All three were framed as "here is how we make it usable." The US narrative still rewards breakthrough and speed. When capital is abundant that difference is invisible. When capital tightens, regulators get more conservative, and payers get tougher — engineering culture wins.
Key takeaways
The hidden theme of ISSCR 2026 was the gene delivery bottleneck, not stem cells. Four keynotes, four scales — molecular, tissue, temporal, system — of the same bottleneck.
The iPSC moat has moved to CMC. Consistency, immune compatibility, unit cost, and sustained supply now decide winners, not differentiation protocols.
AAV's safety ceiling opens a window for new gene delivery platforms, but a window is not a market. Company-level timelines for protein nanoparticles are running years behind the academic narrative.
Organoids are at the model-to-tool dividing line. Predictive, reproducible, quantifiable platforms become pharma suppliers; the rest don't.
Time is an uncommercialized therapeutic variable. If chronopharmacology extends into neurodevelopment, the intervention set expands from drugs into monitoring and behavior.
If you're evaluating gene delivery strategy, CMC pathway, or market-entry timing for a cell or gene therapy program, this is the work LuTra Studio does — translating scientific complexity into executable market decisions.Explore Advisory · Book a consultation
Further reading — the full ISSCR 2026 series
These five pieces are my complete coverage of the 2026 ISSCR Annual Meeting (8–11 July, Montréal) for GBI Monthly, with fuller experimental detail on each keynote. Originals are in Traditional Chinese.
Sources
ISSCR 2026 Annual Meeting, 8–11 July 2026, Montréal — keynote sessions
US FDA, FDA Takes Action on New Boxed Warning for Acute Serious Liver Injury and Acute Liver Failure Following Treatment with Elevidys, 14 November 2025
US FDA, FDA Investigating Deaths Due to Acute Liver Failure Following Treatment with Sarepta's AAVrh74 Gene Therapies, June 2025
Endpoints News, Feng Zhang's protein nanoparticle startup Aera Therapeutics turns to lipid nanoparticles to jumpstart its pipeline, 2 May 2025
Zhang lab, SEND platform, Science, 2021
Lee and Li, Taiwan's Regenerative Medicine Acts take effect 1 January 2026
About the author
Jason Yen-Chun Lu, Ph.D. is the founder of LuTra Studio, advising on biopharma strategy, drug development, biomanufacturing, and AI in biotech, and a columnist for GBI Monthly covering the US biotech industry from Boston. He previously worked at ModeX Therapeutics and Aera Therapeutics, with research training at Cornell University, MIT, and Boston Children's Hospital. His technical focus spans RNA therapeutics, lipid nanoparticles, cell and gene therapy, CMC, and drug development strategy.

