A Platform Bet on Radiation

The most consequential idea in Fiona Marshall’s vision for Novartis research is also the easiest to understand: a radioactive missile that finds cancer cells and detonates. Radioligand therapy, or RLT, works by attaching a radioactive isotope to a small peptide that seeks out a specific protein on the surface of tumor cells. The company’s first approved RLT, Pluvicto, targets PSMA in prostate cancer and has become a blockbuster. But Marshall, President of Biomedical Research at the Swiss drugmaker, believes that drug was merely the proof of concept.

“We don’t just do technology for technology’s sake. We really ground ourselves in what are the disease areas that we want to focus on, where are the highest unmet needs and where can we really bring meaningful differentiation,” Marshall said on Bloomberg Intelligence’s Vanguards of Health Care podcast.

The pipeline now contains more than 20 RLT candidates, and the ambition is explicit: extend the modality to nearly every tumor type. The logic depends on a design constraint that separates RLT from other cancer drugs. Antibody-drug conjugates, the other hot modality in oncology, require their targets to be internalized by cancer cells. RLT ligands don’t. They can hit surface proteins that never enter the cell, which dramatically expands the list of addressable targets. The peptides themselves can be engineered to reach picomolar potency, making them extraordinarily effective homing devices.

Phase Zero: The Kill Decision Advantage

The truly distinctive edge in Novartis’s RLT strategy is speed of failure. Before committing a candidate to years of expensive clinical development, Novartis runs what Marshall calls phase-zero studies. Using PET or SPECT imaging isotopes, researchers can inject an experimental RLT into a small number of human subjects and immediately see whether it binds the tumor, whether it accumulates in normal tissue, and what the radiation dose actually is.

This is not possible with most other drug types. You cannot see an antibody clearing protein aggregates in the brain. You can see whether a radioactive peptide finds its target. The result is that Novartis can evaluate multiple RLT candidates in parallel and kill the losers before investing hundreds of millions of dollars. That portfolio logic changes the risk calculus for a technology that has historically been constrained by manufacturing complexity and radionuclide supply chains.

Lutetium First, Actinium Later

One of the more technical questions facing the RLT field is which radioactive emitter to use. Lutetium-177, the beta emitter used in Pluvicto, delivers radiation through a “crossfire” effect that can damage nearby cells even if they don’t bind the ligand directly. Actinium-225, an alpha emitter, delivers a much more energetic punch that causes catastrophic double-stranded DNA breaks, and it can destroy tumors that have become resistant to lutetium.

Marshall’s answer reveals how Novartis thinks about patient sequencing rather than one-off drug choices. “Alpha is a very powerful emitter. One of the things we want, and it will depend on the antigen… if the expression is very high, you probably don’t need, or maybe not even want actinium. You might actually prefer lutetium because the therapeutic index between normal cells and the tumor cells, that’s what you’re really interested in,” she explained.

The decision comes down to target specificity. PSMA is expressed at very high levels on prostate cancer cells and very low levels on normal tissue, making it suitable for alpha radiation. But HER2, a common target in breast cancer, is also expressed on some normal cells. In that case, the therapeutic window may favor the gentler, more controllable lutetium. Marshall’s prediction: within a single patient, doctors will use lutetium first and actinium later as resistance emerges, a sequential approach that effectively doubles the platform’s addressable treatment duration.

AttributeLutetium (beta)Actinium (alpha)MechanismBeta emission, crossfire effect on adjacent cellsHigh-energy alpha, catastrophic double-stranded DNA breaksBest suited forHigh tumor antigen expression, narrow therapeutic index concernsTumors resistant to lutetium, high tumor specificity antigensExamplePluvicto (PSMA)PSMA candidates, HER2 RLT in developmentClinical logicFirst-line RLTSecond-line after lutetium resistance
The Complementary Modalities: ADCs and siRNA

Novartis is not betting everything on radiation. The company made two acquisitions that round out its oncology toolkit. Mariana, a Cambridge, US-based biotech, brought actinium expertise and phase-zero study infrastructure. Myrix, a UK company, added an antibody-drug conjugate platform built on a novel payload — an NNT inhibitor that interferes with lipid modification of proteins in rapidly dividing cells. Marshall frames ADCs as complementary to RLT: when radiation is not the right answer, a next-generation ADC with reduced toxicity might be.

The second platform bet is siRNA, a technology that silences disease-causing genes by targeting messenger RNA before proteins get made. The liver-targeting version is mature — Leqvio, Novartis’s PCSK9-lowering drug developed with Alnylam, is approved. The frontier is delivery beyond the liver. The Avidity acquisition is the centerpiece: Avidity optimized a transferrin receptor antibody to carry siRNA specifically into skeletal muscle, enabling a drug for facioscapulohumeral muscular dystrophy that has already shown clinical evidence of muscle delivery and downstream gene knockdown.

Marshall is careful to note that not all transferrin antibodies are created equal. “All transferrin antibodies and ligands are not all the same,” she said, explaining that the format and binding affinity determine where in the body the drug goes. The intellectual property is in the optimization, not the concept.

The Antisense Question and a Yearly Convenience

Novartis uses both antisense oligonucleotides and siRNA, and the choice is pragmatic rather than ideological. Antisense drugs can trigger innate immune responses and inflammation; siRNAs generally avoid that problem. But for some targets, first-mover advantage matters more. Pelacarsen, Novartis’s drug targeting Lp(a) — a genetically determined cardiovascular risk factor that statins don’t address — uses antisense chemistry because proving the Lp(a) hypothesis first was worth the trade-off. An siRNA version is in development behind it, potentially offering higher knockdown.

ProgramModalityTargetStatus / RationaleLeqviosiRNA (with Alnylam)PCSK9Approved; liver deliveryPelacarsenAntisenseLp(a)Phase 2–3; first-mover advantage on hypothesisLp(a) siRNAsiRNALp(a)In development; potentially higher knockdownFSHD programsiRNA (Avidity)DUX4Clinical data showing muscle delivery and downstream gene knockdownCharcot-Marie-ToothLipid-modified siRNA (DTX)MuscleIn dose-ascending clinical trials

Chemistry advances are pushing dosing intervals from every three months toward once a year. That matters for adherence in a way that is easy to understate. Marshall cites a sobering statistic: roughly 60 percent of patients don’t take daily statins as prescribed. A once-yearly injection for cardiovascular risk factors could close a massive residual risk gap that daily pills cannot.

Cause-Modifying Neurology Enters the Mainstream

Marshall’s assessment of neurological disease is unambiguous: “Oh yes, for sure” — a new era has arrived. The shift is not driven by one technology but by understanding what actually causes neurodegeneration. Previous generations of drugs treated symptoms; current approaches target disease drivers.

The flagship program is a TREM2 agonist antibody, targeting a genetically validated protein implicated in Alzheimer’s disease. The drug stabilizes microglia, the brain’s immune cells, in a neuroprotective state. It is in clinical trials for both Alzheimer’s and ALS, with CSF biomarkers confirming the drug is engaging its target in the brain. Marshall is candid about the fundamental challenge: by the time symptoms appear, neurons are already lost. The strategy is early intervention, ideally pre-symptomatic, enabled by improving blood biomarkers like phosphorylated tau.

Brain delivery remains the bottleneck. Antibodies typically achieve only about 0.2 percent penetration from blood to brain. Marshall is optimistic that transferrin-based brain shuttles can raise exposure to 2 to 3 percent, and that combining shuttles with more potent antibodies reduces the required dose. For cell-type-specific delivery, she points to gene therapy with cell-specific promoters — a capability strengthened by the Kate Therapeutics acquisition, which brought optimized AAV capsids for muscle.

AI as an Organizational Problem

The shallowest read of AI in drug discovery is that it helps scientists run simulations faster. Marshall’s view is more structural: the bottleneck is not the models, but who sits at the table when experiments are designed.

She has reorganized Novartis Biomedical Research accordingly. A head of data and digital now sits on her leadership team. Every disease-area leader must have a data scientist on their own leadership team. Data scientists now design experiments rather than merely analyzing results after the fact. The computational chemistry group has been renamed “digital design” and includes both wet-lab and dry-lab chemists. A new target identification department, led by a data scientist, uses AI to discover new targets across disease areas.

The quantified result: discovery cycle time compressed from three to four years to eighteen months to two years, a roughly 30 percent reduction. Marshall also cites examples of AI finding molecules for targets previously considered undruggable, and AI-enabled modeling that could support skipping from phase 1 directly to phase 3 in some cases — an approach she says US regulators are receptive to.

The data advantage is real but bounded. Data42, Novartis’s Palantir Foundry-based data lake, contains 30 years of clinical trial data including individual patient-level results and preclinical safety data — crucially, including negative results that public datasets systematically exclude. But Marshall acknowledges the edge is strongest in therapeutic areas where Novartis has deep history, and weaker in areas like anti-infectives where the company has been less active.

On model selection, the company leads with US commercial models but evaluates open-source alternatives through Hugging Face. The token budget is enterprise-wide; scientists are encouraged to use AI liberally without worrying about allocation.

The Quiet Cost of Cutting Basic Science

Marshall’s most pointed commentary concerns US academic science funding. She frames publicly funded research as the foundation of the entire drug industry — not only for discoveries but for training the people industry eventually hires. Her own career arc illustrates the long lag: Heptares, the biotech she co-founded, was spun out of the MRC Laboratory of Molecular Biology and built on cryo-electron microscopy — a technique supported by decades of academic investment that eventually produced drugs now in late-stage trials.

“There can be a disconnect in the time, exactly as you’ve said, of where you make that investment and then later on where you get the economic benefit,” she said.

On a zero-to-ten worry scale, Marshall gives a qualified answer: she is “one of life’s eternal optimists” but acknowledges “some concerns at the moment.” Industry consortium funding for postdocs and PhD students is “a drop in the ocean” compared to government support. Her economic argument is pointed: dementia patients occupy roughly a quarter of UK hospital beds long-term, and effective neurodegenerative drugs would free those beds and reduce the carer burden — a cost-benefit calculation that should make paying for effective therapies an obvious choice.

The San Diego Consolidation and What Comes Next

The interview closes with a signal about how Novartis will operationalize its AI-augmented research. The company is consolidating California operations into a new San Diego facility, replacing the 20-year-old GNF site in La Jolla and absorbing people from Emeryville. The new building will house the acquired biotechs — Avidity, Kate Therapeutics, DTX — under one roof.

On lab automation, Marshall is notably restrained. “I’m not sure I’m that keen on the self-driving, fully automated lab,” she said. Her position is that hands-on experimentation remains a source of creativity, and automation should be applied where it is fit for purpose — screening, for example — rather than universally. It is a striking stance from a leader whose company is betting heavily on AI, and it signals a philosophy: technology accelerates human judgment; it does not replace it.

The unresolved tension for investors is temporal. AI-driven productivity gains compress time-to-candidate, but the revenue payoff arrives years later, after clinical trials, regulatory review, and commercial launch. The US policy environment — Inflation Reduction Act pricing pressure, academic funding uncertainty — could erode the value of faster development just as the gains materialize. Marshall’s optimism is genuine but tempered by structural forces she cannot control.

What to watch: the Pelacarsen phase 2–3 readout for Lp(a) as the first real test of the Lp(a) hypothesis; FSHD siRNA data maturation from the Avidity platform; TREM2 antibody biomarker and efficacy signals in Alzheimer’s and ALS; the HER2 RLT comparison between lutetium and actinium as the template for target selection across the entire RLT pipeline; and whether the 30 percent cycle-time reduction holds across the portfolio, not just in showcase projects. The San Diego facility opening will be the most visible signal of whether the biology-first, AI-accelerated model actually works at scale.