Three Novartis Clinical Setbacks in Eight Days: What CAR-T, Pelacarsen, and Del-Desiran Reveal About Translational Risk

Bad months happen to good companies, but the eight days that opened September 2026 at Novartis looked like more than just bad luck. In a single week, three programs across three modalities—CAR-T, antisense oligonucleotides (ASO), and antibody-oligonucleotide conjugates (AOC)— hit setbacks. The pattern behind them is more instructive than any single failure. The failures stemmed not from a lack of target engagement, but from a fatal mismatch between a working platform and a complex human biological context that standard preclinical models simply failed to resolve.

On September 1, news emerged that Novartis had paused eight clinical trials of its CD19 CAR-T therapy in autoimmune and neurological disease after three patients died following serious cases of immune effector cell-associated hemophagocytic syndrome (IEC-HS); the holds had been initiated on August 24 ^[1]. Three days later, the company reported that pelacarsen, the most advanced Lp(a)-lowering program in clinical development, had failed to reduce cardiovascular events in a Phase III trial of 8,323 patients ^[2]. On September 8, del-desiran, a major asset behind the $12 billion acquisition of Avidity Biosciences, missed its primary endpoint in Phase III ^[3].

Figure 1. Novartis faced three major clinical setbacks across CAR-T, pelacarsen, and del-desiran in early September 2026 ^[4].

What connects them is not incompetence. In each case the technology did what it was engineered to do: rap-cel produced deep B-cell depletion, pelacarsen substantially lowered Lp(a), del-desiran reached muscle and reduced DMPK mRNA. The failures cannot be explained simply by a failure to engage the intended biology but from the boundary between a working platform and a biological context preclinical work had not resolved.

CAR-T in Autoimmune Disease: When a Platform Outruns Its Indication

The CAR-T pause is the most sobering of the three because patients died. Rapcabtagene autoleucel (rap-cel, or YTB323) had become a leading example of CAR-T moving beyond oncology, and the holds covered eight studies in lupus nephritis, systemic sclerosis, ANCA-associated vasculitis, inflammatory myopathies, rheumatoid arthritis, Sjögren's disease, myasthenia gravis, and multiple sclerosis ^[1]. IEC-HS is a severe hyperinflammatory syndrome of uncontrolled immune and macrophage activation; it overlaps clinically with cytokine release syndrome but is a distinct immune effector cell toxicity that can be harder to manage. Bristol Myers Squibb had also paused autoimmune enrollment for its CD19 CAR-T, zola-cel, after separate inflammatory events ^[1]. Rap-cel's oncology studies remained active.

That difference matters. Rap-cel is manufactured on Novartis's T-Charge platform, which keeps cells outside the body for less than two days and preserves a relatively undifferentiated, stem-like phenotype. In cancer that is the advantage: such cells expand more effectively and persist longer. In autoimmune disease, antigen burden and distribution, baseline immune activation, and the surrounding inflammatory environment are fundamentally different. The risk that is acceptable in heavily pretreated cancer can be unacceptable in a young patient with decades of life ahead.

Figure 2. Novel T-Charge manufacturing platform vs. traditional CAR T-cell manufacturing ^[5].

The root of this translational failure often lies in the preclinical infrastructure. Conventional tumor-bearing immunodeficient models are engineered to answer one simple question: does the CAR-T reach and kill the target? By that metric, rap-cel performed flawlessly. However, these models cannot predict how a stem-like CAR-T product will expand and trigger hyperinflammation in a host possessing a fully functional, already-dysregulated autoimmune system. Evaluating this distinct risk requires advanced humanized immune system models on autoimmune backgrounds, where cytokine dynamics and macrophage activation can be measured as primary, predictive readouts.

Pelacarsen and Lp(a): Why Target Engagement Is Not Clinical Validation

Pelacarsen's failure will be debated for years, because Lp(a) looked like one of the best validated untapped targets in cardiovascular medicine, with Mendelian randomization, epidemiology, and biochemistry all aligned. Lp(a)HORIZON enrolled 8,323 patients with established cardiovascular disease and elevated Lp(a), testing pelacarsen against a composite of cardiovascular death, myocardial infarction, stroke, and urgent coronary revascularization, in the overall population with Lp(a) ≥70 mg/dL and a prespecified subgroup at ≥90 mg/dL. Pelacarsen lowered Lp(a), but the trial did not meet its primary cardiovascular endpoint. Chief Medical Officer Shreeram Aradhye put it carefully: lower Lp(a) levels were observed, but the findings did not demonstrate that the reduction translated into lower cardiovascular risk ^[2].

Figure 3. Pathophysiological mechanisms linking elevated Lp(a) to atherosclerosis, thrombosis, vascular inflammation, and aortic valve calcification ^[6].

Three explanations deserve scrutiny, and they are not mutually exclusive. Lifelong genetic exposure to lower Lp(a) may not be equivalent to lowering it pharmacologically for a few years in patients with established atherosclerosis. Background therapy was highly optimized, with mean LDL-C around 66 mg/dL, leaving a narrow margin of residual risk. And the causal contribution of Lp(a) may depend on timing, particle biology, or disease stage in ways circulating concentration does not capture.

Underneath these clinical complexities lies a fundamental preclinical limitation that is easy to overlook: wild-type mice do not naturally carry an LPA gene or produce Lp(a). Human genetics shows Lp(a) is causally relevant, which is not evidence that pharmacologically lowering it in late-stage will reverse atherosclerosis. A useful model also takes more than human apolipoprotein(a), because human apo(a) needs human apoB-100 to form authentic covalent particles, and isoform size, oxidized phospholipid content, and lysine-binding function all shape the biology. Cyagen's conditional ROSA26 hLPA and liver-restricted albumin-promoter hLPA strains, crossed onto hyperlipidemic or humanized APOB backgrounds, make it possible to ask in mice what Lp(a)HORIZON had to answer in 8,323 patients: does suppressing apo(a) alter lesion burden and thrombotic phenotype, or mainly plasma concentration?

Figure 4. Pelacarsen lowers Lp(a) by targeting hepatic LPA mRNA with an antisense oligonucleotide, reducing apo(a) production ^[7].

Del-Desiran and AOCs: When Delivery Works but Function Does Not

Del-desiran was expected to validate antibody-oligonucleotide conjugates (AOCs) as the answer to muscle delivery, and in a narrow sense it may have. The Phase III HARBOR study randomized 159 patients with myotonic dystrophy type 1 (DM1) to del-desiran or placebo every eight weeks over 54 weeks, with video hand opening time (vHOT) as the primary endpoint. The study missed. Novartis reported clinical activity across secondary and exploratory endpoints and a safety profile consistent with previous data, but the market read the topline as a serious setback, sending Novartis's U.S.-listed shares down roughly 12.4% in premarket trading ^[3].

Figure 5. Del-desiran is designed to deliver an oligonucleotide payload to muscle, reduce toxic DMPK RNA, restore MBNL function, and correct downstream mis-splicing in myotonic dystrophy type 1 ^[8].

DM1 is slowly progressive and multisystem, and vHOT is a relatively new functional endpoint, so variability and placebo response are fair questions, though blaming the endpoint alone is too easy. The chain from DMPK knockdown to splicing correction to reduced myotonia to a difference patients notice is long, and 54 weeks may not be enough for its far end to become visible. Preclinical work also has to separate proving delivery from proving that delivery is enough. TfR1-targeting antibodies can be species-selective, so rodent studies may need surrogate binders, and standard DM1 models show molecular correction and improved myotonia without predicting human functional gain. Target-humanized receptor models with conjugate-specific biodistribution studies narrow that gap by testing the actual construct in the clinical receptor context.

Modality Risk Is Falling, Biological Context Risk Is The New Frontier

These three setbacks show where development risk now sits. Modality risk is falling: the field is very good at engineering CARs, stabilizing oligonucleotides, and conjugating payloads. Context risk is not. A platform validated in one indication carries no automatic credit into another, a genetically supported target is not necessarily modifiable at every stage of disease, and a biomarker that moves is not an endpoint that moves. Each gap can be closed earlier, but only if the model interrogates the new context rather than repeating the one that validated the platform.

Familiar and standard models get reused because they are fast, cheap, well characterized, and often produces the “expected” answer. But as Novartis’s eight-days, $12 billion exposure reminds us: the cheapest place to discover a flawed biological assumption is still the animal facility.

Cyagen is engineered to solve context risk. From humanized immune system and CAR-T models for cell therapies, to target-humanized models and biodistribution studies for ADCs and AOCs, to custom genetic strains for targets such as LPA that standard mice lack, we build the preclinical infrastructure required to test the actual clinical context. The right time for that conversation is before the go/no-go decision, not after the topline result.

👉 [Find the Right Preclinical Model for Your Research]

👉 [Explore Our ADC and AOC Development Platform for Translational Research]

👉 [Discuss Your Next Preclinical Study with Our Experts]

Reference:

[1] Russo M. Novartis halts autoimmune CAR-T trials after 3 deaths, as BMS also pauses studies. Fierce Biotech [Internet]. 2026 Aug 31 [cited 2026 Sep 10]. Available from: https://www.fiercebiotech.com/biotech/novartis-bristol-myers-squibb-halt-car-t-cell-trials-due-immune-events

[2] Novartis. Novartis announces Lp(a)HORIZON Phase III topline results for pelacarsen in patients with elevated Lp(a) and established cardiovascular disease (CVD) [Internet]. 2026 Sep 4 [cited 2026 Sep 10]. Available from: https://www.novartis.com/news/media-releases/novartis-announces-lpahorizon-phase-iii-topline-results-pelacarsen-patients-elevated-lpa-and-established-cardiovascular-disease-cvd

[3] Taylor NP. Novartis stock slides as centerpiece of $12B Avidity buyout flunks ‘crucial’ phase 3 test [Internet]. Fierce Biotech. 2026 Sep 8 [cited 2026 Sep 10]. Available from: https://www.fiercebiotech.com/biotech/novartis-centerpiece-12b-avidity-buyout-flunks-crucial-phase-3-test

[4] Taylor NP. Novartis stock slides as centerpiece of $12B Avidity buyout flunks ‘crucial’ phase 3 test. Fierce Biotech [Internet]. 2026 Sep 8 [cited 2026 Sep 10]. Available from: https://www.fiercebiotech.com/biotech/novartis-centerpiece-12b-avidity-buyout-flunks-crucial-phase-3-test

[5] Dickinson MJ, Barba P, Jäger U, Shah NN, Blaise D, Briones J, Shune L, Boissel N, Bondanza A, Mariconti L, Marchal AL, Quinn DS, Yang J, Price A, Sohoni A, Treanor LM, Orlando EJ, Mataraza J, Davis J, Lu D, Zhu X, Engels B, Moutouh-de Parseval L, Brogdon JL, Moschetta M, Flinn IW. A Novel Autologous CAR-T Therapy, YTB323, with Preserved T-cell Stemness Shows Enhanced CAR T-cell Efficacy in Preclinical and Early Clinical Development. Cancer Discov. 2023 Sep 6;13(9):1982-1997. doi: 10.1158/2159-8290.CD-22-1276. PMID: 37249512; PMCID: PMC10481129.

[6] Animati FM, Proietti S, Auletta F, Montone RA, Cappannoli L, Fracassi F, Gaspardone A, Burzotta F. Lipoprotein(a) in Coronary Artery Disease and Aortic Stenosis: Pathophysiology, Clinical Impact, Interventional Implications and Emerging Targeted Therapies. J Clin Med. 2026 Jul 28;15(15):5897. doi: 10.3390/jcm15155897. PMID: 42590001; PMCID: PMC13466283.

[7] Kim KA, Park HJ. New Therapeutic Approaches to the Treatment of Dyslipidemia 2: LDL-C and Lp(a). J Lipid Atheroscler. 2023 Jan;12(1):37-46. doi: 10.12997/jla.2023.12.1.37. Epub 2022 Nov 17. PMID: 36761062; PMCID: PMC9884549.

[8] Kwan TT, Meng Q, Delos Santos N, Johnson NE, Thornton CA, Malecova B, Burke R, Tai LJ, Levin AA, Younis HS, Flanagan WM, Zhu Y. Delpacibart etedesiran improves the molecular pathology of myotonic dystrophy type 1 in the phase 1/2 MARINA study. Mol Ther. 2026 Jul 1;34(7):3872-3884. doi: 10.1016/j.ymthe.2026.03.013. Epub 2026 Mar 11. PMID: 41821312; PMCID: PMC13330057.