08/2026
Challenges facing CNS drug development
Drug development in neurology has an abnormally high failure rate of 94.1% between Phase 1 and approval (in period 2011–20). Candidates fail more, and at later stages, than in other indications, rendering neurological drug development both risky and expensive. Here I try to set out some of the reasons why targeting the CNS in particular is so challenging.
Preclinical development
- Mouse model challenges. All neurological diseases have a subjective component and some, such as mood disorders, are defined by them. It is much harder to prove the presence of anhedonia than it is the presence of a tumour in a mouse model.
- Targeting a molecule does not necessarily mean targeting a function. The chemical identity of a neuron does not solely define its function as other factors, such as gross location and functional connectivity, are also central. This makes it more difficult to target a pharmaceutical product to a specific functional deficit, and increases the risk of adverse effects.
Patient stratification
- Diagnostic categories are behavioural. Brains are complex systems and different patterns of activity can produce similar behavioural outcomes. A therapy that has a consistent effect on biology will produce inconsistent efficacy data if given to patients with different pathological mechanisms.
- Individuals within a diagnostic category may not even share behaviours. The DSM-5 often premises diagnosis on the presence of a subset of common symptoms. Two patients with the same disease could have very different phenotypes.
- Neurodegeneration is irreversible. Interventions must be targeted to pre-symptomatic or early-stage patients for such diseases. The exception may be cell graft therapies, as trialled for Parkinson’s disease.
Clinical evaluation
- Blood–brain barrier. The brain is hard to reach and neurons cannot be replaced, making it difficult to get therapies in contact with neural tissue, and difficult to sample neural tissue to check success. CSF biomarkers still require an invasive procedure, and most CSF proteins do not come from the brain.
- Subjective outcome measures. Few biomarkers are available for neurological disease, meaning endpoints are reported or behavioural. Placebo effects and functional unblinding are common.
- Neurodegeneration is age-related. Older cohorts are more likely to have comorbidities and polypharmacy, adding noise to effect detection. Follow-up periods are long, adding to trial costs.
This is a daunting list. However, some of the features that make the brain a difficult target also offer unique routes to intervention, and some progress has been made.
- Physiological pathways to brain modulation. The brain is the only organ whose structure can be changed by conversation. Pharmacologically enhanced talking therapy, as the new crop of psychedelic candidates promises, may increase the potency of these interventions.
- Location-dependent targeting. Advances in neurotech allow us to target dysfunctional areas based on anatomy rather than chemistry. This has proven successful with deep-brain stimulation, again in Parkinson’s disease, and new non-invasive approaches open a wider range of indications.
- Successful gene therapies. A one-off gene therapy has transformed spinal muscular atrophy, while a similar approach in Huntington’s disease has slowed progression in trials and is now under FDA review. Monogenic disorders are slowly becoming treatable.