The news surfaced quite recently that bitopertin had failed in phase III clinical trials in schizophrenia. Its intended indication was for combination with antipsychotics in the treatment of persistent negative symptoms or sub-optimally controlled positive symptoms associated with schizophrenia. Bitopertin was of interest because it represented an entirely new class of compound. It is a glycine transporter type 1 (GlyT1) inhibitor that increases levels of the neurotransmitter glycine by inhibiting reuptake at the synapse. Glycine acts as a co-agonist of glutamate at N-methyl-D-aspartate (NMDA) receptors. Its use in man has been supported by Positron emission tomography to show increases in glycine availability and evidence of safety in the key area of cardiac conduction (QTc). Moreover, a key phase II trial showed promising findings in schizophrenia. Although I have acted as an advisor to Roche in the past, I have no particular knowledge of the development plan. Nevertheless, it is obvious that failure in phase III is expensive and may well be the trigger to discontinuing development: I do not know.

If the worst happens, it is bad news for applied neuroscience in several different ways and it poses questions that recur in our field. It reflects the concerns that were raised at the ECNP Summit three years ago. Neuroscience can and does provide us well-characterised and novel compounds of defined pharmacology. Do we know what to do with them? Quite simply the answer is often, no. The clinical phase of drug development is a car crash waiting to happen for any drug that is not essentially a ‘me too’. Novel compounds and novel indications may be at a particular disadvantage for a variety of reasons related to recruitment, assessment of outcomes and the perverse incentives of commercial clinical trial practice.

Thus, phase III clinical trials are conducted, often in a hurry, in many sites. As many as 30% of these sites will contribute nothing but noise, whether or not an efficacy signal is present. Yet analysis of primary endpoints is conventionally of the mean effect across all sites and all patients whether or not they actually take the drug. That means trials can fail because a large number of sites fail, not because the drug doesn’t work. As a patient, am I interested in the sites that produce noise or the sites that may show quite impressive treatment effects? Why has there been almost no statistical innovation in how data sets are analysed, at least for public consumption?

If bitopertin has really failed in its assumed first indication, what about alternative indications or experimental uses? There is clearly a need for a more flexible semi-experimental phase in clinical development when what a drug will actually do for patients can be subjected to the guided serendipity of clinical experimentation. Regulation, and the disproportionate application of the costly strangleholds of the European Clinical Trials Directive to experimental studies makes this almost impossible at present. How to reconcile the legitimate interests of the patent holder on the one hand, and the patient on the other, is a central challenge of our time. For too many companies the obvious solution has been to give up altogether.

Best regards,



Guy Goodwin, ECNP President