Harnessing N-Methyl-d-Aspartate Receptors for New Treatment Development in Psychiatry: Positive Lessons From Negative Studies

Use of d-cycloserine is based on its well-replicated role in enhancement of fear learning. However, d-cycloserine is also a mixed agonist/antagonist at the NMDAR, and at high doses (>250 mg) it acts as a net antagonist of NMDARs. For example, at these doses, it worsens psychosis in schizophrenia patients but improves symptoms of treatment-refractory depression (5). Even at low doses, d-cycloserine acts as a “low-efficacy” agonist, as compared with endogenous compounds, such as glycine or d-serine. Whether or not the mixed agonist/antagonist profile and low efficacy of d-cycloserine is beneficial to its clinical effects remains to be determined. Other compounds that could alternatively target NMDARs include glycine and d-serine themselves, glycine transport inhibitors (e.g., bitopertin), and metabotropic glutamate agonists (6).

Another complexity in translating rodent findings to humans is determining the ideal dosing and treatment interval. In the case of d-cycloserine, a relatively wide range of dosing, from 25 mg to 250 mg, appears to be effective. In contrast, the timing of dosing may be critical. The main beneficial effect of d-cycloserine is in transferring learning from short-term, NMDA-independent to longer-term, NMDA-dependent forms of memory (4). Thus, the effects of d-cycloserine may be most robust when it is administered after, not before, CBT sessions.

Even more critically, repeated exposure to d-cycloserine may decrease the effectiveness of subsequent doses, possibly as a result of receptor desensitization or change in subunit composition (7). At present, the minimum time required to recover from desensitization after a single dose is unknown, but intervals of 1 week or more may be required (8). In the Hofmann et al. study, medication was administered 1 hour before desensitization sessions. Since peak brain levels are reached 4–8 hours after administration, this dosing would produce maximal effect only after the CBT session had ended, as desired. Nevertheless, delaying treatment until after the session would minimize pretreatment levels and still allow peak levels to be reached within the reinforcement window. Another advantage of delaying treatment is that medication could be given only after sessions in which significant within-session improvements were observed (4).

As with most academic studies, the Hofmann et al. study was designed to detect a moderate-sized treatment effect (0.5 standard deviations) with a power of 80%. As power increases, the required sample size increases exponentially, making it unfeasible to increase study power beyond this. Nevertheless, the power of 80% means that there remains a 1 in 5 chance that the investigators were simply unlucky and that the negative result was in fact a false negative. Stated another way, just as the possibility of no effect (effect size=0) cannot be excluded, so too the possibility of a moderate-sized effect (effect size=0.5) cannot be excluded.

Another way to interpret the results is in terms of the number needed to treat to achieve remission. In the Hofmann et al. study, remission rates were 10 percentage points higher in the d-cycloserine group than in the placebo group at the posttreatment assessment (34.5% compared with 24.4%), suggesting that if 10 patients were treated with d-cycloserine, one extra patient would show a favorable outcome (i.e., number needed to treat=10). The differences would be even more favorable (number needed to treat=8) if dropouts from the study, which were also higher in the placebo group (N=15) than in the d-cycloserine group (N=7), were also presumed to be nonremitters. These number-needed-to-treat values, if confirmed in larger studies, would be similar to those estimated for use of antidepressants in primary care settings (9). The loss of difference during follow-up is also a concern, but if anticipated, it could potentially be addressed through use of maintenance treatments.

Another take-home lesson is the need for biomarkers even in behavioral treatment research. Increasingly, the pharmaceutical industry is adopting a “fast-fail” approach, in which compounds are not entered into clinical studies until biomarkers have been developed that would permit conclusive interpretation of negative results (10). For example, anxiety signals in social anxiety disorder are linked to overactivation of the amygdala (11). In the Hofmann et al. study, it is unknown whether d-cycloserine was ineffective because it did not affect amygdalar activation or because despite such an effect there was no reduction in symptoms experienced. Without biomarkers that can guide interpretation of negative results, the lessons that can be learned from negative studies remain limited.

Although the Hofmann et al. study did not demonstrate significant between-group differences at follow-up, it did confirm an enhanced rate of improvement. In contrast, the study suggests that repeated dosing of d-cycloserine over days offers little benefit over single doses, and that future studies, which seem to be merited, should use longer intervals—perhaps up to 1 week—between treatments. On theoretical grounds, it might also be preferable to administer d-cycloserine or other NMDAR agonists after, rather than before, CBT sessions and to administer it only after sessions in which significant improvement has been observed. Finally, future research should probably involve fewer subjects studied more intensively, so that mechanisms of both successes and failures can more easily be translated into improved treatment approaches.