The NMDA receptor partial agonist d-cycloserine does not enhance motor learning

Rationale: There has recently been increasing interest in pharmacological manipulations that could potentially enhance exposure-based cognitive behaviour therapy for anxiety disorders. One such medication is the partial NMDA agonist d-cycloserine. It has been suggested that d-cycloserine enhances cognitive behaviour therapy by making learning faster. While animal studies have supported this view of the drug accelerating learning, evidence in human studies has been mixed. We therefore designed an experiment to measure the effects of d-cycloserine on human motor learning. Methods: Fifty-four healthy human volunteers were randomly assigned to a single dose of 250mg d-cycloserine versus placebo in a double-blind design. They then performed a motor sequence learning task. Results: D-cycloserine did not increase the speed of motor learning or the overall amount learnt. However, we noted that participants on d-cycloserine tended to respond more carefully (shifting towards slower, but more correct responses). Conclusion: The results suggest that d-cycloserine does not exert beneficial effects on psychological treatments via mechanisms involved in motor learning. Further studies are needed to clarify the influence on other cognitive mechanisms.

. Additional behavioral results. A) Log-transformed RTs for the first three sequence repeats within the first learning block (block 2). B) Percentage of early responses (i.e. within 200ms of trial onset). C) Accuracy (%) for the initial learning and the two consolidation phases. Error bars show the standard error of the mean, *p<0.055  Table S1. Socio-demographic and questionnaire measurements for the placebo (pla) and the dcycloserine (DCS) groups collected at baseline. Values are means and standard errors; p-values are for two-sided between-subject t-tests. Abbreviations and references for the questionnaires (see also methods): BDI [Beck Depression Inventory, Beck et al. (1996)], Trait anxiety (Spielberger and Gorsuch, 1983), Neuroticism (Eysenck and Eysenck, 1975) Table S2. Visual analogue scale measurements for the placebo (pla) and d-cycloserine (DCS) groups. Data was acquired before the administration of DCS or placebo and after. Values are means and standard errors as indicated by participants by placing a tick on a line of length 100mm. P-values are for two-sided between-subject t-tests of the scores before or after DCS/ placebo administration or the difference (diff., i.e. before minus after).

Supplementary methods and results Study inclusion criteria
Inclusion criteria for participants were: no history of neurological or axis 1 psychiatric disorder as assessed using the Structured Clinical Interview for DSM-IV Axis I Disorders SCID-CV (First, 1996), no first-degree family member with a history of a severe psychiatric disease, passing a basic medical screening and examination for any type of medical illness relevant to the study, no current or recent (6 weeks) CNS-active medication or medication with cycloserine, ethionamide or isoniazid, a body mass index (BMI) between 18-30 kg/m 2 , non-or light-smoking (<5 cigarettes a day), fluent English skills. Female participants were neither pregnant nor breast-feeding. This experiment was collected as part of a larger study in which a third drug (hydrocortisone) was also used (data not reported here).

Assessing initial learning and maximum improvement separately
Our results in the main report showed that learning across the whole experiment was not affected by DCS. To test whether there were maybe more subtle learning differences, we tested whether DCS enhanced either the initial speed of learning or the maximum improvement in performance. To test for effects on the initial speed of learning, we performed an ANOVA over the first four learning blocks (i.e. blocks 2 to 5). To assess the maximum improvement in performance, we compared the difference score between the last and the first learning block between the two groups using an independent-sample two-tailed t-test.
In terms of the early speed of learning, we again found that participants showed learning (ANOVA, effect of block: F(2.2,115.9)=23.5, p<0.001). This early learning appeared to differ between the groups, with the DCS group showing faster learning (ANOVA, block x group: F(2.2,115.9)=3.4, p=0.032). However, a closer look at the data (figure S2A) revealed that this result was not actually due to DCS enhancing the learning speed. Within each block, there were three repeats of the 10element sequence; an analysis of these first three repeats within the first learning block (figure S2A) revealed that, in fact, the placebo group already showed more learning within that first learning block, i.e. block 2 (t(52)=-2.1, p=0.043 for the difference between the third and the first sequence repeat within the first block). This suggested that the steeper difference in RT in the DCS compared to the placebo group when comparing the early learning blocks (figure 1A) should not be interpreted as faster learning in the DCS group. Instead, the data suggests if anything that DCS may have slowed down very early learning within the first learning block.
Next, we analyzed whether DCS affected the overall amount learnt, i.e. the difference between the RT of the last and the first learning block. We did not find a difference between the two groups (t(52)=-0.6, p=0.52). This result did not change when the initial learning performance (block 2) was included as a covariate (F(1,53)=0.026, p=0.87)

Anticipatory responses in early learning
In the analyses above and in the main paper, we included RTs that occurred before the appearance of the cue (see Methods in main paper). The reason for this was that we noted that with learning participants started to respond in anticipation rather than in response to the cues appearing. Figure  S2B shows the percentage of anticipator RTs, i.e. RTs faster than 200ms after cue onset. Both groups started showing more anticipatory responses as they progressed and learned in the task. The cut-off of 200ms was chosen based on RTs in the 'random block' (block 15). Here, we found that the mean