A total of 64 participants completed the study, consisting of 45 patients with schizophrenia and 19 controls. Demographic and clinical data are summarised in Table 1. The patient and control groups did not differ significantly on demographic variables with the exception of years spent in full-time education, which may be expected given that schizophrenia is commonly associated with lower than expected educational achievement (Green 2001; see also Surguladze et al. 2007).
The majority of patients (N = 38) were treated with SGA (clozapine = 10, risperidone = 7, olanzapine = 16, amisulpride = 2, aripiprazole = 3), six patients were treated with FGA (haloperidol = 1, flupentixol = 2, sulpiride = 2, chlorpromazine = 1), and one patient was untreated at time of fMRI testing.
A number of patients were co-medicated with anticholinergic, benzodiazepine, mood stabilising, or antidepressant compounds. Anticholinergic compounds were administered to three patients on FGA and to two patients on SGA. Benzodiazepines were administered to four patients on SGA. Mood stabilisers were administered to one patient on FGA and eight patients on SGA. Antidepressants were administered to two patients on FGA and 15 patients on SGA.
Working memory task performance
Working memory task performance data are shown in Table 2. The patient and control groups did not significantly differ in percent correct trials (F[1,62] = 2.17, p = 0.15, η
= 0.03). There was an effect of load on percent correct trials (F[2,124] = 73.47, p < 0.001, η
= 0.54), indicating fewer correct responses with increasing load, but no group-by-load interaction (F[2,124] = 1.27, p = 0.29, η
The groups also did not significantly differ in the latency to correct trials (F[1,61] = 1.00, p = 0.32, η
= 0.02). As with percent correct responses, there was an effect of load on latency (F[2,122] = 17.11, p < 0.001, η
= 0.22), indicating longer latencies with increasing load, but no group-by-load interaction (F[2,122] = 0.80, p = 0.45, η
Similarly, there was an effect of load on the percentage of omission errors (F[2,124] = 38.24, p < 0.001, η
= 0.38), indicating more omission errors with increasing load, but no effect of group (F[1,62] = 0.04, p = 0.85, η
= 0.001) or group-by-load interaction (F[2,124] = 0.23, p = 0.80, η
BOLD response by group and load
Both groups activated an extensive fronto-parieto-striato-thalamo-cerebellar network during performance of the n-back task. Figure 1 shows the activation of the combined group as a function of load (i.e., increase in BOLD signal with increasing working memory load); the main effect of task (across conditions of load) in the combined sample gave a very similar result.
The main effect of group (across conditions of load) revealed three clusters of increased activation (p < 0.05 corrected cluster level) in patients relative to controls: (1) occipital cortex (BA17; Talairach coordinates of peak voxel, x = −20, y = −88, z = −9; 857 voxels; Z = 4.39) extending into the cerebellum; (2) left middle frontal gyrus (BA9; x = −46, y = 15, z = 36; 1,913 voxels; Z = 4.26) extending into pre- and postcentral gyrus; and (3) right inferior frontal gyrus (BA45/47; x = 46, y = 18, z = 18; 854 voxels; Z = 4.06) extending into right middle frontal gyrus. In contrast, there were no significant increases in controls relative to patients at the corrected or uncorrected cluster level and there were no significant voxels at the chosen height threshold (p < 0.001).
The group-by-load interaction (Fig. 2) revealed two significant clusters (p < 0.05 corrected cluster level): (1) left middle frontal gyrus (BA9; x = −46, y = 17, z = 36; 2,032 voxels; Z = 4.43) and (2) right inferior frontal gyrus (BA45/47; x = 46, y = 18, z = 18; 749 voxels; Z = 3.94) extending into right middle frontal gyrus. A third cluster with a peak in the occipital cortex (BA17; x = −22, y = −86, z = −11; 535 voxels; Z = 4.10) showed the same pattern but narrowly missed the level of statistical significance (p = 0.054, corrected cluster level). These clusters showed stronger increases across load in patients than in controls. On the other hand, there were no significant clusters showing stronger increases across load in controls than in patients at the corrected or uncorrected cluster level, and there were no significant voxels at the chosen height threshold (p < 0.001).
The prefrontal clusters that emerged in this interaction analysis appeared not to be part of the activation seen in response to load across both groups (Fig. 1) but appeared to be neighbouring it. In order to verify this, the group-by-load interaction contrast was masked with the contrast image resulting from the main effect of load; the two prefrontal areas from the interaction analysis also emerged in this masked analysis, indicating that these are voxels that show an additional increase with load in the patient group but not in the controls and not in the combined group.
To better understand the origin of these group-by-load interaction effects, the mean BOLD signal in each of the two frontal clusters that showed a significant interaction effect was extracted as described above and repeated measures ANOVAs with the within-subjects factor of load were run in SPSS separately for the two groups. In the left prefrontal cluster, we found a linear increase in BOLD as a function of load in the patient group (F[2,88] = 6.65, p = 0.002, η
= 0.13) but not in the controls (F[2,36] = 0.35, p = 0.71, η
= 0.02). Similarly, in the right prefrontal cluster, there was a linear increase in BOLD as a function of load in the patient group (F[2,88] = 8.26, p = 0.001, η
= 0.16) but not in the controls (F[2,36] = 0.77, p = 0.47, η
Taken together, this pattern indicates that patients showed a significantly greater BOLD increase in response to working memory load than controls in lateral prefrontal cortical areas that did not show a significant main effect of load.
Association of BOLD with performance level in the patient group
In keeping with the analysis plan, the patient group was split into those with high and low performance along the patient group's median (62.49%), yielding two groups of N = 22 each (the subject with the median score was excluded). N-back data of the two groups are summarised in Table 3. Importantly, the high- and low-performing patient groups did not differ from each other in any demographic or clinical variables (all p > 0.05). However, the two patient groups differed significantly from each other in percent correct response at each level of load (all F > 11.88, p < 0.002), as expected. They also differed with regards to omission errors (all F > 6.82, p < 0.02) but not in reaction time (all F < 1.85, p > 0.18).
At the level of BOLD, each patient subgroup showed similar activation increases in response to load as the combined group (see Online Resource 1). An anatomically unconstrained voxelwise comparison between the two patient groups did not find a main effect of performance and there was no performance-by-load interaction (at p = 0.05, corrected cluster level). A comparison of the prefrontal clusters that had showed group-by-load interactions above similarly did not yield significant main effects of performance for either cluster (both p > 0.79). However, there was a significant performance-by-load interaction for the left (F[2,84] = 4.09, p = 0.02, η
= 0.09) but not right (p = 0.24, η
= 0.03) cluster. In order to better understand this interaction, within-subject ANOVAs of load in the left prefrontal cluster were calculated separately for the two performance groups. These showed that the high- (F[2,42] = 9.63, p < 0.001, η
= 0.31) but not the low-performing patients (p = 0.83, η
= 0.009) showed a statistically significant linear increase in BOLD with load (Fig. 3).
Effects of antipsychotic treatment
The split according to treatment yielded two groups of patients, those treated with FGA (N = 6) and those treated with SGA (N = 38) antipsychotics (N = 1 excluded). N-back performance data of the treatment groups are summarised in Table 3. There were no significant associations of treatment with any demographic variables, clinical variables, or anticholinergic, antidepressant, mood stabiliser, or benzodiazepine treatment (all p > 0.34), and the treatment factor did not significantly interact with the performance variable (χ
2 = 0.89, df = 1, p = 0.35): in the SGA group, there were 17 low- and 20 high-performing patients, and in the FGA group, there were four low- and two high-performing patients. There were no main effects of treatment (all p > 0.18) or treatment-by-load interactions (all p > 0.22) on n-back task performance; therefore, any treatment effects on BOLD reported below are considered to be independent of performance effects.
At the level of BOLD, an unrestricted voxelwise analysis did not yield any significant differences between the two treatment groups or treatment-by-load interactions (all p > 0.05, corrected cluster level). When considering the extracted prefrontal ROIs, no main effects of treatment were seen in either cluster (both p > 0.51). However, there was a significant treatment-by-load interaction (F[2,84] = 5.16, p = 0.008, η
= 0.11) in the left prefrontal cluster. This interaction (Fig. 4) indicated that patients treated with SGA showed an increase in BOLD as a function of load, whereas the patients treated with FGA showed a reduction. The interaction effect appeared similar in the right prefrontal cluster but did not reach statistical significance (p = 0.13). A further split of the SGA into those patients treated with clozapine (N = 10), risperidone (N = 7), and olanzapine (N = 16) did not yield any significant group or interaction effects for the prefrontal ROIs. Therefore, the SGA group as a whole showed a stronger parametric left prefrontal BOLD increase than the FGA group, with no significant differences detected between different SGA compounds.
Association of BOLD with symptom severity
There were no significant correlations of BOLD from the combined n-back activations with PANSS total or subscale scores (all p > 0.05, corrected cluster level). Similarly, there were no correlations between the extracted ROIs and PANSS symptom scores (all p > 0.06).