Authors' reply

Drs Sharan and Bharadwaj object to our representation of Kraepelin's manic–depressive illness with DSM–IV psychotic bipolar I disorder because Kraepelin used the term to refer to a broader spectrum of affective disorders. By this logic our inclusion of patients fulfilling modern diagnostic criteria for schizophrenia rather than dementia praecox should be equally unacceptable to them. However, the Kraepelinian dichotomy continued to stimulate controversy over the past century precisely because the evolution of diagnostic criteria for these syndromes consistently failed to fully separate the disorders on clinical and neurobiological grounds. Thus ‘the Kraepelinian dichotomy’ has come to refer to the distinction between schizophrenia and bipolar disorder (Craddock & Owen, 2005). Furthermore, there is considerable morphometric heterogeneity between bipolar disorder and major depressive disorder (Strakowski et al, 2002), which underlines the need for more homogeneous rather than broader-spectrum affective disorder patient groups for magnetic resonance imaging studies.

Their hypothesis that our failure to identify grey matter abnormalities in bipolar disorder may result from recruiting patients with less-severe illness and a group of healthy volunteers with conditions associated with structural abnormalities is difficult to reconcile with our success in identifying white matter abnormalities in the same patients and typical grey matter deficits in patients with schizophrenia, who were recruited in a similar manner.

Moreover, there is no reason why healthy volunteers would have higher rates of the conditions suggested than the patient groups. Ethnicity is given in the cited associated paper (McDonald et al, 2004). Although type 2 errors are frequently possible, the magnetic resonance sequences used are common for computational morphometry studies and successfully detected differences in patients and healthy volunteers. The ICBM152 template was indeed used, as is standard with the SPM99 (Statistical Parametric Mapping 99) package, to create the customised template. We accept that the risk of type 1 errors would be lower with a single screening analysis of covariance but we hypothesised changes in a voxelwise comparison between each patient group and the control group and thus reported these results.

Although results from computational morphometry have been interpreted variously as volume change, shape change or a result of other processes altering voxel intensity, we dispute the simplistic assertion that region of interest methodologies are ‘more accurate’ – such methodologies have their own difficulties, in particular with interrater reliability and the optimal parcellation boundaries chosen for structures, and the two methodologies are perhaps better viewed as complementary. Region of interest analyses of a similar sample demonstrated that volume deficits of the hippocampus and amygdala characterise schizophrenia but not bipolar disorder (Marshall et al, 2004; McDonald et al, 2006). This is consistent with our computational morphometry study – and with Kraepelin's seminal dichotomy.