Participants

Six right-handed patients with either schizophrenia or schizo-affective disorder diagnosed using DSM–IV ³ criteria were studied. These patients were treated with clozapine (n=3), haloperidol (n=1) or olanzapine (n=1), or no antipsychotic drug (n=1). Informed written consent was obtained from all participants prior to the study.

Imaging

We used a 1.5 T Signa LX scanner (General Electric, Milwaukee, Wisconsin, USA) to obtain T ₁-weighted structural images of 14 contiguous 6 mm slices acquired parallel to the anterior–posterior commissural line. These images were used for anatomical identification and coincided with slices in the functional runs. For functional runs, participants were instructed to depress a button with their right hand to mark the onset of each auditory/verbal hallucination and to release the button at termination of the event; six or seven of these runs lasting 4 min 6 s each were used for each participant. Functional runs – 14 axial–oblique slices of single-shot echoplanar imaging: repetition time (TR) 1500 ms, echo time (TE) 60 ms, flip angle 60°, 64×64 acquisition matrix, voxel size 3.125 mm×3.125 mm×6 mm – were motion-corrected using the statistical parametric mapping algorithm and subjected to a spatial Gaussian filter with 2 pixels full width at half maximum. The time course of button-presses in each scan was convolved with a standard model of the haemodynamic response function. For each individual and each scan, blood oxygen level dependent (BOLD) signal fluctuations in all pixels composing a given slice were correlated with the reference time course at temporal shifts ranging from −4.5 s to 4.5 s in 1.5-s steps, after removing the mean time course across the slice from each pixel time course and from the reference time course. For each time lag and run, correlations were transformed to an approximately normal Gaussian distribution. Gaussian-transformed maps were averaged across runs, yielding a map representing the strength of correlations to the reference button-press time course in terms of standardised z-values. The seven z-maps of correlations from each participant for each time lag were then transformed to Talairach coordinates. At each Talairach pixel and time lag, a t-test was used to assess significance of deviation of z-values from zero. This correlation-based analysis can accommodate shorter inter-hallucination intervals even though the BOLD signal may not return fully to baseline – irregular temporal spacing of hallucinations can improve statistical power, just as ‘jittering’ allows closer event spacing in event-related designs.

Right middle temporal site

The right middle temporal site (Brodmann area 21) was located remarkably close to middle temporal gyrus sites of pre-hallucination activation reported by Lennox et al and Shergill et al. ^{Reference Lennox, Bert, Park, Jones and Morris1,Reference Shergill, Brammer, Amaro, Williams, Murray and McGuire2} Activation in the bilateral superior temporal gyrus (Brodmann area 22) emerged at negative time lags also, but peaked later and was broadly distributed over both positive and negative time lags. Our correlation-based method for mapping BOLD signal time course will produce temporal smearing that broadens with increasing neural activation. The temporal pattern of our data suggests therefore that the more robust bilateral activation in the superior temporal gyrus arose somewhat later – perhaps at hallucination onset – than the middle temporal gyrus activation. Bilateral activation of the latter region has been associated with aspects of verbal comprehension during speech processing distinct from acoustic feature detection referable to the superior temporal gyrus, ^{Reference Giraud, Kell, Thierfelder, Sterzer, Russ, Preibisch and Kleinschmidt4} whereas non-dominant middle temporal gyral activation has been associated with detecting prosodic features of spoken speech. ^{Reference Ethofer, Silke, Erb, Herbert, Wiethoff, Kissler, Grodd and Wildgruber5} One plausible account of our findings is that pre-hallucination activation in the middle temporal gyrus reflecting verbal content and/or prosody is subsequently propagated to the superior temporal gyrus via top-down processing, which generates (hallucinated) acoustic representations.

Left anterior insula

The left anterior insula was close to a site of activation in the left inferior frontal gyrus 9 s prior to hallucination onset identified by Shergill et al, ^{Reference Shergill, Brammer, Amaro, Williams, Murray and McGuire2} who reported expanded activation incorporating the left insula at later times. Left insula activation has been associated with speech articulation, ^{Reference Wise, Greene, Buchel and Scott6} imagining spoken speech of others, ^{Reference McGuire, Silbersweig, Murray, David, Frackowiak and Frith7,Reference Shergill, Bullmore, Brammer, Williams, Murray and McGuire8} and focused auditory attention, ^{Reference Giraud, Kell, Thierfelder, Sterzer, Russ, Preibisch and Kleinschmidt4} suggesting that pre-hallucination insula activation reflects inner speech or auditory imagery generation as previously hypothesised, ^{Reference McGuire, Silbersweig, Murray, David, Frackowiak and Frith7,Reference Shergill, Bullmore, Brammer, Williams, Murray and McGuire8} or enhanced auditory attention. However, pre-hallucination insula activation might instead reflect motor movement required to signal these events. This possibility is suggested by the fact that simple generation of finger movements is preceded by activation in the adjacent Broca's area, which has been postulated to reflect mental preparation. ^{Reference Hunter and Spence9}

Other sites

Evidence of right ventral anterior cingulate and left parahippocampal deactivation preceding hallucination onset was detected. Co-occurring deactivations in these regions have also been linked to heightened vigilance/attention, ^{Reference Lawrence, Ross, Hoffmann, Garavan and Stein10} suggesting a shift in cognitive state preceding auditory/verbal hallucinations. Along these lines, Arieti described a ‘listening attitude’ that predisposes people with schizophrenia to hear ‘voices’. ^{Reference Arieti11} Pre-central activation emerging prior to hallucination onset in our study could reflect either inner speech generation or signalling hallucinations by finger movement.

In summary, activation detected as BOLD signal changes correlated with auditory/verbal hallucination time course at negative time lags may reveal complex brain processes triggering these experiences. Future studies of this type would be advanced by controlling for effects of motor behaviour required to signal hallucination occurrences. ^{Reference Hunter and Spence9}