Functional MRI of Verbal Self-monitoring in Schizophrenia: Performance and Illness-Specific Effects

Participants and Design

Seventy right-handed²² outpatients with schizophrenia, of whom 63 provided usable data. All patients were in a chronic illness phase and on stable doses of antipsychotics for at least 3 months. Twenty right-handed HC matched, on average, on age and sex to the patient group were also studied (see table 1 for participants’ characteristics). All participants had intact hearing.

The study had local ethics committee approval. All participants provided written informed consent.

Assessments

The diagnosis²³ in patients was confirmed, and symptoms were rated using the Positive and Negative Syndrome Scale (PANSS)²⁴ by a trained psychiatrist. The absence of a clinical diagnosis²⁵ in HC was also confirmed. The current IQ²⁶ was measured in all participants.

fMRI Paradigm and Procedure

All participants performed a self-monitoring task while undergoing fMRI. Participants were presented with single words on a computer screen (visible for 750 ms, interstimulus internal 16.25 s), viewed via a prismatic mirror fitted in the radiofrequency head coil, as they lay in the scanner, and were instructed to read each word aloud. The participant's speech was transformed through a software program and a DSP.FX digital effects processor (Power Technology, Brisbane, CA), amplified by a computer sound card, and relayed back through an acoustic MRI sound system (Ward Ray-Premis, Hampton Court, UK) and pneumatic tubes within the ear protectors at a volume of 91 dB (SD 2). The volume of the feedback was sufficient to overcome the bone conduction of the participant's own voice. The verbal feedback was (a) their own voice (self-undistorted), (b) their own voice lowered in pitch by 4 semitones (self-distorted), (c) voice of another person matched on participant's sex (other-undistorted), or (d) another person's voice with the pitch lowered by 4 semitones (other-distorted). The level of pitch distortion was determined based on findings from previous neuroimaging studies of verbal self-monitoring.^20,27 Participants registered their responses regarding the origin of feedback by using the button box with the “self” button press for their voice, the “other” button press for “other” voice, or the “unsure” button if they were unsure about the nature of the feedback. Words “self,” “other,” and “unsure” were displayed on the screen and were outlined in black after each participant's response. Accuracy of the responses was recorded online. Participants’ occasional failures to press a button were recorded as nonresponses. In total, 64 words were presented during the experiment. Each condition occurred 16 times in a pseudorandom order. Participants were familiarized with the experimental procedures prior to scanning.

Image Acquisition

Echoplanar MR brain images were acquired using a 1.5 T GE Signa system (General Electric, Milwaukee, WI). A quadrature birdcage head coil was used for radio frequency transmission and reception. In each of 14 near-axial noncontiguous planes (slice thickness = 7.0 mm, interslice gap = 1 mm) parallel to the intercommissural (ac-pc) plane, T2*-weighted MR images depicting blood oxygenation level–dependent (BOLD) contrast were acquired over 1.1 seconds using a “clustered” acquisition (12–14) (echo time = 40 ms, 70° flip angle), which created a relative silent period of 2.15 seconds for each stimulus within a repetition time of 3.25 seconds and the interstimulus interval of 16.25 seconds, yielding 5 brain volumes for each trial. A clustered acquisition sequence was used to minimize artifacts associated with overt speech during image acquisition.²⁰

Data Analysis

Functional Magnetic Resonance Imaging

Creation of Patient Subgroups

The patient sample was divided, on the basis of their total percentage of correct answers (accuracy rates for the 4 conditions were highly positively correlated), into 2 groups: good and poor performers. The mean percentage of total correct answers in HC minus 1 SD (mean = 78.91, SD = 11.00) was used to define a deficit cutoff score of 67.91 (≤67.91% = deficient performance).

Models and Inferences

We focused on brain activity during trials with correct answers. There were too few errors for the self-undistorted condition in the majority of HC and patients classified as good performers to allow the analysis of brain responses during errors.

fMRI data were analyzed using a 2-stage random-effect procedure.²⁸ The first stage identified subject-specific activations. Although there were relatively fewer correct trials per participant in the poor performance group, this group had sufficient (and the largest) number of participants to allow ample power in further analysis steps. We then identified task-related activations (threshold P < .05 corrected for multiple comparisons at cluster level) using 1-sample t tests across all participants. The second stage of analysis involved separate ANOVAs within SPM for each task condition and then at the levels of source (self-undistorted + self-distorted vs other undistorted + other distorted) and distortion (self-undistorted + other-undistorted vs self-distorted + other-distorted) to identify regions of activity (P = .05 corrected for multiple comparisons at cluster level) differentiating (a) good (HC + patients) from poor performers (patients only) and (b) patients (regardless of performance) from HC.

Next, subject-specific activation values were extracted from voxels showing the maximum group difference in each cluster and explored for possible relationships with performance and symptom scores (within SPSS). The associations with performance and PANSS subscale and total scores were examined using Pearson correlations and with individual PANSS item scores (if detected at the total or subscale level) using Spearman rank-order correlations.

Participant Characteristics

HC and patients were comparable on age and sex distribution, but patients had significantly fewer years of education and lower IQ (table 1).

Performance

Years in education was not associated with performance when examined separately in HC and patients or across the entire sample (P’s > .15). IQ showed modestly positive correlations with percentage of correct answers during the other-distorted condition (r₂₀ = 0.530, P = .016) but not during the other 3 conditions (self-undistorted r = 0.360, self-distorted r = 0.129, other-undistorted r = 0.10; P’s > .10) in controls and during the self-undistorted (r₆₂ = 0.474, P = .016) and other-distorted (r = 0.503, P = .001) but not (P’s > .10) during the self-undistorted (r = 0.205) or other-undistorted (r = 0.197) conditions in patients.

HC showed more accurate performance than patients (F_1,81 = 19.52, P = .001, η² = 0.194). Both groups had more accurate performance during the undistorted than the distorted conditions (F_1,81 = 76.33; P < .001, η² = 0.485). A group × source × distortion interaction was present (F_1,81 = 4.33, P = .04, η² = 0.051), with HC performing significantly better than patients during the self-undistorted, self-distorted, and other-undistorted conditions but not during the other-distorted condition, possibly due to a much reduced accuracy in this condition in HC (table 2).

Complementing the results for percentage of correct answers, patients made more misattributions than HC (F_1,81 = 14.26, P < .001, η² = 0.150). This effect was significantly present only for the self-distorted and other-undistorted conditions (table 2), although the group × source × distortion interaction fell short of formal significance (F_1,81 = 3.78, P = .06, η² = 0.045). Both groups made more errors during the distorted than the undistorted conditions (F_1,81 = 46.21, P < .001, η² = 0.363), especially during the other-distorted condition (source × distortion: F_1,81 = 6.72, P = .01, η² = 0.08).

The patients and HC were not significantly different for the percentage of unsure responses (F < 2.40 for group and group × source × distortion). Both groups made more unsure responses during the distorted, relative to undistorted (F_1,81 = 14.40, P < .001, η² = 0.15), and during the other, relative to self, conditions (F_1,81 = 9.60, P = .003, η² = 0.11).

No significant effects involving the group were detected for the percentage of nonresponses (F < 2.15 for group and group × source × distortion).

Across the entire patient sample, no significant or consistent relationships were found between PANSS total or subscale scores and performance accuracy (r range: −0.20 to +0.28) or error rate (r range: −0.22 to +0.17).

Good and Poor Performance Patients

Twenty-seven patients met the criterion for good and 36 for poor performance. Of 27 patients classified as good performers, 4 patients were excluded due to an odd performance pattern: 3 patients did not make any correct responses during the self-distorted condition and 1 patient made no correct response during the other-distorted condition despite excellent performance during the remaining 3 conditions.

As expected, patients classified as good performers had better performance than poorly performing patients (P’s < .05), but no clinical or demographic characteristics differentiated the 2 groups (P’s > .05) (table 3).

Functional Magnetic Resonance Imaging

Good Performers Vs Poor Performers

Good performers showed greater activity bilaterally in the superior-middle TL during all conditions than poor performers (table 6, figure 2). They also showed greater thalamic and putamen activity during all, except other-undistorted, conditions, and in the IFG and middle occipital gyrus during the self-undistorted condition.

Poor performers showed more activity than good performers in the medial prefrontal and posterior temporal parietal cortices during all conditions (failed to reach corrected significance in the self-distorted condition, thus not presented in table or figure). These differences occurred due to stronger deactivation of these areas (figure 1) in good, relative to poor, performers.

Poor and good performers were also differentiated by right IFG activity. Good, but not poor, performers showed more activity in these regions during the self, compared with other, conditions (regardless of distortion).

Patients Vs Healthy Participants

Patients, regardless of performance, showed less activity than HC in the thalamus (pulvinar) during the self-undistorted condition. They showed more activity bilaterally in the STG during the self-undistorted condition and in the left STG during the other-distorted condition. Subject-specific activation values in these regions did not correlate with age, age at illness onset, illness duration, PANSS scores, IQ, or performance in patients.

Patients showed more activity than HC in the ventral striatum, hypothalamus, and part of the thalamus in the other > self-contrast (table 7, figure 3, source × group). This effect occurred due to greater deactivation of these regions to own voice (note that the caudate showed deactivation during the self-conditions across the whole sample due to this effect mainly in patients, figure 1), combined with nonsignificant activation of the same regions to someone else's voice (not shown) and associated positively with negative symptoms. Within the patient group, greater ventral striatal-hypothamic activity during other > self-contrast (figure 3, source × group) correlated with higher negative symptoms score (r = 0.293, P = .05). At the individual symptom level, blunted affect (ρ = 0.369, P = .01), emotional withdrawal (ρ = 0.332, P = .026), poor rapport (ρ = 0.359, P = .015), and passive social avoidance (ρ = 0.335, P = .025) contributed to this relationship; poor abstract thinking, lack of spontaneity, and stereotyped thinking items were uncorrelated (P > .15).

Patients also showed greater activity in the right TL extending to the right IFG and parietal regions during conditions with undistorted, relative to distorted, feedback (regardless of source); HC showed the opposite pattern although to a lesser degree. This activity change in patients did not correlate with IQ (r = −0.134) but correlated negatively with performance (percentage of total correct r = −0.364, P = .014) and positively with positive symptoms (r = 0.349, P = .019), specifically with hallucinations (ρ = 0.437, P = .003), persecution (ρ = 0.320, P = .032), and disorganization (ρ = 0.320, P = .032).

Behavioral Findings: Impaired Performance in Patients

The data confirm our hypothesis of impaired performance in patients, relative to HC. This deficit, however, was not limited to misattribution of “self-generated” voice (externalizing bias) because the patients also misattributed another person's voice to themselves (internalizing bias). While it is possible that a subgroup of patients show only externalizing bias, studies using similar paradigms to those used in the present study point to a more general monitoring deficit affecting judgments of both self-generated and another person's voice in schizophrenia.^11,29 A possible reason for this might be that participants can perceive their own voice while reading the words across all conditions,^30,31 meaning all conditions involve monitoring of self-generated voice in some way. Furthermore, the performance on our task is likely to be affected not only by the ability to recognize and discriminate between the “self” and “others”² but also to include other influences such as prior experience of having used the word stimuli facilitating recognition in self-generated voice, working memory (WM, remembering the feedback while making a decision), and the ability to switch between and tendency to consider various possible options while making a judgment about origin of the voices.

Neural Findings

Other Observations

Although the right TL dysfunction was associated with both poor performance and positive symptoms, we did not find significant associations between the PANSS symptom dimensions and performance. This observation deserves some discussion because self-monitoring hypothesis of schizophrenia was initially proposed to explain some of the first-rank symptoms, which presume a loss of self-agency.³ Several studies using a range of paradigms have supported the association between the symptoms of delusions of control, thought insertion, and thought blocking and poor self-monitoring^{6–8,71–75} though there are others that did not find such associations.^21,76 Similarly, most,^{10–12,29,77} but not all,⁷⁸ studies using verbal self-monitoring paradigms observed a relationship between positive symptoms, mainly AH and delusions, and impaired self-monitoring. An association between self-monitoring deficit and positive symptoms may more often be found in acutely psychotic samples.