Abstract
Background: There is a great amount of evoked fields known to be marker for impaired cortical processing in schizophrenia compared to healthy controls. The ratio of brain responses occurring 50 ms after paired clicks (S2-evoked M50/S1-evoked M50), for example, serves as a measure of sensory gating. An unusually large ratio is a common finding for schizophrenia patients and is considered as a sign of reduced sensory gating or otherwise dysfunctional organization of the auditory/verbal system as a factor contributing to psychopathology and cognitive dysfunction in schizophrenia. Although, the evoked fields used for calculation of the sensory gating ratios (SGR) are complex functions dependent on many variable factors like signal-to-noise ratio, phase-locking and intra-/inter-cortical coherence which are also known to be impaired in schizophrenia.
Aims: The aim of this study is to investigate the effects of a specific computer-based cognitive training program on the variable factors influencing the averaged evoked response and to clarify the mechanisms controlling the sensory gating ratios improvement suggested in a previous study.
Methods: In a randomized trial, 38 schizophrenia patients (ICD-F20.0 diagnosis) were assigned to CE (20 sessions within 4 weeks) or an established training of a broad range of cognitive functions (Cognitive Package, Cogpack). The inter-trial phase locking (ITPL) and the trial-by-trial variability index (VI) were used to evaluate training-specific effects.
Results: No significant training-induced changes in the variability measures (VI and ITPL) were found in neither of the patients group. Instead a positive relationship between SGR and phase-locking after the training in the patients assigned to the CE training was found. That correlation was neither evident before the training period nor after the training period in the patients group assigned to the Cogpack training.
Conclusions: The improvement of the SGR after a specific cognitive training could be due to synchronization of two networks in schizophrenia: one controlling the amplitude
66 and another controlling the phase-locking. This synchronization could be considered as an adaptive mechanism for correct auditory processing.
Keywords: Schizophrenia, cognitive training, M50, sensory gating, phase locking, response variability
Introduction
There are two main methods for treatment of schizophrenia: pharmacological and psychosocial. Where pharmacotherapy is the prime instrument of treatment, without which most psychosocial treatment could not be performed.
Psychosocial intervention aims to enhance functioning in areas such as independent living, relationships, and work, social skills training, teaching illness management skills, cognitive-behavior therapy for psychosis, and integrated treatment for comorbid substance abuse.
The current pharmacological treatment of schizophrenia has limited effects on cognition in schizophrenia (Marder, 2006; Rund and Borg, 1999) and even less impact on community functioning (Harvey et al., 2004). In order to make good the deficiency of treatment for cognitive impairment in schizophrenia, a range of cognitive remediation programs has been developed and evaluated. For example the Cognitive Exercise (CE) package (PositScience, San Francisco, USA) aims to improve discrimination ability in the auditory system. Another one - Training for emotional intelligence (Training emotionaler Intelligenz – TEI; Kompetenznetz Schizophrenie, Germany) targets the ability to accurately process social information such as emotions in schizophrenia. This new way of treating schizophrenia patients may also be considered as another sub-group of the psychosocial treatment methods.
Recent studies found training induced normalization of the sensory gating ratios (Popov et al., 2010) and verbal memory improvement (Fisher et al., 2009) in schizophrenia. Since it is known that gating ratios are ratios of two dipole amplitudes but the dipole amplitudes are dependent variables of other dipole parameters like signal-to-noise ratio (SNR), dipole position and dipole orientation, it would be interesting to investigate which of the dipole parameters contributing to the amplitude are also changed due to training induced neuroplasticity in schizophrenia. In the current study the brain response variability and the inter-trial phase-locking of the evoked
67 responses were considered as main contributors to the SNR and were investigated for training-induced changes in schizophrenia. The other two parameters – localization and orientation, were fixed through the subjects in order to reduce the dimensions of the data. The same data sets as in the study of training-induced neuroplasticity by Popov and colleges (2010) were analyzed and the results were considered in relation to the amplitudes and sensory gating ratios reported there.
Methods
Participants
Inpatients were recruited at the regional public Center for Psychiatry. Inclusion criteria were an ICD diagnosis of paranoid-hallucinatory schizophrenia (F20.0), age 20–50 years, normal intellectual function, and no history of head trauma with loss of consciousness. Patients meeting inclusion criteria were informed about the training and measurement protocol and were included in the pre-training assessment and random assignment protocol after signing written informed consent. Eleven of 50 patients did not complete the study, six discharged before the end of the training, three quitting the training early, one patient declining electromagnetic measurement, and one patient providing artifact-contaminated data.
Table 1: Demographic and clinical characteristics of schizophrenia patients
Characteristic CE Cogpack Group difference
N 20 18
Age (M±SD) 30.9±7.6 29.6±9.3 0.66a
Education (M±SD) 13±2.5 12.2±2.1 0.28a
Gender: Females/Males 2/18 2/16 0.91b
BPRS (M±SD) 46±10 51±11.8 0.20a
GAF (M±SD) 37±7.5 31±9.5 0.06a
BDI (M±SD) 12±10 16±11 0.40a
Hospitalizations (M±SD) 6±7.4 5±3.8 0.54a
68 Medication, number of patients
receiving:
typical neuroleptics 0 3
atypical neuroleptics 15 10
both 5 5
Immediate recall (M±SD) 5±1.8 6±1.8 0.52a
Learning speed (M±SD) 12±2 12±2.5 0.94a
Delayed recall (M±SD) 10±3 9±4.4 0.72a
Note: BPRS: Brief Psychiatric Rating Scale, BPRS (Lukoff et al., 1986); M±SD: mean ± standard deviation; a Independent two-sample t-test; b Pearson’s Chi-square test
Within the randomized trial (following CONSORT guidelines) patients were randomly assigned to the two training programs. Three exceptions from strict random assignment were necessary as some patients were familiar with Cogpack from previous admissions and/or their regular rehabilitation program and declined to participate in this training. These patients were assigned to the CE-program in order to avoid drop out. Altogether 20 patients completed CE, and 18 patients Cogpack. All patients were on psychoactive medication, 25 receiving atypical, 3 typical and 10 combinations of typical and atypical neuroleptics (see Table 1). Upon treatment assignment, the training groups did not differ in gender distribution, age, education level, global ratings of severity of disorder (Brief Psychiatric Rating Scale, BPRS) (Overall and Gorham, 1962; Lukoff, 1986) and level of function (Global Assessment of Functioning Scale, GAF; DSM-IV), number of previous hospital admissions, type of medication, or overt performance on verbal memory and fluency tests.
For evaluation of auditory sensory gating, 28 healthy subjects were recruited to be comparable to the patient sample in age and gender (Table 1). Subjects were included if they did not meet criteria for a lifetime diagnosis of a mental illness (screened with the MINI interview) (Ackenheil et al., 1999), did not report any history of head trauma with loss of consciousness, and were free of psychoactive medication. For evaluation of sensory gating reliability/stability, 28 healthy subjects participated in the MEG protocol again after 4 weeks. Patients and controls did not differ significantly with respect to
69 gender distribution or age, though controls had more years of education than patients (see Table 2).
Table 2. Demographic characteristics for the group of schizophrenia patients and the group of healthy controls.
Patients Controls Group difference
N 38 28
Age (M±SD) 30.3±8.4 28.3±6.5 0.30a
Gender: Females/Males 4/34 6/22 0.22b
Years of education (M±SD) 12.6±2.3 16.8±2.3 0a Note: M±SD: mean ± standard deviation
a One-way ANOVA; b Pearson’s Chi-square test
Right-handedness (Oldfield, 1971) was confirmed for 32 patients and 24 controls, four patients and three controls were left-handed, and two patients and one control were ambidextrous. All participants gave written informed consent for participation following a complete description of the study protocol and measurements. All participants received a financial bonus (20 €) for participation in the MEG measurement, which lasted for 2 hours. (Participation in the training was introduced as part of the rehabilitation regimen with no additional monetary bonus.)
Study Design, Cognitive Assessments, and Training Protocols
The study design was approved by the Ethics Committee of the University of Konstanz and assigned ClinicalTrials identifier NCT00695708. Before and after training, auditory sensory gating was recorded in a paired-click design (see below), and overt cognitive performance of patients was assessed using a German equivalent of the California Verbal Learning Test (Verbaler Lern- und Merkfähigkeitstest, VMLT (Helmstaedter and Lux, 2001)) and a verbal fluency test (Regensburger Wortflüssigkeitstest, RWT (Aschenbrenner, 2000)). Based on serial learning of 15-word lists, the VMLT assesses immediate recall, learning speed (across 5 repetitions of the word list), and delayed recall (after 30 minutes including the distraction of introducing a second word list). The
70 RWT assesses word production across two minutes.
After the pre-training assessment patients were assigned to either the neuroplasticity-based training program (CE, Cognitive Exercises, Posit Science, San Francisco, USA) or the cognitive training program currently standard on the inpatient unit for patients who receive such training Cognitive Package program (Cogpack, Cognitive Package, Marker Software, Ladenburg, Germany). CE (for details see (Fisher et al. 2009)) consists of 6 computerized exercises: gradually more difficult distinctions between frequency modulation (FM) ‘sweeps’ of auditory stimuli (1) increasing or (2) decreasing in frequency, (3) distinguishing phonemes using synthesized speech, (4) identifying arrays of open and closed syllables in spatial and sequential context, (5) discriminating tone frequencies, and (6) remembering details of a short narrative.
Cogpack includes a much broader array of 64 exercises of visuomotor skills (manipulate, follow, or mark a moving figure with the computer mouse; divide lines or pies;
reproduce or mirror a drawing; catch a bouncing ball), vigilance (scanning, catch falling star, continuous-performance pattern detection), comprehension (character recognition), language (word-finding upon clues, text-content, authors or titles of quoted poems, put words or syllables in order, anagrams), memory (words, images and labels, patterns, signs, addresses, routes, life scenes), logic (mental arithmetic, geometry, numbers and number words, quantities, simple and deductive comparisons, block or series completion), everyday skills (times and dates, compass, geography, money, weights and measures, road signs, license plates, abbreviations). Each Cogpack exercise is available with several variations. Both training protocols were computer-based and adaptive to allow positive reinforcement and to avoid failure. Trainings were also similar with respect to total duration of training (4 weeks) and the effort of training. CE training comprise daily 60- min sessions on 20 consecutive workdays, whereas Cogpack followed the standard protocol as recommended by the founders from the validation studies: it includes a series of task to be accomplished on each of three 60- 90 minutes session per week.
MEG Measurement
One hundred pairs of 3 ms square-wave clicks were presented with a 500 ms inter-stimulus interval and an 8 sec jittered inter-trial interval (ITI offset to onset). Clicks
71 were presented at 50 dB above subjective hearing level, determined separately for each ear, and delivered via non-ferromagnetic tubes of about 5 m length. No task was involved, except that subjects were asked to keep their eyes focused on a small fixation point throughout the measurement.
MEG was recorded while subjects were in a supine position, using a 148-channel magnetometer (MAGNES™ 2500 WH, 4D Neuroimaging, San Diego, USA). Data were continuously recorded with a sampling rate of 678.17 Hz and a bandpass filter of 0.1 to 200 Hz. For artifact control, the vertical and horizontal electrooculogram (EOG recorded from four electrodes placed near the left and right temporal canthus and above and below the right eye) were recorded using a SynAmps amplifier (Neuroscan Laboratories, Sterling, VA, USA). The subject’s nasion, left and right ear canal, and head shape were digitized with a Polhemus 3Space® Fasttrack prior to each session.
Global noise was filtered offline from the MEG data by subtracting external, non-biological noise recorded by 11 MEG reference channels. Before subtraction, reference channels were multiplied with individually calculated fixed weight factors. This noise reduction procedure has little or no influence on biological signals because the distance of the reference sensors to the subject’s head is large (mean = 25.8 cm, std = 6.00 cm, min = 15.5 cm, max = 36.5 cm) relative to the distance between the head and adjacent sensors.
Data Analysis
MEG data were preprocessed using the BESA Research software version 5.3 (www.besa.de). At first the data sets were filtered with Butterworth filter: 1 Hz low cutoff, zero-phase, 12 db/octave and 45 Hz high cutoff, zero-phase, 24 db/octave. In a second step the length of the trials was determined by -750 to 1250 ms (2 seconds length) around the time where the first click appears. A procedure for artifacts rejection was applied where the trials with amplitude greater than 3500 fT and time gradient larger than 2500 fT/s were defined as bad and not included in the further analysis.
Variability in the double click design was defined as trial-by-trial variation of stimulus-evoked responses (Möcks et al., 1988) calculated for each MEG sensor, each time point. The variability index (VI) estimate is the mean magnitude of the difference between every single trial and the average across trials. VI is calculated as a function of time over the entire trial length
72 point t. Results will be reported as VI in femto-Tesla (fT), the square root of the noise power (fT2).
The VI(t) was first averaged over time (-500 to -100 ms for estimation of the VI in the pre-stimulus period; 50 to 150 ms for estimation of the VI for the M50/M100 complex as reaction for the first click-stimulus; 550 to 650 ms for the VI in the reaction to the second stimulus) and then compared between groups (schizophrenia patients versus controls) using the statistical procedures in the Matlab-based FieldTrip toolbox, developed at the Donders Institute for Brain, Cognition and Behavior (http://fieldtrip.fcdonders.nl/). Comparisons were corrected for family-wise error rate by a non-parametric, t-test based randomization test (Maris and Oostenveld, 2007). This procedure effectively controls for multiple comparisons and allows the identification of sensor clusters of significant group differences. A cluster was defined as a set of adjacent sensors (defined as sensors at less than 3.2 cm distance, yielding on average 3 neighbors per sensor) that exhibited similar differences between groups in t-value and magnitude.
Group differences were considered statistically robust for a sensor cluster, whenever the significance level exceeded 95%. Signals of significant sensor clusters were averaged and subjected to a repeated measures analysis of variance (RMANOVA) for group x time comparison with covariates age, years of education, number of accepted trials in the first and in the second MEG measurement.
As a second measure for impaired cortical functionality and failure to build strong evoked responses in schizophrenia patients was taken the inter-trial phase-locking value (ITPL). The ITPL values were calculated again in BESA Research 5.3 with the software module for time-frequency analysis. The ITPL value for a given channel at frequency f and latency t is computed as
( ) ∑ ( )
73 with N = number of trials, k(f,t) = phase in trial k (Rapp and Ktonas, 1977). ITPL is a measure for the consistency of an oscillation's phase across trials. ITPL takes on values between 0 (random phase variation from trial to trial) and 1 (exactly the same phase in each trial). The inter-trial phase-locking analysis was performed in the time frequency domain with time resolution of 25 ms, frequency resolution of 2 Hz, lower frequency cutoff 4 Hz and higher frequency cutoff 50 Hz. In order to calculate the ITPL values a source montage was created. The source montage solves some of the basic problems by inter-person comparisons. Source montage decreases the variation caused by the different positions of the subjects’ heads in the MEG-sensor and, furthermore, reduces the dimension of the problem. The source montage used in this study is based on the thesis that the auditory M50 response happens predominantly in the primary auditory cortex. Two dipoles were seeded in the area of the left and right primary auditory cortices, respectively. Their orientations were chosen to be the mean orientation of the M50 source reconstructions determined in the previous study (study 3). In addition another three regional sources (one in the frontal lobe, one in the visual cortex and one in the motor cortex) were included to pick up background MEG activity (Figure 1). The signal recorded with the MEG sensor was modeled by these five sources and ITPL was computed for each of them. The resulting (source×time×frequency) matrices were send to cluster based analysis in FieldTrip using the same procedure like in the VI case with that difference that neighborhood relations between the sources were not included (the sources are independent contrary to sensor channels) and cluster were built only with respect to time and frequency. The statistical results were used to choose a region of interest in the frequency range. The time range of interest was clear because the orientations of the dipoles were chosen according to the M50 auditory response. When the frequency range of interest was determined clusters in the time range from 50 to 150 ms and in the statistically determined frequency range were taken and averaged over time and frequency in order to get a single value for every source and every subject. The single values were again sent to one-way ANOVA statistical analysis for group comparisons and to repeated measures ANOVA with covariates age, education and number of accepted trials for investigation of the training effects.
74 Figure 1. The source montage built in BESA Research 5.3 used for the calculation of the ITPL values.
Results
Patients compared to controls showed both impaired inter-trial phase-locking values (ITPL) (Figures 2 and 3) and increased variability indices (VI) (Figure 4) in the pre-training measurement. The statistical analysis for the ITPL differences yielded significant clusters only for the dipole source located in the right primary auditory cortex (cluster p-values < 0.025 after 5000 permutations) (Figure 3) and a trend in the dipole source located in the left primary auditory cortex (p=0.06), all other sources did not show any differences in the phase-locking between the both groups (schizophrenia patients and healthy controls).
75 Figure 2. Inter-trial phase-locking (ITPL) values for the sources located in the left- and in the right primary auditory cortices, respectively and plotted for both groups – healthy controls (HC: left) and schizophrenia patients (SZ: right).
The frequency region of interest was determined to be in the interval [4 to 14] Hz. In order to get a single value for the inter-trial phase-locking an average across time and frequency was calculated for the time frequency window [50 to 150]ms × [4 to 14]Hz.
The statistical comparison of the variability index (VI) between SZ and HC yielded similar results in all three temporal sub-intervals of interest: (i) pre-stimulus period [-500 to -100]ms; (ii) M50/M100 time-period for click 1 [50 to 150]ms; (iii) M50/M100 time-period for click 2 [550 to 650]ms (Figure 4).
76 Figure 3. Significant cluster of ITPL values in the time × frequency domain differentiating between SZ and HC for the source in the right primary auditory cortex. In the left picture the difference between the ITPL values in schizophrenia patients and healthy controls is shown. On the right, a masked difference shows only the significant time × frequency regions.
All three statistical analyses yielded more pronounced differences in the left than in the right temporal channels. After the averaging over latencies and frequencies the RMANOVA with between factor group (SZ, HC) and within factors hemisphere (left, right) and latency (pre stimulus, click 1, click2) with covariates age, years of education and number of accepted trials yielded a not significant interaction group × hemisphere × latency (F<1, p=0.9) but a significant group effect (F(1, 61) = 16.6, p=0.0001)(Figure 5:
left) and a significant latency × hemisphere interaction (F(2, 122) = 7.2, p=0.001)(Figure 5: right).
Figure 4. VI differences between SZ and HC in the three time intervals of interest. The channels differentiating the both groups with p<0.025 were labeled with asterisks and these with p<0.05 with “×”.
77 The ITPL values and the VIs came out to be dependent measures (as expected).
The main direction seemed to be that higher VI values corresponded to lower ITPL values where the VIs in the left hemisphere correlated stronger with the ITPL values then the VIs in the right hemisphere (Table 3).
Figure 5: Group effect yielded by the comparison of the VI between schizophrenia patients and healthy controls (left) and latency × hemisphere interaction (right). The bars indicate 95% confidence intervals.
Table 3. Correlation coefficients (r) and p-values (p) for the comparison between VI and ITPL. Significant correlations are marked bold.
N = 66 ITPL(S1) right ITPL(S2) right
VI (baseline) left (r/p) -0.27/0.03 -0.33/0.007 VI (baseline) right (r/p) -0.23/0.06 -0.23/0.06
VI (S1) left (r/p) -0.26/0.04 -0.32/0.009
VI (S1) right (r/p) -0.23/0.06 -0.23/0.06 VI (S2) left (r/p) -0.26/0.03 -0.29/0.02 VI (S2) right (r/p) -0.23/0.07 -0.22/0.08
The RMANOVA analysis for comparison of the M50 auditory stimulus responses to both stimuli (click one and click two) as well as for SGRs yielded not significant interaction group × hemisphere (F<1). Therefore, only values averaged over the
The RMANOVA analysis for comparison of the M50 auditory stimulus responses to both stimuli (click one and click two) as well as for SGRs yielded not significant interaction group × hemisphere (F<1). Therefore, only values averaged over the