Auditory-Based Cognitive Training Drives Short- and Long-Term Plasticity in Cortical Networks in Schizophrenia

Abstract

Schizophrenia patients have widespread deficits in neurocognitive functioning linked to underlying abnormalities in gamma oscillations that are readily measured by the 40 Hz auditory steady-state response (ASSR). Emerging interventions such as auditory-based targeted cognitive training (TCT) improve neurocognitive function in patients. While acute ASSR changes after 1 hour of TCT predict clinical and cognitive gains after a 30-hour course of TCT, the neural substrates of underlying short- and long-term TCT interventions are unknown. To determine the neural substrates underlying TCT-associated ASSR changes, a novel data analysis method was applied to assess the effective connectivity of gamma-band ASSR among estimated cortical sources. In this study, schizophrenia patients (N = 52) were randomized to receive either a treatment as usual (TAU; N = 22) or TAU augmented with TCT (N = 30). EEG recordings were obtained immediately before (T0) and after 1 hour of either computer games (TAU) or cognitive training (TCT; T1), and at 65 ± 15 days (mean ± SD) post-randomization (T2). Results showed increased connectivity from the left ventral middle cingulate gyrus to the left posterior cingulate gyrus, accompanied by decreased connectivity from the left Rolandic operculum (a region that includes auditory cortex) to the right ventral middle cingulate gyrus after 1 hour of TCT. After 30 hours, decreased connectivity from the frontal cortex to a region near the calcarine sulcus were detected. Auditory-based cognitive training drives short- and long-term plasticity in cortical network functioning in schizophrenia patients. These findings may help us understand the mechanisms underlying cognitive training effects in schizophrenia patients and enhance the development of pro-cognitive therapeutics.

Introduction

Schizophrenia patients have pervasive deficits in cognitive functioning that contribute to high rates of psychosocial disability.^1–4 Current treatments do not sufficiently remediate these cognitive deficits and have little effect on the devastating consequences of psychosocial impairment.^5,6 Cognitive functioning requires coordinated information processing within networks of functionally specialized brain areas.^7,8 At the most rudimentary level, action potentials are temporally organized in oscillatory bursts that enhance the reliability of information processing⁹ and the establishment of long-range communication and synchronization.¹⁰

The synchronization of oscillatory activity across distributed cortical circuitry facilitates the connectivity between nodes in large and complex information processing networks in the brain,¹¹ thereby providing a critical mechanism underlying neural coordination.^12,13 This dynamic coordination of the connectivity is evidenced by changes in the strength, pattern, or frequency with which different brain areas are synchronously engaged. In humans, the ~40 Hz gamma-band response appears to be a “preferred resonance frequency” for auditory information processing, first demonstrated using scalp electroencephalography (EEG) stimulated at different rates by Galambos et al.^14,15 and Makeig.¹⁶ This 40 Hz auditory steady state response (ASSR) has emerged as a robust and reliable predictor of higher order cognitive functioning, and has been shown to be impaired in multiple neuropsychiatric disorders including schizophrenia.^17,18

Kwon et al¹⁹ performed the first study of 40 Hz ASSR in schizophrenia patients using EEG and found selectively reduced power and synchronization in response to 40-Hz stimulation. This initial finding has now been replicated and extended in subsequent studies using EEG^20–29 and magnetoencephalography (MEG),^30–34 including findings that relate ASSR dysfunction to important domains of cognitive and psychosocial functioning in schizophrenia patients.^24,26,27 A recent meta-analysis confirmed that the 40-Hz ASSR is a robust index of gamma synchronization deficits in schizophrenia patients.³⁵ Gamma-band ASSR is supported by a distributed network of cortical sources^36,37 including the superior temporal, the orbitofrontal, and the superior frontal cortical regions.³⁶ Findings from our group indicate that gamma-band ASSR deficits in schizophrenia patients are not a static abnormality of these network nodes, but rather reflect a dynamic pathological pattern of information processing at the onset of the auditory steady-state response centered at the superior temporal gyrus that then cascades forward to widespread abnormalities in the engagement of prefrontal brain regions in the later aspects of the response.³⁸

Gamma-band ASSR is widely recognized as a translational measure of auditory system functioning that is increasingly applied as a predictive biomarker in monitoring the therapeutic response to interventions aimed at ameliorating cognitive dysfunction in schizophrenia, such as auditory-based targeted cognitive training (TCT).^39,40 Gamma-band ASSR may be particularly sensitive to TCT since this intervention explicitly aims to improve the accuracy and fidelity of auditory information processing in a “bottom-up” manner.⁴¹ Several studies report that schizophrenia patients show cognitive improvement such as verbal working memory, and verbal learning and memory lasting 6-month to at least 1-year after a course of TCT.^41–43 In this cohort of patients, we previously reported that TCT resulted in significant improvements in verbal learning/memory as well as reductions in auditory hallucinations.⁴⁴ We have recently shown that acute change in gamma-band ASSR measured at the scalp after 1 hour of TCT strongly predicts clinical and cognitive benefit in schizophrenia patients after 30 hours of TCT.⁴⁰ The neural dynamics of TCT-associated changes, however, are not known.

In order to assess these mechanistic underpinnings of TCT, a novel data analysis method was developed and applied to assess the effective connectivity of gamma-band ASSR among estimated cortical sources using multivector autoregressive modeling. This approach applied Granger causality⁴⁵ enables us to show causal interactions among underlying brain regions that generate and maintain 40-Hz ASSR represented as effective connectivity between 2 nodes using a correlation with time delay of EEG phase activity at the pair of nodes. A previous single-dose proof of concept study revealed significant changes in estimated cortical source contributions to another measure of early auditory information processing (ie, mismatch negativity), that were detectable after only 1 hour in schizophrenia outpatients.⁴⁶ Therefore, we hypothesized that TCT would similarly drive changes in the gamma-band ASSR connectivity across cortical networks, including temporo-frontal regions in patients with schizophrenia after 1 hour and 30 hours of TCT.

Methods

Participants and Study Design

Participants included 52 patients with treatment-refractory schizophrenia or schizoaffective disorder (supplementary table 1). Details of the recruitment and ascertainment procedures in this cohort were previously reported,^44,47 as were conventional scalp-level ASSR measures.⁴⁰ Participants were recruited from a community-based inpatient treatment program following an acute care hospitalization. They were enrolled only after being deemed clinically stable by the treatment team. All participants were under public conservatorship by San Diego or Los Angeles Counties. Patients were diagnosed using a modified version of the Structured Clinical Interview for DSM-IV-TR. Antipsychotic medications were prescribed for 51 patients. Exclusion criteria included inability to provide informed consent, limited English proficiency, history of significant neurological illness or head injury, severe systemic illness, or current mania. The Institutional Review Board of University of California San Diego approved all experimental procedures (130874). The current study was conducted in accordance with the Declaration of Helsinki. All participants themselves provided written informed consent. Patients provided assent and signed consent, then the conservator or public guardian reviewed and provided written approval to enroll in the study.

After initial screening and assessments, subjects were randomized to receive either treatment as usual (TAU; medication management, individual and group therapy, and structured social activities; N = 22) or TAU augmented with targeted cognitive training (TCT; N = 30) using a parallel design with stratified random assignment by sex, age, and ethnicity (figure 1). For people assigned to TAU, EEG testing occurred immediately before (T0) and after 1 hour of self-selected computer games (eg, Tetris; T1). No other computer games were presented to subjects for the remainder of the study. For those assigned to TCT, EEG was recorded immediately before (T0) and after 1 hour of the “Sound Sweeps” training which is key component of the full TCT battery (T1). Participants randomized to the TCT group continued to receive a total of ~30 hours of TCT (1 hour of TCT, 3–5 days per week for up to 40 hours), which includes Sounds Sweeps and the other auditory-based exercises. EEG testing again occurred 65 ± 15 days (mean ± SD) post-randomization (T2).

Fig. 1.

Subjects of this study.

Targeted Cognitive Training

TCT was administered on individual laptop computers with headphones. We previously reported details of the TCT exercises.^44,47 Briefly, the BrainHQ (Posit Science Corporation, San Francisco, CA) platform was used for training with access paid by study investigators. Posit Science had no role in the design, analysis, or interpretation of results. These BrainHQ exercises were designed to engage neuroplasticity mechanisms in auditory networks of auditory perception and processing speed (Sound Sweeps, Fine Tuning) and auditory memory (Syllable Stacks, Memory Grid, To-Do List Training, Rhythm Recall). Exercises applied an n-up/m-down algorithm to participant responses to estimate threshold, ensuring that participants were engaging in the exercises and were continuously challenged at an appropriate level (~80% accuracy) as their abilities improved.

Stimuli and Procedures for ASSR

The auditory steady-state paradigm and stimuli are described previously in our studies.^39,40 Briefly, stimuli were 1 millisecond, 85 dB clicks presented via insert earphones at 40 Hz in 500 millisecond trains. A total of 250 click trains with 500-millisecond intervals were presented while participants watched a silent cartoon video. The ASSR recording lasted approximately 4 minutes. EEG recording and preprocessing procedures are shown in supplementary method 1.

Effective Connectivity Analyses

Detailed procedures of effective connectivity analyses were reported in previous studies from our group^38,48,49 and are provided in supplementary method 2. Briefly, a connectivity matrix of 76 anatomical region of interests (ROIs), based on an automated anatomical labeling (AAL) atlas,⁵⁰ was generated for each subject. We analyzed the sample size of N = 21 in both groups at all 3 time points. Group-level analyses identified 43 ROIs with significant dipole density, limiting the size of the connectivity matrix effectively to 43 × 43. The analysis included 35–45 Hz as the frequency range of interest to be consistent with the frequency range in previous gamma-band auditory steady-state response studies.^26–29 Among these graph edges, first-step statistical tests using a threshold of uncorrected P < .01 was used to identify clusters of significant time-frequency points. Across 43 × 43 – 43 = 1806 graph edges, and 318 of them showed time-frequency point clusters that survived multiple comparison correction at P < .005. Theoretically, if one graph edge is associated with one time-frequency point cluster, the expected number of significant clusters due to false positive is 381 × 0.005 = 1.9 clusters.

To further reduce the likelihood of type I error, an additional constraint was applied. Since contrast data recorded at T0 for both TCT and TAU groups, that at T1 for TAU and at T0 for TAU groups, and that at T2 for TAU and at T0 for TAU groups were expected to be the same respectively, a P < .05 threshold (corrected with weak family-wise error rate) was applied as an exclusive mask to both the result of 1-hour cognitive training effects [(TCT_T1 – TCT_T0) – (TAU_T1 – TAU_T0)] and that of full course cognitive training effects [(TCT_T2 – TCT_T0) – (TAU_T2 – TAU_T0)]. Notably, using p< 0.05 for creating an exclusive mask is a much more stringent criterion than lower p-value threshold (eg, P < .0001) since any results that show up in TCT_T0 – TAU_T0, TAU_T1 – TAU_T0, and TAU_T2 – TAU_T0 would be considered effects of non-interest. Thus, by evaluating results from such double subtractions, the possibility that significant results from nonspecific time or generalized treatment effects common to both groups or other effects of non-interest in TCT_T0 – TAU_T0, TAU_T1 – TAU_T0, and TAU_T2 – TAU_T0 are minimized in the service of isolating TCT specific connectivity changes. This conservative strategy resulted in exclusion of 20 potentially significant graph edges (8 for TCT_T0 – TAU_T0, supplementary figure 1; 9 for TAU_T1 – TAU_T0; 3 for TAU_T2 – TAU_T0, supplementary figure 2).

Results

The connectivity matrix of gamma-band ASSR group-difference of initial results (P < .05, corrected; 2-tailed) is shown in figure 2. The predefined P-value threshold of P < .005 (2-tailed) described above revealed 18 graph edges for 1 hour (figure 3 and supplementary movie 1 and 2) and 6 graph edges for full course training effects (figure 3 and supplementary movie 3 and 4).

Fig. 2.

Connectivity matrix of 76 × 76 anatomical region of interests (ROIs). The ROI at the top represents the starting point and the ROI at the side represents the ending point.

Fig. 3.

Effective connectivity of the effect of auditory-based cognitive training. One frame of the effective connectivity movie of the effect of (A) 1-hour auditory-based cognitive training and (B) a full course training at 184 ms and 387 ms after the stimulus onset seen from a sagittal view (supplementary movie 2 and 4). Effective connectivity of the effect of a full course training from the right middle frontal gyrus to a region near the left calcarine sulcus (C). RPDC, renormalized partial directed coherence; A, anterior; P, posterior.

Auditory-Based Cognitive Training Effects After First 1-Hour TCT Exposure

One hour of auditory-based cognitive training effects on effective connectivity among underlying brain regions that generate gamma synchronization in schizophrenia patients are shown in figure 4A and 4B. The 500 ms portion of the ASSR was divided into early vs late portions of the response, consistent with our previous ASSR connectivity results from a large cohort of healthy comparison subjects and schizophrenia patients,³⁸ which showed distinct response dynamics over these time windows. This network figure shows rank order latency of each connectivity along with figure 4E, revealing how 1-hour training effects on connectivity from an early local network to a later broad network in schizophrenia patients.

Fig. 4.

Effect of auditory-based cognitive training on neural networks and temporal dynamics underlying gamma oscillation. Legend: Neural networks formed by 1-hour auditory cognitive training effects at post stimuli 1–250 ms (A) and 251–500 ms (B) and by full course auditory cognitive training effects at post stimuli 1–250 ms (C) and 251–500 ms (D). Temporal dynamics formed by 1-hour auditory cognitive training effects (E) and by full course auditory cognitive training effects (F). Cing Mid Dors, dorsal middle cingulate gyrus; Cing Mid Vent, ventral middle cingulate gyrus; Cing Post, posterior cingulate gyrus; Front Inf, inferior frontal gyrus; Front Mid, middle frontal gyrus; Front Sup; superior frontal gyrus; Occipit Mid, middle occipital gyrus; Rolandic, Rolandic operculum; RPDC, renormalized partial directed coherence; Supp Mot, Supple motor area; Supramarginal; supramarginal gyrus; Temp Sup, superior temporal gyrus.

During the early (ie, onset) time interval of ASSR (ie, ~1–250 ms) after 1 hour of TCT, increased connectivity from the left ventral middle cingulate gyrus to the left posterior cingulate gyrus was identified in the TCT group compared to the TAU group. This increased connectivity was accompanied by decreased connectivity from the left Rolandic operculum (a region that includes auditory cortex and spans Broadmann areas 41 and 42) to the right ventral middle cingulate gyrus, which appears shortly after the onset of the increased connectivity. Given the location of the Rolandic operculum, the decreased connectivity may have suppressed an excessive response to auditory stimuli.

After 1 hour of TCT, connectivity from the right superior frontal gyrus to the right dorsal middle cingulate gyrus was modestly decreased in the TCT group compared to the TAU group during the later sustained portion of the ASSR time window (ie, 251–500 ms).

Auditory-Based Cognitive Training Effects Following a Full Course of TCT

Auditory-based cognitive training effects on effective connectivity underlying gamma synchronization following a full course of TCT are shown in figures 4C, 4D, and 4F. After the full course of TCT, schizophrenia patients showed a stable, well-formed, and simple network consisting of decreased connectivity from the right middle frontal gyrus to a region near the left calcarine sulcus during the later phase of ASSR time window (ie, 251–500 ms).

Comparing Auditory-Based Cognitive Training Effects After First 1-Hour TCT Exposure to Those Following a Full Course of TCT

Comparing auditory-based cognitive training effects after first 1 hour TCT exposure to those following a full course of TCT, the modest decreased connectivity from the right prefrontal gyrus after 1 hour of TCT seems to have been enhanced after a full course of TCT.

Discussion

This study is the first to report how cognitive training alters the connectivity of synchronized gamma-frequency oscillations across neural networks in patients with schizophrenia. After 1 hour of TCT, increased connectivity from the left ventral middle cingulate gyrus to the left posterior cingulate gyrus, accompanied by decreased connectivity from the left Rolandic operculum (a region that includes auditory cortex) to the right ventral middle cingulate gyrus was identified; connectivity from the right superior frontal gyrus to the right dorsal middle cingulate gyrus was modestly decreased. By contrast, after a full course of TCT, strengthened suppression from the frontal cortex to a region near the calcarine sulcus was seen. These results suggest that as early as 1 hour after TCT, an identifiable neural network may be beginning to form that is characterized by relatively inefficient information processing with much more widespread connectivities than a full 30-hour course. We speculate that a 30-hour course of TCT drives a more adaptive and efficient network that is characterized by decreased connectivity from the right middle frontal cortex to a region near the left calcarine sulcus. It may be that the EEG findings produced by 1-hour of TCT are driven by the acute effects of intensive exposure to TCT while the EEG findings of a 30-hour course reflect the long-term neuroplastic effects of a full course of training.

The present results are consistent with previous studies investigating sources underlying the gamma-based ASSR. For example, Tada and colleagues³⁷ investigated cortical surface levels of gamma-band ASSR using invasive electrocorticogram (ECoG) recordings in patients with treatment-refractory epilepsy, and identified prominent gamma-band ASSR responses at the superior temporal gyrus, supramarginal gyrus, postcentral gyrus, sensorimotor area, as well as the prefrontal gyrus. The locations of cortical surface levels of gamma-band ASSR found in the study by Tada and colleagues³⁷ are remarkably consistent with the nodes we detected in the current study using high-density scalp recordings.

Our previous large scale study of a non-overlapping cohort of schizophrenia patients (N = 427) and healthy comparison subjects (N = 293) found reduced gamma-band ASSR in the left superior temporal gyrus in schizophrenia patients, which was modestly correlated with impaired neurocognitive function.³⁶ While in that study schizophrenia patients were not administered cognitive training, our present results further underscore the important contribution of the left temporal gyrus underlying some of the cognitive benefits imparted by this intervention.^41–44

The current analyses identified an unexpected reduced connectivity from the right prefrontal cortex to the left calcarine sulcus, a region centered near the primary visual cortex. Our previous study revealed the changes in the “opposite” direction—abnormal up-regulated connectivity from the left calcarine sulcus to the right prefrontal cortex—underlying early auditory information processing (mismatch negativity) in schizophrenia patients.⁴⁹ We considered that the abnormal occipito-frontal effective connectivity in schizophrenia patients may reflect a process in which higher-order regulatory functions of the frontal lobe are interrupted by excessive and aberrant low-level perceptual/attentional updates imparted by more caudal brain regions in the study. Furthermore, a similar abnormal increase in the occipito-frontal effective connectivity was found in patients with chronic tic disorders,⁴⁸ suggesting that abnormal occipito-frontal connectivity may not be unique to schizophrenia but relevant for understanding multiple neuropsychiatric disorders. Therefore, decreased connectivity from the prefrontal cortex to the calcarine sulcus produced by a full course of TCT may be associated with a suppression of abnormal connectivity from the occipital to the prefrontal regions in schizophrenia patients. Previous studies of TCT in schizophrenia patients have found that gains in global cognition and verbal learning accrued from auditory training are accompanied by modest decreases in visually-based cognitive test performance.⁵¹ While admittedly speculative, it is possible that the observed changes in the occipito-frontal connectivity may be associated with such reductions in visual-cognitive functioning.⁵¹ Further investigating these networks is essential for understanding the neurophysiology of neurocognitive dysfunction and mechanisms underlying TCT response.

Limitations

This study should be considered in the context of several limitations. First, this is a study of a highly symptomatic and functionally impaired cohort of schizophrenia patients, all of whom were receiving complex medication regimens.⁴⁴ Previous studies from this sample suggest that the pro-cognitive benefits of TCT may offset some adverse cognitive effects of polypharmacy and anticholinergic burden.⁵² The complex medication regimens do not allow for an easy disentangling of changes to information flow across cortical networks. Second, we did not perform correlation analyses between the malleability of the connectivity and change of cognitive function after TCT. Future studies are needed to better characterize the neural substrates of ASSR that are associated with TCT-related gains in clinical and cognitive function. Third, to check the reliability of general connectivity patterns in schizophrenia patients, we compared the baseline results from the treatment-refractory inpatients in the current study (N = 52; TAU_T0 + TCT_T0) with both schizophrenia outpatients (N = 426) and healthy comparison subjects (N = 293) from our previous archival and larger datasets³⁸ (supplementary movie 5). Despite large study differences in patient acuity, testing environments, stimulation parameters, and electrode montages, the 3 groups showed similar connectivity patterns in both the early time interval (1–250 ms) in which suppression is dominant as well in the late interval (251–500 ms) in which increased connectivity is dominant. Similarly, it is important to note that in both studies described above, the schizophrenia patients had a chronic and well-established illness; results, therefore, may not generalize to at-risk or early-illness psychosis patients. Fourth, while we applied a conservative approach for isolating TCT-related changes in connectivity above and beyond generalized treatment effects common to both groups, we cannot fully exclude the possibility that either preexisting baseline differences in groups or other nonspecific factors did not contribute to the observed pattern of results. To isolate TCT-related effects from preexisting group differences or generalized time/treatment effects common to both groups, we created and applied a mask to minimize the potential contaminating effects of non-interest in TCT_T0 – TAU_T0, TAU_T1 – TAU_T0, and TAU_T2 – TAU_T0. These masks resulted in the exclusion of connectivities from the occipital to the other regions (supplementary figure 1), which are possibly attributable to modest preexisting group differences. On the other hand, this approach to isolate TCT related effects may be overly conservative and therefore miss potentially important changes in brain regions of interest. In fact, the TAU_T1 – TAU_T0 and TAU_T2 – TAU_T0 (supplementary figure 2) mask excluded connectivities from the middle temporal gyrus, including the auditory cortex, to the other regions. We nevertheless speculate that these nonspecific changes are due to merely repeating the ASSR task immediately before and after the 1st hour of training and the generalized effects of treatment as usual within the context of the long-term care facility common to both groups. Fifth, 64 EEG channels with template head models were used for the analyses in this study. Future studies may fruitfully use individual magnetic resonance imaging (MRI) data and digitized scalp sensor locations rather than template head models and reliance on standardized electrode locations for potentially improved accuracy of source dynamics.

Conclusion

Auditory-based cognitive training drives neuroplasticity in the auditory network underlying the synchronized flow of gamma-band oscillations in schizophrenia patients. Specifically, strengthened suppression from the frontal cortex to a region near the calcarine sulcus was detected after a full course of TCT. These findings support the use of TCT and encourage future investigations of the neural substrates of gamma-band ASSR in the context of TCT and other pro-cognitive therapeutics.

Funding

This study was supported by JSPS Overseas Research Fellowships (D.K.), the Sidney R. Baer, Jr. Foundation, and the VISN-22 Mental Illness Research, Education, and Clinical Center. Swartz Center for Computational Neuroscience is supported by generous gift of Swartz Foundation (New York). The funders had no role in the study design, data collection and analysis, publication decision, or manuscript preparation.