Functional Correlates and Pathophysiology of EAP and Auditory Plasticity in Schizophrenia

Although EAP was once considered an “intact simple function” [24], deficits in pitch processing in schizophrenia were first demonstrated in 1973 in the context of studies of paranoia [5], and have been replicated over recent years by multiple independent groups [25]. EAP deficits are exemplified by elevated TMT thresholds [5,26], and are associated with deficits in more complex information processing that are crucial for economic, occupational, and social functioning for both healthy volunteers [27,28] and schizophrenia [29–31], such as reading [7,32]. EAP deficits also correlate with deficits in global cognition [6,33], AER [8–10,34–37], perceptual music disorders [38,39] and theory of mind [40,41].

As recently reviewed [42], schizophrenia patients show an auditory cortical pattern of EAP deficits. In otherwise healthy controls, isolated bilateral auditory cortex lesions lead to a dramatic increase in pitch TMT. Schizophrenia patients require approximately a 20% difference in pitch (%Δ frequency) to differentiate two tones separated by a brief (300 ms) delay to match performance at 5% Δfrequency in healthy controls [43], a degree of deficit that is as robust (d = 1.25) as that typically seen on more complex tasks such as tests of processing speed or executive dysfunction [43]. Consistent with the functional evidence for an auditory cortex deficit in schizophrenia, reduced auditory cortex volume is specifically seen in schizophrenia [44,45]. These structural and functional EAP deficits appear to be NMDAR dependent, as TMT depends on the formation of a NMDAR dependent ʻechoicʼ memory trace.

The auditory cortex is the center of a complex, distributed, hierarchical network [46] (Figure 1), and was traditionally thought to be “fixed” or nonplastic outside of critical periods early in life [47]. Along with other cortical areas [48,49], however, the auditory cortex remains highly plastic into adulthood [50–52]. AudRem is not a simple measure of either sensory learning or TMT, as AudRem can lead to increased recruitment of neurons responding to the training stimuli [53,54], and hence an increase in the so-called cortical “area of representation” of those stimuli and long term auditory neuroplasticity [55]. Thus, our plasticity outcome of improved (smaller) pitch thresholds between successive auditory stimuli after AudRem has direct relevance to cortical neuroplasticity and can facilitate gains in higher-level cognitive processes [56].

Limitations of Auditory Plasticity Training and Rationale for Combined Treatment

There are no FDA approved pharmacological treatments for cognitive enhancement in schizophrenia, but behavioral interventions have shown promise. Cognitive remediation is commonly defined as “a behavioral training-based intervention that aims to improve cognitive processes [57]”. Cognitive remediation programs vary in the skills they target, with some focused on complex skills like executive functioning, and others targeting more basic skills [58–62]. Posit Science, a program that particularly focuses on AudRem and plasticity, has shown significant, moderate-large (d = 0.56–0.86) effect-size improvements in global cognition compared to a videogame control in schizophrenia [63,64]. Subsequent reports using AudRem have been mixed [14,65,66], and furthermore, the clinical burden of a ~50-h treatment administered 3 to 5× a week limits feasibility.

A recently published 12-week study that randomized 103 subjects to either AudRem or a more general cognitive remediation [56] is exemplary of both the promise and limitations of AudRem. At baseline, subjects had clinically significant cognitive deficits, exemplified by a MATRICS [67] composite T score (MCCB) of 20.2. Fifty-three subjects (51.5%) had impaired EAP, defined as a baseline TMT score of <70% correct. EAP deficits were treatable, as within the impaired EAP subgroup, there was a significant, specific effect on auditory cognition (p = 0.04, d = 0.73). Baseline TMT predicted overall MCCB improvement (r = 0.77, p < 0.001), suggesting that subjects with baseline EAP deficits will especially benefit from AudRem.

Nonetheless, the efficacy of AudRem on its own is limited. Even in this study [56], overall improvements were small, with subjects improving to a mean MCCB T score of 25. While this exceeds the known practice effects [68] (~2 T score), it remains 2.5 standard deviations below normal (T score = 50). This is consistent with meta-analyses of cognitive remediation trials [57,69] that suggest that up to 45% of people with schizophrenia demonstrate minimal improvement after undergoing a therapeutic dose (≥32 h) of cognitive training [70]. Our group [16,17], and others [12,71–74] have proposed that the efficacy of AudRem may be enhanced by combining with adjunctive medication, particularly an NMDAR modulator. As detailed in the next several sections, NMDAR functioning in schizophrenia is reduced, but not absent. Although the present project is not designed to assess synergistic effects, our model asserts that d-serine combined [71] with AudRem may be an improvement over either one alone.

Rationale for Studying the NMDAR Target for Enhancing Auditory Neuroplasticity

As previously reviewed [20], in addition to a well-characterized role of brain NMDAR function in the pathophysiology of schizophrenia symptoms [75–77], NMDAR antagonists such as ketamine reproduce core neuropsychological abnormalities of schizophrenia, including auditory cognitive deficits [21,78,79] and MMN [80]. By contrast, similar effects are not observed during exposure to dopaminergic (e.g., methylphenidate [81]) or serotonergic [82,83] agents, further supporting NMDAR models.

The NMDAR is the primary glutamate receptor and facilitates temporal summation of slow excitatory postsynaptic potentials, allowing sustained neural excitation [84]. Thus, NMDAR also have a well characterized role in neuroplasticity, serving as a critical trigger for plasticity-related functions such as acquisition and retention of information [85,86], and long-term potentiation (LTP) and depression (LTD) [84,87,88] throughout the brain. The NMDAR is blocked by a magnesium ion at rest and is dually voltage and ligand gated, and therefore able to detect coincident presynaptic and postsynaptic activity. These unique properties allow the NMDAR to integrate information from multiple pathways and make them critical for memory formation.

As recently reviewed [42], in addition to broader cortical NMDAR deficits, both preclinical and postmortem studies directly implicate NMDAR dysfunction in auditory plasticity deficits in schizophrenia, particularly in impaired pitch processing. Preclinical studies suggest that NMDAR blockade during auditory learning can impair brain plasticity [55,89,90] and NMDAR stimulation produces improvements in both auditory LTP and MMN in healthy rats [91], providing further preclinical support for equating MMN with cortical neuroplasticity.

Rationale for Early Auditory Processing Electroencephalography (EEG) as a Measure of Target Engagement of NMDAR Agonists

MMN [92,93] is a neurophysiological response elicited most commonly in the context of an auditory oddball paradigm in which a sequence of repetitive standards is interrupted infrequently by a physically different oddball (deviant) stimulus. Deviants may differ from standards in one or more physical and/or abstract dimensions, including pitch or frequency [94,95]. As recently reviewed [42], MMN requires intact auditory cortex functioning, and auditory cortex volume is related to MMN [96]. Support for using MMN as a target engagement biomarker stems from well documented, highly reliable deficits in schizophrenia [33,97–102] and linkages to NMDAR dysfunction in the auditory cortex across rodent [103,104], monkey [105,106] and human [80,107–111] investigations. Moreover, similar to EAP, MMN deficits are highly predictive of poor functional outcomes [9,112–116], as recently confirmed in a large, 1415 subject cross-sectional study of schizophrenia [6]. MMN is increasingly conceptualized as reflecting the “prediction error” evoked when the deviant differs from the standard stimulus [117–121], and thus can also be considered a neurophysiological proxy of plasticity. Recent studies of AudRem in both established [122] and early schizophrenia [115] show that MMN changes are correlated with cognitive improvement.

The utility of MMN as an NMDAR target engagement biomarker is further supported by recent studies showing that MMN deficits in schizophrenia may be sensitive to NMDAR modulating compounds, including d-serine [15,123], glycine [124] and N-acetylcysteine [125]. Our recent findings [123,126] suggest that MMN may have positive and negative predictive value in predicting the efficacy of novel NMDAR agonists and plasticity treatment.

In addition to MMN, we will also assess event-related oscillation (ERO) responses as functionally relevant, target engagement biomarkers, particularly auditory θ. Electrophysiological activity is divided conventionally into discrete θ (4–7 Hz), α (7–12 Hz), β (12–24 Hz) and γ (>24 Hz) bands, which reflect differential underlying local-circuit processes [127]. Stimulus-induced responses, including MMN and N1, are typically associated with increases in θ [128]. In contrast to θ and its association with sensory processing, reduction of β activity (termed, event-related desynchronization) has been associated with “higher-level” cognitive processing in frontoparietal networks [129,130]. In our published work, we recently demonstrated that ERO may be especially sensitive to early auditory deficits, and that θ activity during MMN was predictive of symptomatic and functional impairment [128]. Similar findings are seen preclinically [131].

Specific Rationale for Use of d-Serine to Enhance Plasticity

d-Serine is a direct, full agonist at the d-serine/glycine modulatory site of the NMDAR [22,23]. Activation of the NMDAR requires binding of both glutamate and concurrent binding of d-serine or glycine [132]. Although direct enhancement of NMDAR signaling by glutamate itself can produce excitotoxicity, stimulation via the d-serine site offers a safer method for facilitating activity [20,77]. Furthermore, the d-serine site is not fully saturated in cortical/subcortical regions [133], suggesting that exogenous d-serine may be beneficial. In addition to its more general role in NMDAR modulation, d-serine also has a specific role in LTP/LTD and long-term plasticity [134,135] and synaptogenesis [136]. Schizophrenia patients have well documented, functionally relevant deficits in d-serine [137,138].

Specific advantages of d-serine as a plasticity enhancer include well classified pharmacokinetics (PK) [123,138], and a short half-life (tmax ~30–60 min) allowing for practical administration ~30 min before sessions, thus allowing for plasticity assessment during peak levels. d-serine has shown efficacy in enhancing plasticity outside of critical periods early in life [47], including studies showing efficacy of d-serine in geriatric animals [139] and geriatric [140] and adult healthy volunteers [141]. Chronic use of d-serine does not lead to tachyphylaxis [15,123,142,143], suggesting utility in repeated use, which is essential for a sustained AudRem study. A recent report [72] suggests a specific relationship between d-serine and AudRem, finding a positive correlation between increased d-serine levels and improved global and auditory cognition within the active AudRem group, but not in the sham AudRem group.

The majority of d-serine studies have used a low (30 mg/kg, ~2 g/day) dosage, with a significant, but small effect size improvement at this dose in meta-analyses [137]. This provides proof of concept, but suggests 30 mg/kg may be inadequate to fully engage the target [144,145]. Safety, efficacy and pharmacokinetics of higher dose d-serine (≥60 mg/kg, ≥4 g/day) was recently studied, finding dose-dependent improvement in symptoms and cognition [138]. A significant dose effect for cognition was supported by significantly greater improvement at ≥60 mg/kg vs 30 mg/kg dose for the MCCB composite (p = 0.017), with specific improvement in auditory cognition (p = 0.035). Pharmacodynamic analysis also supports a dose effect, as higher peak serum levels predicted greater MCCB improvements in this study. Subsequent double-blind studies in both established [123] and early, clinically high risk for schizophrenia groups [146] further support the efficacy of higher dose d-serine. A meta-analysis [123] including high dose studies demonstrates a moderate-large (d = 0.7) effect size for negative symptoms, improving on meta-analysis exclusively in low doses.

Two recently published studies support the utility of d-serine to study and enhance plasticity and AudRem, finding that both intermittent and sustained d-serine treatment modulates MMN [15,123]. The first [15] serves as the model for the design of the R61–R33 project. In this study, 21 schizophrenia patients received either d-serine 60 mg/kg or placebo with three 1× weekly sessions of AudRem program. ERO responses were recorded during the AudRem, while MMN was measured pre-post training sessions. A significant d-serine treatment effect vs placebo was seen for pitch discrimination after AudRem (behavioral plasticity), along with the EEG plasticity measures MMN (Figure 2A) and activity in the θ and β ranges. Furthermore, following just two d-serine treatments, schizophrenia outcomes for behavioral plasticity were normalized vs. controls. Consistent with prior studies [122], schizophrenia subjects receiving AudRem alone tended to show worsening in MMN. A relationship between improvements in plasticity and MMN was seen (r = −0.34, p = 0.034, Figure 2C), while changes in θ also correlated significantly with plasticity improvements (r = −0.39, p = 0.002).

Open in a separate window

Figure 2.

Voltage topography maps for mismatch negativity (MMN) for indicated group for Baseline (left) and Final (right) shown at peak latencies for intermittent (A) and sustained (B) treatment. Analyzed electrode noted by red circles. Analyzed electrode noted by red circles (Fz). (C) Scatter plot for % change in behavioral plasticity during AudRem vs change in MMN amplitude. Modified from [15,124]. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Sustained effects of d-serine on MMN, along with relative effects on NMDAR target engagement has also been recently shown [123,126]. Across two separate, double-blind, placebo controlled, NMDAR-related clinical development programs, MMN showed positive and negative predictive value in predicting efficacy of novel NMDAR agonists. In these studies, 44 individuals with schizophrenia were treated with placebo, d-serine (60 mg/kg/day) or bitopertin, a selective glycine transport type 1 (GlyT1) inhibitor [147] (10 mg) for 4–6 weeks. For d-serine, a significant, large effect size improvement vs. placebo for MMN was seen (Figure 2B), along with intercorrelated improvements in clinical symptoms. By contrast, bitopertin did not significantly affect either symptoms or MMN, consistent with negative Phase II and III studies for bitopertin 10 mg [148,149]. This suggests that bitopertin 10 mg may have failed because of inadequate target engagement, and further emphasizes the need to conduct target engagement studies for dose finding prior to Phase II.

Limitations of Alternative NMDAR Modulators

As recently reviewed [16,17], most prior studies combining NMDAR agonists with cognitive remediation were conducted with d-cycloserine (DCS), a d-serine derivative which is clinically available because of its use as an anti-tuberculosis agent [150], a function which is unrelated to its NMDAR effects. In low doses, DCS acts as an agonist at the same site as d-serine. Recent studies of low dose DCS in healthy volunteers [151–153], anxiety disorders [154], and schizophrenia [155–157] have shown proof of concept efficacy for plasticity enhancement [155–158]. However, meta-analysis [159,160] shows only small effect size improvements (d = 0.25), and low dose DCS is a low potency agonist [161], and higher dose DCS acts as an NMDAR antagonist [162]. Thus, DCS may result in receptor desensitization [163] or psychosis in schizophrenia. Other promising alternatives, such as CTP-692 [164], sarcosine [165] and d-amino acid oxidase inhibitors [166,167] are presently unavailable for general study. The NMDAR partial antagonist memantine has inconsistent effects on MMN [168] and cognition [169] that may be due to dopaminergic [170], not glutamatergic mechanisms [16], and thus would not unambiguously assess the NMDAR target. Thus, d-serine balances efficacy, availability and safety, and is thus the best available agent for assessing NMDAR-based plasticity enhancement and target engagement.

R61 Design

The R61 is conducted in cohorts of 15 subjects in which 12 subjects are randomized to double-blind d-serine for each of the three treatment visits and 3 are randomized to double-blind placebo for each of the visits (Figure 3). The first cohort of 15 will receive 80 mg/kg or placebo, the second cohort receive 100 mg/kg or placebo, followed by a third cohort which receive 120 mg/kg or placebo. FDA approval is required after each cohort. Thus far, the first and part of the 2nd cohort have been safely completed.

Open in a separate window

Figure 3.

Study Flow Chart. *Three 15-subject cohorts will be conducted sequentially, beginning with a d-Serine dose of 80 mg, followed by a 100 and 120 mg/kg cohort. A 60 mg/kg cohort will be added in case of safety concern in 80 or 100 mg/kg cohort.

After informed consent, the subjects undergo full medical and psychiatric screening, which occurs over an up to 31-day period. After satisfying initial inclusion/exclusion criteria (Table 1), subjects complete baseline cognitive and behavioral assessments with a clinical rater.

Subjects are randomized to either receive d-serine or placebo. Subjects will receive the same drug assignment/dose for each of the three treatment days. Randomization is stratified by baseline TMT [26], to allow for potential sub-group analysis. After a negative urine pregnancy test for fecund women and the Columbia Suicide Severity Rating Scale (C-SSRS) [171], each of the three treatment visits begins with a pre-treatment EEG capping. Baseline EEG is collected prior to treatment day 1 only. Subjects then receive d-serine or placebo and AudRem begins 30 min after study drug administration to allow for training during peak d-serine levels. EEG is recorded during sessions to assess ERO, including θ. Immediately after AudRem, subjects complete post-treatment MMN. A d-serine level is drawn after the session using established methods to allow for functional pharmacodynamics readout [138], along with urinalysis and clinical laboratory assessments. The SAFTEE assessment is used to assess general side effects [172].

Neuroplasticity-Based AudRem Program

We utilize an AudRem program that was originally developed for use in developmental dyslexia [18]. Through 2018, the published studies of this AudRem program include 1137 subjects, including 331 from a patient population (dyslexia: n = 237 subjects; schizophrenia: n = 61 and ADHD: n = 33), amongst 16 publications in total [15,18,173–186]. While most of the replications are in dyslexia by the same group (n = 12 published papers), the reliability/validity of the program has been independently replicated by 5 independent groups [15,175,179,184]. Our study [15] was the only one in schizophrenia, and the only one to assess NMDAR mechanisms. While most studies, including our own [15], have focused on acute changes over a few sessions, independently published [180,184] data also show sustained working memory improvements with extended training on similar AudRem programs, supporting our design. In summary, across disorders with impaired auditory plasticity (dyslexia and schizophrenia), all studies that use this or other similar AudRem programs demonstrate the same pattern of improvement over time in behavioral and neurophysiological plasticity.

AudRem sessions are administered once per week ± 2 days. Participants are presented with paired tones (e.g., Stimulus 1 (“reference”) and Stimulus 2 (“test”): S1 and S2) and indicate which tone is higher in pitch (frequency). In the first pair, the between tone ratio is 50% (e.g., 1000 ± 500 Hz), and the difficulty level is adjusted to maintain a steady (~70% correct) level of performance. When the reference (S1) remains constant, highly significant improvement is seen. An added advantage of our AudRem program is that its simplicity minimizes the confound of consolidating a failed or unsuccessful trial, a concern in enhancing extinction learning [187]. EEG analysis will be conducted using previously published methodology [15]. Plasticity will be operationalized as improved (smaller) pitch thresholds between successive auditory stimuli after AudRem.

Baseline Measures

In order to assess the functional relationships of plasticity, a number of measures will be collected at baseline only, and will be used for correlational analysis under AIM 2 and milestone 2, as detailed below. These measures will be evaluated for use as outcome measures in the R33.

Auditory cognition is defined by the Verbal Memory domain of the MCCB. Cognition is also secondarily assessed using the overall MCCB and remaining individual domains. EAP is assessed with the TMT [35]. The baseline TMT task differs from the primary behavioral plasticity outcome by being a fixed, static measure and does not provide feedback or dynamically change its difficultly based on subject performance. This task consists of pairs of tones, and within each pair, tones are either identical or different in frequency by specified amounts in each of the five blocks (2.5, 5, 10, 20 or 50%).

Reading assessments includes the Wide Range Achievement Test (WRAT) [188], a single-word reading test that assesses premorbid reading level; the Comprehensive Test of Phonological Processing (C-TOPP) [189], which measures phonological processing and the Woodcock Johnson Tests of Achievement, 3rd edition (WJ) [190], which tests comprehension of written language. Social cognition is assessed with the AER [10] task and the Sarcasm task [40]. Psychiatric symptoms are assessed with the Positive and Negative Symptom Scale (PANSS) [191] and general function with the University of California San Diego Performance-Based Skills Assessment (UPSA) [192].

Study Drug and Maintenance of the Blind

This study is conducted under IND 122821, which specifically allows for the present design and dosing >60 mg/kg. d-Serine or placebo are administered as a solution, prepared (pre-mixed) in water and prescribed only as “study medication”. d-Serine is dosed by weight (e.g., 80 mg/kg for the first cohort) and dispensed to subjects in identical appearing bottles. An artificial sweetener is used as placebo. Medication is dispensed on the day of the visit by unblinded pharmacist who is otherwise uninvolved in the study. Envelopes containing treatment assignment are kept in the pharmacy and are unsealed only in case of medical necessity and blind-breaks will be on an individual subject basis only. The pharmacist does not participate in assessing any dependent variable and conveys no information about drug status to patients or staff except in a medical emergency.

Permitted medications: Subjects will receive concurrent antipsychotics. Patients are allowed to receive the following adjunctive medications during the course of the study: anticholinergic agents; beta-blockers; mood stabilizers, antidepressants; and anti-anxiety agents. As needed doses of clinically determined benzodiazepines or antipsychotics are permitted. Given a possible detrimental effect on cognition, patients are asked to not take these adjunctive medications the night before or on the day of testing/training sessions if clinically feasible.

Plasma d-serine levels: Venous blood samples (10 mL per blood draw) are drawn after the AudRem session for assay.

Milestones (Go/No Go Criteria)

In the R61-R33 mechanism, emphasis is placed on the development of target engagement biomarkers to ensure that clinical trials produced an adequate test of the underlying hypothesis. The study go/no-go milestones are presented below. Failure to demonstrate target, functional and safety milestones will lead to project termination.

Additional Considerations

Additional measures will be used to inform the design of the R33 study, but not as R61 Go/No-Go criteria. These include (1) relative effects of dose, (2) effect size estimates for secondary EEG outcomes and (3) pharmacodynamics assessment. In general, assuming target engagement, relationship with function and safety milestones are reached by at least one active arm, the dose showing the largest effect size improvement will be utilized for the R33 phase. The 120 mg/kg dose will be used if effect sizes and safety are equivalent between doses.

Statistical Analysis Plan & Power

Before any specific statistical techniques are applied, we will examine all variables at all time points for illegitimate values, outliers, and other inconsistencies. Distributions of demographic variables and other clinically important baseline variables will be examined and summarized by means, standard deviations, minima, and maxima for continuous measures and proportions for categorical measures. We will make every effort to obtain all data to reduce or eliminate missing data issues. Intent-to-treat analysis will be implemented for all estimation and testing. Tests will be two-sided and statistical significance determined by p < 0.05.

This study was funded by the National Institute of Mental Health (R61MH116093) to JTK.