|Home | About | Journals | Submit | Contact Us | Français|
In an effort to reveal susceptibility genes, schizophrenia research has turned to the endophenotype strategy. Endophenotypes are characteristics that reflect the actions of genes predisposing an individual to a disorder, even in the absence of diagnosable pathology. Individual endophenotypes are presumably determined by fewer genes than the more complex phenotype of schizophrenia and would, therefore, reduce the complexity of genetic analyses. Unfortunately, despite there being rational criteria to define a viable endophenotype, the term is sometimes applied indiscriminately to characteristics that are deviant in affected individuals. Schizophrenia patients exhibit deficits in several neurophysiological measures of information processing that have been proposed as candidate endophenotypes. Successful processing of sensory inputs requires the ability to inhibit intrinsic responses to redundant stimuli and, reciprocally, to facilitate responses to less frequent salient stimuli. There is evidence to suggest that both these processes are “impaired” in schizophrenia. Measures of inhibitory failure include prepulse inhibition of the startle reflex, P50 auditory evoked potential suppression, and antisaccade eye movements. Measures of impaired deviance detection include mismatch negativity and the P300 event-related potential. The purpose of this review is to systematically evaluate the endophenotype candidacy of these key neurophysiological abilities. For each candidate, we describe typical experimental procedures, the current understanding of the underlying neurobiology, the nature of the abnormality in schizophrenia, the reliability, stability and heritability of the measure, and any reported gene associations. We conclude with a discussion of the few studies thus far that have employed a multivariate approach with these candidates.
In an effort to finally reveal susceptibility genes ensconced in the human genome, schizophrenia research has turned, at a remarkable pace, to the endophenotype strategy. Endophenotypes are characteristics, usually assessed in a laboratory, that reflect the actions of genes predisposing an individual to a specific disorder, even in the absence of any diagnosable pathology. As relatively simple, well-defined and quantifiable biobehavioral characteristics, individual endophenotypes are presumably determined by fewer genes than the more complex phenotype of schizophrenia. Ideally, therefore, endophenotypes could serve as dissected components of the complex schizophrenia phenotype, reflecting fewer genes and thereby reducing the complexity of the genetic analyses required to identify contributing genes.
Investigations of endophenotypes of psychopathology have accelerated because of a growing recognition of the promise of the approach, as described by Gottesman and Gould1 in their influential and oft-cited discussion of the strategy. At the time of their report, a MEDLINE search for “endophenotype” identified 62 entries between the years 2000 and 2002, compared with 16 articles prior to 2000. For 2003 through June 2006, an additional 109 publications with keyword endophenotype have been added to MEDLINE, reflecting a substantial increase over all prior years. The psychopathology research field is now abuzz with the endophenotype term. Unfortunately, the term is sometimes applied indiscriminately to any characteristic observed to be “deviant” in affected individuals. There is a risk, with the science striving to advance so quickly, that the rigorous critical evaluation of candidate endophenotypes will fall to the wayside. There are several criteria, rationally stemming from their proposed use as tools for revealing the genetic and neurobiological underpinnings of psychopathology, that candidate measures should meet, if they are to be considered viable endophenotypes.1–4 While there is some variability in specific criteria, it is generally agreed that the ideal candidate endophenotype would exhibit the following features:
Individual endophenotypes fulfilling these criteria may constitute independent risk factors that identify different unrelated types of genetic risk. Conversely, if multiple deficits tend to coaggregate in the families of patients with a specific disorder, it would suggest that the combined set of deficits reflect a single, common variant of genetic risk for the disorder. An alternative to the use of a single endophenotype is to combine endophenotypes or create a composite, multivariate endophenotype, in order to better identify genetic risk.5–7 Although there is a growing interest in the investigation of multivariate endophenotypes,7–10 most research to date has centered on the evaluation of individual candidate endophenotypes.
If individual candidates are to form a multivariate endophenotype, it is arguably important that each candidate stand alone as a viable endophenotype. Schizophrenia patients exhibit well-documented deficits in several laboratory-assessed neurophysiological abilities that have been proposed as candidate endophenotypes. Observed deficits extend from the earliest preattentive stages of information processing to relatively late higher cortical processes. However, it has been suggested that a breakdown in the processes that regulate the inflow of information from the environment is fundamental.11 Successful processing of sensory inputs requires the ability to screen out or inhibit intrinsic responses to redundant or irrelevant inputs and, reciprocally, to enhance or facilitate responses to deviant, novel, or salient stimuli. There is evidence to suggest that both these discrete but related processes are “impaired” in schizophrenia. Among measures of inhibitory failure, those most commonly studied in schizophrenia are prepulse inhibition (PPI) of the startle reflex, P50 auditory evoked potential suppression, and AS eye movements. Among measures of impaired deviance detection, mismatch negativity (MMN) and the P300 event-related potential (ERP) have been most commonly studied. While there have been reviews of particular aspects of the literature related to the endophenotype candidacy of each of these neurophsyiological measures,12,13 to our knowledge, there have been no reviews describing how well each individual candidate fulfills all the above criteria.
The primary purpose of the current review is, therefore, to offer a systematic evaluation of the endophenotype candidacy of the key neurophysiological abilities implicated in schizophrenia. For each candidate, we describe typical experimental procedures, current understanding of the underlying neurobiology, the nature of the abnormality in schizophrenia, the reliability, stability and heritability of the measure, and any reported specific gene associations. As a reflection of the current state of the field, we first review the extensive literature on each individual candidate endophenotype. We then conclude with a discussion of the few studies thus far that have employed a multivariate approach with these candidates.
PPI is the normal reduction in startle that occurs when a startling stimulus is preceded 30–300 ms by a weak prestimulus.14 PPI is deficient in schizophrenia patients and unaffected relatives,15–19 suggesting that it may be a trait marker for individuals at risk for developing the disorder (rather than a marker of schizophrenia per se).
The assessment of PPI requires appropriate stimulus delivery time-locked to the acquisition of a valid measure of startle response magnitude. Though startle can also be elicited via a robust, abrupt visual, or tactile (air puff or electrical) stimulus, acoustic stimuli allow a high degree of experimental control and precision, and avoid logistical complexities of delivering electrical shocks to acutely psychotic individuals. The motor response most often used to assess startle and PPI in humans is the contraction of orbicularis oculi (OO) via electromyographic (EMG) measures during the first 250-ms epoch after startle stimulus onset.
There are many variants of startle/PPI acquisition procedures. For one common test procedure, subjects sit in a comfortable chair in a sound-attenuated room with a constant, neutral visual field (because any weak “noise” or visual stimulus can serve as a prepulse). Because of potential lateralized differences in the neural regulation of PPI, and its deficits in clinical populations,16 bilateral eyeblink measures are optimal. EMG electrodes are positioned below and lateral to each eye over OO, with a ground electrode behind one ear. Common blink acquisition and scoring parameters are described elsewhere.20 Startle stimuli are most commonly presented binaurally through headphones. Importantly, most studies reporting PPI deficits in schizophrenia populations use an “uninstructed” test session, in which test subjects are informed that they will hear noises, but are not instructed to perform any task related to those noises, that might engage volitional or attentional mechanisms.19
Stimulus parameters and test session design are selected based on the specific experimental questions of highest priority. Typically, in our studies, a test session includes a total of 74 active and 18 blank “no-stim” trials, and lasts 23.5 minutes, beginning with a 5-minute acclimation period with 70-dB (A) sound pressure level (SPL) white noise that continues throughout the session. Startle stimuli are 40-ms 115-dB (A) SPL noise bursts. Prepulses are discrete 20-ms noise bursts 15 dB above background, with onset 30, 60, or 120 ms prior to pulse onset. Figure 1 illustrates the effect of a prepulse on the magnitude of the startle response.
Preclinical studies have demonstrated that PPI in rodents is regulated by cortical structures (mesial temporal cortex and medial prefrontal cortex [PFC]) and subcortical structures (striatum, pallidum, and pontine tegmentum).21 This limbic cortico-striato-pallido-pontine circuitry converges with the primary startle circuit at the level of the nucleus reticularis pontis caudalis (NRPC).
Studies of the neurobiology of PPI in humans reveal both similarities and differences from findings in animal models. For example, evidence of profound PPI deficits in patients with Huntington disease (HD)22,23 support the role of the striatum in the regulation of PPI in humans and parallel findings of PPI deficits after excitotoxic lesions of the striatum in rats and in mice transgenic for the HD gene.24,25 Neuroimaging studies in schizophrenia patients and normal comparison subjects confirm both the role of limbic cortico-striato-pallido-thalamic (CSPT) circuitry in the normal regulation of PPI21,26 and the association between CSPT abnormalities and PPI deficits in patients. Also consistent with a number of preclinical findings is the fact that PPI deficits are observed in other clinical populations, ranging from Tourette syndrome to seizure disorders, with known or presumed pathology in limbic CSPT circuitry.19
On the other hand, pharmacological studies of PPI in normal humans raise questions as to whether the PPI neural circuit “blueprint” in rodents can be easily translated across species.21,27 Thus, while the indirect dopamine (DA) agonist, D-amphetamine, reduces PPI in both normal humans and some rodent strains, the N-methyl-D-aspartate (NMDA) antagonist ketamine, and the serotonin releaser methylene-dioxy-methamphetamine have opposite effects on PPI across species—disrupting PPI in rodents but increasing PPI in clinically normal humans. A similar pattern is observed with the mixed DA agonist/NMDA antagonist amantadine. Direct DA agonists potently disrupt PPI in rats, but their effects in humans are not as convincing. Both atypical antipsychotics and nicotine have been reported to increase PPI across species but perhaps only under certain experimental conditions.
Schizophrenia patients exhibit abnormal PPI despite having relatively normal responses to startling stimuli, as indicated by measures of startle amplitude, latency, and latency facilitation. Empirically, these PPI deficits indicate that in the immediate aftermath of a stimulus, the central nervous system in schizophrenia is overly responsive to a second stimulus. Conceptually, this deficit of time-locked, automatic inhibition puts the information contained in the initial stimulus at greater risk of being degraded, thereby disrupting its appropriate cognitive or behavioral impact. Importantly, there is no direct evidence that the sensory processing of the prepulse is degraded in patients nor is there direct evidence that this presumed degradation of prepulse processing causes impairments in the cognitive functions that depend on the intact sensory processing of the prepulse. However, PPI deficits in schizophrenia patients are not modality specific,20 and they correlate with distractibility28 and thought disorder29,30 measures and more global scales of functioning.31
Some forms of PPI exhibit robust test-test reliability in normal comparison subjects, over intervals of several months.32,33 Thus, intraclass correlations exceeding 0.90 have been observed for PPI elicited over 3 consecutive months, using relatively intense prepulses34 and 30- to 120-ms prepulse intervals. With shorter prepulse intervals, and/or weaker prepulse intensities, PPI is less reliable over time.34,35
Relatively few studies have assessed the longitudinal stability of PPI deficits in schizophrenia patients. While PPI deficits have been reported by several groups in relatively stable, medicated schizophrenia cohorts,16,19,36–40 others report that PPI deficits occur only in patients who are in an acute symptomatic phase of a psychotic episode and resolve in these same patients as symptoms subside.41 One complexity in interpreting such “state-related” changes in PPI relates to the role of antipsychotic medications in both symptomatic improvement and potentiation of PPI (see below).
Very convincing evidence for the genetic control of PPI comes from studies in rodents. For example, Francis et al42 reported that mice generated via embryonic transplants into a different maternal strain exhibited PPI phenotypes of the embryo and not of the maternal uterine or rearing environments. Thus, PPI appears to be linked closely to genotype in mice. PPI levels differ substantially across mouse strains,43 and strain-specific PPI phenotypes are stable over multiple generations. One published study has directly assessed heritability of PPI in humans. Anokhin et al44 assessed PPI in 40 monozygotic (MZ) and 31 dizygotic (DZ) female twin pairs and used structural equation modeling to demonstrate heritability that accounted for over 50% of PPI variance. Menstrual cycle phase was not matched in these twin pairs, suggesting that heritability may have actually been higher, given the likely added variance in this sample based on the differential hormonal “state” influences on PPI (see below).45,46
Cadenhead et al16 reported PPI deficits in first-degree relatives of schizophrenia probands. Preliminary analyses17 using an “impaired” criterion of PPI (>1 SD below control means) revealed an “impaired PPI” status in 8/12 unaffected siblings of schizophrenia probands vs 5/25 controls (P < .005); analyses using impaired schizophrenia probands and their unaffected siblings revealed a relative risk of impaired PPI of 3.13. Pearson correlation between sib-pairs was .66. This finding of impaired PPI in unaffected siblings of schizophrenia patients was recently replicated by Kumari et al18 in larger samples. Preliminary analysis of the Consortium on the Genetics of Schizophrenia (COGS) data sample, for 284 individuals, found PPI heritability to be significant at 0.24 (P < .05) (T. A. Greenwood, D. L. Braff, K. S. Cadenhead, M. E. Calkins, D. J. Dobie, R. Freedman, M. F. Green, R. E. Gur, R. C. Gur, G. A. Light, J. Mintz, K. H. Nuechterlein, A. Olincy, A. D. Radant, L. J. Seidman, L. J. Siever, J. M. Silverman, W. S. Stone, N. R. Swerdlow, D. W. Tsuang, M. T. Tsuang, B. I. Turetsky, N. J. Schork unpublished data).
The initial report of PPI deficits in schizophrenia patients15 predated the use of atypical antipsychotics. Subsequent clinical reports have suggested that atypical antipsychotics are associated with greater and potentially “normalized” PPI levels in schizophrenia patients.19,31,40,41,47 Because an overwhelming number of schizophrenia patients are currently treated with atypical antipsychotics, it is possible that PPI deficits in this population are a “vanishing” biomarker. Alternative strategies for understanding the biology and clinical implications of deficient sensorimotor gating have become increasingly important, including the use of “optimized” stimulus features (eg, 60-ms prepulse intervals),48 broadband stimuli and prominent background noise,49 unmedicated schizophrenia41,50 and schizophrenia “spectrum” patients,16,51 unaffected family members, and populations of “low-gating” controls.48
“Progating” effects of nicotine have been reported in measures of both PPI and P50 ERP suppression.19,52–54 If nicotine use per se is associated with increased PPI, then the higher rates of smoking among schizophrenia patients would be expected to diminish patient vs control group differences in PPI measures and to diminish measures of heritability.
Sex differences and menstrual cyclicity in PPI have been reported by several different groups45,46,55–58 and may have important implications for the interpretation of PPI differences in schizophrenia vs control populations. Typically, “open” subject recruitment in studies favors ascertainment of male patients and female controls. Because men exhibit more PPI than women, this ascertainment bias artificially diminishes control vs schizophrenia group differences. Menstrual cyclicity of PPI adds uncontrolled variance that may differentially affect control vs patient samples, and without carefully timed measurements, this “state-sensitive” cyclicity of PPI must diminish apparent heritability. Nevertheless, PPI deficits have been reported in both male and female schizophrenia patients.37,45,49
Three lines of evidence implicate specific genes or gene regions in the regulation of PPI. First, PPI deficits in both HD23,58 and 22q11 deletion syndrome59 suggest that the genes affected in both these disorders modify brain circuitry that regulates PPI. In both cases, animal models with homologous genetic defects also exhibit PPI deficits.25,60 Second, quantitative trait loci (QTLs) have been identified either through interval mapping in inbred rat strains (QTLs on chromosomes 2 and 18)61 or recombinant congenic mouse strains (5 QTLs across chromosomes 3, 5, 7, and 16)62 or through the use of chromosome substitution strains in mice (2 QTLs on chromosome 16).63 Third, reverse genetic approaches have identified a long list of genes which, when inactivated via constitutive or conditional knock out techniques, are associated with a reduction in PPI.64
Most genes associated with lower vs higher levels of PPI may be unrelated to reduced PPI in schizophrenia or other disease states. The most potent physiological influence on acoustic PPI is hearing threshold because an organism that cannot detect a prepulse will not exhibit PPI. Thus, many or most candidate “PPI genes” identified via gene inactivation or mapping strategies in inbred and recombinant rodents will likely be associated with hearing threshold. Beyond the level of sensory registration, the most potent physiological influence on PPI is exerted at the level of the pedunculopontine nucleus (PPTg), which mediates PPI via its impact on the NRPC.65 For the same reasons noted for hearing threshold, genetic studies of PPI will likely be influenced strongly by genes coding for the normal function of the PPTg—a structure that does not play a central role in models of the pathophysiology of schizophrenia. In contrast, the PFC—which is suspected to be a critical substrate for some symptoms of schizophrenia—is likely to be 3 or 4 synapses removed from the primary startle circuit and, in a normal human or rodent, genes controlling the PFC will likely contribute only weakly to any gene mapping “signal” based on levels of PPI.
Conversely, we know that the HD gene exerts a powerful control on the viability of striatal circuitry that regulates PPI and that this gene is strongly associated with low or absent levels of PPI in humans and animal models. Yet, despite this, it is almost certain that no gene mapping efforts based on the PPI phenotype in normal rodent strains, or reverse genetic models based on constitutive or conditional knockout strategies (absent the insertion of the HD gene), will ever link the HD gene to low levels of PPI. Importantly, PPI is a measure that reflects the normal function of specific brain circuits, and genes associated with PPI deficits in neuropsychiatric disorders are likely to be ones that contribute uniquely to dysfunction in those brain circuits. Such genes will be most effectively identified through the use of affected and at-risk individuals, in whom deficient PPI is used as an endophenotype—a surrogate marker for neural circuit dysfunction that contributes to the vulnerability for the expression of the full clinical phenotype.
The P50 wave is a midlatency auditory evoked potential that exhibits reduced amplitude, or suppression, when a second click sound is presented 500 ms after an initial click. This paired-click experimental procedure is referred to as the conditioning-testing paradigm.66 Although the functional significance for this response suppression has not been clearly established, it is thought to reflect the brain's inhibitory control of its response to stimuli.11 In this model of response suppression, the first click stimulus initiates or conditions the inhibition and the second tests its strength. Thus, the conditioning response is generally maximal because no inhibitory circuits have yet been activated, and the test response is smaller because of the action of these inhibitory mechanisms. Alternative explanations of P50 suppression have been proposed, such as a protracted refractory or recovery period following the first click response. This explanation is contradicted, however, by the finding that the P50 test response actually increases when the interstimulus interval becomes very short (eg, 100 ms).67 The generally accepted measure of the inhibitory gating effect is the ratio of the test amplitude to the conditioning amplitude.
Subjects typically are supine, with their neck supported to minimize myogenic artifacts. The auditory stimuli are clicks, delivered 50 dB over the subject's auditory threshold.68 The stimuli are presented in trains of pairs, with an intrapair interval of 0.5 seconds and an interpair interval of 10 seconds. Subjects are instructed simply to remain awake, with their eyes open and fixed at a distant target; a behavioral response is not required. Electroencephalographic (EEG) activity is monitored, recorded and processed according to previously published methods.68–70 Trials containing EEG or electro-oculographic (EOG) activity greater than 30 μV within the first 80 ms following the auditory stimulus are rejected. The operationalization of this criterion replaces an earlier procedure that relied on the technician's judgment to accept or reject individual trials.
The conditioning P50 wave is identified as the most positive peak of the average auditory evoked potential, occurring 40–80 ms after the conditioning stimulus, measured relative to the preceding negative trough. To be accepted, a P50 wave must be >0.5 μV, and any corresponding EOG activity must be lower in amplitude than the P50 wave itself. The test wave is identified as the most positive peak occurring after the test stimulus, within ±10 ms of the latency of the conditioning response. The ratio of the P50 test and conditioning amplitudes defines the strength of the inhibitory response, with a larger value indicating less robust inhibition or suppression. Figure 2 illustrates the test and conditioning responses in both schizophrenia patients and healthy control subjects.
The P50 is a relatively small ERP component, and questions have been raised concerning the effect of noise on the detection and measurement of the P50 signal. To address this question, a detailed trial-by-trial statistical analysis of P50 measurement has been performed.71 This analysis demonstrated that normal subjects had average evoked P50 conditioning responses that were significantly larger than the prestimulus EEG activity. Their test responses, in contrast, were not significantly different from the prestimulus EEG. Schizophrenia subjects had conditioning waves that were both similar in amplitude to those of controls and significantly different from the prestimulus activity. However, the patients' test responses were also significantly larger than the prestimulus EEG. In the single-trial analysis, selection of the largest wave in the 40–80 ms poststimulus interval resulted in identical amplitudes for prestimulus, conditioning and test periods in both normal and schizophrenia subjects. This demonstrates that, although the signal-to-noise ratio is too low to reliably identify the P50 wave in individual trials, averaging increases the signal-to-noise ratio to a level that is adequate to reliably distinguish the P50 response from background activity.
P50 suppression is regulated by wide-ranging neural circuitry, prominently involving hippocampal structures72 as evidenced from depth recordings in humans using an electrode that penetrated the hippocampus73 and single-neuron recordings.74 The penetrating recordings showed that the hippocampus generates a response 50 ms poststimulus, and the neuronal recordings showed that the response to repeated stimulation quickly diminishes. These findings in humans have been modeled in rats to determine the underlying synaptic mechanisms. Ablation of the CA3-CA4 area of the hippocampus, in rats, eliminates the P20-N40 complex, considered the analog of the human P50.75 One major input pathway to CA3-CA4 is from the medial septal nucleus through the fimbria-fornix. Lesion of this cholinergic pathway causes CA3 and CA4 neurons to lose their gating response to sensory stimuli. Cholinergic inputs to CA3-CA4 are further implicated by the fact that antagonists of the lower affinity nicotinic receptors, such as α-bungarotoxin, block the gating response76; antagonists of high-affinity nicotinic receptors, such as mecamylamine, or muscarinic receptor antagonists have no effect on gating.
CA3-CA4 interneurons might be the final mechanism for suppression, as they are activated by this cholinergic input. These interneurons release γ-aminobutyric acid (GABA) onto the pyramidal neuron, which depresses its membrane potential so that it cannot discharge. GABA activates GABAA- and GABAB-type receptors on the pyramidal neuron, which together inhibit neuronal firing for up to 300 ms. A still longer inhibition may involve presynaptic GABAB receptors located on the perforant pathways' synaptic terminals on the apical dendrites of the CA3 neurons. Although GABA is released primarily at the cell body of the CA3 neurons, if enough GABA is released by burst of interneuron activity, it may diffuse to the apical dendrites and contact GABAB receptors on the presynaptic terminals of the perforant pathway fibers, inhibiting their release of the excitatory neurotransmitter glutamate. Then the perforant path can no longer send sensory information to the CA3 pyramidal neuron, so the response to the test stimulus is diminished. Activation of the interneurons' nicotinic receptors by cholinergic medial septal inputs could provide the additional burst activity for sufficient GABA release.
It is important to note that sensory gating has also been observed in other brain regions. Using a combination of hippocampal depth electrodes, subdural strip, and grid electrodes in epilepsy patients, evidence of sensory gating was found in the hippocampus, the temporoparietal region (Brodmann's area 22 and 2) and the PFC (Brodmann's areas 6 and 24), with the neocortical habituating responses peaking around 50 ms and hippocampal responses peaking around 250 ms poststimulus.77 This suggests that sensory gating may be a multistep process with an early temporoparietal and prefrontal phase and a later hippocampal phase. The extent to which the gating deficits observed in schizophrenia are mediated by hippocampal vs other cortical or subcortical inputs remains a matter of some debate.
That brain cholinergic systems regulate at least some of these gating deficits is supported by findings that P50 suppression abnormalities in both schizophrenia patients52 and their relatives78 resolve temporarily after administration of the cholinergic nicotinic receptor stimulant, nicotine. Further evidence for nicotinic involvement in schizophrenia is the decreased expression of the α-7 nicotinic receptor as evidenced by 124I-α-bungarotoxin binding or immunoreactivity.79–83 Noradrenergic neurotransmission may also be involved in the P50 auditory evoked response. Administration of yohimbine affects the release of norepinephrine by the blockade of alpha-2 presynaptic noradrenergic receptors that normally inhibit norepinephrine release. Yohimbine causes loss of auditory evoked potential gating in humans and rats.84,85 Although norepinephrine is probably not a major determinant of the abnormality in inhibition in schizophrenia, there is some correlation with plasma 3-Methoxy-4-Hydroxyphenylglycol levels.86
Initial studies by Adler et al,87 Freedman et al,88,89 and Siegel et al90 showed that there was a failure of P50 suppression in schizophrenia patients, consistent with theories of failed inhibitory function.91 The deficit has been associated with diminished performance on neuropsychological measures of attention.92 Some groups have failed to replicate this finding, while others have replicated it but found it to be unrelated to self-reports of sensory disturbance.93–95 A common misconception, in this regard, is that normal inhibition means that healthy subjects do not hear the second sound. The role of inhibition is not to block all sound or other stimuli from reaching the hippocampus. Rather, it blocks weaker stimuli so that the responses to strong stimuli are emphasized. In the absence of inhibition, the hippocampus becomes hyperactive and then it can no longer respond to stimuli, a phenomenon called “occlusion.” Evidence for occluded responses to novel stimuli in the hippocampus, in schizophrenia, comes from several types of neuroimaging studies.96 The inhibition of weaker stimuli is also easily overcome by making the stimuli more relevant.97
P50 suppression deficits in schizophrenia patients are persistent98 and found in both acutely ill and more stable schizophrenic outpatients.87,88,99 The P50 deficit is present in both predominantly positive symptom and negative symptom patients,100 though some studies have reported that the phenomenon is significant mainly in the disorganized/undifferentiated patients compared with the paranoid subtypes.101 P50 suppression deficits are also present in schizotypal patients.102
Unmedicated schizophrenia patients have unusually small P50 waves, and the amplitude of the waves is normalized by neuroleptic treatment.103 However, although the P50 increases in amplitude during treatment, it increases for both the conditioning and the test responses, so that the P50 ratio remains abnormally high. Individuals treated with clozapine, though, exhibit normalization of their P50 ratios coincident with improvement in their clinical symptoms.104,105 Clozapine, which releases acetylcholine in the hippocampus, may thereby indirectly act on the nicotinic receptor to normalize the P50 ratio, as people with schizophrenia also decrease the number of cigarettes they smoke while taking this medication.106 Clozapine also has the property of 5-Hydroxytryptamine 3 (5HT3) antagonism, which activates the nicotinic receptor. Acute administration of odansetron, a highly selective 5HT3 antagonist, induces a similar effect of improvement in P50 auditory gating in schizophrenia.107 Direct α-7 nicotinic receptor agonism with 3-(2,4-dimethoxybenxylidene) anabaseine also normalizes the P50 ratio in a double-blind, placebo-controlled study of 12 people with schizophrenia.108 Other atypical neuroleptics which may not have direct nicotinic agonism or 5HT3 antagonism have more variable effects on the P50 ratio. In 2 separate investigations,109,110 risperidone had only a marginal effect on P50 suppression, while olanzapine normalized it in one study110 and partially improved it in the other.109 A third study111 of 14 schizophrenia patients assigned to double-blind treatment with haloperidol or olanzapine found no group differences in P50 ratio. Thus, while risperidone does not appear to highly influence gating, the data on olanzapine remain inconsistent.
Problems in test-retest reliability have been noted by several investigators.112 Although conditioning and test amplitudes were reliably measured, the test-to-conditioning ratio was not. All cerebral evoked potential measurements include both a signal and a noise component. A ratio measurement that includes noise in both the numerator and the denominator will approach an asymptote of 1 as the noise increases.113 This problem is exacerbated by the relatively low signal-to-noise ratio of the P50 compared with other auditory evoked potential components. Though a reliable test-to-conditioning ratio is difficult to obtain due to its inherent mathematical properties, ie, test and conditioning are not independent, so that the shared variance between test and conditioning cannot be completely eliminated,112,113 the test-to-conditioning ratio nevertheless has the greatest power to distinguish normal subjects from schizophrenic subjects.113 Furthermore, the schizophrenia P50 suppression deficit has been reliably reproduced in large numbers of subjects across multiple sites.99,114–119
Waldo et al120 measured P50 suppression in 13 normal subjects, 10 days apart. P50 suppression was 66.5% on day 1 and 69.5% on day 10. There was no significant change in the gating of auditory test responses over this period. The change between days for individual subjects was generally within the 8% mean variability observed with repeated recordings in other groups of normal subjects. Griffith et al68 recorded 10 schizophrenia patients and 10 healthy subjects on 3 occasions. The intraclass correlation was 0.73. Additionally, Hall et al121 measured P50 suppression in 19 MZ twin pairs on 2 separate occasions and found a P50 ratio intraclass correlation of 0.66.
Young et al122 examined P50 suppression in 15 normal MZ twin pairs and 12 normal DZ twin pairs. The upper limit of the heritability estimate (calculated as twice the difference between the MZ and DZ intraclass correlations) was 1.0. The 95% confidence limit for the lower limit of heritability was 0.44. Hall et al121 examined P50 suppression in 40 healthy MZ twin pairs and 30 DZ twin pairs and reported a heritability estimate for the P50 test-to-conditioning ratio of 68%. Interestingly, preliminary heritability analysis of the COGS data sample (n = 201) revealed a nonsignificant heritability estimate of 0.07 for the test-to-conditioning ratio. However, the difference in amplitude between the P50 responses to the first and second clicks, which is an alternate way of measuring suppression, had quite high heritability (h2 = 0.53, P = .005) (T. A. Greenwood, D. L. Braff, K. S. Cadenhead, M. E. Calkins, D. J. Dobie, R. Freedman, M. F. Green, R. E. Gur, G. A. Light, J. Mintz, K. H. Nuechterlein, A. Olincy, A. D. Radant, L. J. Seidman, L. J. Siever, J. M. Silverman, W. S. Stone, N. R. Swerdlow, D. W. Tsuang, M. T. Tsuang, B. I. Turetsky, N. J. Schork unpublished data).
The P50 auditory evoked potential endophenotype has been linked with a genetic marker at the locus of the CHRNA7, the gene coding for the α-7 subunit of the nicotinic receptor.123 Furthermore, the presence of a single nucleotide polymorphism in the 15q14 gene CHRNA7 5′ core promoter is significantly associated with P50 suppression deficits (P = .008).110 Association of CHRNA7 polymorphisms with P50 gating has been replicated, but the specific allelic associations differ, which suggests that responsible mutations have not yet been unambiguously identified.124–126
The AS task, first described in 1978127 utilizes the intrinsically precise and quantifiable nature of oculomotor performance to assess a specific aspect of oculomotor/cognitive function.
To perform the AS task, subjects are seated in a dark room with their heads stabilized, and their eye position is ascertained with high precision, typically using infrared oculography or EOG, while they watch a specialized sequence of dot movements. A single AS trial commences with fixation on a central point, followed by an unpredictably located stimulus to the left or right. Rather than looking at this stimulus, the subject is asked to look at the mirror image location on the opposite side of the screen. Thus, the participant must inhibit an unwanted, reflexive saccade to the stimulus. A cue then appears signaling the location of a correct response and then the target returns to central fixation for the next trial.
Duration of each task component, the number of potential different AS target locations, and maximal and minimum distances from center, vary across studies. There may be a gap, simultaneous offset and onset, or overlap between central fixation and the AS cue stimulus, and this factor appears to affect the magnitude of schizophrenia-control differences.128 Direction of the first major saccade determines whether the subject has made an incorrect AS (sometimes referred to as a “prosaccade” or reflexive error) vs a correct AS. Figure 3 illustrates the infrared tracings associated with correct and incorrect AS performance. The summary variable is typically the proportion of incorrect (error) saccades, or sometimes proportion of correct AS, over all trials.
Correct performance of the AS task requires accurate perception, ability to transform location information to a mirror image representation, and suppression of a reflexive visually guided saccade to the AS stimulus as well as nonspecific task demands such as motivation, ability to comprehend the task, and willingness to hold one's head still.
While the oculomotor system is composed of many cortical and subcortical areas, certain areas appear crucial for accomplishing the unique demands of the AS task. Sensory transformation of location information appears to occur in the lateral interparietal area,129,130 although PFC may also be involved.131 Damage to dorsolateral PFC causes decreased frequency of correct AS132 and specific neurons in PFC activate specifically during AS.133–136 Amemori et al137 have argued that this phenomena reflects working memory. Finally, single-unit recordings in monkeys demonstrate that supplementary eye field neurons are active during tasks that require generation of saccades based on internal representations of location including the AS task.138 Thus, parietal, prefrontal, and supplementary oculomotor areas appear uniquely relevant to AS performance.
Neuroimaging and performance data indicate that basic sensorimotor processes involved in visually guided saccade generation appear essentially normal in schizophrenia.139 Because schizophrenia patients generally make at least 50% of correct AS responses, and often will generate corrective saccades despite an initially incorrect response, they clearly understand the task itself and appear to have sufficient motivation. Raemaekers et al,140 using functional magnetic resonance imaging (fMRI) found that while schizophrenia patients activated frontal and parietal areas normally during an AS task, they failed to activate the striatum as much as controls and concluded that dysfunction of a frontal striatal saccade suppression network explains the poor performance of schizophrenia patients. Patients with lesions of dorsolateral PFC and/or abnormalities of the caudate exhibit increased AS error rates, while patients with cortical lesions of the frontal eye fields, supplementary eye fields, posterior parietal cortex, and temporal cortex do not.141 This suggests that the increased error rate observed in schizophrenia patients, described below, is consistent with dorsolateral prefrontal cortical dysfunction.142,143 Imaging144,145 and ERP146 studies in schizophrenia patients and healthy individuals have tended to support this conclusion, though not ubiquitously.147
In the late 1980′s, Fukushima et al148 reported that schizophrenia patients evince a greater number of inappropriate reflexive saccades to the target in an AS task than nonpsychiatric controls. Since then, more than 50 studies have consistently reported this effect.149–159 Notably, there have been no investigations, to our knowledge, failing to find that schizophrenia patients generate more errors than controls. Moreover, particular AS task manipulations have been reported to enhance the magnitude of effect.128,160–163
Inconsistencies in the literature have been noted regarding the diagnostic specificity of this AS dysfunction to schizophrenia patients,143,164 with some reports of increased AS error rates in members of other psychiatric disorders, such as major depression165 and bipolar disorder.166 Conversely, evidence has been cited suggesting that increased AS error rates do not characterize performance of mood or anxiety disordered patients.167 Although the question is unresolved, increased error rates in mood disorder patients would not be deleterious to the endophenotype status of AS performance; eye movement dysfunction in some mood disorder patients may reflect shared genetic susceptibility influencing mechanisms that contribute to both clinical state and poor AS performance.168
While error rate has been most extensively investigated in schizophrenia, other parameters, including latency, gain, and accuracy, have also been examined. Anomalies in response latency may provide evidence of visual processing inefficiencies. Schizophrenia patients have been reported to exhibit longer latencies on correct AS responses than control subjects,128,160,169 potentially reflecting compensatory slowing due to difficulties inhibiting unwanted reflexive saccades.160 Although less commonly measured, latencies to error responses suggest that they do not differentiate schizophrenia patients from nonpsychiatric controls.160,161 Latency to correct responses thus appears to tap an aspect of performance that is distinct from latency to incorrect responses, highlighting the importance of differentiating the 2, a distinction that is not routinely made in current investigations.
Several studies have reported reduced spatial accuracy of AS in schizophrenia patients.128,150,158,169,170 This abnormality may implicate parietal and prefrontal cortical control of sensorimotor coordinate transformations150 or impairment in generating saccades from internal representations in the supplementary motor area.
A number of studies have shown good test-retest reliability of AS error rate over periods ranging from several months to many years among patients with schizotypy,171 schizophrenia patients,172,173 first-degree relatives,173 and controls.174 Our recent analysis of data from the COGS sample of schizophrenia patients (n = 103) and community comparison subjects (n = 138) suggests high within-session reliability of the AS paradigm at 7 sites (range = 0.77–0.96).175 Moreover, there were no significant cross-site differences in performance, suggesting that high-quality AS data can be obtained across multiple sites using standardized measures, equipment, and training procedures.175
Recent studies of the test-retest stability of AS error rates have suggested high temporal stability in schizophrenia patients (r = .87, test-retest interval = 2.78 years),172 and in a mixed group of schizophrenia patients and their relatives (r = .73, test-retest interval = 1.82 years).173 The results are consistent with 2 other reports examining performance in groups of psychiatric patients (r = .75, test-retest interval = 1 year,176 r = .90, test-retest interval = 1 week177). It has been reported that first-episode,158,178–180 remitted,142 and unmedicated181 schizophrenia patients all manifest AS deficits, further suggesting that the deficits are not merely a reflection of clinical state or chronicity of illness. The majority of studies examining the relationship between medication and AS error rates have described the results as nonsignificant.140,148,161,180,182–187 However, there is some evidence that AS performance in schizophrenia may actually improve with nicotine administration155,188 and with some medications, including risperidone189 and cyproheptadine treatment.190 Nonetheless, in general, the AS deficit observed in schizophrenia is temporally stable and does not appear to be attributable to clinical variables such as medication exposure, current symptomatology, and chronic illness. Thus, the available evidence is consistent with the trait stability of the AS deficit in schizophrenia patients.
Malone and Iacono,191 using a large sample of identical and fraternal healthy twin girls, found a high heritability of 0.57 for AS performance; no other published studies have examined heritability. However, preliminary analysis of the COGS data (n = 340) found a very similar heritability estimate of 0.49 (P < .0001) (T. A. Greenwood, D. L. Braff, K. S. Cadenhead, M. E. Calkins, D. J. Dobie, R. Freedman, M. F. Green, R. E. Gur, G. A. Light, J. Mintz, K. H. Nuechterlein, A. Olincy, A. D. Radant, L. J. Seidman, L. J. Siever, J. M. Silverman, W. S. Stone, N. R. Swerdlow, D. W. Tsuang, M. T. Tsuang, B. I. Turetsky, N. J. Schork unpublished data). Numerous studies have investigated the familiality of poor performance in schizophrenia by evaluating performance of first-degree biological relatives of schizophrenia patients. The presence of increased AS error rates in relatives has been described as inconsistently demonstrated.151,192,193 Two recent meta-analyses quantitatively evaluated the magnitude of the relative-control difference. Levy et al193 reviewed selected studies (k = 9) using the “standard” (nonoverlap and nongap) version of the AS task and obtained a moderate mean magnitude of effect between relatives and controls (mean Cohen d = 0.43). Calkins et al159 included 17 independent groups of relatives and all AS studies regardless of AS paradigm. Meta-analysis yielded an increased AS error rate in relatives vs controls at a moderate to large magnitude of effect (mean Cohen d = 0.61). The results of both meta-analyses suggest that, on average, relatives of schizophrenia patients produce a greater number of AS errors than controls.
In their review, Levy et al193 conducted moderator analyses and concluded that schizophrenia relatives appear impaired in the AS task because studies use more stringent inclusion criteria for controls than relatives, in effect leading to the spurious appearance of a deficit in relatives. However, because there are so few studies in this realm, Calkins et al159 conducted a reanalysis of primary data142 in which they varied inclusion and exclusion criteria and found impairment even in medically and psychiatrically healthy relatives who were screened comparably to controls. More recently, Ettinger et al150 compared psychiatrically healthy controls with comparably screened siblings of schizophrenia patients and obtained an effect size of 0.49. Thus, the impairment observed in relatives does not appear attributable to inclusionary criteria practices, at least in these 2 investigations. Instead, the differences across studies may lie not in the controls or their selection criteria, but in the relatives, perhaps vis-à-vis proband or relative inclusion criteria.159 Nonetheless, this methodological issue underscores the importance of carefully addressing and analyzing the potential impact of comorbid psychiatric and medical conditions on AS performance in both relatives and comparison subjects.
A small number of studies suggest that, like schizophrenia patients, relatives tend to demonstrate longer latencies to correct trials.161 Two studies examining spatial accuracy of AS in relatives reported reduced AS gain in healthy siblings150 and in unaffected MZ cotwins of patients with schizophrenia,149 compared with healthy controls. These results are suggestive that, in addition to error rates, AS gain and latency are worthy of further investigation as candidate endophenotypes.
The only genetic study to include the AS task reported linkage at D22S315 on chromosome 22q11–12 in 8 schizophrenia multiplex families when relatives were identified by either an AS deficit or a P50 sensory gating deficit.194 The composite endophenotype identified substantially more relatives as affected than did either of the endophenotypes alone, likely enhancing the power to detect linkage. While the linkage finding has yet to be replicated, the results are particularly notable because the linked region is the site of the catechol-O-methyltransferase (COMT) gene, variations of which have been linked to performance on candidate endophenotypes believed to reflect PFC abilities in schizophrenia patients and their relatives.195 Given the apparent involvement of the dorsolateral prefrontal cortex in the accurate performance of the AS task, the association between AS performance (and/or P50) and chromosome 22q could be partially explicable by COMT effects. However, given the small size and lack of replication, this result must be regarded as preliminary.
MMN is an auditory ERP component that is elicited when a sequence of repetitive standard sounds is interrupted infrequently by deviant, “oddball” stimuli (eg, infrequent stimuli that differ in duration or pitch from the more frequently presented stimuli). The MMN occurs rapidly: following deviant stimuli, the response onset can be as early as 50 ms and peaks after an additional 100–150 ms. Physiologically, MMN is the first measurable brain response component that differentiates between frequent and deviant auditory stimuli and reflects the properties of an automatic, memory-based comparison process.196
MMN has many advantages for psychiatric and cognitive neuroscience studies, including the exploration of the neural substrates of schizophrenia and its treatment.197–199 First, MMN can be rapidly assessed and is highly stable in normal subjects.200–202 In a longitudinal study of schizophrenia patients retested after 1 year, Light and Braff203 found extremely high MMN reliability coefficients (intra-class correlations ~0.90). Second, the mismatch response appears to reflect a predominantly automatic process: it is not under subject control, requires no overt behavioral response from subjects, and can be elicited while subjects perform other mental activities in parallel without apparent interaction or interference.204 In this context, well-defined MMN waveforms can be obtained from sleeping infants,205–207 adults,208,209 patients with extremely severe brain injuries, and even comatose patients.210–212 Because MMN occurs even in the absence of conscious and effortful attention, it appears to index a preattentive form of sensorimemory.204 While later ERP components occurring 300–500 ms after stimulus presentation (eg, P3b) are also sensitive to changes in stimulus characteristics and sequencing, they are only elicited in response to attended stimuli and are therefore associated with attention-dependent and active cognitive processes. Attention-dependent cognitive functions assessed by traditional neuropsychological tests or long-latency ERP methods (eg, P3b) can be markedly influenced by motivational factors, level of task engagement, performance incentives, self-monitoring, and emotional factors.213–218 In contrast, preattentional cognitive measures such as MMN offer promise for accurately characterizing the integrity of sensory network dysfunction free of attentional or motivational artifacts in studies of neuropsychiatric patient populations.199,219
In the prototypical MMN paradigm, an unchanging standard tone is presented repeatedly with a brief interstimulus interval (eg, 500 ms). The relatively rapid stimulus presentation ensures that the echoic memory trace of the preceding stimulus is still active when the subsequent stimulus is presented. The repeating standard tone is replaced infrequently (eg, 10% of trials) by a deviant tone that differs in one physical attribute. This is typically a change in either the pitch or the duration of the tone, though stimulus intensity has also been used to define deviance. Subjects are not instructed to attend to or respond to the tones in any way. In most cases, attention is specifically directed away from the tones through the use of a visual distracter, such as a video, or an active visual attention task. This ensures that the automatic detection of deviance at the level of echoic memory is not obscured by responses related to controlled or directed attention. Figure 4 illustrates the auditory ERP responses to the standard and pitch deviant stimuli, in patients and healthy control subjects. Measurement of MMN is carried out on the difference waveform constructed by subtracting the auditory evoked potential response to the standard tone from that of the deviant. MMN appears as a prominent negative potential on this difference waveform. For a pitch deviant, its peak latency is typically between 100 and 200 ms poststimulus, but this is more variable for a duration deviant. The amplitude of MMN is a function of the magnitude of the physical difference between the standard and deviant stimuli.
In the auditory domain, maximal mismatch responses are evident at frontocentral scalp recording sites with phase reversal at posterior scalp electrodes (eg, mastoids).204 Magnetoenchephalography, high-density EEG, functional imaging, and studies of patients with discrete brain lesions indicate that the auditory MMN is generated within the primary and secondary auditory cortices with possible contributions from bilateral dorsolateral prefrontal cortices.220–230 In addition, MMN is often utilized to probe frontotemporal brain systems across a range of developmental and neuropsychiatric disorders.199,211,231–235
Previous studies have demonstrated that NMDA may play a crucial role in both the generation of MMN and the MMN deficits observed in schizophrenia (see below). NMDA receptor antagonists selectively diminish MMN generation in awake monkeys,236 and subanesthetic doses of ketamine, an NMDA antagonist, selectively decrease MMN in healthy human volunteers without affecting other ERP activity.237 Umbricht et al238 also found that lower baseline MMN was significantly associated with psychotic-like behavioral effects and experiences induced by subsequent ketamine administration. Thus, MMN may serve as a neurophysiological “assay” of NMDA receptor functioning in models of schizophrenia.
Deficits in MMN represent a remarkably robust finding in schizophrenia research. Shelley et al239 first identified MMN deficits in schizophrenia patients using deviant stimuli that differed in duration (ie, duration MMN) from standard stimuli. Since that time, there have been several published reports of reduced MMN in schizophrenia patients utilizing various stimulation parameters (eg, pitch, duration, and intensity stimulus manipulations) and conditions.198,240 In a recent meta-analysis performed by Umbricht and Krljes,240 the mean effect size for the schizophrenia deficit was ~1.0—a large deficit according to common standards. Also, though it was not a statistically significant difference, the effect size was approximately 40% larger for studies that used a duration deviant, compared with studies that used a frequency deviant. While the meaning of this remains unclear, it likely implicates task-specific neural mechanisms that underlie the schizophrenia deficit and is therefore a very intriguing, if still only suggestive, difference. Importantly, in contrast to most other physiological indices, MMN deficits appear to be relatively specific to schizophrenia. Bipolar, major depressive241,242 and obsessive-compulsive disorder patients243–245 all have normal MMNs, though there are reports of MMN deficits among chronic alcoholics.246
There is substantial evidence to indicate that MMN has good test-retest reliability.202 In a study of 15 healthy individuals tested on 2 separate occasions 1–27 days apart, Tervaniemi et al247 examined the reliability of MMN for deviant stimuli that varied in duration, pitch, or intensity. Reliability was greatest for the duration deviant (r = .78) and lowest for the pitch deviant (r = .53). Kathmann et al200 reported very similar estimates from a study of 45 subjects tested 2–4 weeks apart. Test-retest reliability, in this case, was >0.8 for a duration deviant and ~0.5 for a pitch deviant. Escera et al248 observed reliabilities of 0.72 and 0.80 for a duration-deviant MMN, depending upon whether the peak or the mean within a defined time interval was used to measure MMN amplitude. Kujala et al201 noted test-retest correlations of 0.60–0.75, depending on the degree of deviance of the infrequent stimulus. Only one small study (n = 14)249 reported correlation coefficients that were described as unacceptable. Most importantly, the one study reporting the results of repeat testing in schizophrenia patients203 found intraclass correlations of ~0.90 after 1 year.
In schizophrenia patients, MMN deficits do not appear to be ameliorated by first-generation antipsychotic medications,250 risperidone,251 olanzapine,252 or clozapine.229,250 Similarly, clinical changes from acute to post-acute phases of illness do not correspond to a “normalization” of MMN deficits in chronic patients.253 In chronic schizophrenia patients, MMN deficits are highly associated with impairments in real-world functioning and level of independence in community living situation.203,254,255
There are no studies that have attempted to formally estimate the heritability of MMN. We know of no studies that have examined MZ vs DZ twin-pair correlations or considered the differences between intrafamilial and interfamilial associations. However, there are reports of specific genetic associations (see below), which indicate a degree of genetic modulation of MMN. Also, animal models of the MMN indicate NMDA-mediated differences in auditory deviance processing among genetically distinct inbred mouse strains.256 It is reasonable to expect, therefore, that the normal heritability of MMN will ultimately prove to be comparable to that of the other physiological measures considered here.
Whether or not this is true for the heritability of the schizophrenia abnormality, though, is not clear. Clinically unaffected family members of schizophrenia patients,257,258 children at risk for developing schizophrenia,259,260 and recent-onset patients261,262 have all been reported to have reduced MMN amplitudes. MMN would appear, therefore, to be a specific schizophrenia-related endophenotype8,258 for studying the complex genetics of the disorder.12,13 However, there have also been reports of normal MMNs in unaffected family members.263 Moreover, in contrast to the virtually universal finding of abnormal MMNs among chronic schizophrenia patients, MMNs have been reported to be normal in first-episode patients.262,264 Longitudinal follow-up suggests that the MMN deficit may emerge over time in concert with the progressive temporal lobe volume loss that occurs over the early course of the illness (D. F. Salisbury, personal communication). The degree of MMN impairment may also be modulated by the level of premorbid educational achievement in first-episode patients.262 Additional studies are therefore needed to clearly delineate the nature of the MMN abnormality, its prevalence among putatively prodromal or first-episode schizophrenia subjects, and its utility for predicting conversion to psychosis in individuals at genetically high risk for developing the disorder.265
Deletions of chromosome 22q result in complex congenital syndromes that frequently include schizophrenia-like psychoses. Two studies of adolescents with 22q deletions have now demonstrated that this is associated with diminished MMN.266,267 The 22q deletion includes the COMT gene, which codes for the enzyme that deactivates DA, leading to regionally specific hyperdopaminergia. Reduced MMN was associated with the presence of the MET allele of the VAL108/158MET polymorphism of the COMT gene in the remaining copy on the unaffected chromosome.267 This suggests that genetic modulation of dopaminergic activity can affect MMN, a finding that is of obvious relevance to schizophrenia. Importantly, the limited evidence of association between COMT and schizophrenia suggests that the VAL, rather than the MET, allele may confer an increased risk of disease.195,268 However, it is difficult to make specific inferences concerning the physiological effects of the VAL/MET polymorphism in schizophrenia because this is contingent upon both the background level of dopaminergic activity and the presence or absence of other modifying genes.269 The only other genetic association study reported no relationship between ApoE allelic variation and MMN among older individuals with mild cognitive impairment.270
The P300 event-related brain potential is an index of endogenous cognitive processes, typically elicited by infrequent sensory stimuli that are either task relevant or novel.271 It receives its name from its appearance as a large vertex-positive component with peak latency approximately 300 ms after stimulus presentation. Occurring after the obligate evoked potential response to the physical attributes of a stimulus, it reflects a variety of cognitive processes elicited by a change in the sensory environment. These include directed attention, the contextual updating of working memory, and the attribution of salience to a deviant stimulus. Since its discovery in 1965, it has been widely investigated, with a multitude of studies examining the clinical and psychological correlates of P300 amplitude and latency, in both healthy and clinical populations.272
Although it can be elicited by stimuli in any sensory modality, it is the auditory P300 that has been most widely studied in schizophrenia patients. The prototypical experimental paradigm has been the oddball task in which an infrequent tone, designated as the “target,” is randomly interspersed within an ongoing train of a different repeating tone, designated as the standard. Subjects are instructed to indicate their perception of each target by making a button press or other response. The task is designed to be quite simple, with a relatively slow stimulus presentation rate (1- to 2-second interstimulus interval) and a large pitch difference (eg, 1000 Hz standard and 2000 Hz target). Schizophrenia patients typically achieve >90% correct target identifications. Figure 5 illustrates the auditory ERP responses to target and standard tones, in patients and healthy control subjects. A variant of the oddball task, known as the “3-tone” or “novelty” P300 paradigm, adds an additional infrequent stimulus which is not the designated target; instead, it is distinctive, variable, and somewhat intrusive in its physical attributes but is intended to be ignored by the subject.
It is now clearly established that the P300 is not a unitary phenomenon. Rather, it is a composite representation of the activity of temporally overlapping but anatomically and functionally distinct neural generators. Experimental task manipulations have elucidated at least 2 functionally discrete subcomponents. The P3a subcomponent, which is elicited by stimuli that are novel or unexpected (as in the 3-tone or novelty paradigm), occurs slightly earlier, has a frontocentral topographic scalp distribution and appears to reflect attentional orienting processes.273 P3b, in contrast, is elicited by stimuli that are task relevant and contextually salient (as in the oddball paradigm). It occurs later, has a parietal scalp maximum, and reflects cognitive processes associated with stimulus evaluation and response formation. Intracranial electrophysiological monitoring and fMRI studies have similarly discerned multiple sources of P300-like ERP activity, including the hippocampus,274–277 thalamus,277,278 inferior parietal lobe,275,277 superior temporal gyrus, and frontal lobe.275,277,279 It is unlikely, though, that deeper sources such as the hippocampus or thalamus contribute substantially to the P300 as measured on the scalp. Convergent evidence from ERP source localization and fMRI activation studies, as well as recordings from patients with focal neurological lesions, suggests that P3b scalp activity arises primarily from the inferior parietal cortex, particularly the supramarginal gyrus, while the P3a reflects the activity of lateral prefrontal and superior temporal areas.280 However, specific neural circuits or neurotransmitters underlying the P300 response are not clearly defined.
Reduced amplitude of the auditory oddball P300 response is perhaps the most robust physiological abnormality observed in schizophrenia patients, having been replicated repeatedly with virtually uniform consistency.272,281–283 While prolonged P300 latency has also been reported,284 this appears to be a much more equivocal and less reliable finding. This may be due, in part, to differences in patient populations across studies. A recent meta-analysis of 104 studies reported a significant effect size of 0.59 for auditory P300 latency (compared with an effect size of 0.89 for amplitude).285 However, this analysis also reported that the latency effect size diminished with illness duration, making it a less reliable finding among studies of chronic patients. Investigations have now begun to move beyond the simple documentation of a cohort deficit to elucidate the clinical, familial, and neuroanatomical correlates of the amplitude decrement. There is now considerable evidence that this represents a trait abnormality that is independent of medication status, duration of illness, or symptom severity.286–289 Although state-dependent relationships have been reported between P300 amplitude and measures of negative symptomatology,290 positive symptomatology,291 treatment status,292 and stage of illness,293 it is clear from both longitudinal and meta-analytic studies that P300 amplitude does not normalize in patients, even when treated with newer atypical antipsychotic agents.285,286,289,292–296 Consistent with this idea of a trait abnormality, McCarley and associates have reported that the greatest P300 amplitude separation between schizophrenics and normals is observed at left temporal electrode sites291,297 and that this focal decrement is correlated with a decreased left superior temporal gyrus volume.298 Although these studies have focused primarily on the P3b subcomponent, there is some evidence to indicate that P3a is also reduced in schizophrenia patients299,300 and that this deficit is associated with decreased gray matter volume in the frontal lobe.301
It is important to note that, despite the highly reproducible and persistent nature of the schizophrenia P300 abnormality, the deficit is not specific to this disorder. Consistent with its multifactorial role in information processing and its distributed neural substrate, the P300 response is disrupted in a variety of neuropsychiatric disorders that included disturbed cognition. Reduced P300 amplitude has been observed in, among others, Alzheimer's disease,302 alcoholism,303 bipolar illness,304 and unipolar depression.305 It is notable, though, that in Alzheimer's disease the decreased amplitude is also associated with marked latency prolongation, in bipolar illness it is associated with a different scalp topography, in depression it is a state-dependent abnormality evident only during acute depressive episodes, and in alcoholism it is more abnormal in the visual than the auditory domain. These variations suggest that different neural mechanisms may underlie the deficit in different disorders.
Despite its multidimensional character, scalp measurements of P300 amplitude exhibit very good test-retest reliability.296,306–311 Studies of healthy control subjects with test-retest intervals of up to 2 weeks have documented reliabilities ranging from 0.81 to 0.91.296,306,307 One study of 32 subjects retested after 4.5 months reported test-retest correlations of 0.79–0.81.309 Two studies with even longer intertest intervals (14 and 24 months) had test-retest reliabilities of 0.59 and 0.61, respectively.310,311 In a study of both schizophrenia patients and controls tested twice over the course of 1–3 years, Turetsky et al289 reported reliabilities of 0.86 for controls and 0.61 for patients.
As noted above, many studies have reported persistent longitudinal deficits in schizophrenia, despite active treatment and marked symptom reduction. Comparisons across sites, however, are hampered by a lack of uniformity in both the experimental protocol parameters and the methods used to measure P300 amplitude. Different laboratories focus on different electrode sites. Some measure peak amplitude at a single time point, while others use the area under the curve within a specified time interval. Still others use sophisticated decomposition strategies to estimate individual subcomponent amplitudes. As a common metric, we examined the patient/control amplitude ratio at Pz. For a large sample from our laboratory at the University of Pennsylvania, this value was 0.64. Comparable values extracted from 5 published studies were 0.59,312 0.61,304 0.69,313 0.77,294 and 0.80.314 This indicates highly consistent findings across independent laboratories.
There is strong evidence of a genetic contribution to P300 amplitude among healthy individuals. The most convincing data are from a study by O'Connor et al.315 In a sample of 59 MZ and 39 same-sex DZ twin pairs, heritability was estimated to be 0.60. Consistent with this, Polich and Burns316 reported a correlation of 0.64 between MZ twin pairs, compared with –0.20 for unrelated matched pairs. A recent large-scale study of adolescent twin pairs reported that additive genetic factors accounted for 48%–61% of the variance in P3 amplitude.317 A second similar study found that familial factors accounted for 30%–81% of the variance among adolescent twin pairs. In this case, the familial covariance could be attributed primarily to genetic factors among the males but to shared environmental factors among the females.318 In another study of 10 healthy families, each consisting of 2 parents and 2 children, Eischen and Polich319 reported Fisher z-transformed correlation coefficients of ~ +0.40 for within-family associations, but ~ –0.02 associations between unrelated individuals. Only one smaller study320 failed to find evidence of heritability; MZ and DZ sib-pairs had intrapair correlations, respectively, of 0.50 and 0.35, but this was a nonsignificant difference.
There is also substantial evidence that the schizophrenia trait abnormality is, at least in part, genetically mediated.260,321–325 Studies of individuals who share a portion of the genetic diathesis for schizophrenia, by virtue of being either the full siblings or offspring of schizophrenia patient probands, have P300 amplitude decrements that are similar to, though less severe than, those of their ill relatives.260,287,321–324,326 Frangou et al,322 in the Maudsley family study, reported standardized z scores of –1.03 in 57 family members compared with healthy controls. In a study notable for its methodology of examining MZ twin pairs both concordant and discordant for schizophrenia, Weisbrod et al327 observed decreased amplitudes in both affected and unaffected cotwins of the patient probands, compared with healthy twin pairs. Only a handful of studies have considered topographic differences that reflect the relative contributions of P3a and P3b subcomponents to this familial deficit. Kimble et al325 observed z score measurements of –0.48 at Pz and –0.68 at Fz in 15 first-degree relatives. They argued that the greater frontal deficit reflected heritable impairments specifically in the attentional processes underlying P300. Another study that deconstructed the scalp P300 into its discrete P3a and P3b subcomponents326 also found that the familial deficit was evident for the frontal P3a subcomponent (z = –0.71) but not for the parietal P3b. These 2 findings are consistent with behavioral findings from high-risk family studies, which also suggest that abnormalities in attention are indicators of biological susceptibility.328 A more recent study, however, reported that both schizophrenia patients and their unaffected siblings had increased frontal P300 amplitudes, along with the expected decrease in the parietal P300 response.329 So, although the evidence for a P300 abnormality in the unaffected first-degree relatives of patients is quite strong, the relative contributions of P3a and P3b remain unclear.
Although evidence supporting the viability of P300 as a physiological endophenotype is strong, there is only limited knowledge of specific genetic contributions to either the generation or the disruption of the ERP response. Most of what is known is derived from the Collaborative Study on the Genetics of Alcoholism studies of the visual P300 in the context of alcohol risk and may not, therefore, be applicable to schizophrenia. Nevertheless, these studies demonstrated relatively strong linkages (LOD Score > 2.3) between P300 amplitude decrements and areas of chromosomes 2, 5, 6, and 17.330,331 Of these, only the region of chromosome 6, which contains the dysbindin candidate gene, has also been implicated in schizophrenia.332 A specific association has also been reported, among children at risk for alcoholism, between reduced auditory P300 amplitude and the A1 allele of the DRD2 DA receptor on chromosome 11.333 The relationship between genetic determinants of dopaminergic function and P300 is further supported by the association between P300 amplitude and the Ser9Gly polymorphism of the DRD3 DA receptor in healthy subjects.334 These findings are not specifically linked to either schizophrenia or the schizophrenia P300 deficit. Nevertheless, the obvious importance of DA to both the symptomatology and treatment of schizophrenia makes them intriguing.
The only evidence of a specific genetic association to reduced P300 amplitude in schizophrenia comes from the study of a large family pedigree with a balanced translocation of the long arm of chromosome 1 and the short arm of chromosome 11.335 This translocation disrupts the DISC1 gene at the chromosome 1 breakpoint and is strongly linked to schizophrenia (LOD Score = 3.6). Among the members of this family, those with the translocation exhibited reduced P300 amplitudes compared with both familial noncarriers and unrelated control subjects. This association was observed even among carriers of the translocation who exhibited no psychiatric symptoms, strongly implicating P300 amplitude as an endophenotypic marker of the DISC1 genetic vulnerability.
The ideal neurophysiological endophenotype is one that exhibits a robust and stable deficit in both patients and unaffected family members and shows strong evidence of both heritability and cosegregation with illness within pedigrees. It also should be easily and rapidly measured with minimal subject demands, demonstrate excellent test-retest and across-site reliability, and, preferably, reflect a discrete neurobiological mechanism that is both informative for the pathophysiology of the disorder and regulated by a limited number of genes. Each of these 5 candidate endophenotypes has been shown to be abnormal in schizophrenia patients. For 4 of the 5, there is also strong evidence that the abnormality is heritable and present in unaffected family members of schizophrenia probands. The one exception to this is MMN, which is not unequivocally impaired in either newly diagnosed patients or unaffected first-degree relatives. It should be noted, though, that there is very little evidence that any of these measures actually cosegregate with the illness within individual pedigrees. This is a reflection of the lack of such comprehensive family studies, rather than an indication of negative findings.
Of the 4 remaining measures, 2—PPI and P50—have the distinct advantage of being preattentive indices of relatively discrete neural mechanisms that can be assessed without any observable patient response. They, therefore, require much less motivation, cooperation, or comprehension from the subject. However, they are both influenced by state-dependent factors that add additional nongenetic noise to their measurement. The most notable of these is their tendency toward normalization by atypical antipsychotic medications, which could confound a quantitative trait linkage analysis. The 2 remaining candidate endophenotypes—AS and P300—both appear to be highly stable trait measures that are reliably assessed and impervious to the effects of treatment. However, these require a level of subject cooperation that presents a challenge for assessment in the most severely ill patients. Also, these are more complex behavioral tasks that rely upon a more distributed neural network and, therefore, a presumably more complex genetic architecture. It is ironic that, of all these measures, MMN perhaps best fulfills the combined criteria of simplicity, reliability, and state independence. It is therefore extremely important that the status of this deficit in unaffected family members, high-risk individuals, and newly diagnosed patients be clarified using a standardized methodology.
As indicated in the introduction to this review, there is growing interest in the concept of multivariate endophenotypes. A composite endophenotype comprised of multiple measures may exhibit greater experimental stability and test-retest reliability than a single endophenotype. It may thus be a more robust marker of genetic vulnerability than any one measure. Also, to the extent that one measure can act as a surrogate for another, then the relative merits of each might allow them to be selectively applied under different circumstances. For example, one measure of inhibitory failure (eg, PPI) might be assessed in an unmotivated patient taking a typical antipsychotic, while another (eg, AS) might be measured in a cooperative patient taking an atypical agent. However, the extent to which these putative endophenotypes overlap with each other and denote the same genetic vulnerability is an issue that remains relatively unaddressed. Only rarely have multiple measures been acquired in the same patient samples. One study that assessed 4 of these endophenotypes (PPI omitted) in schizophrenia patients, relatives, and controls replicated the deficits but found no meaningful correlations across measures, with the one exception of a robust association between P50 and AS.8 Another comparably structured study, though, failed to find any association between these 2 indices.153 There are data demonstrating a similar lack of association, in schizophrenia, between PPI and AS,152 as well as both animal336 and human337,338 data suggesting a dissociation between PPI and P50. It is telling that a recent study of P50, MMN, and P300 in healthy MZ and DZ twins9 found, once again, that each of these 3 measures was highly heritable but that they shared virtually no genetic contributions.
The strong implication of these aggregation studies is that, although these measures may share a common conceptual framework (ie, inhibitory failure or impaired deviance detection), they probably reflect different neurobiological and genetic substrates. Although no one measure is an ideal physiological endophenotype neither is any one of them redundant. Rather, each is likely to denote an independent contribution to the overall genetic vulnerability to schizophrenia. In this case, individuals who are impaired on more than one measure are more likely to be those who have the highest genetic loading for the illness and to be most informative for genetic linkage and association studies.9 Conversely, individuals who are impaired on different measures may reflect different variants of genetic risk that could assist in the identification of distinct genetic subtypes of schizophrenia. Future studies should therefore focus on the assessment of multiple endophenotypic measures in the same individuals and families, despite the methodological difficulties that this would entail. Detailed investigations of the interrelatedness of these measures will enable us to better address questions regarding both the possible heterogeneity and underlying etiologic mechanisms of the disorder. Large-scale multisite investigations, such as the National Institute of Mental Health–funded COGS, described elsewhere in this Issue,339 will be in an ideal position to perform such analyses.