Introduction

Among the paths to understanding the neurobiology of schizophrenia, one heavily traveled, has been the study through preclinical and clinical models of sensorimotor gating and its neural and genetic substrates. A laboratory paradigm frequently used to operationally measure sensorimotor gating is prepulse inhibition of the startle reflex (PPI). Medline lists over 1400 published reports utilizing the key word “prepulse inhibition” and over 580 that also include the key word “schizophrenia”. Research using PPI to probe the neural and genetic bases of schizophrenia has crossed every level of the “top‑down” and “bottom‑up” investigations of this disorder—from studies of the psychological implications of PPI to those assessing the control of PPI by signal transduction pathways and the genes that regulate them. Arising implicitly and explicitly from such a broad application of the PPI paradigm have been assumptions and expectations that we hope to examine critically in this review. In so doing, we hope to offer some perspectives on both potentially productive directions of this work, and the degree to which some assumptions and expectations may, or may not, be reasonable.

Historical overview

The popularity of PPI as an experimental paradigm for understanding schizophrenia comes from its conceptual linkage to clinical observations that schizophrenia patients are unable to optimally filter or “gate” irrelevant, intrusive sensory stimuli (Bleuler 1911; Kraepelin and Robertson 1919; McGhie and Chapman 1961; Venables 1964). These clinical observations led to the formulation of a construct—“gating deficits” in schizophrenia—that has been extended to refer to deficient inhibition of both sensory and cognitive information. The PPI paradigm was developed as a measure of automatic or preconscious inhibition in normal comparison subjects, as one variant of numerous paired-pulse paradigms in which the presentation of a lead stimulus led to the reduced perceptual or motor response to a second stimulus (Peak 1939; Graham 1975) (Fig. 1). Braff et al. (1978) first merged the construct and its operational measurement by identifying PPI deficits in schizophrenia patients, a finding that has since been replicated by many independent groups and [as reviewed previously (Braff et al. 2001b) and below], has become among the most influential paradigms in the field of schizophrenia psychophysiology. A comprehensive review through the year 2000 of all reports linking PPI deficits to schizophrenia in clinical populations is found in Braff et al. (2001b); reports subsequent to this date are listed in Table 1. Animal studies first linked this finding to a neurochemical (DA) and anatomical (ventral striatum) substrate (Sorenson and Swerdlow 1982; Swerdlow et al. 1986), and subsequent reports centered these substrates within an extended forebrain and pontine circuit that regulates PPI in rodents (Koch and Schnitzler 1997; Swerdlow et al. 1992, 2000a; see Table 4). Animal studies have identified developmental (Geyer et al. 1993; Lipska et al. 1995; see Table 3) and genetic (Carter et al. 1999; Ralph et al. 1999; Geyer et al. 2002; see Table 3) influences on PPI and have led to predictive models for antipsychotic development (Swerdlow et al. 1994) that have been modified and widely applied towards antipsychotic discovery. A comprehensive review through the year 2000 of all reports using PPI in models predicting antipsychotic properties is found in Geyer et al. (2001); reports subsequent to this date are listed in Table 2.

Fig. 1
figure 1

Schematic representation, adapted from Swerdlow et al. (1994), of stimuli used to elicit PPI in laboratory measures (a). b shows superimposed tracings of electromyography of the right orbicularis oculi in an adult male subject, from sequential trials that included either a prepulse [20 ms noise burst 4 dB over a 70-dB(A) background] followed 100 ms later by a 118-dB(A) 40 ms startle noise pulse (solid black area), or the startle pulse alone (open area). Tracings in (b) begin at pulse onset. The amount of inhibition generated by the prepulse can be appreciated visually by subtracting the solid area from the open area

Full size image
Table 1 Studies of PPI in schizophrenia patients and related groups, ca. 2001–2007a
Full size table
Table 2 Examples of studies using PPI to assess or predict antipsychotic properties, ca. 2001–2007
Full size table
Table 3 Model organisms, ca. 2001–2007
Full size table
Table 4 Examples of studies providing anatomically-specific information regarding the neural substrates of PPI, ca. 2001–2007
Full size table

This quantitative physiological abnormality in schizophrenia patients, conceptually linked to an intuitive clinical construct and neurochemical, anatomical, developmental, and genetic substrates, has provided a powerful focus for scientific developments. With the rapid expansion and broad application of variations of PPI measures, new expectations for its use to inform us about the biology of schizophrenia have at times outpaced critical thinking and falsifiable hypotheses about the relative strengths vs limitations of these complex studies. Here, we hope to enumerate some of these expectations and the future promises and potential limitations of PPI studies.

Human studies: What can our field realistically expect to learn about schizophrenia based on studies of PPI in humans?

Diagnosis

As an isolated measure, PPI is not a “diagnostic instrument”. There is substantial variability and significant overlap in PPI distributions among normal and disordered populations. In addition, there are many different disorders in which affected individuals are characterized by reduced PPI, on average, compared to a normal comparison population (cf. Braff et al. 2001b). The reason for the “non-pathognomonic” nature of PPI deficits is simple: the amount of PPI exhibited by any organism at any given moment reflects activity at many different levels of integrated cortico–striato–pallido–thalamic (CSPT) circuitry and its output via the pontine tegmentum. Low levels of PPI can result from normal variations at several levels of this circuitry; alternatively, disease processes can impact different levels of this circuit, with synergistic effects on pontine activity that mediates PPI. Conceivably, disease processes might even impact this circuitry in such a way as to bias it towards elevated levels of PPI, and compensatory or allostatic changes within feedback or downstream elements of the circuitry might offset the effects of otherwise PPI-disruptive disease processes. Thus, absolute levels of PPI—either low or high—are neither diagnostically nor neurophysiologically specific.

A corollary of this fact—that PPI is not “diagnostic”—is that no simple qualitative value of “normal” or “deficient” can accurately be applied to any particular level of PPI, particularly among clinically normal individuals. It is common in the literature (including our own reports) to describe relatively low levels of PPI as “deficient”, “impaired”, or “poor”. In fact, we know of no clear adaptive or functional advantage of higher vs. lower levels of PPI among clinically normal individuals. Perhaps, this idea is most easily conveyed in the comparison between clinically normal men and women: on average, under specific stimulus conditions (e.g., 20 ms white noise prepulses, 10 dB over a 70-dB(A) white nose background, 100 ms before a 115-dB(A) 40 white noise pulse), men exhibit more PPI than do women (Swerdlow et al. 1993b, 2006f; Kumari et al. 2004; Aasen et al. 2005). Furthermore, there is some evidence that among normal women, PPI shifts across the menstrual cycle (Swerdlow et al. 1997; Jovanovic et al. 2004). Clearly, there is no basis for describing PPI in women vs. men as “deficient”, nor for describing luteal- vs. follicular-phase PPI as “impaired”. Similarly, drugs that increase PPI in normals cannot be accurately claimed to “improve” PPI.

At a more basic level, at any given moment in time, individuals are not characterized by a single “PPI” value, in the same manner in which they might be characterized by other quantitative traits such as height, Q–T interval, or fasting glucose level. One of PPI’s strengths as an experimental measure is its exquisite sensitivity to stimulus parameters and test conditions [as described for the startle reflex by Davis 1984]. The inhibition generated by prepulses under different stimulus conditions likely reflects different underlying physiological substrates. Thus, under a variety of test/stimulus conditions, the same clinical population might conceivably exhibit PPI levels that are reduced, equal to, or elevated, compared to normal comparison subjects. An instructive example from preclinical studies of PPI is found in the report that inbred Brown Norway (BN) rats exhibit “deficient” PPI compared to outbred Sprague Dawley (SD) rats, based on measurements with 100 ms prepulse intervals (Palmer et al. 2000). Subsequent studies reproduced this finding, but also demonstrated that at shorter prepulse intervals, the opposite relationship existed: BN rats exhibited significantly more PPI compared to SD rats (Swerdlow et al. 2006a, 2008). Thus, depending on the stimulus parameters, populations can exhibit either relatively reduced or excessive PPI.

PPI is also highly sensitive to state variables and influences, such as medications (Table 1), cigarette smoking (Table 1), fatigue (van der Linden et al. 2006), stress (Grillon et al. 1998), and hormonal status (Swerdlow et al. 1997; Jovanovic et al. 2004). While some of these variables and influences can be controlled under experimental conditions, the notion of using such a sensitive measure in isolation as a diagnostic tool is not realistic. This being said, one potentially valuable strategy in the characterization of clinical populations is the use of PPI in combination with multiple other measures of forebrain inhibitory function, such as P50 event-related potential (ERP) suppression (“P50 gating”; Adler et al. 1982) and antisaccade deficits (Radant et al. 2007), to identify multiple measures and patterns of normal vs. deficient function (Cadenhead et al. 2002; Braff et al. 2008; Sugar et al. 2007). PPI and P50 gating are both deficient but correlate weakly, if at all, in schizophrenia patients (Braff et al. 2007b); similarly, PPI and antisaccade performance are both deficient but do not correlate significantly in schizophrenia patients (Kumari et al. 2005b). Thus, these measures apparently assess forebrain inhibitory processes that are dissociable and nonredundant. More importantly, there are patients who exhibit normal levels of some but not other gating measures (and presumably normal function within brain circuitry regulating some but not other measures), and subpopulations of patients who exhibit different profiles in these deficits (Kumari et al. 2005b; Swerdlow et al. 2006f; Braff et al. 2007b). These subpopulations may reflect different patterns of brain dysfunction and conceivably distinct genetic substrates and treatment sensitivities (Braff et al. 2007a).

Symptoms, course, and outcome

Can we predict the clinical course or even clinical features of schizophrenia based on PPI levels? There is no compelling data to suggest that among schizophrenia patients, levels of PPI predict clinical course, nor are there consistent robust relationships between lower levels of PPI and higher levels of specific symptoms of schizophrenia, or cumulative positive or negative symptoms scores (Table 1). Certainly, there is much interest in determining whether, with repeated or longitudinal measures, a change in PPI predicts or accompanies clinical deterioration or improvement, including the prediction of illness onset in prodromal subjects (Cadenhead 2002; Addington et al. 2007; Cannon et al. 2008). Very few studies have collected longitudinal measures of PPI in schizophrenia populations with adequate sample size and duration to be informative, although some are in progress. One might predict a relationship between PPI and psychosis in extreme conditions, such as the shift from euthymic to manic bipolar disorder, but even in this case, studies have been limited to cross-sectional comparisons, and results across studies have not been consistent (Perry et al. 2001; Rich et al. 2005; Barrett et al. 2005; Carroll et al. 2007). Duncan et al. (2006a, b) did detect an association between lower levels of PPI, and greater levels of psychotic symptoms and psychological discomfort among unmedicated schizophrenia patients.

Interestingly, while robust relationships between PPI and the most common clinical indices of schizophrenia have been hard to detect, reports have identified significant correlations between PPI and a number of relatively complex clinical measures, ranging from quantitative Rorschach ink blot indices of thought disturbance (Perry and Braff 1994) to scales of distractibility and attention (Karper et al. 1996). One report (Swerdlow et al. 2006f) identified a significant positive correlation between PPI and global functioning levels (GAF score) in schizophrenia patients, but this relationship was evident only among male patients, and the correlation—while highly significant (p < 0.005)—accounted for a relatively modest amount of the total PPI variance. In addition, PPI levels were associated with levels of independent living, also perhaps reflecting its relationship to global functioning. As a result, more sophisticated and sensitive analyses of PPI, related gating measures, and function in schizophrenia patients are being pursued (Light et al. 2007a; Braff et al. 2007a). Studies have detected modest but statistically significant relationships between PPI and measures of executive function in some patient groups [e.g., children with 22q11DS (Sobin et al. 2005a, b)]. A preliminary qualitative article by Butler et al. (1991) noted a nonsignificant trend toward greater tactile (but not acoustic) PPI among six (predominantly male) patients with schizophrenia and low levels of Wisconsin Card Sorting Test perseverative responses than among nine (predominantly female) patients distinguished by high levels of Wisconsin Card Sorting Test (WCST) perseverative responses. Kumari et al. (2007a) recently reported a significant (p < 0.03) correlation between tactile PPI and WCST perseverative responses in male schizophrenia patients. Significant positive relationships between acoustic PPI and working memory as well as other formal indices of neurocognitive function have been detected among clinically normal individuals (Bitsios et al. 2006; Light et al. 2007b, 2008; Csomor et al. 2008), although no such relationships have been reported for schizophrenia patients.

The relative insensitivity of PPI to clinical state speaks of the importance of trait features of this measure, which may reflect more “hard-wired” anatomical and genetic determinants. The fact that some relationships can be detected between PPI and relatively global measures of function in schizophrenia patients, but not between PPI and clinical state per se, is consistent with the hypothesis that the causal link between genes and functional outcome in schizophrenia reflects the impact of forebrain circuits that regulate basic gating mechanisms, more than those that control the expression of specific symptom states (Light et al. 2004; Braff and Light 2004; Light and Braff 2005). Thus, while diagnosis in schizophrenia will remain symptom-based for the foreseeable future, it could be argued that studies of the biology of schizophrenia and its relationship to functional outcome may be best advanced through quantitative measures of forebrain inhibitory function such as PPI.

Treatment

As PPI deficits in schizophrenia reflect dysfunction in forebrain circuitry and are linked to both cognitive and functional deficits in schizophrenia patients, can PPI or its potentiation by drugs in patients be used to predict individualized treatment for this disorder? Certainly, in terms of preclinical predictive models, PPI has been quite powerful, as discussed below. In schizophrenia patients, cross-sectional data and some longitudinal findings demonstrate that antipsychotic treatment is associated with elevated (i.e., “normalized”) PPI and that this association is most robust with atypical antipsychotics as a class, compared to first generation antipsychotics (Table 1). Of course, interpreting medication effects in most of these reports is difficult because patients are uniformly being treated with complex multidrug regimens across a range of doses, and medication compliance is known to be poor among schizophrenia outpatients (Lieberman et al. 2005). A recent controlled study with a multidrug cross-over design detected PPI-increasing effects of olanzapine (but not risperidone or haloperidol) in chronically ill schizophrenia patients (Wynn et al. 2007). Findings of PPI-increasing effects of both quetiapine and clozapine in clinically normal, “low-gating” subjects suggests that the PPI-increasing effects of these drugs in schizophrenia patients may not reflect disorder-specific processes (Swerdlow et al. 2006a; Vollenweider et al. 2006). We do not know if the PPI-enhancing effects of these drugs, and conceivably some of their clinical benefit, may reflect their ability to optimize function within spared (intact) gating mechanisms, rather than their ability to correct or normalize activity within dysfunctional mechanisms.

Still, it is reasonable to ask whether the ability of drugs to normalize PPI in patients, or to increase PPI in “low-gating” normals, might reflect their impact on brain processes and resulting cognitive abilities that ultimately would have clinical utility and perhaps cognitive-enhancing effects in schizophrenia. While clinically effective antipsychotics (particularly atypical antipsychotics) are associated with increased PPI in patients and low-gating normals (Table 1), PPI is also increased in non-patients by ketamine and methylenedioxymethamphetamine (MDMA; discussed below; Duncan et al. 2001; Abel et al. 2003; Vollenweider et al. 1999), neither of which would be on anyone’s list of likely antipsychotic agents. Nicotine is associated with increased PPI in schizophrenia patients (Kumari et al. 2001; Swerdlow et al. 2006f), but despite the hypothesis that smoking reflects a form of “self-medication” in schizophrenia patients, there is no clear evidence for either antipsychotic or cognitive-enhancing effects of nicotine in these patients. While there is an active quest by many groups to develop cognitively enhancing nicotinic receptor-specific agonists, based on the putative relationship between the alpha-7 nicotinic receptor subtype and schizophrenia (Freedman et al. 1997), there is presently no evidence that such compounds either increase PPI or enhance cognition in patients. Thus, screening compounds as effective antipsychotics based on their PPI-enhancing effects in clinical or special populations is likely to yield both true and false positives. At this point, there is an inferential, but not empirical, basis for using PPI enhancement as a basis for predicting the ability of a compound to enhance cognition and real-world daily functioning in schizophrenia. Clearly, this is an area of active investigation, and such empirical evidence might emerge based on these efforts.

A reliable, robust quantitative phenotype

While the realistic expectations for PPI as a clinically useful biomarker may be somewhat limited, it is very realistic to expect that PPI will continue to be a valuable tool for investigating brain functions relevant to several neuropsychiatric disorders, including schizophrenia. The many strengths of PPI as an experimental measure have been reviewed elsewhere (Braff et al. 2001b), and none of the realistic limitations described above detract from its attributes as an objective, quantifiable, reliable, robust, neurochemically and parametrically sensitive cross-species measure of a neurobiologically important process. Nonetheless, even in its use as an investigative experimental tool in humans, there should be a realistic assessment of what we can and cannot expect from PPI.

Two types of studies speak strongly to the general reliability of this quantitative phenotype. First, test–retest reliability has been established for PPI in normal comparison subjects (NCS), across days (Abel et al. 1998; Swerdlow et al. 2001c; Flaten 2002), weeks, and months (Cadenhead et al. 1999; Ludewig et al. 2002). More recently, 1-year retest data collected in 68 schizophrenia patients yielded intra-class correlations of 0.75 (30 ms)–0.89 (120 ms; Light et al. 2007a), suggesting a very high stability of this phenotype in patients. Second, a multisite study of PPI in NCS was conducted, using carefully standardized equipment, test methods, and inclusion/exclusion criteria. No significant differences in PPI were detected across seven geographically dispersed test sites, despite some modest methodological drift that was detected via rigorous quality assurance efforts (Swerdlow et al. 2007). Thus, within individuals, and across test samples, PPI appears to be a reliable phenotype.

While PPI is a reliable phenotype, at least among NCS, it is not reasonable to expect that every schizophrenia patient will exhibit a “deficient-PPI” phenotype. In fact, as noted above, there is no way to test this possibility because there is no absolute value that defines “deficient” PPI. Under commonly used test conditions, there is substantial overlap in the distribution of PPI values, between schizophrenia patients and community comparison subjects (cf. Braff et al. 2001b). Clearly, there are schizophrenia patients who have higher levels of PPI compared to many NCS. The overlapping group distributions with this measure likely reflect the many influences on PPI, other than schizophrenia-related pathology, such as sex, hormonal status, smoking, withdrawal from caffeine or nicotine, fatigue, and medications. There are also normal interindividual differences in activity within brain circuitry (e.g., in the pallidum, pons, or cerebellum) that regulates PPI, but is not primarily involved in schizophrenia. With typical testing parameters, NCS vs. unmedicated patients or patients receiving only typical antipsychotics, group separation in mean percent PPI might be reasonably expected to reach 1 SD (e.g., Kumari et al. 1999; Ludewig et al. 2003; Swerdlow et al. 2006f), which corresponds to 55% nonoverlap. However, when patients taking atypical antipsychotics are included, group separation drops dramatically, to about 0.3 SD (e.g. Swerdlow et al. 2006f)—or 21% nonoverlap. This latter fact is particularly important, given that upwards of 90% of schizophrenia patients in most current open-enrollment studies report taking atypical antipsychotic medications [although true compliance is likely lower (Dolder et al. 2002; Lacro et al. 2002)].

In addition to medication status, studies have reported many other variables in patient selection that influence group separation in comparisons of schizophrenia patients vs. NCS. One issue that may ultimately impact the utility of PPI as a quantitative phenotype is its potential sensitivity to ascertainment bias. As noted above, PPI correlates positively with global function in schizophrenia patients. Thus, on average, studies of lower functioning patients will detect greater separation vs. NCS, and those of higher functioning patients will detect less group separation. For this reason, investigators are considering the impact of study designs that select for higher- vs. lower-functioning schizophrenia patients, such as those that require a proband within an intact family structure (and who thus may be relatively higher functioning) vs. those utilizing patients without intact families, who are often homeless or medically indigent (Calkins et al. 2007).

Perhaps equally important as the selection of patients is the selection of NCS. Comparison samples differ substantially across studies and can range from generally healthy, young college students, to “professional controls”, who are often low-functioning and unemployed, beyond their activities as test subjects in biomedical research. The latter group is more likely to have histories of disorders that are associated with reduced PPI, such as anxiety disorders (OCD, panic disorder or post-traumatic stress disorder) or “cluster A” personality disorders; they may also be more likely to carry vulnerability genes for neuropsychiatric disorders, take psychotropic medications that influence PPI, and have histories of substance use or brain trauma that might impact PPI-regulatory brain circuitry. Much has been written about the considerations in selecting a “matched”, “representative”, “normal” or “supernormal” comparison group in biomedical research (e.g., Roy et al. 1997; Calkins et al. 2004), and without belaboring this point, these same considerations apply to studies of PPI and may greatly impact group separation in comparisons of control vs. schizophrenia populations.

As reviewed in Braff et al. (2001b) and elsewhere, the amount of separation between schizophrenia and NCS populations in PPI is highly dependent on testing conditions, and specifically, on stimulus parameters. Thus, if all else is equal, schizophrenia-linked PPI deficits are most pronounced under conditions in which prepulse salience, often based on its intensity over background, is within a “dynamic range”: not too high, but not too low. For example, most studies find this “sweet spot” of maximal schizophrenia vs. NCS separation using discrete white noise prepulses 8–16 dB over a 70-dB(A) background, with about 60 ms prepulse intervals [or stimulus onset asynchronies (SOAs; Table 1)]. Some studies failing to detect PPI deficits in schizophrenia samples have used prepulses in the absence of a background white noise, effectively creating very large prepulse intensities of 25–40 dB(A; Hazlett et al. 2003, Wynn et al. 2004, 2005). In addition to prepulse intensity relative to background, prepulse frequency (e.g., tone vs. white noise), duration (discrete vs. continuous) and other variables (including the use of binaural vs. mono-aural stimuli) may contribute to maximizing the group separation in PPI between schizophrenia and NCS populations (Braff et al. 2001a; Hsieh et al. 2006; Kumari et al. 2005b, 2007b).

As noted above, the temporal “sweet spot” for detecting automatic (uninstructed) PPI deficits in schizophrenia patients appears to occur with prepulse intervals between 30 and 240 ms, depending somewhat on other stimulus characteristics. The temporal range around 60 ms appears to be most sensitive in several studies (Braff et al. 1978, 1992, 2005; Weike et al. 2000; Leumann et al. 2002; Swerdlow et al. 2006f) and may be the range in which PPI deficits are most resistant to normalization by antipsychotic medications. Interestingly, this interval sits at the juncture between preconscious and conscious information processing, based on perceptual detection thresholds (Libet et al. 1979; Kanabus et al. 2002). The possibility that PPI in this temporal range may be most deficient in schizophrenia suggests that automatic inhibitory mechanisms may be most “porous” at a critical barrier between preconscious processing and conscious awareness. While clearly a point for more systematic analysis, such a notion suggests a biological mechanism that is syntonic with psychological models for the intrusion of unedited, preconscious content into conscious awareness in this disorder (Libet et al. 1979; Gray 1995; Swerdlow 1996; Grobstein 2005).

A useful tool for probing the neurobiology and genetics of gating deficits in schizophrenia

Perhaps the most realistic expectation is that PPI is and will remain a useful tool for studying the neurobiology of information processing abnormalities in schizophrenia. While the PPI deficit “signal” in genetic studies of schizophrenia has been blunted by the widespread use of atypical antipsychotics, investigators are increasingly well informed about the many other factors affecting the measurement of PPI and the detection of schizophrenia-associated deficits, and in this way are better positioned to study the basis for these deficits at the levels of their neurobiological and genetic substrates. These studies will be aided by special populations, including “low-gating” normals (Swerdlow et al. 2006a; Vollenweider et al. 2006) and asymptomatic relatives of schizophrenia probands (Kumari et al. 2005b), and by patients with related disorders, such as 22q11 deletion syndrome and unmedicated “prodromal” individuals (Sobin et al. 2005a, b).

As a relatively robust and reliable quantitative phenotype, PPI will be used to map genes associated with deficient sensorimotor gating in schizophrenia probands and families (Swerdlow et al. 2007; Greenwood et al. 2007). The strength of this “endophenotype” approach to understanding disease genetics has been described by many, including Gottesman and Gould (2003), Gould and Gottesman (2006), and Braff et al. (2007a), and largely reflects the fact that the quantitative laboratory measure (in this case, PPI), is closer to the underlying biology (i.e., aberrant neural circuits and their regulation by disease genes), compared to the more variable clinical phenotype (Braff et al. 2007a). There are a small but growing number of examples in which this strategy has proven successful, in identifying genes that confer risk for colon cancer (Leppert et al. 1990) and Type II diabetes (Scott et al. 2007). Whether this strategy can succeed in identifying vulnerability genes for more complex neuropsychiatric disorders is a question at the core of several large ongoing investigative efforts.

Gains will likely be made through the combined use of PPI with sophisticated neurocognitive, neuroimaging, and genetic/genomic tools in schizophrenia and normal populations. It is realistic to expect that these various applications will converge in a top‑down or bottom‑up fashion, i.e., to link: (1) genes with (2) brain substrates that cause (3) gating deficits responsible for (4) neurocognitive disturbances and (5) the resulting daily functional impairment in schizophrenia. Based on the genes and brain substrates identified in these studies, one might reasonably expect that novel treatments will be identified, perhaps acting on intracellular G-protein-coupled signal transduction mechanisms that have already been implicated in the regulation of PPI (van den Buuse et al. 2005a; Kelly et al. 2007; Swerdlow et al. 2006d; Culm et al. 2004; Svenningsson et al. 2003), and which may also be abnormal in some schizophrenia patients (cf. Catapano and Manji 2007). There are also mature lines of research suggesting that novel treatments may target neuropeptides, such as neurotensin (Kinkead et al. 2005; Feifel et al. 2004), that potently regulate PPI and its dopaminergic control, or may target specific dopamine receptors subtypes that regulate PPI via relatively localized effects within mesolimbic and limbic–fronto–striatal circuits (e.g., Zhang et al. 2006). At some stage, it is reasonable to expect that the development of any one of these or other novel treatments might be guided by their effects on PPI in control or clinical populations.

A surrogate measure for neural processes with wide-reaching psychological implications

The frontal, limbic, and mesolimbic circuitry that regulates PPI also regulates many higher-order psychological processes. Thus, PPI can be viewed as a simple surrogate “readout” of activity in this circuitry—an experimentally generated signal from the forebrain, detected through efferents descending through a “pontine portal”. Alternatively, PPI can be viewed as a measure of a fundamental psychological process—sensorimotor gating—with broad-reaching implications for the structure of complex behavior and thoughts. In truth, both views are at least partly accurate, under specific uses of the PPI paradigm.

“Gating” can be a very specific process when operationalized in the laboratory, but is less precisely defined when used as a psychological construct. How broadly can we extrapolate from the laboratory measure of one type of gating—sensorimotor gating—to other forms of automatic inhibition of sensory, cognitive, or motor information? There is credible evidence that PPI correlates significantly with a form of perceptual “gating”, measured by the degree to which the prepulse reduces the perceived intensity of the startling stimulus (Peak 1939; Swerdlow et al. 2005b). On the other hand, PPI does not correlate strongly with the most structurally similar form of “gating”—sensory gating—measured by suppression of the P50 auditory event-related potential (ERP; Light et al. 2006; Hong et al. 2007). Nor does PPI in normal humans correlate strongly with other measures thought to assess inhibitory processes that contribute to forms of “cognitive gating”, such as latent inhibition (Murphy et al. 2001; Leumann et al. 2002; Peleg-Raibstein et al. 2006a, b) or visuospatial or semantic priming (Swerdlow et al. 1995b). Certainly, there is little evidence that PPI assesses processes that are strong determinants of normal personality structure and dimensions (Swerdlow et al. 2003d). At the least, it is important to recognize that the construct of “gating” is applied to many different processes and that it is reasonable to expect PPI to be informative about some, but not all or even most of these processes.

Summary: human studies

Human studies of PPI will continue to provide one important level of information within a top‑down or bottom‑up understanding of the biology of schizophrenia. PPI offers great promise as a quantitative phenotype for genetic studies and will be used in combination with other measures to connect an aberrant physiological signal (impaired startle inhibition) with its underlying neural substrates (via neuroimaging studies) and with its consequences in terms of cognitive deficits (via neurocognitive measures) and real-life impairment (via functional measures). It is realistic to expect that as we gain a better understanding of its modulating variables and optimal experimental methods, PPI in humans will continue its evolution, started in 1978 (Braff et al. 1978) from an isolated laboratory-based psychophysiological phenomenon, into a productive clinical research tool for understanding psychopathology. As we learn more about PPI, our scientific approaches to its use will continue to become more sophisticated, and we will be better positioned to take full advantage of what it can tell us about normal and abnormal brain functions.

Animal studies: What can our field realistically expect to learn about schizophrenia based on studies of PPI in laboratory animals?

Etiology

Two general applications of animal studies of PPI will be considered here: (1) the use of PPI to evaluate models or model organisms relevant to the etiology of schizophrenia; and (2) the use of PPI to “map” the neural substrates of deficient PPI in schizophrenia.

Model organisms, created via genetic, developmental, surgical, pharmacological, or immune manipulations, have been a mainstay of studies of the etiology, pathophysiology, and treatment of schizophrenia. Of course, schizophrenia—as defined clinically—is a uniquely human disorder (least we ascribe to rats the ability to have “two or more voices conversing with one another or voices maintaining a running commentary on the [rat’s] thoughts or behavior,” or the ability to conceptualize that “alien thoughts have been put into his or her mind...”, or to have homologous complex social cognitive deficits; APA 2000). However, investigators can apply schizophrenia-linked constructs to these models and test whether the resulting animal reproduces laboratory-based phenotypes exhibited by schizophrenia patients. The degree to which these phenotypes are reproduced in the model organism provides a level of validity to the construct, even if it is specific to the laboratory-based phenotype, rather than the broader clinical disorder.

For example, given a particular schizophrenia candidate gene “X”, it is reasonable to ask whether manipulations of gene “X” produce an animal that exhibits reduced levels of PPI compared to a wild-type animal. If so, then the gene “X” mutant would be a valid model for PPI deficits in schizophrenia. Such an approach has been taken with many different animal models (Table 3). There are obvious limitations to the specificity and sensitivity of this approach, which could be deduced from the above discussions of the PPI findings in humans.

Because deficient PPI is not unique to schizophrenia populations, there is no a priori justification for claiming that such a mutant specifically models the PPI deficits in schizophrenia, rather than OCD (Swerdlow et al. 1993a; Hoenig et al. 2005), Tourette Syndrome (Smith and Lees 1989; Castellanos et al. 1996; Swerdlow et al. 2001b), Blepharospasm (Gomez-Wong et al. 1998), or a number of other conditions. The specificity of the linkage of the model with schizophrenia, and hence with PPI deficits in schizophrenia, must come from the construct. For example, the finding of PPI deficits in a murine model of 22q11 deletion syndrome (22q11DS) links this model to PPI deficits in schizophrenia (Paylor et al. 2006; Sobin et al. 2005a, b), on the basis of the clinical relationship between 22q11DS and schizophrenia. Without this clinical relationship, this would just be a mouse with low PPI, and the model would most likely be a “false positive” for the schizophrenia phenotype.

Certainly, it is unlikely that most genes associated with low vs. high levels of PPI will be related to reduced PPI in schizophrenia or any one other disease states. This is because the most potent influences regulating baseline PPI involve physiological substrates that are probably not relevant to schizophrenia. For example, a very potent determinant of acoustic PPI is hearing threshold, as an organism that cannot hear a prepulse will not exhibit PPI. Thus, many candidate “PPI genes” identified via gene inactivation or mapping strategies of drug-free PPI in inbred and recombinant rodents will likely be associated with hearing threshold. Beyond the level of sensory detection, the most potent neural control of baseline PPI is exerted by the pedunculopontine nucleus (PPTg) (Swerdlow and Geyer 1993a), which mediates PPI via its impact on the nucleus reticularis pontis caudalis (NRPC; Koch et al. 1993). 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 any model for the pathophysiology of schizophrenia. In contrast, the prefrontal cortex (PFC)—which is viewed as a critical substrate for some core symptoms of schizophrenia (e.g., cognitive disorganization, deficient working memory, executive functioning, abstract reasoning, cognitive flexibility and context processing, and negative symptoms)—is likely to be three or four synapses removed from the primary startle circuit; in a normal human or rodent, genes controlling the PFC will likely contribute only weakly to a genetic “signal” based on levels of baseline PPI.

One might argue that a finding of PPI deficits provides additional validation that a particular model reproduces one of the quantitative phenotypes associated with schizophrenia. But as noted above, there is no definitive evidence that PPI deficits—or the neural abnormalities that produce them—are necessary for the expression of the broader schizophrenia phenotype. Rather, it is almost certainly true that there are large numbers of functionally impaired, symptomatic schizophrenia patients who exhibit levels of PPI in the “normal” range. Thus, rejecting animal models on the basis of “normal” PPI levels would likely result in a number of “false-negative” models—i.e., ones in which some features of the model accurately recreate important aspects of the biology of schizophrenia, but do not result in reduced PPI.

Perhaps the most realistic expectation of PPI in the assessment of animal models of schizophrenia is that it can provide validation for specific existing constructs—i.e., that the construct can reproduce PPI deficits exhibited by a significant subgroup of the heterogeneous population of schizophrenia patients. On the other hand, “normal” or unaltered PPI should not be used as the basis for rejecting a model: even in the presence of “normal” (i.e., wild-type, sham lesioned or placebo-treated) PPI levels, it is very possible that a model might be highly informative about the biology of schizophrenia.

Animal studies are also used to explicate the neural regulation of PPI, as a means of understanding the neural basis of PPI deficits in schizophrenia and other disorders. In this case, the manipulations are selected not necessarily based on a “construct” of schizophrenia, but rather based on the extant PPI neural “map”, and the understanding of anatomical and neurochemical properties of that map. In general, the organism used in these studies is not a schizophrenia “model” per se, but is more akin to a canvas on which a neural map can be painted. A reasonably comprehensive understanding of this “map”, ca. 2000, is found in Swerdlow et al. (2001a), and an updated list of studies of “PPI anatomy” is found in Table 4.

Much can be gleaned about PPI and its broader context by considering two facts related to its anatomical substrates. First, PPI remains intact after acute trans-collicular decerebration in the rat (Davis et al. 1982). In other words, the expression of unimodal acoustic PPI in rats does not require any part of the forebrain, and therefore, it must be mediated at or below the pons. The prepulse does not (and by physical and temporal constraints, cannot) “travel” to the forebrain to generate its inhibitory impact on the simple startle reflex (see discussion in Swerdlow et al. 2001a). Second, PPI can be regulated, and even eliminated, by subtle pharmacological manipulations at the most rostral tip of the forebrain [e.g., D1 receptor blockade within the medial prefrontal cortex (Ellenbroek et al. 1996; Shoemaker et al. 2005; Swerdlow et al. 2005c)]. Thus, brain substrates at the furthest point from the PPI “mediating” circuitry in the pons are capable of potently regulating the amount of inhibition generated by the prepulse, presumably via tonic, “thermostat”-like stimulus-independent changes in activity within descending circuitry.

These two facts lead to a simple conclusion: while PPI is mediated via the pons, it can be regulated by the forebrain. A relative loss of PPI in clinical populations, and in the animal models that are used to study them, can be a consequence of aberrant activity within this descending circuitry—somewhere “between” the cortex and pons—or within substrates that impinge upon it. The efforts to “map” this PPI-regulatory circuitry, point-to-point, from cortex to pons, are aimed to help investigators identify candidate substrates that contribute to the loss of PPI in patient populations and candidate targets for therapeutic interventions. Of the many words of caution related to this use of animals to “map PPI”, two will be noted here.

First, rodent brains and human brains are not the same. Thus, a map of neural circuitry regulating PPI in rodents cannot be expected to translate exactly to human brains. Indeed, it is surprising how much overlap is suggested across species, based on neuroimaging findings in humans (Kumari et al. 2003a, 2005, 2007a; Postma et al. 2006), and based on examples of localized neuropathology associated with PPI deficits in brain disorders such as HD and in rat and murine models of this disorder (Swerdlow et al. 1995a; Carter et al. 1999; Van Raamsdonk et al. 2005). These findings notwithstanding, it is clear that species differences will be most pronounced in phylogenetically newest regions, some of which—e.g., frontal cortex—may be of most relevance to schizophrenia. As we attempt to interpret these circuit maps at higher levels of resolution to guide drug development—i.e., beyond simple efferent/afferent patterns, and down to the receptor- and subcellular levels—these cross-species differences may become increasingly important. A number of these differences are already suggested based on simple pharmacological challenge studies, described below.

Second, all rodent brains are not the same. Strain differences in PPI, and in sensitivity to drug effects on PPI, are quite remarkable across inbred and outbred rat strains, and across inbred and outbred mouse strains. These differences must reflect differences in the PPI-regulatory brain circuitry, potentially at any level from the presence of different cell types within a larger circuit organization, down to differences in the activity of specific enzymes within signal transduction pathways. Inbred Brown Norway rats exhibit significantly more PPI at short prepulse intervals and significantly less PPI at long prepulse intervals, compared to outbred Sprague Dawley (SD) rats (Swerdlow et al. 2006a). These differences are heritable (Swerdlow et al. 2008), and must reflect genetically mediated differences in brain organization. Albino SD and hooded Long Evans (LE) rats differ significantly in their sensitivity to the PPI-disruptive effects of dopamine (DA) agonists (e.g., Swerdlow et al. 2004a, 2006d) and in the expression of DA-regulatory enzymes [e.g., catechol-o-methyl transferase (COMT)] and signal transduction enzymes (e.g., protein kinase) within the nucleus accumbens (Shilling et al. 2008). Which of these strains provides an anatomical/neurochemical “map” of PPI that is most informative about human PPI circuitry, and hence, about PPI circuit abnormalities in schizophrenia? The answer is likely to differ, based on the neural systems and levels of resolution being studied, and the models being applied.

Treatment

It is reasonable to expect that studies of PPI in laboratory animals will continue to play a major role in the discovery and development of novel therapeutics for schizophrenia. As noted above, there is no compelling empirically based reason to expect that increased PPI per se might be desirable or functionally enhancing, nor that the ability of a drug to increase PPI in schizophrenia patients should be necessary or sufficient for clinical benefit. Despite this caveat, there is clear empirical evidence that the ability of drugs to “normalize” PPI levels after they have been reduced experimentally by specific drugs or perhaps by other manipulations (e.g., developmental manipulations) strongly predicts clinical utility and even potency of antipsychotic agents (Swerdlow et al. 1994; Swerdlow and Geyer 1998; Fig. 2). Towards this end, PPI has been used in several different types of predictive models, which differ in their sensitivity, specificity, logistical complexity, and even in the types of antipsychotics that they appear to identify. These issues are reviewed in Geyer et al. (2001), and an update of studies using PPI for its predictive validity since 2000 are found in Table 2.

Fig. 2
figure 2

Evidence supporting the predictive validity of one “rapid-throughput” animal model of PPI deficits. In these studies (Swerdlow et al. 1994), PPI was disrupted in adult male Sprague–Dawley rats by the mixed D1/D2 agonist, apomorphine (0.5 mg/kg sc). The ED50 of a number of drugs to reverse this apomorphine effect correlated significantly with their clinical potency. Subsequent studies have identified many other clinically effective antipsychotic agents from different chemical classes that prevent the PPI-disruptive effects of apomorphine in rats [see Table 2 and Geyer et al. (2001)]. A small number of potential “false-positive” compounds have also been detected, primarily in other species or strains. Other predictive models have been developed using PPI as a dependent measure, as described in the text and Table 2, each with different sensitivity, specificity, and logistical complexities

Full size image

The four most common variations of the PPI paradigm in models predictive of antipsychotic effects involve the use of (1) DA agonists (Fig. 2), (2) NMDA antagonists, (3) isolation rearing (IR), and (4) neonatal ventral hippocampal lesions (NVHLs). While each of these variations is based on a biological “construct” for the etiology of schizophrenia, i.e., hyperdopaminergia, hypoglutamatergia, and specific neurodevelopmental insults, they have all been applied towards predicting antipsychotic properties in novel compounds. In truth, only the former two variants are well suited to traditional “rapid throughput” drug screens, based on the amount of time and resources necessary for the developmental models, and the relatively small (and often strain- or sex-dependent) effects of isolation rearing on PPI (Weiss et al. 1999, 2000; Powell et al. 2002). In each of these variations, the ability of a drug to “normalize” PPI is interpreted as evidence for antipsychotic potential. Some second generation antipsychotics, such as clozapine, quetiapine, and olazapine, tend to increase PPI in otherwise intact animals (Swerdlow and Geyer 1993b; cf. Geyer et al. 2001), particularly in mice, adding some interpretative complexity to their ability to normalize PPI after pharmacological, developmental, or surgical manipulations. In fact, the ability to enhance baseline PPI is a signal that has been used as a predictor of antipsychotic potential in mice, in some normally “low-gating” mouse strains (cf. Ouagazzal et al. 2001a), rat strains (Feifel et al. 2001, 2004), and even in normal “low-gating” humans (Swerdlow et al. 2006a; Vollenweider et al. 2006).

Beyond the dopamine system, some new targets of antipsychotics have emerged in recent years, based in part on studies using variations of PPI paradigms as predictive models. Examples of these targets include (but are not limited to) selective 5-HT2C receptor agonists (Marquis et al. 2007), CB1 cannabinoid receptor antagonists (Nagai et al. 2006), neurotensin-1 receptor agonists (Shilling et al. 2003, 2004; Caceda et al. 2005), selective adenosine A(2A) receptor agonists (Wardas et al. 2003), alpha-7 nicotinic receptor agonists (Suemara et al. 2004), and selective histamine H3 receptor antagonists (Fox et al. 2005; Table 2). It should be emphasized that in some cases, these targets were identified based on PPI assays with less compelling predictive validity, such as the ability of compounds to increase basal PPI levels in mice, or to normalize PPI after its disruption by 5HT agonists or NMDA antagonists. These assays may have strong sensitivity, particularly for identifying compounds with potentially novel mechanisms, but they also may lack specificity for detecting antipsychotic properties, at least in comparison to assays based on the ability to block the PPI-disruptive effects of apomorphine and perhaps other DA agonists (Fig. 2). We will have to await clinical evidence to determine whether these reports reflect “false positives” of these models.

PPI has only more recently begun to be used in models for detecting preventative or neuroprotective interventions, to identify strategies that would prevent the neuropathological and clinical consequences of a vulnerability gene or developmental insult involved in the prodrome and onset of schizophrenia. Some studies are approaching such an application, using early neuroimmune challenges to yield PPI deficits during adulthood (e.g., Borrell et al. 2002), or using sustained early life antipsychotic exposure to blunt the PPI-disruptive effects of developmental insults (Powell et al. 2006a, b). Assuming that these models succeed, it remains to be determined how one would test or apply such interventions in a clinical setting.

A reliable, robust, quantitative phenotype

In any given rodent species and strain, both PPI and its drug sensitivity are quite robust and reliable phenotypes. Within a range of 30–120 ms prepulse intervals, and 2–16 dB noise prepulses over a 65- to 70-dB(A) noise background, and 105–120 dB(A) noise pulses, PPI in rats exhibits a magnitude and parametric sensitivity that are strikingly similar across a number of studies from different laboratories and, conveniently, are also quite similar to those exhibited by humans. Similarly, PPI-disruptive effects of a number of simple manipulations (e.g., administration of a direct DA agonist) have been replicated across laboratories to the point that they have become “standard assays”, in predictive models for antipsychotic development. The PPI-disruptive effects of more complex manipulations, including early developmental lesions or isolation rearing, tend to be more variable across laboratories (discussed above), perhaps due to the complexities (and hence variability) of the methods and uncontrolled sources of variance. Some differences in reports of PPI drug sensitivity and sensitivity to developmental manipulations clearly seem to result from differences in rat strain or even supplier (e.g., Swerdlow et al. 1998, 2000b, 2003a, 2004a), and these differences are being explicated at the levels of heritable differences in neural substrates regulating PPI.

Some disparities in reported drug or other manipulation effects on PPI may also reflect differences in the recording properties of a variety of “home built“ and commercially available startle response acquisition systems. While there is no “gold standard” for such an apparatus, there are a number of characteristics that should be evaluated in interpreting whether response measurements “obey the laws of physiology”, e.g., intensity- and interval-dependence of PPI, and relative insensitivity of PPI to weight differences across animals. These issues are reviewed in Geyer and Swerdlow (1998).

Startle and PPI data can be deceptively complex, and some disparities in reported effects on PPI in rodents undoubtedly reflect these complexities and resulting interpretative differences across studies. Despite the impressive degree of automation in laboratory measures of PPI, one cannot automatically enter startle data into an equation and reasonably expect the calculated percent PPI to be informative. For example, we have previously reviewed the importance of considering the impact of changes in startle magnitude on changes in PPI (Swerdlow et al. 2000a). Simply put, the only unambiguous changes in sensorimotor gating are ones that can be demonstrated in the absence of changes in startle magnitude. In this case, reduced sensorimotor gating reflects a diminished impact of the prepulse on startle magnitude and, hence, an increase in startle magnitude on prepulse + pulse trials only. Any other related pattern of results, involving significantly reduced or increased startle magnitude on pulse-alone trials, must be interpreted in the context of additional supportive evidence. Such evidence might come from the use of low and high pulse intensities or from subgroups of rats that are matched based on comparable levels of startle magnitude.

Another interpretative issue that has been discussed in several recent reports relates to the potential impact of prepulse-induced startle activity on PPI and its modification by drugs or other experimental manipulations (Yee et al. 2004; Swerdlow et al. 2004c). A stimulus is only considered a “prepulse” in relationship to a second stimulus. By any other metric, it is simply a stimulus and can elicit motor activity including a startle reflex, depending on its properties. If the prepulse intensity exceeds the startle threshold, a “prepulse + pulse” configuration is better described as a “paired-pulse” configuration, and the resulting decrement in the startle response elicited by the second pulse is described as “paired-pulse inhibition”, comparable to the phenomenon used to study “blink excitability” (e.g., Kimura and Harada 1976; Valls-Sole et al. 2004). The similarities and differences of PPI and paired-pulse inhibition have been described for a small number of drug effects (e.g. Swerdlow et al. 2002a), but relatively little is known about this relationship for the long list of manipulations that have been applied towards PPI studies.

The interpretative ambiguities created by “prepulse-elicited startle” are most relevant to conditions in which the prepulse exceeds startle threshold. In a rat, for 20 ms noise prepulses over a 70-dB(A) noise background, this threshold is generally between 12 and 15 dB, although the precise value varies with strain, sex, age, and other factors. Other prepulse characteristics, including frequency (pure tone vs. white noise), duration, and configuration (continuous vs. discrete) can impact its motor-inhibiting and activating properties. For the vast majority of published PPI studies, prepulses are used at levels that elicit no or little detectable motor activity; even relatively intense prepulses (e.g., 10–15 dB salience, based on the stimulus conditions described above) might elicit a motor “signal” that is <1% of the total startle signal (Swerdlow et al. 2004c). In fact, this signal is comparable to that detected on “NOSTIM” trials, i.e., when no motor activity is recorded in the absence of stimulus delivery, suggesting that this small signal reflects ongoing motor activity rather than a prepulse-elicited motor response (e.g. Swerdlow et al. 2004c; Weber and Swerdlow 2008). Importantly, only a small fraction of studies utilize prepulses with supra-threshold intensities, and among these, most also utilize much weaker prepulses as internal comparisons. PPI is used to assess many things, and in some cases, a range of prepulse intensities is used to create a complete parametric characterization for purposes unrelated to drug effects (e.g., QTL analyses). Clearly, in these cases, the use of intense prepulses is not a “confound”, but simply a way to fully characterize a phenotype.

It is argued that potentially confounding effects might arise if a drug or other manipulation lowers startle threshold and, hence, transforms a non-startling prepulse into one that elicits a motor response (Yee et al. 2004). Specifically, a potentially confounding interaction might arise if increases in prepulse-evoked motor responses diminished the prepulse’s inhibitory effects on a subsequent startle response. In fact, there is no reason to predict such an effect: full startle responses elicited by an S1 in a paired-pulse paradigm do not interfere with the inhibitory impact of S1 on the startle response to S2 (e.g., Swerdlow et al. 2002a), so there is no credible reason to predict that such interference would result from a prepulse-evoked response that is 100-fold less intense. Nonetheless, under drug conditions, a number of control comparisons can be conducted—analogous to those used to understand the impact on PPI of drug-induced changes in startle magnitude—to determine whether drug effects on prepulse-evoked motor activity and PPI can be “dissociated”. We might predict that a common drug receptor (e.g., D1 or D2) might mediate two processes (reduced PPI and increased prepulse-induced motor activity), via effects within different brain substrates. Similar to changes in startle magnitude, a given drug might elicit either increases, decreases, or no change in prepulse-induced motor responses, yet have a consistent effect on PPI (e.g., Weber and Swerdlow 2008); even in cases where drug-induced changes in prepulse-induced activity are detected, they amount to shifts of less than 1% of the total “signal” of the startle response and, as noted above, are comparable to changes observed in “NOSTIM” activity. Thus, while it is a reasonable precaution to consider measuring prepulse-elicited motor activity to ascertain whether it is significantly greater than ongoing background motor activity, and whether it might potentially interact with the startling effects of the startle pulse, in our experience, such an exercise amounts to “much ado about [almost] nothing” (Swerdlow 2005).

A useful tool for modeling the neurobiology and gating and its deficits in humans

The most compelling contribution of animal studies of PPI towards the understanding of the basis for PPI deficits in schizophrenia comes in the ability to directly manipulate neural and genetic substrates and test hypotheses in a controlled experimental setting. The challenges of extrapolating such findings across species are not trivial, as discussed above in relationship to neural circuit maps. Still, for understanding the contribution to PPI deficits in schizophrenia of pathology in medial prefrontal cortex, hippocampus, amygdala or ventral striatum, or of specific candidate genes or early developmental insults, cross-species studies are a unique, powerful tool.

PPI studies have also identified neurobiological bridges across species that may reveal potential limitations of these studies and, perhaps, more generally of animal models of schizophrenia. For example, several drugs potently disrupt PPI in rats and yet increase PPI in normal humans. This is most notable because the drugs in question—ketamine (Abel et al. 2003; Duncan et al. 2001), MDMA (Vollenweider et al. 1999) and under some conditions, DA agonists (Bitsios et al. 2005)—have pharmacological and clinical properties that are central to models for the pathophysiology of schizophrenia. These findings raise both experimental and conceptual issues.

At an experimental level, drug doses, routes of administration, and pharmacokinetic/dynamic properties differ substantially across species. As one example, amphetamine reliably decreases PPI in rats only at doses above 2 mg/kg administered subcutaneously (Mansbach et al. 1988; Sills 1999; Swerdlow et al. 2006d), while the oral dose of amphetamine given to normal humans in PPI studies rarely exceeds 0.29 mg/kg (20 mg total; e.g., Hutchison and Swift 1999; Swerdlow et al. 2003b). Species differences in drug effects might also reflect contextual differences in the test setting. Humans volunteer and are paid for study participation, have the test conditions explained by a supportive research assistant, swallow a pill, and sit in a comfortable chair during testing; by contrast, rats are removed from a cage, injected with a drug, and then placed alone in a plastic tube inside an unfamiliar box where they are exposed to loud, unexpected noises. One might imagine that drug effects on a fight-or-flight reflex (startle) might differ in these two conditions, independent of species. Furthermore, while the parametric properties of PPI (e.g., sensitivity to prepulse intensity and interval) are strikingly similar across species, drug effects might reveal some cross-species differences in these parametric effects. For example, at 120 ms prepulse intervals, ketamine has opposite effects on PPI in rats (disrupts PPI; Mansbach and Geyer 1989) and humans (increases PPI; Abel et al. 2003; Duncan et al. 2001); on the other hand, ketamine can increase PPI in rats at shorter prepulse intervals (e.g., 30 ms; Mansbach and Geyer 1989). Our group has detected similar species- and interval-dependent effects with the NMDA antagonist, memantine (Swerdlow et al. 2003c, 2005a). Conceivably, NMDA-related mechanisms of drug effects on gating at 30 ms in rats might best approximate those at 120 ms in humans.

However, this explanation does not address the conceptual dilemma created by the fact that psychotomimetic drugs increase PPI in normal humans, while schizophrenia is associated with reduced PPI. While PPI deficits in schizophrenia might possibly reflect the consequences of sustained deficiencies in glutamatergic activity in the context of developmentally aberrant neural connections, it does not follow that such effects would be reproduced by an acute challenge of an NMDA antagonist to a normal individual with normal neural connectivity. Furthermore, one might easily imagine that acute drug effects on an intact brain might enhance sensorimotor gating via a mechanism that is very distinct from (e.g., “upstream” or “downstream” from) those responsible for reduced gating in the brain of a schizophrenia patient. Nonetheless, faced with these discrepant effects of psychotomimetic drugs on PPI, it is difficult to know whether the failings lie in the cross-species translation of the PPI model, in the validity of the acute ketamine/glutamate antagonist model of schizophrenia, or both.

An additional challenge in building neurobiological bridges of PPI studies across species comes from the human side of the bridge—from the observations that drug effects on PPI in humans can differ significantly, depending on basal levels of PPI. A number of drugs—including amphetamine (Swerdlow et al. 2003b), pergolide, amantadine (Bitsios et al. 2005), quetiapine (Swerdlow et al. 2006a), and clozapine (Vollenweider et al. 2006)—have been demonstrated to have effects that differ significantly (and in some cases, are arithmetically opposite) in normal humans with low vs. high PPI levels, relative to the overall test population. Similar findings may be emerging from animal studies, e.g., among inbred strains with low basal levels of PPI (cf. Ouagazzal et al. 2001a). How we interpret this “rate dependency” of drug effects on PPI in humans and laboratory animals and what it means about the many reported drug effects on PPI that have not considered or tested the impact of basal PPI levels, are issues that remain to be resolved.

While this discussion has focused primarily on cross-species comparisons between rodents and humans, and we discussed earlier the strain differences in PPI that have been detected in both rats and mice, it is also worth noting that there are also a number of important cross-species differences in PPI and its parametric and pharmacological sensitivity between rats and mice. Just as one example, while PPI is disrupted by DA agonists in both rats and mice, there is some evidence that this effect primarily reflects activation of D2 receptors in rats (Swerdlow et al. 1994; cf. Geyer et al. 2001), but of D1 receptors in mice (Ralph-Williams et al. 2003a; Ralph and Caine 2005). Within a restricted set of stimulus parameters (particularly prepulse intervals), infusion of D2 agonists into the nucleus accumbens decreases PPI in rats and increases PPI in mice (Mohr et al. 2007). This issue is not yet settled, as mice lacking D2 receptors are insensitive to the PPI-disruptive effects of d-amphetamine (Ralph et al. 1999), and some mouse strains exhibit “rat-like” PPI sensitivity to D2 agonists (Ralph and Caine 2007). Nonetheless, enough data exists that we can be fairly confident that a similar drug effect on PPI in rats and mice does not necessarily reflect a common underlying brain substrate. This raises the dilemma that when modeling the loss of PPI in schizophrenia, we are almost certainly studying very different neurobiological substrates, depending on the model species; this makes it very difficult to identify a clear, a priori rationale for selecting one species over another.

A surrogate measure for neural processes with wide-reaching psychological implications

Models of higher cognitive processes are only now being developed in rodents. Given the limited size and processing capacity of the frontal cortex in mice and rats vs. primates, and its relatively weaker contribution to the organization of behavior, there is reason to be skeptical that rodent models of higher cognitive processes will provide meaningful homology to human cognition. Nonetheless, mice and rats are amenable to complex conditioning schedules and are capable of performing choices and sophisticated behavioral sequences, and it is certain that studies will assess the potential relationship of PPI to these processes (e.g., Roegge et al. 2007; Depoortere et al. 2007a, b; Garner et al. 2007; Paine et al. 2007). Extrapolating these findings to humans will present many challenges. In general, the farther forward one moves in the brain, the greater the anatomical and functional differences between rodents and humans. For example, one might imagine a scenario in which “cognitive” control in rodents involves a prominent role for subcortical (e.g., basal ganglia) functions that overlap with PPI-regulatory circuitry, while in humans, higher cognitive control is “encephalized” to discrete frontal circuits that participate less in the regulation of startle gating.

There is already some evidence for both convergence and divergence of PPI and other operational animal models of “gating”, in terms of their underlying neural substrates. For example, contemporaneous measures of PPI and N40 gating—an animal model of P50 ERP gating in humans—revealed that apomorphine, phencyclidine, and DOI each disrupt PPI and reduce ERP responsivity to the S1 stimulus in the N40-gating paradigm, but do not specifically disrupt N40 gating per se (Swerdlow et al. 2006b). Some overlap has been reported in the pharmacological sensitivity of PPI and [some of the various forms of] latent inhibition to DA agonists and NMDA antagonists (Mansbach and Geyer 1989; Bakshi et al. 1995; Razoux et al. 2007), although many conditions lead to a loss of PPI in rats but leave latent inhibition intact (e.g., amphetamine withdrawal (Peleg-Raibstein et al. 2006a, b) and D2 blockade in the basolateral amygdala (Stevenson and Gratton 2004)). Thus, neurobiological mechanisms of PPI cannot be assumed to be common to experimental measures of either sensory or cognitive gating in rats. The potential overlap in the neurobiology of PPI and higher-order functions in rats, such as working memory, is an area of ongoing investigation. At present, there is no compelling evidence that such an overlap exists or that PPI is informative about higher cognitive functions in rodents.

Summary: animal studies

Animal models will remain an important tool in developing and testing hypotheses for the pathogenesis of brain disorders. As a reliable, quantitative “read out” of relatively well-defined neural circuitry, measures of PPI in laboratory animals will continue to be used to test and validate these hypotheses and to generate important new hypotheses regarding cellular mechanisms and therapeutic strategies. PPI models provide predictive validity in drug discovery and development, both as rapid through-put screens and as components of more biologically sophisticated models involving developmental, immunologic, and genetic manipulations. Areas of convergence and divergence are being identified in the cross-species pharmacology of PPI; areas of convergence will be exploited so that human drug effects can be predicted and understood based on PPI drug effects in rodents and their underlying cellular and molecular substrates. Finally, the relationship of PPI to higher-order learning processes is being explored in rodents, and the findings will be used to generate and test hypotheses regarding the interplay of sensorimotor gating and cognition in normal and disordered humans.

Conclusions

The construct of gating deficits in neuropsychiatric disorders has empirical support and intuitive appeal, and serves as a unifying heuristic for understanding the psychological and neural substrates shared by otherwise apparently unrelated disorders. PPI is an operational measure of basic, brain-based gating processes. It is robust, reliable, easily quantified, and versatile as an experimental tool, and is abnormal in several brain disorders including schizophrenia, that are characterized by clinical evidence of impaired gating of sensory, cognitive, motor of affective information. PPI can be measured across species and is regulated in laboratory animals by neurochemical, anatomical, developmental, and genetic substrates that can be systematically studied and used as the basis for developing and testing hypotheses for the biological basis of PPI deficits in patients.

For all of these reasons, studies of PPI in humans and laboratory animals have multiplied and expanded, and this measure is being used to explore many new questions at many different levels of analysis. While our field does not yet face the floods of the “Sorcerer’s Apprentice” (von Goethe 1779), it is clear that findings have amassed at an exponential rate and are testing our collective ability to critically integrate results, to identify areas of consistency, redundancy, and disagreement. Based on a review of the present literature, we reached several conclusions: (1) in humans, PPI is not “diagnostic”; levels of PPI do not predict clinical course, specific symptoms, or individual medication responses; (2) in preclinical studies, PPI is valuable for evaluating models or model organisms relevant to schizophrenia, “mapping” neural substrates of deficient PPI in schizophrenia, and advancing the discovery and development of novel therapeutics; (3) across species, PPI is a reliable, robust quantitative phenotype that is useful for probing the neurobiology and genetics of gating deficits in schizophrenia. In this review, we also identify some realistic expectations of this paradigm, describing its considerable strengths but also limitations, and stress some interpretative issues for consideration as we move forward with this powerful tool for translational neuropsychiatric research.