Weaker situations: Uncertainty reveals individual differences in learning: Implications for PTSD

Abstract

Few individuals who experience trauma develop posttraumatic stress disorder (PTSD). Therefore, the identification of individual differences that signal increased risk for PTSD is important. Lissek et al. (2006) proposed using a weak rather than a strong situation to identify individual differences. A weak situation involves less-salient cues as well as some degree of uncertainty, which reveal individual differences. A strong situation involves salient cues with little uncertainty, which produce consistently strong responses. Results from fear conditioning studies that support this hypothesis are discussed briefly. This review focuses on recent findings from three learning tasks: classical eyeblink conditioning, avoidance learning, and a computer-based task. These tasks are interpreted as weaker learning situations in that they involve some degree of uncertainty. Individual differences in learning based on behavioral inhibition, which is a risk factor for PTSD, are explored. Specifically, behaviorally inhibited individuals and rodents (i.e., Wistar Kyoto rats), as well as individuals expressing PTSD symptoms, exhibit enhanced eyeblink conditioning. Behaviorally inhibited rodents also demonstrate enhanced avoidance responding (i.e., lever pressing). Both enhanced eyeblink conditioning and avoidance are most evident with schedules of partial reinforcement. Behaviorally inhibited individuals also performed better on reward and punishment trials than noninhibited controls in a probabilistic category learning task. Overall, the use of weaker situations with uncertain relationships may be more ecologically valid than learning tasks in which the aversive event occurs on every trial and may provide more sensitivity for identifying individual differences in learning for those at risk for, or expressing, PTSD symptoms.

Introduction

Most individuals experience a trauma event at some point in their lives, but very few develop posttraumatic stress disorder (PTSD). For example, the lifetime prevalence of trauma exposure is nearly 90% (Kilpatrick et al., 2013), but the lifetime prevalence of PTSD is less than 10% (Breslau et al., 1991; Kessler et al., 1995; Kilpatrick et al., 2013). Traumatic events, alone, would seem to have very little predictive validity for the development of PTSD in that the experiencing of a traumatic event is necessary, but not sufficient, for the development of PTSD. Thus, the experience of trauma is less important than who experiences the trauma.

Individual differences in personality temperaments can play a role in responses to stress and/or trauma. Yehuda and LeDoux (2007) put forth a call for research that focuses on identifying pre- and posttraumatic risk factors that explain the development of the disorder. Various individual differences have been put forth as risk factors for the development of PTSD, including age, gender, family history of psychopathology, cognitive factors (such as lower IQ), childhood adversity, preexisting avoidant personality, and poor social support (Bromet et al., 1998; DiGangi et al., 2013; Sayed et al., 2015). This review will focus on the effects of one specific risk factor, preexisting avoidant personality temperaments, which have been found to reveal individual differences in learning tasks interpreted through the perspective of strong versus weak situations.

Strong versus weak situations

Lissek et al. (2006) proposed applying the social psychological concept of strong and weak situations (Mischel, 1977; Monson & Snyder, 1977; Ickes, 1982) as a framework for identifying individual differences in studies of fear and anxiety. In this context, a strong situation would involve experimental parameters in which salient or aversive events are reliably predicted by unambiguous stimuli resulting in uniform responses with minimal individual differences. By contrast, a weak situation would involve more uncertain experimental parameters in which stimuli are less predictive of less-salient or less-aversive events, resulting in an increase in response variability indicative of individual differences. Lissek et al. (2006) provided examples from psychobiological responses, such as startle response (Grillon et al., 1998, b; Morgan et al., 1995) and skin conductance response (Pitman and Orr, 1986), in which participants expressed equal levels of fear to acute threats of shock (i.e., a strong situation) but differential psychophysiological responses to reduced levels of threat (i.e., a weak situation).

In addition to the factor of stimulus salience, Lissek et al. (2006) proposed that uncertainty may contribute to the sensitivity of weaker situations to identify individual differences between individuals with anxiety disorders and control samples. They argued that weaker situations characterized by some degree of uncertainty provide more opportunity to reveal individual differences, including learning patterns between patient populations and healthy controls.

Furthermore, Beckers et al. (2013) argued that weaker situations may reveal differences within nonclinical populations between individuals with low versus high vulnerability for the development of anxiety pathologies. Emmelkamp et al. (2014) further argued that the ability of experimental models of psychopathology to successfully differentiate between clinical and nonclinical populations, such as social anxiety disorder (Lissek et al., 2008) and panic disorder (Lissek et al., 2009; Michael et al., 2007), but not within nonclinical or subclinical populations who express risk factors may be due to the use of fear conditioning tasks, which can be interpreted as stronger situations rather than weaker situations. Pittig et al. (2018) also hypothesized that a stronger situation in which one stimulus precisely predicts a subsequent aversive event would result in almost all participants acquiring strong fear while a weaker situation in which the conditioning task involves some degree uncertainty may reveal differences between those individuals at risk and not at risk for developing an anxiety disorder.

Fear conditioning as a stronger situation

Most PTSD learning studies have utilized fear conditioning, which can generally be interpreted as a stronger situation in which a highly salient and aversive stimulus is reliably predicted by some cue. For example, Watson and Rayner (1920) demonstrated that a fear response could be classically conditioned with the classic example of Little Albert who learned to fear a white rat that was paired with a loud, aversive noise. The next century of work with fear conditioning has found that fear conditioning of autonomic responses (measured as changes in heart rate or skin conductance) to a conditioned stimulus (CS) predicting an aversive unconditioned stimulus (US) electric shock is facilitated in humans expressing PTSD symptoms (Orr et al., 2000; Peri et al., 2000; Blechert et al., 2007). Within this fear conditioning literature, impaired extinction of previously conditioned fear responses has emerged as a consistent learning mechanism for PTSD (Sijbrandij et al., 2013; Briscione et al., 2014; Zuj et al. 2016; Norrholm and Jovanovic, 2018; Zuj and Norrholm, 2019). In addition, an overgeneralization of fear to stimuli similar to those trained has been identified with anxiety and PTSD (Lissek and Grillon, 2015; Kaczkurkin et al., 2017; Cooper et al., 2022).

However, Lonsdorf and Merz (2017) detailed how much of the work on fear conditioning has focused on average responding, which revealed the basic and universal principles of fear conditioning resulting in a detailed understanding of the mechanisms of fear conditioning within the amygdala and related structures in both rodents and humans (LeDoux, 2000; Maren, 2001; Kim and Jung, 2006; Milad and Quirk, 2012; Fullana et al., 2020). This work did not necessarily identify individual differences between patient and control populations or between individuals who may or may not be at increased risk for the development and maintenance of PTSD. Lonsdorf and Merz (2017) argued that individual differences had been generally interpreted as noise compared with the signal of the overall group response, but more recent work has shifted focus from a previous focus on average responding uncovering general mechanisms of fear acquisition and extinction to individual difference factors (Bonanno and Mancini, 2012; Gazendam et al., 2015; Galatzer-Levy et al., 2017).

One recent study that did address individual differences in fear conditioning based on possible risk factors was done by Stegmann et al. (2019) who hypothesized that individual differences in fear generalization exist in a healthy population. They conducted a study by using differential fear conditioning phase with female faces as the CSs and a loud female scream US followed by a generalization test with morphs of the original faces in 20% steps. This analysis resulted in the identification of five clusters of individuals that systematically differed in skin conductance responses, differentiation between conditioned threat and safety, and linearity of the generalization gradients. Furthermore, the five clusters differed on psychometric measures, including trait-anxiety, anxiety sensitivity, agoraphobic cognitions, social anxiety, and behavioral inhibition, and lower scores in general self-efficacy. Thus, a nonclinical population exhibited individual differences in fear conditioning and generalization that can be attributed to individual differences in anxiety-related temperaments and tendencies. This finding also has been confirmed by Sep et al. (2019) who conducted a meta-analysis of anxious personality traits and fear generalization in healthy participants who indicated that trait anxiety increases vulnerability to anxiety disorders via fear generalization. However, neither of these studies addressed the issue of strong and weak situations.

Lonsdorf and Merz (2017) called for future experiments to test the theory of the weak situation for identifying individual differences in fear and anxiety. However, there has been very few fear conditioning studies exploring the concept of the weak situation with less salient stimuli and uncertainty. One fear paradigm that has received attention in manipulations of uncertainty, which can be interpreted as a weaker situation, is overgeneralization of fear in which a specific CS is trained with an aversive US. Following this training, other stimuli similar to the trained CS are presented as tests of how much the response to trained CS generalizes to the similar but untrained stimuli. In one very recent study by Zhao et al. (2022), participants were conditioned with a colored circle (CS+) paired with a shock at various reinforcement rates of 50%, 75%, or 100% and to an unpaired colored circle (CS−). The participants were then tested for generalization to a range of novel-colored circles that varied in hue along the green–blue dimension. The continuous reinforcement group showed a flatter generalization gradient, while the partial reinforcement groups showed increased generalization. However, these results were not discussed within the strong versus weak situation hypothesis. However, evidence supporting the strong versus weak situation has come from several fear conditioning studies that, while not directly addressing this hypothesis, have manipulated variables, including stimulus salience and uncertainty, which would render fear conditioning tasks as weaker situations.

Manipulations of stimulus salience in fear conditioning

For example, variables, such as alterations of cue salience, reducing shock intensity, or alternative aversive cues, have been studied (Duits et al., 2015). Tasks can be interpreted as relatively stronger or weaker based on stimulus salience and uncertainty. Fear conditioning can be interpreted as a stronger situation in that a highly salient aversive cue, such as an inescapable electric shock, is reliably preceded by a cue, such as a tone or light. However, shock intensity can be increased, which results in greater conditioned responding as well as slower extinction (Phillips and LeDoux, 1992). In Mobbs et al. (2019), Michael Fanselow hypothesized that a very weak shock produces more flexibility in a rat’s behavior, which is lost as the shock intensity is increased. This loss of flexibility is interpreted as a fear state. In a meta-analysis of classical fear conditioning (Duits et al., 2015) provides examples of shock and nonshock stimulus (picture, sound, odor, air blast to the throat) in which the shock stimulus has a greater effect on acquisition. However, there were no overall differences between patient and control participants in acquisition of fear responses to the CS+, which may indicate a ceiling effect based on the aversiveness of the stimuli. This ceiling effect could be interpreted as the highly salient cues resulting in a loss of the ability to detect individual differences due to these tasks being stronger situation. Thus, the stronger salience cues typically used in fear conditioning may limit the ability to identify individual differences in the context of anxiety and PTSD.

Manipulations of uncertainty in fear conditioning

In addition to alterations in cue salience as a method for producing strong versus weaker learning situations, manipulations of uncertainty, such as schedules partial reinforcement have been studied in fear conditioning. Typically, the use of schedules of partial reinforcement with fear conditioning has been utilized for the purpose of extending the time required to acquire (Ho and Lipp, 2014) or extinguish (Phelps et al., 2004; Tronson et al., 2012) conditioned fear. Lonsdorf and Merz (2017) explained that often it is desirable for fear extinction to take longer when extinction is the focus of the experiment, because CRs typically extinguish very rapidly when shifting from continuous (100%) reinforcement during CS-US paired acquisition training to CS alone trials during extinction training (Vansteenwegen et al., 1998). However, schedules of partial reinforcement also have the undesired effect of reducing contingencies between stimuli and have been found to weaken the development of the behavioral fear responses, such as freezing (Cain et al., 2005). In addition, schedules of partial reinforcement reduce response frequency (Flora and Pavlik, 1990; Svartdal, 2003; Huang et al., 1992), as well as CR amplitudes (Dunsmoor et al., 2007). Furthermore, schedules of partial reinforcement reduce physiological measures, such as resulted in reduced differential skin conductance response (Grady et al., 2016), as well as reduced learning-related responses in the amygdala in humans (Dunsmoor et al., 2007). Because many fear conditioning studies sought to elucidate the details of the neural circuitry underlying fear conditioning, it becomes obvious why they did not include schedules of partial reinforcement and tended to avoid less-salient, aversive cues due to a desire maximize behavioral, physiological, and neural responses to fear conditioning. Thus, support for strong versus weak situations from fear-conditioning studies is limited. However, some recent evidence supporting the idea of weaker situations revealing individual differences has come from work that utilized learning tasks other than fear conditioning. These tasks have all included some aspects of uncertain relationships between stimuli as represented in Fig. 1.

Fig. 1

Schematic representation of the three learning tasks interpreted as weaker situations. Weaker situations involve some degree of uncertainty which results in the identification of individual differences in fear and anxiety studies. Uncertainty is manipulated in these tasks through the use of schedules of partial reinforcement in which only half of the training trials were paired trials and probabilistic rather than deterministic relationships. The uncertainty aspect of a weaker situation reveals individual differences in the form of enhanced learning for individuals with PTSD symptoms as well as those with avoidant personality temperaments which are risk factors for the development of PTSD

Three weaker situation tasks

In the work reviewed here, the learning tasks that involve some degree of uncertainty include eyeblink conditioning, avoidance learning, and a computer-based task. Specifically, the eyeblink and avoidance tasks involve schedules of partial reinforcement in which the stimuli are paired on only half of the training trials. The computer-based, categorization task involves probabilistic categories in which an item is a member of that category only 80% of the time rather than a deterministic relationship in which an item is always a member of the same category.

As mentioned previously, this review will focus on one type of risk factor for PTSD, avoidant personality temperaments. Specifically, behavioral inhibition or BI has been the focus of a series of experiments in humans and rodents in the past decade or so that have utilized these learning tasks that can be interpreted as weaker situations. BI is defined as a relatively stable temperamental tendency to avoid or withdraw from the unfamiliar (Kagan et al., 1987; Morgan, 2006). In the work reviewed here, enhanced learning with behaviorally inhibited humans and rodents. This work will detail results from both nonclinical populations (i.e., undergraduates), expressing behavioral inhibition, as well as clinical populations, including veterans and military personnel exhibiting PTSD symptom that have come from studies utilizing uncertainty and weak learning situations.

Weaker Task 1: classical eyeblink conditioning

The first learning paradigm to be discussed is classical eyeblink conditioning, which has a long history in the study of anxiety and PTSD. Eyeblink conditioning is a relatively benign paradigm, combining a neutral tone conditioned stimulus (CS) with a mild air puff to the eye or periorbital shock unconditioned stimulus (US) that elicits an eyeblink response (i.e., the conditioned or unconditioned response; CR or UR). Following several pairings of the CS and US, the CS comes to elicit a CR or an eyeblink that occurs before the onset of the US. Eyeblink conditioning studies involve either delay or trace conditioning. In delay conditioning, the CS precedes and co-terminates with the US while in trace conditioning, the CS precedes the US, but terminates before the onset of the US. Thus, there is a trace period between the CS and US when no stimulus is present. The work presented in this review will focus on delay conditioning.

Most eyeblink conditioning studies in human and nonhuman animals involve nearly 100% paired trials with the exception of an occasional CS alone test trial. In other words, there is a strong contingency or predictive relationship between stimuli. Eyeblink conditioning can be made a weaker situation by the addition of some form of uncertainty based on alterations of contingency or the predictive relationship between stimuli. Uncertainty can be introduced by the use of a schedule of partial reinforcement in which only a portion of the trials involve paired training and the remainder involve presentations of an unpaired stimulus. Based on the hypotheses of weak versus strong situations put forth by several researchers (Lissek et al., 2006; Beckers et al., 2013; Emmelkamp et al., 2014), the utilization of schedules of partial reinforcement should increase the identification of individual differences in those expressing PTSD symptoms as well as those at increased risk for developing PTSD symptoms based on personality temperaments.

Eyeblink conditioning has a long history in the study of anxiety effects on learning starting with seminal human eyeblink conditioning studies by Kenneth Spence and colleagues (Spence and Taylor; 1951, 1953; Taylor, 1951, 1956; Farber and Spence, 1953; Spence and Farber, 1953, 1954; Spence and Beecroft, 1954; Spence and Weyant, 1960; Spence and Spence, 1966). Specifically, anxiety was measured by using the Manifest Anxiety Scale (MAS; Taylor, 1953), a 50-item, true/false inventory, which assessed the personality trait of overt or conscious anxiety. Using the MAS, healthy, college-aged individuals in the top quartile of scores were classified as high MA, whereas individuals in the lowest quartile were classified as low MA. These studies consistently reported that high MA individuals exhibited facilitated eyeblink classical conditioning compared with low MA individuals. Spence and Spence (1966) interpreted these differences in terms of MA being “an emotionally based drive” suggesting that anxious participants have increased drive to detect stimuli and produce responses necessary for conditioning to occur.

Recently, eyeblink conditioning studies of anxiety have been renewed and expanded to explore PTSD with more modern constructs of anxiety. One such construct that has been compared to manifest anxiety (Spielberger and Rickman, 1988) is trait anxiety, which is measured by the State Trait Anxiety Inventory (STAI; Spielberger et al., 1983). The STAI assesses trait anxiety, which is a relatively stable temperament involved in how an individual perceives and responds to anxiety separately from state anxiety, which is a participant’s current anxiety level that can change with situations. Trait anxiety is related to PTSD in that individuals with high-trait anxiety are more likely to display hypervigilance (Eysenck and Calvo, 1992), whereas higher-trait anxiety is exhibited by individuals with PTSD symptoms than individuals without PTSD symptoms (Orsillo et al., 1996; Casada and Roache, 2005, 2006).

Much like the findings of individual differences in eyeblink conditioning from studies with manifest anxiety, humans expressing trait anxiety exhibit faster acquisition of conditioned eyeblinks than those not expressing trait anxiety (Holloway et al., 2012; Caulfield et al., 2013). This work with trait anxiety and eyeblink conditioning was extended to include a related construct, behavioral inhibition (BI), which is positively correlated to trait anxiety (Caulfield et al., 2013).

Individual differences based on behavioral inhibition

As mentioned previously, behavioral inhibition is defined as a relatively stable temperamental tendency to avoid or withdraw from the unfamiliar (Kagan et al., 1987; Morgan, 2006). In the work described, BI has been measured with the Adult Measure of Behavioral Inhibition (AMBI; Gladstone and Parker, 2005), which measures an individual’s tendency to avoid new stimuli or social situations. Childhood levels of BI are measured with the Retrospective Measure of Behavioral Inhibition (RMBI; Gladstone and Parker, 2005), which measures childhood memories of avoiding unfamiliar situations. Carr et al. (1997) suggested that avoidant coping strategies are a core vulnerability for how behaviorally inhibited individuals develop into PTSD.

Behaviorally inhibited adolescents (Caulfield et al., 2015) as well as young adults (Caulfield et al., 2013; Holloway et al., 2014) exhibited enhanced delay eyeblink conditioning. These studies also demonstrated that this enhanced conditioning was not due to differences in the reflexive responding to the air puff US. Thus, anxiousness (i.e., state anxiety) or responsivity to aversive stimuli were not responsible for observed differences in eyeblink acquisition. These findings with nonclinical samples of adolescents and undergraduates were extended by Myers et al. (2012a) who found that veterans who self-reported high levels of childhood BI as measured by the RMBI exhibited faster acquisition of conditioned eyeblinks. Myers et al. (2012b) had previously found that veterans with high scores on the AMBI and RMBI had more severe PTSD symptoms. Thus, behavioral inhibition and PTSD symptoms have been found to replicate the enhanced eyeblink conditioning previously reported for manifest and trait anxiety with the weak situation learning task of eyeblink conditioning.

A rodent model of behavioral inhibition also has been used to further explore the individual differences in learning based on behavioral inhibition. The inbred Wistar Kyoto (WKY) rat strain exhibits inhibited temperaments compared with outbred Sprague Dawley (SD) rats, which do not express inhibited temperament (for a detailed review, see Jiao et al., 2011a). In brief, WKY rats exhibit several behavioral tendencies linked to PTSD, including hypervigilance as measured by an increased acoustic startle response (Glowa and Hansen, 1994; Servatius et al., 1998; McAuley et al., 2009), and greater pre-pulse inhibition (Conti et al., 2002; McAuley et al., 2009), and a deficit in open field activity (Pare, 1994; Ferguson and Cada, 2004; McAuley et al., 2009). WKY rats also express exaggerated autonomic responses to stress that were similar to findings from behaviorally inhibited children (Smoller et al., 2003, 2005).

In addition to these individual differences in inhibited behavioral tendencies, WKY rats acquire conditioned eyeblink responses faster and to a greater degree than noninhibited SD rats with a tone CS and a periorbital shock US (Beck et al., 2011; Ricart et al., 2011; Janke et al., 2015). Well-trained WKY rats also exhibit slower extinction on CS alone trials compared with SD controls (Beck et al., 2011). These findings that WKY rats exhibit enhanced delay eyeblink conditioning fit with findings of enhanced eyeblink conditioning with various behaviorally inhibited populations, including adolescents, undergraduates, and veterans.

Overall, the studies highlighted indicate that individual differences in eyeblink conditioning are evident in humans expressing various levels of anxiety (i.e., manifest and trait anxiety) or humans and rodents expressing avoidant personality temperaments (i.e., behavioral inhibition). In addition to these explorations of individual differences in classical eyeblink conditioning that are based on avoidant personality temperaments, such as manifest anxiety, trait anxiety, and behavioral inhibition, recent studies have extended this idea of a weak situation to include some aspect of uncertainty.

Uncertainty in eyeblink conditioning

One uncertain form of eyeblink conditioning that has been tested with behaviorally inhibited individuals and rodents involves an avoidance omission task in which the US is omitted on trials in which a CR was performed to the CS. This omission eyeblink conditioning task includes three groups: a delay group with all CS-US paired trials; an omission group in which the US is omitted on trials on which a CR was performed to the CS; and a yoked control group that received the same schedule of CS and US delivery as a member of the omission group who had the same level of behavioral inhibition. Thus, the delay group had no uncertainty in that every trial was an CS-US paired trial, the omission group had control of the US in that the performance of a CR on a particular trial resulted in the omission of the US on that trial, and the yoked controls had a high degree of uncertainty in that they received a random trial order of some CS-US paired trials and some CS alone trial. A finding of more CRs in the omission group compared with the yoked controls is evidence of avoidance responding in the omission group.

This omission form of eyeblink conditioning was initially tested with behaviorally inhibited WKY rats (Ricart et al., 2011), which performed more conditioned responses in the omission condition than yoked controls. This pattern was not observed in SD controls, which did not differ in conditioned responding between the omission and yoked situations. This finding was interpreted as evidence of WKY, but not SD, rats performing avoidance responses in the omission group. This finding also was interpreted such that behaviorally inhibited WKY rats express a greater sensitivity to negative reinforcement or inhibition from partial reinforcement enhanced avoidance than noninhibited SD rats.

Based on the findings of avoidance responses in an eyeblink conditioning omission task by behaviorally inhibited WKY rats, but not noninhibited SD rats, this task was subsequently tested with behaviorally inhibited individuals (Holloway et al., 2014). Overall, behaviorally inhibited individuals exhibited enhanced acquisition of eyeblink CRs compared with noninhibited individuals. Individuals in the omission and yoked control groups also exhibited less CRs than individuals in the 100% CS-US paired training condition; however, there were no differences in CR performance between omission or yoked training. Thus, unlike WKY rats, behaviorally inhibited individuals did not exhibit avoidance responding in the omission eyeblink conditioning task. However, this human eyeblink conditioning omission study resulted in a novel finding that the enhanced acquisition of conditioned eyeblinks was more evident in the omission and yoked conditions than in the 100% CS-US paired training condition. The finding of enhanced learning in these uncertain conditions was theorized by Holloway et al. (2014) to be due to the effects of situations in which a random number of the trials were tone alone rather than CS-US paired presentations based on the individual patterns of conditioned responding in the omission condition, which resulted in schedules of partial reinforcement.

Much like the early work on anxiety, eyeblink conditioning studies from the 1950s and 1960s tested the effects of schedules of partial reinforcement with humans. These human eyeblink conditioning studies of partial reinforcement intermixed CS (tone) alone trials into CS-US paired training based on the US air puff being defined as the reinforcing event in eyeblink conditioning (Leonard and Theios, 1967). Human eyeblink studies of schedules of partial reinforcement have reported variable results. Whereas some studies produced slower acquisition of conditioned eyeblink with schedules of partial reinforcement with CS alone trials compared with 100% CS-US training (Reynolds, 1958; Ross, 1959; Hartman and Grant, 1960; Ross and Spence, 1960; Runquist, 1963; Perry and Moore, 1965), other studies found no detrimental effects of CS alone schedules of partial reinforcement (Humphreys, 1939; Grant et al., 1950; Hake and Grant, 1951; Grant and Schipper, 1952; Moore and Gormezano, 1961; Price et al., 1965; Foth and Runquist, 1970). Thus, schedules of partial reinforcement would be expected to slow CR acquisition or have no effect on acquisition compared with the 100% CS-US training.

Based on the finding of greater enhancement of conditioning by behaviorally inhibited individuals in the omission and yoked conditions, the effects of specific schedules of partial reinforcement on eyeblink conditioning with behaviorally inhibited individuals were tested. Specifically, Allen et al. (2014) tested the effects of specific schedules of partial reinforcement in which half the trials were either CS alone or US alone trials. Both schedules of partial reinforcement degraded CR acquisition compared with 100% CS-US training. However, behaviorally inhibited individuals exhibited enhanced acquisition of conditioned eyeblinks with a medium effect size in both schedules of partial reinforcement compared with noninhibited individuals.

As shown in Fig. 2, the enhancements observed with the schedules of partial reinforcement were greater than the enhancement observed with 100% CS-US training. In addition to enhanced acquisition, behaviorally inhibited individuals exhibited a small, but significant, effect of slowed extinction (i.e., more CRs during CS alone trials) than noninhibited individuals when training was switched from the 50% CS partial reinforcement schedule to extinction training with 100% CS alone trials. This finding was consistent with prior reports (Longenecker et al., 1952; Perry and Moore, 1965; Newman, 1967; Leonard, 1975) that CS alone training produces a slowing of extinction termed a partial reinforcement extinction effect or PREE when CS alone trials are presented following CR acquisition with a schedule of partial reinforcement that included tone alone trials. This slowed extinction was similar to that reported for behaviorally inhibited individuals conditioned with 100% CS-US paired trials (Beck et al., 2011).

Fig. 2

Acquisition and extinction of conditioned eyeblinks with a tone CS and air puff US in behaviorally inhibited and non-inhibited individuals. A Behaviorally inhibited and non-inhibited individuals did not differ in CR acquisition with 100% CS-US paired trials or extinction with CS alone training. B Behaviorally inhibited individuals expressed more CRs than non-inhibited individuals during acquisition with a 50% CS alone schedule of partial reinforcement and extinction with CS alone trials. C Behaviorally inhibited individuals expressed more CRs than non-inhibited individuals during acquisition with a 50% US alone schedule of partial reinforcement but did not differ during CS alone extinction trials. Figure adapted from Allen et al. (2014)

This line of human eyeblink conditioning with schedules of partial reinforcement with behaviorally inhibited individuals (i.e., undergraduates) has been extended to individuals with PTSD symptoms. A delay eyeblink conditioning task with a 50% CS alone schedule of partial reinforcement was tested with active-duty Coast Guard personnel exhibiting PTSD symptoms and veterans self-reporting symptoms of PTSD (Handy et al., 2018, 2020). Both military personnel and veterans identified with PTSD symptoms exhibited a medium effect of more conditioned eyeblinks than non-PTSD personnel, which fits with the findings from behaviorally inhibited individuals (Allen et al., 2016). In addition to enhanced acquisition, Handy et al. (2018) found that personnel with PTSD symptoms exhibited more conditioned responses to tone-alone extinction trials with a medium effect size, which resulted in slower extinction (i.e., PREE) than non-PTSD personnel, which supports findings of slowed extinction in behaviorally inhibited individuals (Beck et al., 2011; Allen et al., 2014).

Overall, schedules of partial reinforcement with eyeblink conditioning revealed enhanced acquisition and slowed extinction of conditioned eyeblinks in behaviorally inhibited individuals and individuals expressing PTSD symptoms. Thus, schedules of partial reinforcement can account for the enhanced learning observed in omission and yoked training (Holloway et al., 2014). Overall, it appears that uncertainty concerning contingency between stimuli reveals individual differences in learning for some individuals or organisms, particularly those with inhibited or avoidant temperaments, as well as symptoms of PTSD through a weaker learning situation.

In addition to the effects of schedules of partial reinforcement, another possible explanation for the enhanced acquisition observed by Holloway et al. (2014) in behaviorally inhibited individuals trained in the omission and yoked conditions is uncertainty in trial timing. The omission of the US on some trials would result in variations in the timing between CS-US paired trials termed the intertrial interval or ITI, which could alter the learning rates of members of the yoked control group. Specifically, the lengthening of ITIs has been reported to enhance acquisition of conditioned eyeblinks as well as slowed extinction to subsequent CS alone training in humans (Spence and Norris, 1950; Prokasy et al., 1958; Prokasy, 1965).

Allen et al. (2016) investigated the effects of extending and varying the ITI for eyeblink conditioning with behaviorally inhibited individuals. Participants were trained with either a short ITI (range 25-35 s), a long ITI (range 52-62 s), or a variable long ITI (range 25-123 s). Overall, behaviorally inhibited individuals exhibited enhanced conditioned eyeblink responses compared with noninhibited individuals. Training with a variable long ITI, but not the short or long ITI, produced a medium sized effect of enhanced acquisition of CRs.

This finding indicates that making the conditioning parameters more uncertain by varying the ITI is more effective in identifying enhanced acquisition of conditioned eyeblinks in behaviorally inhibited individuals than simply extending the ITI. Thus, uncertainty in the form of variability in trial timing as well as schedules of partial reinforcement may be better at revealing enhanced learning exhibited by behaviorally inhibited individuals than standard paired training with consistent trial timing. The manipulations of classical eyeblink conditioning with schedules of partial reinforcement as well as variable trial timing have attempted to accentuate individual differences in associative learning with a weak situation. Whereas these learning enhancements generalize across several unique populations of participants, ranging from adolescents and undergraduate students to active duty military and veteran populations, these results strongly suggest future studies include similar uncertain learning conditions to maximize the utility of the eyeblink conditioning paradigm in the study of stress and anxiety.

Overall, these weaker learning situation studies utilizing uncertainty, including omission and yoked training, schedules of partial reinforcement, and varying trial timing provide further support for the usefulness of uncertainty in identifying individual differences in associative learning. Given that an enhancement of acquisition of conditioned eyeblink as well as a slowing of extinction has been shown to generalize across several unique cohorts of participants, ranging from undergraduate students to active duty military and veteran populations, these results strongly suggest future studies impose similar weak situation learning conditions involving uncertainty to maximize the utility of learning paradigms, such as eyeblink conditioning in the study of individual differences in learning in those at risk for PTSD or expressing symptoms of PTSD.

Weaker Task 2: avoidance learning

A second example of uncertainty being applied to produce a weaker task, which has revealed individual differences in learning associated with behavioral inhibition, is avoidance learning with rodents. In this avoidance task, a tone warning signal precedes an aversive outcome, such as an electric foot shock. Initially, the rat learns to make an escape response (i.e., a lever press) that terminates the aversive foot shock. As the rat learns the association between the warning signal and the shock, it learns to make an avoidance response by pressing the lever during the warning signal, but before the onset of the shock, which omits the shock for that trial.

Individual differences based on behavioral inhibition

A balance of escape or avoidance responses is expressed based on dispositions of different rat strains. Thus, the avoidance task reveals individual differences in avoidance learning. Specifically, behaviorally inhibited WKY rats avoid faster and to a greater degree than noninhibited SD rats (Servatius et al., 2008; Beck et al., 2010; Jiao et al., 2011b; Perrotti et al., 2013). WKY rats do not express greater pain sensitivity than SD rats to foot shock. Therefore, the enhanced avoidance evident in WKY rats appears to be due to differences in associative and avoidance learning (Servatius et al., 2008; Fragale et al., 2016; Spiegler et al., 2018). These findings of enhanced avoidance learning with WKY rats came about from 100% tone-shock paired trails. However, this avoidance task can be made a weaker situation by the inclusion of a manipulation of uncertainty.

Uncertainty in avoidance learning

Based on the combined findings of enhanced avoidance learning in WKY rats and enhanced acquisition and slowed extinction of conditioned eyeblinks in behaviorally inhibited individuals trained with schedules of partial reinforcement, Miller et al. (2020) examined the effects of a schedule of partial reinforcement on the acquisition and expression of avoidance in WKY and SD rats. Standard training with a fixed 60-s tone which predicted shock on 100% of trials was compared with a schedule of partial reinforcement in which the tone was presented without the foot shock on half of the trials using a pseudorandom schedule based on the trial parameters of Allen et al. (2014). This schedule of partial reinforcement introduced an element of uncertainty in that was not apparent on each trial whether the tone would be followed by the shock. Overall, there was a large effect in that WKY rats acquired avoidance responses more rapidly and at higher asymptotic levels than SD rats, which replicated prior findings of WKY rats trained with 100% tone-shock trials (Servatius et al., 2008; Beck et al., 2010; Jiao et al., 2011b; Perrotti et al., 2013).

Both WKY and SD rats exhibited reduced avoidance responding in the 50% tone-alone condition compared with the performance of each strain in the 100% tone-shock condition. This finding is consistent with the results of avoidance studies of partial reinforcement (Davenport et al., 1971; Olson, 1971; Olson et al., 1971), as well as classical eyeblink conditioning studies with schedules of partial reinforcement (Reynolds, 1958; Ross, 1959; Hartman and Grant, 1960; Ross and Spence, 1960; Runquist, 1963; Perry and Moore, 1965; Allen et al., 2014; Allen et al., 2018a, b). In the 50% tone alone schedule, WKY rats avoided at robust levels, whereas SD rats did not acquire significant levels of avoidance and escaped rather than avoided the shock. As shown in Fig. 3, there was a large effect in that WKY rats performed better in the 50% tone-alone condition than SD rats in the 100% tone-shock condition.

Fig. 3

Acquisition of avoidance lever press responses with behaviorally inhibited Wistar Kyoto (WKY) rats and non-inhibited Sprague Dawley (SD) rats. WKY rats trained with a schedule of partial reinforcement with 50% tone alone trials expressed more avoidance responses than SD rats trained with 100% tone shock trials. Figure adapted from Miller et al. (2020)

When the total number of tone-shock pairings experienced early in training were accounted for WKY rats in the 50% tone alone condition and WKY rats in the 100% tone-shock condition did not differ in the level of avoidance exhibited on the last day of training. Therefore, WKY rats appear less sensitive to the detrimental effects of a schedule of partial reinforcement than SD rats.

Schedules of partial reinforcement revealed individual differences that were similar to the findings from the eyeblink conditioning studies in that behaviorally inhibited WKY rats enhanced avoidance learning. While a schedule of 50% CS alone trials reduced overall avoidance responding, WKY rats exhibited similar levels of avoidance in both continuous and partial schedules of reinforcement while noninhibited SD controls failed to achieve avoidance. The uncertainty inherent in a schedule of partial reinforcement revealed individual differences in the form of a bias that behaviorally inhibited WKY rats tend to avoid based on the expectation of the impending shock, whereas the noninhibited SD rats tend to perform escape responses based on the actual experience of the shock.

Overall, uncertainty produced by the inclusion of schedules of partial reinforcement has resulted individual differences in that behaviorally inhibited organisms exhibit enhanced eyeblink conditioning and enhanced avoidance learning. Both of these learning tasks involve associative learning with schedules of partial reinforcement, which have revealed greater individual differences than those evident in 100% paired training. Another form of uncertainty beyond partial reinforcement also has been explored with behaviorally inhibited individuals completing a computer-based learning task.

Weaker Task 3: probabilistic category learning

A third learning task that has been applied to the study of behavioral inhibition and PTSD is a computer-based category learning task in which a simple keystroke serves as the response resulting in a loss or gain of points rather than a physically aversive stimulus. While some avoidance studies in humans have utilized computer-based tasks in which the aversive event is a mild electric shock (Lovibond et al., 2008; 2009, 2013; Delgado et al., 2009) and an unpleasant visual or auditory stimulus (Dymond et al., 2011) that can be interpreted as a stronger situation, other computer-based tasks have used less aversive outcomes, such as a loss of points (Molet et al., 2006; Sheynin et al., 2013) or money (Schlund et al., 2011), which can be interpreted as a weaker situation. One computer-based avoidance task, which included both acquisition and extinction of avoidance responses, has been tested with anxiety vulnerable populations (Sheynin et al., 2013, 2014a) as well as veterans with PTSD symptoms (Sheynin et al., 2017). This computer-based task can further be interpreted as a weaker situation; it also included some degree of uncertainty, including probabilistic relationships, as well as ambiguous feedback.

Uncertainty in category learning

A computer-based task in which points are gained or lost was used to explore how behaviorally inhibited individuals make decisions or strategize to maximize benefit and minimize loss in a learning task with some degrees of uncertainty. Sheynin et al. (2013) used a computer-based categorization task (Bodi et al., 2009) in which participants learn to classify shapes into two categories to obtain a reward (i.e., gain points) or to avoid a punishment (i.e., lose points). The task had some degree of uncertainty based on the probabilistic nature of the categories. Membership of any individual shape in one of two categories was probabilistic in that a particular shape was a member of a category only 80% of the time. That is, if an individual categorized a particular shape into the correct category on every trial, they would be correct on 80% of the trials and incorrect on 20% of the trials.

In addition to the probabilistic nature of categories, there is some ambiguity or uncertainty based on the lack of feedback provided on some trials. When a shape is correctly categorized on a reward trial, the individual receives feedback that they were correct and have gained 25 points. However, no feedback is provided on incorrect reward trials. When a shape is incorrectly categorized on a punishment trial, the individual receives feedback that they were incorrect and have lost 25 points. No feedback is provided for a correct response on punishment trials. Thus, there is a degree of uncertainty in that no feedback could be interpreted by the participant as either an indication of failure to receive a reward or successful avoidance of a punishment. This ambiguous feedback allows for the exploration of the tendencies of behaviorally inhibited individuals to continue making previously rewarded responses or explore new responses (Myers et al., 2013).

Sheynin et al. (2013) found that behaviorally inhibited undergraduates expressed both enhanced reward and punishment learning in this probabilistic category learning task. These findings differed slightly from the results of Bodi et al. (2009) in which reward learning was correlated with the personality trait of novelty seeking while punishment learning was associated with the personality trait of harm avoidance (HA), which has a moderate positive relationship to BI (Allen et al., 2017). HA is defined as a tendency to respond strongly to aversive stimuli and learn to avoid punishment, novelty, and nonreward (Nixon and Parsons, 1989), as well as excessive worrying (Cloninger, 1986). Thus, both BI and HA are expressed through similar behaviors, such as avoidance and worrying in the face of uncertainty about aversive outcomes, which may have resulted in enhanced learning exhibited in the probabilistic category learning task.

Furthermore, Sheynin et al. (2013) explored whether behaviorally inhibited individuals would withdraw or opt out of a trial if given the option. Behaviorally inhibited individuals opted out more than noninhibited individuals overall. This pattern of behavior was evident on punishment, but not reward, trials. Thus, behaviorally inhibited individuals would rather skip a punishment trial with no points loss rather than take the chance of losing points. Thus, behaviorally inhibited individuals prefer to avoid uncertain situations in which an aversive outcome is possible (i.e., a loss of points). Thus, individual differences between behaviorally inhibited and noninhibited individuals were evident in a weaker situation task in which the aversive outcome was only a possible loss of points, the relationships between stimuli and categories was uncertain (i.e., probabilistic), and feedback was ambiguous on some trials.

The finding of enhanced reward and punishment learning in a nonclinical undergraduate sample exhibiting behavioral inhibition were extended to veterans expressing PTSD symptoms (Myers et al., 2013) who outperformed veterans without PTSD symptoms on reward, but not punishment, trials. Myers et al. (2013) further analyzed the patterns of responding from the veterans using a computational model of reinforcement learning adapted from a gain loss model (Frank et al., 2007) to simulate participant’s responses to uncertain feedback. Estimations of the reinforcement value of the no-feedback outcome were significantly greater in the veterans without PTSD symptoms than the veterans with PTSD symptoms. This finding suggests that the veterans without PTSD symptoms were more likely to interpret this uncertain outcome as positively reinforcing (i.e., signaling successful avoidance of punishment), whereas veterans with PTSD symptoms interpreted the uncertain outcome as more neutral. One prediction of the Myers et al. (2013) model is that task manipulations that vary the relative value of uncertain feedback might affect the pattern of behavior observed in veterans with severe PTSD symptoms versus those without PTSD symptoms. However, the idea that veterans with severe PTSD symptoms tend to value ambiguous feedback differently than veterans with few or no PTSD symptoms suggests a mechanism through which uncertainty can contribute to the facilitated associative learning often observed in PTSD patients.

The effects of probabilistic relationships on behaviorally inhibited individuals and individuals expressing PTSD symptoms in this computer-based probabilistic category learning task align with the enhanced learning with schedules of partial reinforcement evident with classical eyeblink conditioning and avoidance learning. The uncertainty of the probabilistic relationships between objects and categories as well as the ambiguity in a lack of feedback on some responses can be interpreted as a weaker learning situation, which resulted in enhanced learning in behaviorally inhibited individuals and individuals expressing PTSD symptoms.

In general, tasks that involve some degree of uncertainty have been found to produce a stereotypic enhancement of classical eyeblink conditioning, avoidance learning, and a probabilistic category learning task. These combined findings support future studies utilizing these types of task along with schedules of partial reinforcement or probabilistic relationships for identifying individual learning differences in clinical and nonclinical populations involving anxiety and PTSD.

Shared neural substrates of learning, PTSD, and uncertainty

Beyond the utility of the three tasks that have manipulated uncertainty for exploring individual differences in behavioral responding, these tasks offer the opportunity of exploring the underlying neural substrates and the mechanisms through uncertainty can modulate learning by anxiety vulnerable individuals and those expressing PTSD symptoms. As previously described, the dominant learning paradigm for exploring PTSD has been fear conditioning. Ledoux and Pine (2016) have promoted fear conditioning as having the potential benefit for developing treatments for anxiety disorders based on the conservation of brain-behavior relationships across species. The neural substrates of fear conditioning and PTSD have been well defined in the amygdala and associated regions (Maren, 2001). Neural substrates associated with the three learning tasks detailed in this review, eyeblink conditioning, avoidance learning, and probabilistic category learning also are associated with PTSD as well as uncertainty.

For example, the detailed understanding of the behavioral parameters of eyeblink conditioning and the consistency of neural substrates across mammalian species, which allow for the direct comparison of results from animal models to both nonclinical and clinical populations. Thus, eyeblink conditioning has been put forth as an “exceptional platform for the integrative clinical-cognitive-neuroscience investigation into psychopathology” (McFall et al., 2002, p. 260). Our detailed understanding of the neural substrates for eyeblink conditioning can provide possible mechanisms through which the effects of uncertainty can be expressed behaviorally through enhanced associative learning with these weak situation tasks leading to the development of anxiety disorders and PTSD. While less detailed, the neural substrates underlying avoidance learning and probabilistic category learning will be discussed.

As will be detailed, neural substrates found to underlie the learning of eyeblink conditioning, avoidance, and probabilistic categories also have been identified in PTSD and uncertainty.

Cerebellar substrates of eyeblink conditioning and avoidance learning

One of the strengths of applying classical eyeblink conditioning to the study of PTSD is the detailed understanding of its neural substrates. Since the 1980s, Richard F. Thompson and colleagues identified the neural circuits in the cerebellum and brainstem, which are required for the acquisition and expression of conditioned responses in delay eyeblink conditioning (for detailed reviews of this work refer to Thompson and Steinmetz, 2009; Freeman, 2015). The studies establishing the cerebellum as the site of plasticity for delay eyeblink conditioning across several mammalian species were done initially in rabbits but have since been replicated in rats (Lee and Kim, 2004) and most importantly in humans (Gerwig et al., 2007). Specifically, a cerebellar role in eyeblink conditioning has been confirmed through functional MRI brain imaging in humans (Dimitrova et al., 2002; Cheng et al., 2008; Thurling et al., 2015; Kent et al., 2020).

The neural substrates of lever press have not been so widely explored as those of eyeblink conditioning. However, there is some evidence that avoidance involves the cerebellum in that removal of the cerebellum in rats has been found to disrupt passive avoidance conditioning in a shuttle box task (Dahhaoui et al., 1990; Guillaumin et al., 1991). Selective ibotenic acid lesions of the deep cerebellar interpositus nucleus prevented the learning of the lever press avoidance task but had no effect on an appetitive version of the lever press task (Steinmetz et al., 1993). Further study of the role of the cerebellum in avoidance learning is needed to explore the neural substrates of enhanced eyeblink conditioning and avoidance learning.

Cerebellar role in PTSD

The cerebellum is typically considered a motor structure involved in balance, coordination, monitoring and correcting ongoing movements, and associative learning, such as classical eyeblink conditioning. However, recent work has found a role for the cerebellum in affective and cognitive effects (Schmahmann, 1998, 2010; Bergmann, 2000; Strick et al., 2009) that are involved in anxiety disorders and PTSD (Caulfield and Servatius, 2013; Carletto and Borsato, 2017; Hilber et al., 2019). Strong support for a cerebellar role in PTSD has come from functional and structural imaging studies that have identified a possible role for the cerebellum in PTSD (Fernandez et al., 2001; Bonne et al., 2003; Blithikioti et al., 2022). Specifically, individuals with PTSD have been found to exhibit cerebellar hyperactivity, including increased resting state activity (Wang et al., 2016), as well as increased cerebellar blood flow at rest and in response to threat-related stimuli (Bremner et al., 1999, 2003; Osuch et al., 2001; Bonne et al., 2003; Pantazatos et al., 2012). In addition, patients with PTSD exhibit smaller cerebellar volume (Baldacara et al., 2011; Sui et al., 2010; Sussman et al., 2016), as well as altered connectivity between cerebellum and the cerebral cortex and hippocampus (Rabellino et al., 2018). Thus, these changes in the cerebellum in PTSD may underlie the differences observed with eyeblink conditioning and avoidance learning in behaviorally inhibited individuals and those expressing symptoms of PTSD.

Cerebellar role in uncertainty

Beyond a role in PTSD, the cerebellum also has been implicated with uncertainty. Dreher and Grafman (2002) observed increasing activation of the cerebellum in random relative to fixed timing in a task-switching experiment. In addition, there is speculation for a cerebellar role (specifically the left medial and lateral cerebellum Lobule VI) in decision making in uncertain conditions (Blackwood et al., 2004). Jakobs et al. (2009) theorized that cerebellar activation may increase as temporal uncertainty increased. Guo et al. (2013) found cerebellar activation in a reward task involving uncertainty when making a risky decision. Gray matter volume of the left cerebellum predicted risk taking behavior and risk tolerance more so than the amygdala (Quan et al., 2022). Specifically, the left cerebellar volume may be associated with decisions when the probabilities are not explicit (i.e., uncertain) and must be determined through experience.

Overall, there is convergent imaging and clinical evidence that eyeblink classical conditioning, avoidance learning, anxiety disorders, and PTSD, as well as uncertainty, all involve the cerebellum. While there has been increased attention given to the cerebellum in the context of PTSD and anxiety-related disorders (Moreno-Ruis, 2018), this imaging work has not included behavioral measures. The inclusion of behavioral measures, such as eyeblink conditioning to imaging studies, would improve the ongoing work on exploring the role of the cerebellum in the development and maintenance of PTSD, especially under uncertain conditions.

Hippocampal substrates of eyeblink conditioning and avoidance learning

A second brain region implicated in eyeblink conditioning and avoidance learning is the hippocampus. While the cerebellar circuitry is necessary and sufficient for delay eyeblink conditioning, the hippocampus, modulates delay eyeblink conditioning, and avoidance learning. Hippocampal lesions or damage do not disrupt delay conditioning in rabbits (Schmaltz and Theios, 1972; Shohamy et al., 2000) or humans (Gabrieli et al., 1995). In fact, hippocampal lesions have been found to accelerate acquisition of conditioned eyeblinks in delay conditioning in rabbits (Schmaltz and Theios, 1972) and rats (Christiansen and Schmajuk, 1992; Lee and Kim, 2004; Port et al., 1985). Although the hippocampus is not required for simple delay eyeblink conditioning, it normally contributes to acquisition of the CS-US association. A hippocampal role in delay eyeblink conditioning is based on the finding that disruption of the hippocampal activity by electrical stimulation or pharmacological manipulation can disrupt acquisition of the delay eyeblink conditioning (Berry and Thompson, 1979; Solomon et al., 1983; Allen et al., 2002). Overall, the hippocampus has a complex relationship to delay eyeblink conditioning such that lesions do not disrupt eyeblink conditioning, can actually accelerate CR acquisition while disruption of hippocampal function can retard CR acquisition.

The hippocampus also plays a modulatory role in lever press avoidance learning. Much like findings with eyeblink conditioning, hippocampal aspiration lesions were found to facilitate acquisition of avoidance in shuttle avoidance (Olton, 1973; Black et al., 1977), as well as in a lever press avoidance task (Schmaltz and Giulian, 1972). In addition, a hippocampal role has been put forth in extinction of avoidance. Specifically, Pang et al. (2011) found that disruption of medial septal GABAergic inputs to the hippocampus had no effects on acquisition of lever press avoidance but did impair extinction. More recently, Cominski et al. (2014) reported that selective ibotenic lesions of the hippocampus impair extinction in SD rats. The hippocampal lesioned SD rats exhibited slowed extinction, which is similar to that observed in intact WKY rats. Cominski et al. (2014) also reported that WKY rats had smaller hippocampal volumes than SD rats. In addition, these WKY rats exhibited impaired synaptic plasticity as demonstrated by a lack of long-term potentiation (LTP) with parameters that produced LTP in SD rats. Thus, alterations in the hippocampus may underlie differences in extinction of avoidance responses between behaviorally inhibited WKY rats and noninhibited SD rats.

Hippocampal role in PTSD

The hippocampus has also been implicated as playing a role in PTSD. Patients with PTSD symptoms exhibit reduced hippocampal volume (Gurvits et al., 1996; Villarreal et al., 2002), which has been localized to the CA3/dentate gyrus region (Wang et al., 2010). To determine whether this loss of hippocampal volume was a result of trauma and subsequent PTSD, as hypothesized by Bremner (2001), or was a preexisting risk factor for the development of PTSD, Gilbertson et al. (2002) compared the hippocampal volume of combat exposed veterans with PTSD to their twin with no combat exposure. The reduced hippocampal volume was evident in both twins suggesting that decreased hippocampal volume exists before trauma exposure and diagnosis of PTSD. A subsequent study found impaired learning in a hippocampal-dependent configural task was associated with reduced hippocampal volume (Gilbertson et al., 2007). However, more recent studies have found that hippocampal volume differences with PTSD are due to current rather than lifetime PTSD symptoms (Apfel et al., 2011) and that environment has more of an effect than genetics (Bremner et al., 2021). Thus, debate continues on whether reduced hippocampal volume is a preexisting risk factor for PTSD or a symptom developed subsequent to trauma. Whether hippocampal volume differences occur before or after trauma, it is possible that this abnormality can alter learning, which may account for enhanced eyeblink conditioning and avoidance learning exhibited by behaviorally inhibited WKY rats. The finding that hippocampal damage or disruption can facilitate the acquisition of conditioned eyeblinks and avoidance responses may give insight into possible mechanisms for the development of PTSD through hyperconditioning.

Hippocampal role in uncertainty

In addition to its role in PTSD, the hippocampus also has a role in uncertainty (Harrison et al., 2006; Vanni-Mercier et al., 2009; Soltani and Izquierdo, 2019). One aspect of uncertainty that is a well-established function of the hippocampus is novelty detection (Knight, 1996; Kumaran and Maguire, 2007; Barbeau et al., 2017). When a novel stimulus is experienced, it has some degree of uncertainty until its relationships are learned. Performance in classical conditioning tasks is dependent on stimulus novelty in that conditioning of a response is quicker when a novel stimulus is used, compared with when a previously experienced stimulus is used, an effect known as latent inhibition (Lubow, 1973). Activity in the medial septum also is dependent on novelty. Neurons in the medial septum respond strongly when novel, but not familiar, stimuli are presented in an eyeblink conditioning task (Berger and Thompson, 1977). There is evidence that the septum also is activated by probabilistic outcomes (Monosov and Hikosaka, 2013). Septohippocampal modulation determines what information is stored and recalled in the hippocampus (Baxter et al., 1999; Buhusi and Schmajuk, 1996; Hasselmo and Schnell, 1994; Hasselmo, 1995). For example, Hasselmo and Schnell (1994) proposed that signals sent from hippocampal neurons to the septal region indicate how well stimulus representations are encoded. If a stimulus has strong representation in memory, learning about that stimulus is not necessary. If a stimulus has no representation or a weak representation in memory, the hippocampus needs to learn a representation of that stimulus and thus sends signal to the septum which in turns activates hippocampal learning. Thus, the medial septum is involved in novel, possibly uncertain stimuli, as well as probabilistic relationships, and may modulate how the hippocampus learns and thus how eyeblink conditioning, avoidance learning may be accelerated in behaviorally inhibited individuals and those expressing PTSD symptoms.

Overall, there are independent lines of research supporting hippocampal involvement in eyeblink conditioning, avoidance learning, PTSD, and uncertainty. Further experimental and computational studies should explore how mechanisms within the hippocampal system may underlie the enhanced learning demonstrated with weak situation learning tasks that include uncertainty.

Striatal substrates of probabilistic category learning

The probabilistic category learning task involves another neural substrate, the striatum, which also has been implicated as playing a role in PTSD and uncertainty. The basal ganglia/striatum is necessary for probabilistic category learning (Shohamy et al., 2008), as well as reward and punishment learning (Delgado, 2007). Striatal activity is associated with reward and punishment learning in humans (Salamone, 1994; Ungless et al., 2004; Heekeren et al., 2007) and damage to the dorsal striatum impairs punishment learning but not reward learning (Palminteri et al., 2012). Striatal activity related to reward and punishment learning is modulated by dopamine. For example, tasks involving separate reward and punishment trials, such as those reviewed here, have been found to be disrupted by striatal damage to or pharmacological manipulation of brain dopamine systems. (Palminteri and Pessiglione, 2017). Dopaminergic drugs selectively affect learning to obtain reward but not punishment learning (Pessiglione et al. 2006; Frank et al., 2007, Bodi et al., 2009). Specifically, Bodi et al. (2009) found that L-DOPA increased the relationships between reward learning and novelty seeking and reduced the relationship between punishment learning and harm avoidance in the probabilistic computer-based task discussed in this review (Sheynin et al., 2013). Thus, there is evidence for a role for the striatum and dopamine in the computer-based task found to be enhanced in behaviorally inhibited individuals and those expressing PTSD symptoms.

There also is some limited evidence for a striatal role in eyeblink conditioning and avoidance learning. Neural recordings in the neostriatum (i.e., the caudate nucleus) revealed stimuli-related as well as CR-related activity in eyeblink conditioning (White et al., 1994). However, Huntington-related damage to the striatal system disrupted the timing but not acquisition of conditioned eyeblinks (Woodruff-Pak and Papka, 1996). Limited evidence for a striatal role in avoidance also has been put forth. For example, White and Rebec (1993) reported activity in the caudate putamen and nucleus accumbens related to stimuli and responses in a lever release avoidance task. Dorsomedial striatal lesions disrupted enhanced lever pressing in a schedule of partial reinforcement (Torres et al., 2016). Therefore, there is evidence that the striatum plays a role in the three types of learning, classical eyeblink conditioning, avoidance learning, and probabilistic reward and punishment learning, discussed in this review.

Striatal role in PTSD

Various structural abnormalities of the striatum have been linked to PTSD. These include altered volume of the caudate (Looi et al., 2009; Herringa et al., 2012), as well as microstructural changes (Waltzman et al., 2017). In addition to structural differences, alterations in striatal activity have been reported for PTSD. For example, PTSD has been associated with reduced blood flow (Herringa et al., 2012) and activity in the striatum (Lucey et al., 1997; Lindauer et al., 2004) but also increased blood flow (Nardo et al., 2011) and activity in the striatum (Geuze et al., 2007; Linnman et al., 2011). Thus, abnormalities in the striatum may underlie some of the symptoms of PTSD.

Striatal role in uncertainty

The striatum also plays a role in uncertainty. Intolerance of uncertainty has been associated with increased striatal volume (putamen) in the context of obsessive-compulsive disorder and generalized anxiety disorder (Kim et al., 2017). Furthermore, dorsal striatal neurons signal uncertainty in reward learning (White and Monosov, 2016) while uncertainty produced reduced activity in the ventral striatum (Hebart et al., 2014). Buzzell et al. (2016) found that these reductions in ventral striatal activity predicted slowing of responses following uncertain responses. Dopamine also plays a role in uncertainty. Schlösser et al. (2009) examined the effects of dopaminergic agonists on uncertainty and found that dopaminergic agonists shifted activation from cortical regions to the hippocampus and cerebellum. Thus, alterations in dopamine based on uncertainty may recruit the brain areas involved in eyeblink conditioning and avoidance learning and result in the enhanced learning evident in anxiety vulnerable individuals as well as individuals reporting PTSD symptoms.

Overall, the neural substrates (i.e., cerebellum, hippocampus, and striatum) that underlie tasks of classical eyeblink conditioning, avoidance learning, and probabilistic category learning also are known to play a role in PTSD and uncertainty. The overlap in functions in these brain areas across PTSD and uncertainty combined with the behavioral results summarized in this review may provide a road map for utilization of the weak situation tasks for the exploration of how anxiety- and trauma-based psychopathologies develop and can be treated.

Future work

The work discussed in this review can serve as a foundation for future work incorporating the manipulations of uncertainty that produce a weaker learning situation that is more adept at identifying individual differences those expressing, or at risk for, PTSD. Future work can further explore aspects of uncertainty in the three learning tasks summarized in this review, eyeblink conditioning, avoidance learning, and probabilistic category learning, as well as apply some of the strategies from these studies to other learning paradigms, such as fear conditioning.

The studies involving uncertainty with eyeblink conditioning can be extended in several ways. For example, while enhanced eyeblink conditioning with schedules of partial reinforcement has been demonstrated in a non-clinical sample of undergraduates self-reporting behavioral inhibition as well as clinical samples of military personnel and veterans with symptoms of PTSD, the effects schedules of partial reinforcement on eyeblink conditioning have yet to be tested with the behaviorally inhibited WKY rat strain. If WKY rats exhibit the same degree of enhanced acquisition and slowed extinction in eyeblink conditioning as have been observed in behaviorally inhibited humans, this would provide an animal model to further explore neural mechanisms within the cerebellar and hippocampal systems for the development of PTSD. Furthermore, parametric studies of a range of various schedules of partial reinforcement could reveal at what point contingencies less than 100% continuous reinforcement are uncertain enough to produce a weaker learning situation.

Further work with avoidance learning with partial reinforcement can increase our understanding of the effects of uncertainty. For example, the finding of enhanced acquisition of avoidance responses with behaviorally inhibited WKY rats can be extended to include extinction training. The Miller et al. (2020) avoidance study with a schedule of partial reinforcement included acquisition, but not extinction, of avoidance responses. Behaviorally inhibited WKY rats exhibiting a slowing of extinction (i.e., a partial reinforcement extinction effect) would be expected based on the finding from classical eyeblink conditioning studies (Allen et al., 2014). Future work could also explore how behaviorally inhibited WKY rats would differ from SD rats when switching between continuous and partial reinforcement. As previously mentioned, a study by Grady et al. (2016) revealed that a schedule of partial reinforcement followed by continuous CS–US pairings yielded the strongest CR during fear acquisition training as well as delayed extinction. This pattern of partial reinforcement followed by continuous reinforcement should be tested in eyeblink conditioning as well as avoidance learning with WKY rats.

Findings of enhanced avoidance learning in WKY rats also can be expanded to explore the effects of uncertainty on avoidance in behaviorally inhibited individuals and individuals expressing PTSD symptoms. Behaviorally inhibited individuals have demonstrated enhanced avoidance learning in a computer-based avoidance task (Sheynin et al., 2014a) as well as slowed extinction (Sheynin et al., 2014b). These findings have been replicated with Veterans exhibiting symptoms of PTSD (Sheynin et al., 2017). However, these studies have only tested a 100% contingency between the warning signal and the aversive outcome (points loss). Future work should test the effects of schedules or partial reinforcement or probabilistic relationships in this computer-based avoidance task.

Work with the probabilistic category learning task can also be expanded. The studies summarized in this review included probabilistic categories (i.e., individual stimuli are in one category 80% of the time and the other category 20% of the time) but not deterministic categories (i.e., stimuli are always in the same category). Parametric comparisons of probabilistic and deterministic relationships could reveal at what point uncertainty (i.e., reductions in contingency) produces the individual differences in enhanced reward and/or punishment learning discussed in this review. Other probabilistic learning tasks also could be tested with behaviorally inhibited individuals and those reporting PTSD symptoms.

Applications to fear conditioning

Several aspects of weaker situations from the three reviewed tasks can be applied to fear conditioning for the purpose of identifying individual differences. For example, further investigations of cue salience across conditioning paradigms would be of interest. Specifically, the effects of the salience of the stimuli of fear conditioning (electric shock or another aversive stimulus) could be directly compared with the less-aversive stimulus of eyeblink conditioning, the corneal air puff. In addition, the use of the manipulations of uncertainty discussed in the current review, including schedules of partial reinforcement, lengthening and varying of the intertrial interval (ITI), and the use of probabilistic relationships may further reveal individual differences in fear conditioning studies with those expressing, or at risk for, PTSD. Of particular interest would be to determine whether the effects of lengthening and varying the intertrial interval reported by Allen et al. (2016) reveal individual differences in fear conditioning. A review of the literature indicated that this ITI manipulation has not been tested in fear conditioning.

One major expansion of fear conditioning would be to test the behaviorally inhibited WKY rat strain as well as behaviorally inhibited humans with various forms of fear conditioning. It would be of interest to determine if the individual differences observed between behaviorally inhibited WKY rats and noninhibited SD controls would generalize to the acquisition, extinction, and generalization of fear conditioning. In addition, the identification of individual differences in behavioral inhibited human populations would extend the theory that avoidant personality temperaments are a risk factor for PTSD. Overall, the work reviewed indicates that manipulations of uncertainty could be utilized to increase the identification of individual differences, both between patient and control populations as well as between those at risk and not at risk for pathologies, such as PTSD. The addition of uncertainty manipulations to fear conditioning for the sake of identifying individual differences rather than for experimental manipulations to slow or extend acquisition or extinction training could drive future work in fear conditioning.

Weaker situations and uncertainty: applications to therapies

The weaker situation learning tasks discussed in this review also may be applicable to therapies for PTSD and anxiety disorders. Learning tasks, such as eyeblink conditioning or computer-based tasks involving point loss, may be more palatable to clinical populations than fear conditioning involving an electric shock. Weaker situations reveal individual differences that could then be targeted by therapy to reduce enhanced learning or alter extinction rates

Based on the evidence that uncertainty modulates learning that may result in the development and maintenance of PTSD provides a possible mechanism for unlearning the associations that underlie these psychopathologies. The inclusion of some aspects of uncertainty into therapies or treatments for PTSD may improve the outcome for individuals who do not benefit from more continuous therapies. For example, the finding that extinction in eyeblink conditioning was slowed in behaviorally inhibited individuals and individuals with PTSD symptoms trained with a schedule of partial reinforcement is important given the use of exposure (i.e., extinction) training for PTSD and anxiety disorders (Joseph and Gray, 2008). Empirical and computational studies could guide the development of more effective treatments or preventive interventions (Myers et al., 2013). The use of manipulations of uncertainty in therapies could be beneficial given that typical experimental procedures that investigate avoidance learning tend to use 100% contingencies and may miss clinically relevant processes (LeDoux and Pine, 2016; Xia et al., 2017) in that fear disorders may depend on more probabilistic than continuous reinforcement (Erlich et al., 2012).

Conclusions

The purpose of this review was to provide evidence from learning tasks that included manipulations of uncertainty, which are interpreted as weaker situations that identify individual differences in individuals expressing, or at risk for, PTSD. Lissek et al. (2006) hypothesized that weaker situations that involve less salient stimuli and some degree of uncertainty would be better at identifying individual differences between patient and control groups than strong situations that involve more salient and reliable stimuli. This hypothesis has been supported by the enhanced eyeblink conditioning and category learning demonstrated by veterans and military personnel expressing PTSD symptoms compared with individuals without PTSD symptoms.

Furthermore, Beckers et al. (2013) as well as Emmelkamp et al. (2014) championed the ability of weaker situations to reveal differences within nonclinical populations between individuals with low versus high risk for the development of anxiety pathologies. This hypothesis was supported by enhanced eyeblink conditioning and category learning with nonclinical undergraduate populations identified as behaviorally inhibited. In addition to the findings from behaviorally inhibited humans, enhanced eyeblink conditioning and avoidance learning was demonstrated in an animal model of behavioral inhibition—the WKY rat strain. As summarized in Fig. 4, individuals with anxious or avoidant personality temperaments, such as behavioral inhibition learn faster, and in some cases extinguish slower, when trained under uncertain conditions. This enhanced learning and slowed extinction may indicate learning mechanisms that are related to increased vulnerability to the development of PTSD in a limited number of individuals.

Fig. 4

Schematic representation summarizing the relationships between avoidant or inhibited personality temperaments, manipulations of uncertainty, enhanced learning, and increased vulnerability for the development of PTSD

Furthermore, this review has detailed evidence supporting the value of manipulations of uncertainty to further reveal enhanced learning to greater degree than evident with 100% paired training. In fact, it has been suggested that some degree of uncertainty in learning tasks may be more ecologically valid for the study of PTSD than learning tasks in which the aversive event occurs on every trial (Beckers et al., 2013; Pittig et al., 2018; Miller et al., 2020). In the real world, traumatic events are uncertain, and thus, experiments exploring the underlying learning mechanisms of these psychopathologies should certainly include some aspect of uncertainty. Overall, the findings summarized support the continued study of uncertainty and weaker situation tasks as originally proposed by Lissek et al. (2006) to identify individual differences in learning in individuals at risk for PTSD and expressing PTSD symptoms.