On the basis of sex: Differences in safety discrimination vs. conditioned inhibition

Abstract

Inaccurate discrimination between threat and safety cues is a common symptom of anxiety disorders such as Post-Traumatic Stress Disorder (PTSD). Although females experience higher rates of these disorders than males, the body of literature examining sex differences in safety learning is still growing. Learning to discriminate safety cues from threat cues requires downregulating fear to the safety cue while continuing to express fear to the threat cue. However, successful discrimination between safety and threat cues does not necessarily guarantee that the safety cue can effectively reduce fear to the threat cue when they are presented together. The conditioned inhibitory ability of a safety cue to reduce fear in the presence of both safety and threat is most likely dependent on the ability to discriminate between the two. There are relatively few studies exploring conditioned inhibition as a method of safety learning. Adding to this knowledge gap is the general lack of inclusion of female subjects within these studies. In this review, we provide a qualitative review of our current knowledge of sex differences in safety discrimination versus conditioned inhibition in both humans and rodents. Overall, the literature suggests that while females and males perform similarly in discrimination learning, females show deficits in conditioned inhibition compared to males. Furthermore, while estrogen appears to have a protective effect on safety learning in humans, increased estrogen in female rodents appears to be correlated with impaired safety learning performance.

1. Introduction

It is well-established that disorders such as anxiety and Post-Traumatic Stress Disorder (PTSD) pose a large burden on mental health resources [1]. These disorders are significantly more prevalent in females than males [2]. Despite the pronounced population differences, research explicitly examining both male and female subjects is limited, leading to less effective treatments for females affected by these disorders. The need to address sex as a biological variable is beginning to be addressed more rigorously by research funding agencies, and there is a growing body of research investigating male and female subjects in disorders of maladaptive fear.

Maladaptive fear can result from incorrectly attributing danger to an otherwise non-threatening stimulus. A large body of literature has been directed towards investigating how fear can be downregulated in situations where expressing fear is maladaptive. Much of this research has focused on the mechanisms of fear extinction, in which a cue that was once predictive of threat is no longer a reliable predictor of threat. These studies first condition subjects to fear a conditioned stimulus by pairing it with an aversive outcome (i.e. CS+) and then, during extinction training, present the same stimulus without the aversive outcome. However, because these studies typically do not present any additional cues during training, it is not possible to assess whether the subject’s fear response is specific to the training cue or generalized to other cues (Table 1).

Table 1.

Comparing different paradigms involving safety learning towards a cue.

CUE DRIVEN SAFETY	Extinction of a fear CS+	CS− only	CS− vs CS+ discrimination	CS−/CS+ conditioned inhibition
Requires learning to downregulate fear to a cue?	YES	YES	YES	YES
Requires learning to downregulate fear to a cue that was predictive of threat?	YES	NO	NO	YES
Requires learning to discriminate between CS + associated with threat and CS− associated with safety?	NO	NO	YES	YES
Requires learning to downregulate fear when the CS− is presented together with the CS+?	NO	NO	NO	YES

An increasing number of studies are exploring and testing the idea that a cue that is explicitly unpaired with an aversive outcome (i.e. CS−) is serving as a safety signal (reviewed in [3]). Some of these studies use a between-subjects design in which one group of subjects are trained with one cue explicitly unpaired to the aversive stimulus (CS−), while a separate group of subjects are trained with one cue paired with the aversive stimulus (CS+). With this approach the explicitly unpaired group (CS−) shows downregulated fear during the CS while the paired group (CS+) shows upregulated fear during the CS, compared to the non-CS period. Since each group is trained to only one cue, neither group is required to discriminate between cues (Table 1).

Alternatively, several studies have employed a within-subjects design in which each subject is trained with a cue paired with an aversive stimulus (CS+), as well as another cue that is explicitly unpaired (CS−). This type of training requires learning to discriminate between the CS + associated with threat and the CS− associated with safety (Table 1). However, discrimination does not explore if the subject is able to utilize the CS− to downregulate fear elicited by the CS +. To assess this possibility, the CS+ and CS− need to be presented simultaneously to examine how the fear response differs compared to either cue presented alone. If the fear response to the CS−/CS+ combination is significantly lower than the fear response to the CS+, the safety CS− is referred to as a conditioned inhibitor (Table 1). There are relatively fewer studies exploring conditioned inhibition as a method of safety learning. Successful discrimination between a fear CS+ and a safety CS− does not necessarily guarantee that conditioned inhibition will be present but is most likely a requirement.

While cue-driven safety, where a cue results in a learned reduction in fear responding (Table 1), has been studied extensively in male subjects, only recently have studies emerged investigating behavioral and neurobiological differences in both male and females (detailed below). Sex differences in extinction have been extensively reviewed elsewhere and will not be the focus here [4,5]. Recent reviews have also thoroughly investigated the underlying mechanisms of, as well as explored sex differences, in learned safety [3,4]. Instead, the focus here will be regarding sex differences in discriminating between threat and safety cues, and conditioned inhibition when threat and safety cues are both present, in both humans and rodents. In rodents, differing behavioral strategies in males vs. females in various learning paradigms are well-documented [6–10]. These strategies in females, such as fast-paced movements known as ‘darting’, increased context generalization, or risk avoidance at the cost of metabolic need, may be evolutionarily beneficial [6–8]. It is essential to note that differing behaviors in females vs. males do not necessarily indicate cognitive deficits. Rather, these behaviors reflect risk mitigation strategies for self-preservation [9]. How might these differences in strategy play out specifically within the framework of safety learning? We propose that while behavioral out-comes and strategies may be similar between males and females when discriminating between threat and safety cues, in some cases females exhibit a strategy shift when undergoing conditioned inhibition. We will investigate individual differences in behavioral performance as well as explore some potential underlying mechanisms of these sex-specific behaviors.

2. The fear regulation network and associated estrogen receptors

Learned safety is governed via brain regions associated with fear regulation. For an extensive review of safety circuitry and mechanisms underlying fear suppression, we direct you to “Know Safety, No Fear” [3]. In brief, the primary brain regions associated with fear regulation include the prelimbic and infralimbic cortices of the prefrontal cortex (PFC), the hippocampus (HPC), as well as the amygdala [3]. In males, the prelimbic (PL) and infralimbic (IL) cortices of the PFC mediate fear expression and inhibition, respectively [11–14]. Within the basolateral amygdala, subpopulations of neurons specifically responsive to safety-associated cues were identified in male rats [15]. Elevated fear expression is associated with higher neuronal activity in the lateral and basolateral amygdala (LA and BLA) in females compared to males [16]. Furthermore, during generalization of contextual fear, females, but not males, demonstrate an increase in c-Fos expression in the amygdala [7]. The hippocampus can be subdivided into two regions: the dorsal hippocampus (dHPC), important for spatial and contextual memory, as well as the ventral hippocampus (vHPC), which is involved in both contextual and emotional memory [17]. While males show increased activation of the dHPC during a context fear generalization task, females instead show increased activation of the amygdala [7]. Additionally, recent work has demonstrated projection specific activation of vHPC-prelimbic cortex projections during fear suppression in male mice [18]. Overall, a systematic and in-depth investigation of this circuit comparing males against females during learned safety is still incomplete.

Sex differences in neuronal activity within the safety circuit likely result in sex differences in learned safety behavior. Activity in these brain regions may vary due to differing responses to estradiol (estrogen) expression, which fluctuates in females. Estradiol acts through activation of the estrogen receptor beta in brain regions associated with fear regulation [19]. These receptors are minimally expressed within the prefrontal cortex and amygdala, but strongly expressed within the hippocampus [19]. Estrogen signaling can also alter dendritic spine formation, short and long-term plasticity, and modulate glutamate receptor trafficking (reviewed in [20]. It is unsurprising that the presence of estrogens can also rapidly alter rodent behavior in working memory tasks [20]. Due to cyclical fluctuations in estradiol, differing responses in estrogen signaling cascades in females may alter activity within the safety network, leading to differential behaviors. For example, female mice in estrus phases high in estradiol exhibit enhanced fear extinction recall (reviewed in [5]). Below we will examine how estradiol may contribute to sex differences in fear discrimination and conditioned inhibition within the framework of safety learning.

3. Discrimination between threat and safety cues

3.1. Human studies

Learned safety via discrimination between a threat cue (CS+) and a safety cue (CS−) requires learning to downregulate fear in response to the CS− , while maintaining a robust fear response to the CS+. Healthy male subjects have consistently been shown to demonstrate fear to a CS+ but not to a CS− [21–26]. In a study with both female and male human subjects, discrimination between threat and safety cues was observed, however distinct sex differences were not analyzed [18].

Patients with clinical disorders, such as PTSD or anxiety, often express generalized fear to non-threatening situations. Consistent with this observation, males with high PTSD symptoms (PTSD+) (n = 13) generalized fear between fear and safety cues compared to healthy (n = 14–28) or low-PTSD symptomatic subjects (n = 14) [24,27]. A lack of discrimination in PTSD + veterans was attributed not to decreased startle response to the CS+, but an increased fear response to the safety signal itself [27]. Although PTSD + patients were unable to suppress the startle response to the CS−, CS+ vs. CS− contingency recognition was intact, indicating impaired suppression of the physiological fear response but intact cognitive awareness to the meaning of the CS− [24]. Interestingly, patients with panic disorder (n = 19) have been shown to discriminate between a CS+ and a CS− that was dissimilar to the CS+, but showed generalized startle potentiation to cues that were increasingly similar to the CS+; this generalization effect was not seen in control subjects (n = 19) [25].

Although one might hypothesize learned safety via discrimination is impaired in clinical populations, this is not always the case. In contrast to the above reports, several studies report accurate discrimination between a CS+ and CS− in healthy and PTSD/panic disorder subjects [21, 22,25,28]. These studies included 10–19 control and 10–19 subjects with PTSD/panic disorder [21,22,25,28]. It may be the case that learned safety via discrimination is intact in clinical populations if the cues used in the paradigm are directly trauma-related. Taken together, these studies suggest that safety via discrimination is consistently learned in healthy populations and may be impaired in patients with PTSD/panic/anxiety disorders.

While studies examining sex differences in discrimination find that both males and females differentiate between fear and safety signals, some work has shown that females discriminate to a lesser extent than males [29,30]. For example, in school-age children with PTSD symptoms, males (n = 45) had larger skin conductance and fear-potentiated startle responses than females (n = 41), though both females and males discriminated between the CS+ and CS− [29]. As this study was conducted prior to puberty, it is unlikely that the skin conductance response and startle differences were due primarily to differences in estrogen (discussed further below). In healthy adults, Lonsdorf et al. described similar findings: differentiation between CS+ and CS− in both males and females, but more discrimination in male subjects [30]. As discussed further below, Lonsdorf found little evidence for sex differences based on cycle phase or contraceptive use, though cycle phases were self-reported and there were a limited number of subjects per phase (n = 18 follicular phase, 22 luteal, 172 on hormonal contraceptives). Taken together, these studies indicate that subtle differences in fear and safety discrimination exist between males and females, but they may not be dependent upon estrogen.

Several studies noted above examined discrimination in both male and female subjects but did not explicitly investigate sex differences [21, 18,25,28]. Again, these studies largely found that in non-clinical populations, discrimination between fear and safety signals were intact between sexes. It is likely many challenges exist in recruiting sufficient age, trauma, education etc. matched subjects in human studies. Though the number of male and female participants were noted in each study, individual values were not represented. While these studies were likely underpowered to determine sex differences, it may be advantageous for future studies to group data but plot individual values by sex.

Hypersensitivity of the amygdala is often expected in PTSD + patients (reviewed in [31]). For example, compared to non PTSD + controls, PTSD + subjects had significantly higher amygdala activation during visual discrimination between negative and neutral stimuli [21, 32], but see [22,28]. In these studies, PTSD + subjects were male [32] or mixed male/female groups with insufficient power to test for sex differences (10–14 subjects total) [21,28] and [22]. Differences in amygdala activation in PTSD + vs. PTSD− patients in these studies could be due to trauma type, testing paradigms including content of images presented, as well as sex-age matching between studies. Currently, research explicitly examining sex differences in amygdala activity during discrimination tasks in clinical populations vs. healthy subjects warrants further investigation.

3.1.1. Estradiol and fear discrimination in humans

The literature discussed in the previous section suggests that overall, while both males and females discriminate between CS+ and CS−, males demonstrate more pronounced discrimination [29,30]. One potential explanation for this is fluctuations in estradiol in females may lead to differences in fear discrimination (i.e. more generalization to the CS−). During the follicular phase of the menstrual cycle, estradiol is low, while during the luteal phase of the cycle, estradiol is high [33,34]. In humans, estradiol is hypothesized to have a protective effect on learning, particularly on fear extinction [5,35,36]. Some studies have found that in patients with PTSD, low estradiol may be associated with more severe symptoms, as well as worse extinction learning and recall (reviewed in [35]). Given the potential protective effect on fear extinction, estrogen may also have a protective effect on safety learning via discrimination. Indeed, women in the luteal phase (n = 15) demonstrated an initial increase in CS+vs. CS− discrimination in SCR compared to men (n = 19) or women taking oral contraceptives (n = 15), though all groups discriminated between the cues [37]. Similarly, in another study, women in the luteal phase (n = 14) discriminated between the CS+vs. CS− compared to women in the follicular phase (n = 14), who did not discriminate between the stimuli [33]. Furthermore, women with high estradiol (n = 17) had higher activation of the vmPFC compared to women with low estradiol (n = 17), which was associated with improved fear extinction of a CS+ [38]. While these studies find a protective effect of estradiol on learned safety, another study examining discrimination using self-reported cycle phase found no differences in discrimination [30]. In women with PTSD, women with low estrogen (n = 22) showed higher fear potentiated startle than women with high estrogen (n = 22) [33]. Even though it appears higher estradiol is protective and beneficial to extinction recall, there are some reports suggesting a detrimental effect of estradiol during the luteal phase on PTSD symptoms (reviewed in [35]. Overall, these results do suggest that, in healthy adult women, the presence of estrogen may be protective during fear vs. safety discrimination; however, this effect has not always been replicated.

3.2. Rodent studies

Studies of learned safety in rodents via fear discrimination demonstrate differing results between sexes depending upon cued vs. contextual conditioning. Sex differences in rodent context discrimination and contextual fear memory have been reviewed elsewhere and are beyond the scope of this review [10]. When investigating discrimination in cued conditioning, studies in rodents largely conclude that males effectively discriminate between CS+ and CS− cues [15,39–42]. The circuitry and potential mechanisms underlying safety learning via discrimination are thoroughly reviewed in [3].

In studies that have examined learned safety via discrimination in female rodents, the majority of findings have demonstrated that both male and female rodents discriminate between fear and safety cues [39, 43,44], but see [45]. There is, however, some nuance within the results of each study that do in fact demonstrate sex differences exist in the degree of discrimination. For example, during a fear and safety discrimination test, both female and male C57/B6 mice froze significantly more to a fear cue than a safety cue, but freezing to the fear cue was higher in male mice compared to females [43]. Using a fear/safety discrimination paradigm and examining sex differences in safety learning, both male and female rats froze less to safety cues presented alone than they did to the fear cue [40,44]. Similarly, fear versus safety cue discrimination was also intact in both male and female mice that were bred for high alcohol preference (HAP mice) (Müller, Adams, Sangha, & Chester, this issue). In contrast to human studies of discrimination [29,30], Foilb et al. described greater cue discrimination during acquisition and recall in females as compared to males [44,46]. This result is consistent with work from Day et al., demonstrating that early in training, females, but not males discriminated between a CS+ and CS− [39]. However, as training progressed, males, but not females discriminated between CS+ and CS− , indicating that perhaps females had a lower threshold to generalize between the cues [39]. The studies evaluated here indicate that differences in pain sensitivity between males and females do not account for discrepancies in freezing to the CS+. At differing shock intensities, there were no differences in jumping/darting between male and female rats [40]. A study determining threshold current for flinching or vocalization in female and male rats found a significant effect of response type but not of sex (ie: no sex differences in the threshold current needed to induce either flinching or vocalization) [39]. Though not evaluated in this review, previous research in C57/B6 mice has demonstrated no sex difference in pain thresholds to footshock [47].Taken together, these results demonstrate successful discrimination in both male and female rodents, though some sex differences exist in the extent and time course of discrimination.

In addition to assessing freezing or startle attenuation as measures of fear, darting behavior in females can also be assessed as an informative measure of fear [6,48]. This behavior, fast paced darting movements, was first characterized in a large-scale study assessing differing types of fear responses in male and female rats and was shown to be more pre-dominant in female subjects during fear cues [6]. Importantly, estrous phase was not correlated with darting behavior [6]. Greiner et al. also showed increased darting behavior in females during a fear cue late in training compared to males [40]. Moreover, this study demonstrated reduced darting, and freezing, levels in females during the safety cue compared to the fear cue indicating females discriminated between the fear and safety cues [40]. Darting behavior was not demonstrated in male rats, reinforcing the need to assess different behavioral responses in female vs. male subjects as measures of fear [40,48,9,10]). In contrast to the Greiner study, Gruene et al. noted a significant decrease in freezing in female darters compared to female non-darters [40,6]. Notably, while darting during CS + presentation has not been correlated with estrus phase, at least one study has found differences in freezing to a CS + in high vs. low estrogen rodents [45]. It is essential that in future work differences in behavioral presentations between males and females are appreciated and assessed.

As discussed above, in humans hypersensitivity of the amygdala is expected in PTSD + patients. Similarly, increased activity of the amygdala is expected during fear learning. As expected, in rodents increased activation of the BLA (as measured by expression of the immediate-early gene c-Fos) is noted following context and cued fear conditioning [7,46]. When c-Fos expression in BLA was examined, no significant differences were noted between males and females. Generalization to context in females was not associated with sex differences in BLA activation [7], nor was increased freezing in males compared to females in a CS+ vs. CS− cued discrimination task associated with sex differences in BLA activation [46] (this issue). Of note, sex differences in activation of the CeA and BNST during discrimination were found, indicating potential brain regions involved in sex differences in safety learning [46].

3.2.1. Estradiol and fear discrimination

Similar to work in humans, few rodent studies have explicitly examined the role of estrogen specifically in fear and safety discrimination. The rodent estrus cycle is divided into four stages: metestrus, diestrus, proestrus and estrus. Estrogen is highest during the proestrus phase [34]. It should be noted that estrus phase was not monitored during the Greiner, Day or Foilb studies discussed above, and it was assumed that estrus phase was distributed equally amongst the female subjects. Human studies suggest a somewhat protective effect of estrogen on safety discrimination. One study found female, but not male mice generalized to context, though there was no effect of estrus phase [7]. More recent work in cue discrimination found females with high estrogen during testing generalized between a CS+ and CS−, while low estrogen females and males discriminated between the cues [45]. Of note, when females were instead assessed for estrus phase during training, both low and high estrogen females, as well as males, discriminated between CS+ and CS− cues during testing [45]. In contrast to the Trask study, recent work from Foilb et. al. found no effect of estrus phase on freezing to the cue in females [46]. This discrepancy may be due to differences in the number of shock presentations during training.

Several studies have also assessed the effects of estrogen replacement on gonadectomized rodents. In active avoidance, both ovariectomized (OVX) females and OVX females with estrogen replacement discriminated between contexts one day following training [49]. However, at longer retention intervals OVX + estrogen females generalized between contexts while OVX females continued to discriminate [49]. In fear-potentiated startle, OVX females and OVX + estrogen females successfully discriminated between CS+ and CS− cues, though there was a delay in discrimination learning in the OVX + estrogen group [50, 51]. In males, the presence of estrogen in gonadectomized (GDX) rats increased startle to the CS+ compared GDX males without estrogen [50]. Taken together, results from both Lynch and Toufexis demonstrate that while CS+vs. CS− discrimination occurs in both OVX and OVX + estrogen females, the time course of discrimination learning is altered in gonadectomized rodents with estrogen replacement.

The discrepancies in the effects of estrogen between these studies may be due to the differences in paradigms – both the Keiser and Lynch studies utilized context fear conditioning, Trask examined freezing during cue discrimination, and Toufexis used acoustic startle response. Each study examined differing measures of fear including freezing, latency to cross and percent change in startle from baseline. Finally, while Keiser and Trask explicitly examined the effect of estrus phase on discrimination, Toufexis and Lynch examined estrogen replacement in rodents. Regardless, these results appear to conflict with the human literature, where high estradiol offered a somewhat protective effect on discrimination [33].

3.3. Summary of safety discrimination in males vs. females

In the human studies reviewed, both males and females in non-clinical populations discriminate between fear and safety cues. Here, it appears that the safety learning strategies between males and females are largely similar, resulting in learned fear to a CS+ and learned safety to a CS−. Although discrimination between the fear and safety cues is present in both sexes, females discriminate between fear and safety cues to a lesser extent than males. Furthermore, in clinical populations, both male and female subjects with PTSD or panic disorder generalized more between fear and safety cues [24,27,29,33]). Similar conclusions can be applied to rodent studies, where many studies have demonstrated male and female rodents successfully discriminate between fear and safety cues (but see [45]). What appears to differ between male and female rodents is the magnitude of discrimination between fear and safety cues, with males showing more discrimination between fear and safety cues than females. However, there still remains the possibility in some studies this could be driven by higher overall freezing in males to the fear cue (e.g. [39,43,44]).

This modest generalization noted in females could be due to differences in estradiol concentration. Several of the human studies reviewed reported effects of cycle phase on discrimination. Overall, there was a protective effect of high estrogen on discrimination in females [33,37, 38]. In contrast to these studies, when estradiol was examined in rodent studies of discrimination, there was largely no effect on discrimination [49–51]. In one case high, but not low, estrogen led to cue generalization [45]. The differential effects of estrogen on discrimination between species (and even studies) may be due to differences in experimental paradigms (see Table 2). Alternatively, an evolutionary benefit may exist for female rodents with high estrogen to generalize between fear and safety cues [45]. Perhaps that particular benefit is not necessary for female humans. Thus, the effects of estradiol on discrimination may be species-specific.

Table 2.

Summary of Reviewed Works.

CS+ vs. CS− discrimination	Females	Males	Type of Stimuli	Discrimination or Conditioned Inhibition	Behavioral Measure	Behavior Summary
Human
[29]	Yes, children, trauma+, (n = 41)	Yes, (n = 45)	Colored shapes, airblast to larynx, noise	Discrimination	Skin conductance response; Fear-potentiated startle	F discriminated early, not late; M discriminated early and late
[30]	Yes, (n = 261) (172 contraceptives, 89 free cycling (18 follicular, 22 luteal)	Yes, (n = 116)	Geometric symbols, shock	Discrimination	Fear ratings; US expectency	F & M discriminated between CS+ and CS− (increased discrimination in F compared to M)
[33]	Yes, trauma+, (n = 44) (22 low E2, 22 high E2)	No	Colored lights, airblast to larynx, noise	Both	Fear potentiated startle; US expectancy	Low Estrogen (E) did not discriminate in startle; High E discriminated in startle; Low and High E discriminated in US expectency
[33]	Yes, (n = 28) (14 follicular, 14 luteal)	No	Colored lights, airblast to larynx, noise	Both	Fear potentiated startle; US expectancy	Follicular (low E) did not discriminate in startle; Luteal (high E) discriminated in startle; Both follicular and luteal discriminated in US expectency
Rodent
[43]	Yes (n = 14)	Yes (n = 14)	Tone, shock	Discrimination	Freezing	F & M discriminated between CS+ and CS−
[39]	Yes (n = 8)	Yes (n = 8)	Tone, shock	Both	Freezing	F discriminated (early, but not late in training); M discriminated (late, but not early)
[44]	Yes (n = 24)	Yes (n = 24)	Tone, shock	Both	Freezing	F & M discriminated between CS+ and CS− (F more than M)
[40]	Yes (n = 20)	Yes (n = 16)	Tone, light, shock	Both	Freezing, Darting	F & M discriminated between CS+ and CS−
[50]	Yes (n = 10 each group)	Yes (n = 10)	Light, white noise, fan, tone, shock	Both	Startle attenuation	OVX discriminated between CS+ and CS−; OVX + E2 discriminated late; but not early in training, M discriminated between CS+ and CS−
[45]	Yes, (Low E n = 20, high E n = 5)	Yes (n = 10)	Tone, shock	Discrimination	Freezing	F low estrogen discriminated; High estrogen generalized; M discriminated
CS+/CS− conditioned inhibition	Females	Males	Type of Stimuli	Discrimination or Conditioned Inhibition	Behavioral Measure	Behavior Summary
Human
[33]	Yes, trauma+, (n = 44) (22 low E2, 22 high E2)	No	Colored lights, airblast to larynx, noise	Both	Fear potentiated startle; US expectency	Low E did not inhibit startle; High E inhibited startle; Low & high E inhibited expectency ratings to CS+/−
[33]	Yes, (n = 28) (14 follicular, 14 luteal)	No	Colored lights, airblast to larynx, noise	Both	Fear potentiated startle; US expectency	Follicular did not inhibit startle; Luteal inhibited startle; Follicular & luteal inhibited in expectency ratings
Rodent
[39]	Yes (n = 8)	Yes(n = 8)	Tone, shock	Both	Freezing	F did not inhibit; M inhibited
[44]	Yes (n = 12)	Yes (n = 12)	Tone, shock	Both	Freezing	F & M inhibited
[40]	Yes (n = 20)	Yes (n = 16)	Tone, light, shock	Both	Freezing; Darting	F did not inhibit freezing; M inhibited freezing; F inhibited darting
[50]	Yes (n = 10 each group)	Yes (n = 10)	Light, white noise, fan, shock	Both	Startle attenuation	F OVX inhibited; OVX + E2 did not inhibit, M inhibited

4. Conditioned inhibition when threat and safety cues are both present

Successful discrimination between a fear CS+ and a safety CS− does not necessarily guarantee that the safety CS− can effectively reduce fear to the fear CS+. When both a fear CS+ and a safety CS− occur simultaneously and results in a reduced fear response when compared to just the presence of the fear CS+, this is referred to as conditioned inhibition [52]. There are relatively fewer studies exploring conditioned inhibition as a method of safety learning.

4.1. Human studies

With conditioned inhibition, in addition to requiring subjects to learn to discriminate between a CS+ that was associated with threat and a CS− that was associated with safety, they also need to learn to downregulate fear when a CS+ is presented in compound with a CS−. In some studies, participants discriminated between fear and safety cues, and demonstrated conditioned inhibition to compound fear/safety (CS+/−) cues [18,23,24]. One study examined both male and female participants with low or high anxiety, and found that low, but not high anxiety subjects expressed conditioned inhibition of eyeblink conditioning [53]. Similar to findings from Grillon, while healthy males and males with low PTSD symptoms exhibited conditioned inhibition during presentation of a CS+/− [23,24], males with high PTSD symptoms did not discriminate between the CS + alone and the CS+/− [24], i.e. a lack of conditioned inhibition. Although all groups (PTSD−, low and high symptom PTSD) gave the CS− a ‘safe’ expectancy rating, the eyeblink startle potentiation in PTSD + patients during the CS+/− cue remained high [24]. Taken together, these results demonstrate that similar to studies of discrimination, healthy subjects are able to suppress fear in a conditioned inhibition paradigm, while clinical populations may show some deficits.

4.1.1. Estradiol and conditioned inhibition

Two studies discussed above included both males and females in studies of conditioned inhibition. However, potential sex differences were not addressed in the results [18,53]. A study that did explicitly investigate the effects of estrogen on conditioned inhibition used fear potentiated startle to demonstrate that healthy women in the luteal phase showed conditioned inhibition, while women in the follicular phase did not, regardless of similar expectancy ratings of CS contingency between groups [33]. This indicated a deficit in top-down control in responding, rather than a lack of understanding of cue contingencies. The same experimental procedures were then performed in individuals with trauma classified as either high or low estrogen. In this second experiment, high estrogen trauma + women demonstrated conditioned inhibition during presentation of the CS+/−, while low estrogen trauma + women did not. These trauma groups also gave higher expectancy ratings of shock to the safety signal than the control groups [33]. This is in contrast to studies of high anxiety/PTSD + males, who do not express conditioned inhibition [24,53]. Consistent with studies of fear versus safety discrimination, the results of Glover et al. demonstrated a protective effect of estrogen on conditioned inhibition. This buffering effect of estrogen appears to wash out in studies that do not control for cycle phase.

4.2. Rodent studies

While discrimination tasks explicitly require differentiation between a CS+ and a CS−, conditioned inhibition studies examine the inhibitory properties of the safety signal (the CS−) via summation or retardation testing [52]. Summation tests present the CS+ in compound with the CS−, while retardation tests pair a conditioned inhibitor (previously trained CS−) with an aversive stimulus. In both cases, if the CS− is a true conditioned inhibitor, the conditioned response should be opposite (or less than) the response to the CS+ alone. Indeed, initial studies of conditioned inhibition found exactly that in male rats [54,55]. Further studies in male rats demonstrated conditioned inhibition of freezing or fear potentiated startle during discriminative conditioning (e.g [4,15, 18,40,44,56,57]. and others). Expression of safety in male rodents is associated with alterations in neuronal activity within the amygdala in both between [58] and within-subject designs [15]. Successful conditioned inhibition also depends upon neuronal activity in the mPFC and vHPC [13,18]. Pre-exposure to stress demonstrate mixed results with one study showing impaired conditioned inhibition of eyeblink conditioning in male rats [59], while another showed intact conditioned inhibition of freezing in male rats [60]. For an extensive review, largely based on male rodent data, on neuronal circuitry and mechanisms underlying suppression of fear during conditioned inhibition, the authors direct you to a recent review from Sangha et al. [3].

The limited rodent literature examining sex differences in safety learning largely finds that females do not suppress fear during conditioned inhibition paradigms [39,40,61], but see [44]. Furthermore, one study showed that neither males nor females bred for high alcohol preference showed significant conditioned inhibition unless subjected to a juvenile stressor (Müller, Adams, Sangha, & Chester, this issue). Recently published work from Greiner et al. examined sex differences in reward, fear and safety discriminatory conditioning in Long Evans rats. Male, but not female rats learned to inhibit freezing during presentation of a compound fear + safety cue (Fig. 1A), adapted from [40]. While females did not exhibit conditioned inhibition of freezing, they did suppress darting behavior during the compound cue by the fourth day of training [40]. Darting was not assessed in the Day or Aranda-Fernandez studies, which also found a lack of conditioned inhibition in females. These results suggest the need to assess differing behaviors in females vs. males, as perhaps there are sex differences in learning pathways of conditioned inhibition.

Fig. 1.

A novel analysis of sex differences in freezing behavior during conditioned inhibition in Long Evans rats from Greiner et al [40]. (A) Male rats showed a significant reduction in freezing to the fear+safety, safety, and reward cues compared to the fear cue. Female rats did not show a significant reduction in freezing to the fear+safety cue compared to the fear cue. Individual differences in freezing in males (B) and females (C) to the fear cue compared to the fear+safety cue. Most males froze less to the fear+safety cue compared to the fear cue, while most females did not.

Here, we present a novel analysis of data from the Greiner study. We examined individual differences in freezing during the fear cue compared to the fear + safety cue on the final day of discriminatory conditioning (Fig. 1B & C). Of note, most (~70 %) male rats froze less during the fear + safety cue than they did to the fear cue (Fig. 1B). However, approximately 30 % of the males exhibited no difference in freezing between cues. These animals could be grouped as ‘suppressors’ vs. ‘non-suppressors’ of freezing. Interestingly, when the female data were broken down in the same manner, the groups of ‘suppressors’ vs. ‘non-suppressors’ are almost the opposite. Here, ~25 % of the females inhibited freezing to the fear + safety cue, while ~75 % did not (Fig. 1C). It would be interesting to assess if there is a correlation between the ‘suppressor’ and ‘non-suppressor’ groups with their darting behavior during the task. It is possible that the female ‘suppressors’, and their darting behavior, may have been in a specific estrus phase. However, estrus phase data was not collected in this study so it is not possible to correlate the female ‘suppressors’ to a specific estrus cycle; this is an interesting future direction. Another unexplored question is if there are sex differences in the neuronal mechanisms underlying suppression and non-suppression of freezing during conditioned inhibition. We recently observed alterations in IL single-unit activity associated with suppressed freezing during presentation of the fear + safety cue (Ng, K.H. and Sangha, S., personal communication), though this data was not specifically broken down by suppressors vs. non suppressors of freezing. We hypothesize that alterations in IL activity would be reduced in female ‘non-suppresors’ performing the same task.

While the above work and others found a lack of conditioned inhibition in females using summation or retardation tests [39,40,61], at least one study presents data to the contrary [44]. This finding is likely due to differences in paradigms. Primarily, the Foilb study consisted of separate training and testing sessions, where the summated fear + safety cue was presented during testing [44], while the Greiner study presented the summated fear + safety cue within each discrimination session [40].

4.2.1. Estradiol in conditioned inhibition

In the human studies of conditioned inhibition discussed here, there was evidence of some protective effect of estrogen on expression of conditioned inhibition. The rodent work examined here included one study specifically examining estradiol in conditioned inhibition. Toufexis et al. examined conditioned inhibition using a fear-potentiated startle paradigm in gonadectomized female and male rats [50,51]. Although gonadectomized (GDX) female rats suppressed startle during CS+/− presentation, GDX + estrogen females did not [50,51]. Additionally, GDX male sham and GDX male + estrogen inhibited startle in a CS+/− test [50]. This indicates that estradiol may play a role in the ability to inhibit fear [50]. The Greiner, Day and Foilb studies did not explicitly investigate individual differences of estrus phase vs. behavioral performance. It is possible that some of the individual differences described in Fig. 1C could be explained via differences in estrus phase.

In addition to estrogen, there may be a role of neuropeptides such as oxytocin, vasopressin and corticotropin releasing factor (CRF) in sex differences in safety learning, though this area is understudied. For example, there are sex differences in CRF stress response that play a role in anxiety disorders such as PTSD ([62,63]; Risbrough & Stein, 2006). Sex differences also exist in oxytocin and vasopressin signaling [64]. Although CRF, oxytocin, vasopressin and other neuropeptides have been moderately investigated in studies of discrimination (not discussed in this review), their roles in conditioned inhibition have not been explored.

4.3. Summary of conditioned inhibition in males vs. females

Taken together, the above human studies demonstrate that males and females exhibit learned safety via conditioned inhibition. However, specific sub-groups may show impairment in this type of safety learning. For example, PTSD + males, as well as traumatized females with low estrogen do not transfer the safety of a CS− cue to a CS+/−presentation, which indicates that despite contingency awareness, they are unable to downregulate their fear when threat and safety cues are presented together. Consistent with work reviewed above on discrimination, traumatized females with high estrogen are indeed able to learn safety via conditioned inhibition. Some of the sex differences found in studies of clinical populations could relate to differences in the types of trauma leading to PTSD, where compared to men, women are more likely to have high impact trauma beginning at younger ages [65]. Lacking in the above-reviewed literature is a study specifically examining conditioned inhibition in human males vs. females. Furthermore, while the effects of estrogen have been explored in conditioned inhibition, the role of other neuropeptides such as oxytocin, vasopressin or corticotropin releasing factor that may have an effect on sex differences have yet to be examined.

In the literature reviewed here, we find that while male rodents exhibit conditioned inhibition, females may not, but see [44,50]. An analysis of individual rats in the Greiner study (Fig. 1) demonstrates that some female rats do indeed show conditioned inhibition of freezing during presentation of the CS+/−. The neuronal mechanisms underlying the general lack of fear suppression in the majority of females in this study have yet to be explored. The work from Toufexis using fear-potentiated startle suggests that females in a high estrogen phase may demonstrate less conditioned inhibition than males. The question remains if a similar result would be found in paradigms other than fear-potentiated startle. Furthermore, while darting behavior was not examined in several of the studies reviewed, females did demonstrate conditioned inhibition of darting in at least one study [40]. This suggests that future studies should continue to focus on examining multiple behaviors in female and males in order to fully assess sex differences in safety learning.

5. Future directions: parallel studies in humans and clinical models

Moving forward, it is essential to investigate how sex differences in human and animal safety studies converge as well as differ. One way to ensure consistency between studies is to conduct rodent and human studies in parallel. This work has begun in several labs [18,38,66]. For example, in an fMRI study on humans designed to mimic rodent models of learned safety, safety conditioned subjects had decreased amygdalar and increased dlPFC responses compared to fear conditioned subjects [66]. However, while the study investigated both male and female subjects, data was not presented between sex, nor was cycle phase noted [66]. Another study investigated how variance in estrogen levels affects activation of the vmPFC and amygdala during extinction in female rodents and humans [38]. In female rodents in metestrus, activation of estrogen receptors facilitated extinction recall and was correlated with increased activity (as measured by expression of the immediate-early gene c-Fos) in the vmPFC with decreasing activity in the amygdala [38]. In women, an increase in estrogen was associated with extinction memory retention as well as increased activation of the vmPFC during extinction recall [38]. Taken together, these results indicate that estrogen could serve a protective function against elevated fear and anxiety.

More recently, an elegant set of parallel studies across species by Meyer et al. investigated neural activity during inhibition of threat responding in both healthy human adults as well as mice [18]. By recording neural activity (via calcium imaging in mice and fMRI in humans), Meyer et al examined functional connectivity between the ventral hippocampus, amygdala and prefrontal cortex [18]. In rodents, as well as in humans, the group found that neural activity in the ventral hippocampus increased during conditioned inhibition relative to a threat cue alone. Furthermore, the ventral hippocampus activity was highest during the onset of a compound cue (threat + safety) in neurons projecting to the infralimbic cortex, indicating that perhaps this connection is required for circuit processing of cue ambiguity. In parallel, humans also demonstrated interactions between the ventral hippocampus and dACC (PL/IL homologue) during conditioned inhibition. Importantly, the study did not find functional connectivity between the amygdala and vHPC during conditioned inhibition in humans or mice. While this study provides a fresh look into similarities in safety processing between mice and humans, it does so solely in male mice, and does not segregate human data between male and female subjects. Because females have differing behavioral responses to threat inhibition [8,10,40,48], we hypothesize that the circuit of parallel neural activity in female mice and humans during threat inhibition likely differs from the presented data.

6. Conclusions on Safety Discrimination VS. Conditioned inhibition

In this review, we explored how behavioral differences in males and females alter as subjects take on more complex safety learning tasks. Learned safety via discrimination requires downregulation of fear to a safety cue while continuing to express fear to a threat cue. In this case, both sexes in human and rodent studies largely exhibited learned safety. In human populations, subjects with high symptoms of fear/anxiety disorders were more likely to generalize between fear and safety cues, though cue perception (fear or safety) remained intact. Similarly, the reviewed studies largely concur that in cue discrimination tasks, male and female rodents differentiate between fear and safety cues. A selective summary of reviewed studies is presented in Table 2. However, males appear to show comparatively ‘more’ discrimination than females. This could be due to overall higher initial freezing to the CS+ in male groups. Discrimination between cues in both sexes lasts over multiple days of training and testing [40,44], but see [39]. Estrogen may have an effect on discrimination, though studies are limited [45,50]. Conversely, the reviewed work suggests that while males suppress freezing during conditioned safety via either retardation or summation tests, females do not (but see Foilb). Moving forward it will be important to acknowledge the differences in cue-driven safety behaviors since, as reviewed here, accurate discrimination between the fear and safety cues does not automatically guarantee that the safety cue can effectively reduce fear elicited by a fear cue. The conditioned inhibitory ability of a safety cue to reduce fear is typically assessed via a fear/safety compound cue and compared to the fear expressed to the fear cue alone. While the studies examined in this review largely provide behavioral differences between males and females in learned safety, the neural mechanisms and circuitry underlying these behaviors remain to be explored (but see [3] for a proposed circuit mediating learned safety). Activity analyses via the expression of the immediate-early gene c-Fos suggest that sex differences exist within the amygdala as well as the BNST during discrimination tasks [7,46]. Sex differences in in vivo recordings from these regions during a combined discrimination and conditioned inhibition paradigm may yield interesting results for the field. It is likely the neural mechanisms mediating fear versus safety cue discrimination will differ from those mediating conditioned inhibition during a combined fear/safety cue.