The association between acute stress & empathy: A systematic literature review

Empathy is a fundamental component of our social-emotional experience. Over the last decade, there has been increased interest in understanding the effects of acute stress on empathy. We provide a first comprehensive-— and systematic — overview identifying emerging patterns and gaps in this literature. Regarding affective empathy, there is abundant evidence for stress contagion — the ‘spillover ’ of stress from a stressed target to an unstressed perceiver. We highlight contextual factors that can facilitate and/or undermine these effects. Fewer studies have investigated the effects of acute stress on affective empathy, revealing a nuanced picture, some evidence suggests acute stress can block contagion of other ’ s emotions; but again contextual differences need to be considered. Regarding cognitive empathy, most studies find no conclusive effects for simplistic measures of emotion recognition; however, studies using more complex empathy tasks find that acute stress might affect cognitive empathy differentially for men and women. This review provides an important first step towards understanding how acute stress can impact social-togetherness, and aims to aid future research by highlighting (in)congruencies and outstanding questions.


Introduction
Humans are social animals, wired to engage with and understand others in our social world (Depue and Morrone-Strupinsky, 2005). Indeed, not only is understanding the feelings and thoughts of others critical to navigating social life, such understanding helps foster and maintain social connections (de Waal and Preston, 2017). Two well-established routes to gather information about others-commonly grouped under the umbrella term "empathy"-are by simulating/mirroring others' emotional states (affective empathy/ contagion) and by mentalizing about their thoughts and feelings (cognitive empathy). Unfortunately, there are a myriad of factors that can undermine these processes: Some perceivers are less skilled, some targets are less readable, and, even assuming perfectly skilled perceivers and readable targets, numerous situational factors can interfere with the accurate decoding of others' internal states. Our objective is not to review all the factors that can influence empathy, as there are numerous comprehensive reviews on this topic (Baron-Cohen, 2002;Bérubé et al., 2021;Gonzalez-Liencres et al., 2013;Preston et al., 2020). Rather, we focus on one factor that has received increasing research interest: the experience of acute stress. We begin by presenting a brief overview of the acute biological and psychological stress response, and of empathy. We then turn to the literature on acute stress and affective empathy/ contagion. We begin by reviewing seminal research demonstrating basic contagion processes in humans-that is, the transfer of an affective state like stress or pain from one individual to another. While this work does not test the effect of stress per se on affective empathy, it is important for understanding the basic processes underlying affective empathy/ contagion. We then review research that specifically manipulates acute stress in perceivers to probe the effects of that stress on the perceiver's affective empathy for a target in distress. We then review the literature investigating the effects of acute stress on cognitive empathy (see supplemental materials for search queries used in the systematic literature review, as well as a PRISMA flow chart). Finally, we conclude with a discussion of emerging patterns and directions for future research. homeostatic balance (Lazarus, 2006). The acute stress response encompasses the synchronized activation of several biological systems. Two of these systems are frequently investigated for their effects on human behaviour: the fast-acting sympathetic branch of the autonomic nervous system, or sympathetic nervous system (SNS), and the slower acting hypothalamic-pituitary-adrenal (HPA) axis. Activation of the SNS results in, among other things, the release of catecholamines and increased heart rate. Activation of the HPA axis triggers a hormone cascade: specifically, the central release of corticotropin-releasing hormone triggers the release of adrenocorticotropic hormone in the anterior pituitary, which, in turn, stimulates the release of the downstream glucocorticoid cortisol from the adrenal gland. Through a feedback-loop cortisol reaches the central nervous system where it binds to glucocorticoid and mineralocorticoid receptors and inhibits further release of cortisol by suppressing the release of corticotropin-releasing hormone (Chrousos, 2009).
Central glucocorticoids exert their effects through non-genomic (rapid), as well as genomic (slow) processes (de Kloet et al., 2008;McEwen et al., 2016). Research indicates that the immediate non-genomic effects change the way the situation is appraised and attended to (de Kloet et al., 2005;Sapolsky et al., 2000), including how information about the situation is retained and retrieved. In this regard, attentional processes become more selective (Chajut and Algom, 2003;Hermans et al., 2011), resulting in a narrowing of attention, and reduced interference from irrelevant and distracting information (Plessow et al., 2011;Putman and Roelofs, 2011;Sänger et al., 2014), which allows the organism to cope with the stressor(s). In addition to this narrowing of attention, it is thought that acute stress elicits a shift from more cognitively demanding information processing to more habitual response patterns (Schwabe and Wolf, 2009;Wirz et al., 2018). This shift should enable more rapid information processing and, thus, might be beneficial for dealing with stressors (and ultimately, survival), especially when quick decisions are necessary. However, this shift can also come at a cost, as mental flexibility is diminished under habitual-favouring strategies (Nitschke et al., 2020a;Shields et al., 2016;Vogel et al., 2016).
In addition to these processes, research indicates that acute stress results in an attentional shift to what is salient in the environment (Hermans et al., 2014). This shift may have important consequences for social information processing in general, and for empathy in particular. On the one hand, given the inherent salience of social information, such a shift would be expected to (further) augment social information conveyed in the environment. Moreover, to the extent that attention is biased to potential threats in the environment, we might expect that the processing of others' negative emotions to be especially enhanced (through vicarious means or by making cognitive inferences). On the other hand, if demands are particularly high and, possibly, surpass available resources, such a shift could result in a more ego-centric focus (in an effort to regulate one's emotions); in this case, the identification and/or sharing of others' emotions might be more challenging.

Empathy (a brief overview)
Empathy is a multifaceted construct, and definitions of empathy can vary substantially between researchers (Batson, 2009;Coll et al., 2017;Davis, 1983;Kogler et al., 2020;Stietz et al., 2019). Here, we broadly define empathy as a set of abilities (and motivations) that can help us identify, understand, and respond appropriately to another person's experience (Baron-Cohen, 2002;de Waal and Preston, 2017).
One way to understand another's emotional state is through affective empathy, which is defined as the ability to share an emotional experience with another person; this encompasses such processes as simulation, neural resonance, and emotion contagion. For example, when witnessing someone in distress, there will be at least a partial transfer of that negative experience to the perceiver (e.g., seeing someone else step on a sharp object might trigger surprise and an unpleasant feeling in the perceiver). It is thought that this experience sharing gives the perceiver valuable insight into the emotional state of the other person, and might help the perceiver to relate to the target's experience and thus respond appropriately. Another way to understand another's emotional state is through cognitive empathy; this encompasses such processes as emotion recognition, perspective taking, and mentalizing more broadly. Rather than simulating the other's emotional state, cognitive empathy is thought to rely on more deliberate attempts to reason about the other's mental state by, for example, using rule-based knowledge (i.e., principles about mental states and how those states govern behaviours) and/or by referencing the self (i.e., "how would I feel in that situation?") (see, e. g., Mitchell, 2009). In addition to helping the perceiver relate to a specific target's experience, cognitive empathic abilities help reduce uncertainties in social interactions more generally by allowing us to understand the intentions (and feelings) of others (FeldmanHall and Shenhav, 2019).
Importantly, these two facets of empathy are not just conceptually distinct from one another, but partly rely on non-overlapping neuralcomputations (Schurz et al., 2020). Affective empathy/ contagion is generally thought to be a relatively automatic, spontaneous process (Preston and de Waal, 2002;Prochazkova and Kret, 2017). 1 As alluded to above, affective empathy largely relies on more "bottom-up" processes, such as increased activation in the amygdala, in thalamic regions, and in the anterior cingulate cortex; conversely, cognitive empathy largely relies on more "top-down" processes, such as increased activation in the dorso-lateral and ventro-lateral prefrontal cortices, and in the temporoparietal junction. There are, though, some common areas of activation, including the anterior insula and the anterior middle cingulate cortex (for reviews: de Waal and Preston, 2017;Lamm et al., 2019). Notably, research indicates that more ecologically valid (i.e., naturalistic) empathy tasks, and, in particular, empathy tasks evoking complex social cognition that require the individual to draw inferences from various sources, have been shown to result in coactivations of both affective and cognitive networks (Schurz et al., 2020). Indeed, while these two systems are thought of as (somewhat) distinct, they likely work in coordination and are difficult to parse in everyday experience.
The overall functions of these empathic abilities and behaviours are thought to increase social coherence, social bonding, and ultimately to enable group survival (for a review: de Waal and Preston, 2017). In this regard, it has been proposed that understanding others' thoughts and/or emotions can augment empathic concern and sympathy (i.e., feeling for the other person), and ultimately to prosocial, helping behaviours (however, for two more critical discussions see: Decety, 2021;Sassenrath et al., 2021). In this way, empathy can be seen as a complex set of competencies that help us cope with, and interact in a complex social environment. Given the central role that empathy plays in the human experience, and the ubiquity of stress, it is no wonder that researchers have become interested in understanding whether and how stress influences empathic responding.

Acute stress and affective empathy/contagion
In essence, affective empathy is when we observe someone else in a particular affective state and experience a similar affective state; this can occur in response to real-life social interactions (e.g., a close other is crying, and we feel sad), but also in response to more abstract phenomena (e.g., we see a character in a movie crying, and we choke up). A basic paradigm for studying this in the lab involves a "perceiver," who witnesses another person (a "target") undergo an emotional, stressful, or painful experience; researchers then measure the "contagion" of the target's affective state (emotion, stress, or pain) in the perceiver; this 1 The emphasis here is on "relatively" as research has shown that even automatic processes, such as affect sharing, are not beyond motivational and cognitive control (Duffy and Chartrand, 2015;Lamm et al., 2007;Weisz and Cikara, 2020;Weisz and Zaki, 2018). contagion is typically quantified as the correlation between the perceiver's and the target's biological and/or psychological response (see Box 1), but sometimes just focuses on physiological and/or psychological changes in the perceiver's state. In addition to this basic paradigm, researchers have also studied the simultaneous experience of an acutely stressful experience in two or more individuals. Here, the degree of coactivation is of interest as it implies a shared (stressful) experience. A higher degree of activation would thus indicate greater degree of affect sharing. A third paradigm involves a stressed perceiver observing another person's emotions. Finally, more recently, researchers have tried to go beyond basic contagion processes to probe differential effects in targets and perceivers. Here, the focus is on whether exposure to the distressed target leads to an allocentric, other-directed experience or an egocentric experience, in which the perceiver's own feelings take precedence. See Table 1 for an overview of effects. We begin by reviewing research demonstrating affective empathy/stress contagion processes in humans that are initially unstressed themselves observing others in distress-that is, the (dis)stress that occurs when observing someone else in distress. We then turn to work that specifically manipulates acute stress in perceivers. While the former does not test the effect of stress on affective empathy/ contagion (because stress in the perceiver is the measure of emotion contagion), these studies are important for understanding the basic processes underlying affective empathy/ contagion-does the stress of a target spill-over to an observer, when the observer is unstressed-as well as factors that can moderate the effect and, as such, lay the foundation for subsequent research that manipulates acute stress in initially unstressed perceives. (Table 2-5).

Evidence of affective empathy/stress contagion
In now seminal studies on affective empathy/contagion in humans, Singer et al. (2004) and Jackson et al. (2006) had participants observe pictures of others in pain while undergoing functional magnetic resonance imaging (MRI); results showed that observing others in pain (versus not) elicits the activation of brain areas commonly involved in the experience of first-hand pain providing evidence for the "neural resonance" of other's affective states. Since these initial investigations there have been numerous studies showing that observing others in pain leads to emotion processing that is similar to the way people process self-experienced pain (for reviews : Jauniaux et al., 2019;Lamm et al., 2019Lamm et al., , 2011. While pain and stress are distinct processes, these studies provide some of the first evidence that an aversive experience can spill-over from a target to a perceiver. More direct measurement of affective empathy/contagion comes from research looking at the effects of observing others in distress on the perceiver's hormonal stress-response (For reviews: Engert et al., 2019;White and Buchanan, 2016). For example, Buchanan et al. (2012) manipulated acute stress with a modified version of the Trier Social Stress Test (TSST; see Box 1); they then measured stress reactivity in the TSST panelists-that is, the "judges" (n = 20; 11 women), while the judges observed the participants (n = 152; 55 women) undergo the TSST. Results showed that the TSST judges' stress response (measured by cortisol reactivity) was positively correlated with the participants' stress response (cortisol reactivity). This study thus provides the first evidence of stress-contagion in perceivers, in the form of a (partially) synchronized HPA axis response. One caveat, however, is that the TSST judges were not neutral; rather, they were actively stressing the participants, which raises questions about interpretation (e.g., maybe when judges were more severe, they were also more stressed, as were the participants).
This issue was addressed in a study by Engert et al. (2014) and colleagues, who had neutral participants (n = 211) observe another participant (n = 151) undergoing the TSST. Consistent with Buchanan et al. (2012), results showed that perceivers experienced a significant increase in cortisol; moreover, perceiver's cortisol levels were correlated with the targets' cortisol levels, suggesting contagion. This study also manipulated how perceivers witnessed the targets by having some perceivers watch the TSST through a one-way mirror, and others watch through video-feed; although both modalities were effective in increasing perceivers' cortisol levels, the effects were stronger in the one-way mirror condition, suggesting that more salient and/or palpable experiences have a stronger effect on emotion contagion. 2 Another interesting observation about this study is that additional analyses revealed that closeness/familiarity moderated the emotion contagion effect: 40% of romantic partner dyads showed an increase in cortisol compared to only 10% of the stranger dyads; this observation that is consistent with the well-documented finding that familiarity facilitates empathy (de Waal and Preston, 2017).
In addition to being in an established relationship, other factors appear to influence the stress-familiarity-emotion contagion effect. For example, Schury et al. (2020) had participants work in groups of 4-5, and were randomly assigned to either the "shared identity" or "personal identity" condition. In the shared identity condition, participants i) wore the same colour T-shirt, ii) sat at the same table, iii) had to identify similarities between the group members, and iv) were told that their performance would be scored on a team-level; by contrast, in the personal identity condition, differences between participants were emphasized. Following this manipulation, one participant from each group (n = 28) was randomly assigned to undergo the TSST, while being observed by the other group members (n = 89). Results showed that 25% of perceivers in the shared identity condition showed an increase in cortisol, compared to just 7% of perceivers in the personal identity condition. Of note, a similar study by the same research group (Erkens et al., 2019) failed to show this familiarity/similarity effect, with only 17% of participants-regardless of the identity manipulation-showing an endocrine response to a stressed target. However, the lack of an effect in this study may have been because participants watched a pre-recorded video-taped session of another person (confederate) undergoing a stressor, whereas Schury et al. (2020) had participants sit in the same room where the TSST took place, quietly observing the stressful situation. This methodological difference between the two studies is consistent with Engert et al.'s 2014 finding that emotion contagion (cortisol resonance) was higher when participants observed the stressed individual through a one-way mirror, rather than through a video-feed.
Other evidence for affective empathy/emotion contagion comes from the literature on autonomic synchrony, and mimicry, in motherinfant dyadic interactions (for a review Palumbo et al., 2017). For example, Waters et al. (2020Waters et al. ( , 2017Waters et al. ( , 2014 investigated the effects of mothers' stress on infants' autonomic response. In a first study, mothers (n = 69) were separated from their infants before receiving negative, positive, or neutral feedback. When stressed mothers (i.e., those who received negative feedback) were reunited with their infants, the infants showed increased autonomic arousal that mirrored that of their mothers (i.e., greater ANS covariation in the stress condition) (Waters et al., 2014). In a follow-up study (Waters et al., 2017), they added a twist: the mothers (n = 105) experienced the same negative, positive or neutral feedback; however, upon their return, the infants were either placed in the mother's lap ("close touch"), or not. Infants in the close-touch condition showed greater covariance with autonomic markers over time, compared to infants who did not sit on their mother's lap, again an observation that supports the notion that more palpable experiences have stronger effects on emotion contagion (cf. Engert et al., 2014). Similar effects were observed by Manini et al. (2013). In this study, the researchers used the "mishap paradigm" in which a child played with a toy that broke while playing, an experience that lead to increased facial thermal levels (thought to indicate negative affect/distress); results 2 When focusing only on the partner dyads (cf., Engert et al., 2018), the video and mirror conditions were equally effective in eliciting a cortisol resonance in the perceiver.
showed that mothers (n = 9), compared to strangers (n = 9; all female), who observed their child in this stressful situation matched the child's facial thermal levels, suggesting an increase in stress-resonance (cf. Ebisch et al., 2012). Lastly, Waters et al. (2020) found that when parents (n = 114; 55 in suppression group) tried to suppress their emotions ("try not to show any emotions'') to their child following a stressor (TSST), it

Box 1
Manipulating Stress in the Laboratory.
Acute stress is a multifaceted process. Common indicators of an acute stress response are: an increase in glucocorticoids (resulting in elevated salivary cortisol); an increased activation of the sympathetic nervous system (SNS; resulting in increased heart rate, electrodermal activity, or salivary alpha amylase); and markers of emotional arousal (i.e., psychological distress).
The most commonly used method to induce stress in the laboratory is the Trier Social Stress Test (TSST; Kirschbaum et al., 1993), a "psychosocial" stressor designed to elicit a robust biological and psychological response-characterized by increased SNS activation, HPA stimulation, and feelings of distress. The basic paradigm is that participants are informed that they will be undergoing a mock job interview; they are asked to select a job that they would ideally like to have one day and are given a short period of time (e.g., 10 min) to prepare their argument for why they would be a good candidate for the job. Following this, the participant delivers their argument to two "behaviour analysts" (research confederates) who are instructed to remain neutral and provide no verbal or non-verbal feedback to participants. To further augment the stress response, participants believe they are being videotaped. Finally, after the mock job interview, participants must complete an arithmetic task under time pressure, also in front of the analysts. Notwithstanding the public speaking component, the defining characteristic of the TSST is its social-evaluative component as participants believe they are being judged. This social component may also have implications for subsequent social behaviours; as Dickerson and Kemeny note, because social stressors highlight aversive interpersonal interactions as the source of distress (Dickerson and Kemeny, 2004), they might affect social behaviours that follow. Many variants of the classic TSST exist, for example adaptations for the use in groups, which might highlight the already salient social nature of the situations (von Dawans et al., 2011), or for the use in a neuroimaging environment, such as the Montreal Imaging Stress Task (Dedovic et al., 2005).
Another common procedure to elicit stress in the laboratory is the cold pressor task (CPT; Hines and Brown, 1936). Here, participants are instructed to immerse their hand in ice-cold water (typically 4 • C, or 39.2 • F), sometimes repeatedly. As such, the CPT relies on an aversive physical experience (i.e., pain) to elicit the stress response. This immersion of the hand into ice-cold water triggers a robust vasopressor effect, and subsequent sympathetic nervous system (SNS) activation, resulting in increased heart-rate and blood pressure. Of note, whereas the TSST is a psychosocial stressor, the CPT relies on an aversive physical experience (i.e., pain) to elicit the stress response; while both paradigms are undoubtedly stressful, evidence suggests that they do not produce identical stress profiles (and, thus, might not be comparable). For example, the CPT typically fails to elicit a robust HPA axis (i.e., cortisol) response, which is in stark contrast to the TSST (Schwabe et al., 2008). There are, however, social variants of the CPT, coined the social-evaluative CPT (Schwabe et al., 2008;Schwabe and Schächinger, 2018), or the Maastricht Acute Stress Test (Smeets et al., 2012). Here, the aversive experience of immersing the hand in cold water is combined with a social-evaluative component; and this social-evaluative CPT has been shown to result in an activation of the HPA axis, making this version of the CPT more comparable to the TSST.
An important factor to consider when interpreting findings from acute stress studies is that some effects of acute stress tasks last much longer than the actual task. For example, the TSST elicits an immediate SNS response, which dissipates rather quickly; this is followed by a slower cortisol response that can last for up to 40-minutes after stress induction (Goodman et al., 2017); it is therefore possible to measure effects of acute stress after a longer delay. Although importantly, the acute stress response unfolds in several "stages" (Hermans et al., 2014;Joëls et al., 2012) and we can expect to find different effects of glucocorticoids depending on whether they are measured after a short or a long delay (de Kloet et al., 2008;Henckens et al., 2011;Schwabe, 2017). The timing of the dependent variable in relation to the stress task therefore matters. Of note, some studies reviewed here assessed empathy more than 40 min after the stress induction (e.g., Graumann et al., 2021;Wingenfeld et al., 2018). Given the time course of the cortisol response, results from short delay (i.e., < 40 min after stress induction) studies, and long delay (i.e., > 40 min) studies should be interpreted differently, or at least with greater consideration when comparing findings. In addition, studies often use modified versions of stress tasks (for various reasons), and these modified versions do not always result in similar magnitude of the stress-responses as the original paradigms. The degree of dissociation from the perceiver's experience to the target's.
Notes: Three contexts in which affect sharing can occur in the context of acute stress. The initial state of the actors involved is indicated in the S columns: An X indicates no initial stress; a check-mark indicates stress on the onset of the task. The contexts: A) Stress Contagion from a stressed target to an initially unstressed perceiver. B) Simultaneous experience of a stressor; how does a shared experience relate to affect sharing. C) The effects of acute stress in a perceiver on affect sharing with a target (which emotions are unknown at the onset). seemingly paradoxically increased their child's autonomic stress response (although notably only in mother-child dyads, not for fathers). This observation points to the potential buffering effect that affect sharing can have for the stressed individual. Indeed, parents that suppressed their emotions were rated as less warm and less engaged during a subsequent interaction, compared to the non-suppression group.
One question arising out of the affective empathy/contagion research is the extent to which the perceiver's distress reflects affect sharing (that, presumably, should lead to sympathy and concern for the other) versus a rapid modeling of others' experiences, but with an egocentric evaluation of that affective state-that is, personal distress. In fact, there is some evidence for the latter, at least for stranger dyads. Dimitroff et al. (2017) had participants (n = 63; 41 women) observe targets who experienced different levels of stress: specifically, targets were 1) actively stressed (via modified TSST), 2) recovering from stress (i.e., recovering post TSST), or 3) not stressed; they then measured both the perceiver's and target's cardiac activation patterns, as indexed by changes in interbeat-intervals (IBI) from baseline. (Note: IBIs are a continuous measure of heart rate, indexed by beats per minute, such that shorter IBIs reflect faster/higher heart rate). When perceivers observed a target who was recovering from stress, their cardiac activation pattern was similar to that of the target (i.e., both perceivers and targets showed longer IBIs); however, when perceivers observed a stressed target, their cardiac activation pattern diverged from that of the target (i.e., perceivers showed longer IBIs (importantly, also in relation to baseline), whereas stressed targets showed shorter IBIs). Importantly, while (↑) correlated SNS response in suppression group for mothers-child (7-11 years), no linkage for fathers; no effect in control group (no-suppresion).
Notes: a observers only; b a total of 20 TSST panelists observed a total of 112 speakers, thus each TSST panelist had several dyadic observations; b defined as a cortisol increase of a minimum of 1.5 nmol/l from baseline; d mothers and strangers. TSST= Trier Social Stress Test; HPA= hypothalamic-pituitary-adrenal axis; ANS= autonomic nervous system; SNS= sympathetic nervous system; PNS= parasympathetic nervous system; IBI= Interbeat-Intervals; HR= heart rate; VC= ventricle contractility; PEP = pre-ejection period. Higher pain ratings in the friend condition, compared to strangers or alone condition (Exp-1); HPA suppression results in higher pain ratings in the stranger condition, compared to control (Exp-2). A shared experience between the stranger prior to stress/pain exposure also increased pain ratings for the stranger (Exp-3).
Nahleen et al. Notes: a stress elicited in the perceiver and in the target. CPT= Cold Pressor Task. TSST-G= group version of the Trier Social Stress Test. HPA= hypothalamic-pituitaryadrenal axis; ANS= autonomic nervous system; SNS= sympathetic nervous system; sAA= salivary alpha amylase; EDA= electrodermal activity.
typically stress is associated with short IBIs (i.e., increase in cardiac activation), longer IBIs (i.e., a decrease in cardiac activation) as well can indicate stress in a perceiver (Hagenaars et al., 2014;Lang et al., 1993), in particular for more passive stressors (i.e., video stimuli). In addition, longer IBIs may also be indicative of heightened attentional processes (Bradley et al., 2001;Graham and Clifton, 1966). Thus, these findings might suggest that observing a target in distress may induce personal distress, or at least an anticipatory response to aversive content (as depicted by the stressed targets) that is self-relevant. A similar divergence effect was reported by Young et al. (2017) who had participants (n = 50; 27 women) watch a series of short movie clips of characters in neutral, fearful, or painful situations. Participants who reported the tendency to feel vicarious pain for others, and rated at least some of the video-clips as distressing, were more likely to slow their breathing in anticipation of the pain-situation depicted, rather than mimic the breathing patterns of the target in the video-clip. Of note, this "freezing" response to aversive or stressful stimuli has previously been reported (Hagenaars et al., 2014), and could indicate self-protective avoidance ↑ Stress group showed increased activation of pain areas of the brain (pain matrix), compared to control. Notably, also in one of the control conditions (anesthesized hand). Nitschke et al. (2020b) 73 m,f TSST facial mimicry ↓ Acute stress lead to reduced smile mimicry; no change for frown mimicry; cortisol as moderator (*within-subject design) Notes: TSST= Trier Social Stress Test; MIST= Montreal Imaging Stress Test.
For men in the stress condition cortisol led to higher accuracy, relative to low levels of cortisol (median splits; n = 8); For women the opposite pattern was observed. RMET → Stress and control group did not differ on RMET scores. Wolf et al. (2015) 99 (49  stress Stress led to higher empathic accuracy for men, compared to control. Higher EA was positively associated with cortisol response; For women no effect was observed. Notes: a sub-sample, healthy controls only, part of a larger study also including borderline patients and patients with Cluster C personality disorders (e.g., anxiety, fear) b sub-sample, healthy controls only, part of larger study protocol also including borderline diagnosed women; c same sample as Wingenfeld et al. (2018 behaviour (cf. Nazarewicz et al., 2015). 3 Although Dimitroff et al.'s and Young et al.'s findings suggest that personal distress (from watching others in distress) may interfere with pure emotion contagion, an alternative explanation comes from work by Stellar et al. (2015). Here, the researchers had participants (4 studies; n = 299) watch video clips of either stressed individuals, or individuals in neutral or positive situations. Watching stressed individuals resulted in decreased heart rate, relative to the other conditions, and increased respiratory sinus arrhythmia (increased activation of the parasympathetic nervous system). As others have pointed out (Eisenberg et al., 1989;Keltner et al., 2014;Porges, 2007), the authors interpret the increase in vagal activity as a necessary condition for emotional concern ("compassion") for others and, in turn, for prosocial actions. Specifically, increased activation of the parasympathetic nervous system might be associated with a reduction in arousal in the perceiver and, in this way, might enable a shift in attention to the person in need (Stellar et al., 2020), rather than experiencing emotion contagion. 4

Evidence of affect sharing/contagion in simultaneously stressed perceivers
In the following section we review the effects of a simultaneous stressful experience in perceivers and targets-i.e., both are undergoing a stressful experience at the same time. Following up on research demonstrating empathy/contagion in mice (Langford et al., 2006), Martin et al. (2015) aimed to study stress contagion in humans and had friend-dyads and stranger-dyads undergo the cold pressor task (CPT; see Box 1), in which participants repeatedly immersed their hand in cold (4 • C, 39℉) water for 30 s (recall from the previous section that contagion effects are stronger with familiar other). After each immersion dyad (n = 36), participants rated how much pain they were experiencing as an index of emotion contagion. Consistent with the emotion contagion studies described above (and Langford et al., 2006), pain ratings were significantly higher when a friend was present compared to when a stranger was present, and compared to the alone condition, indicating greater emotion contagion among familiar others. Of note, in another experiment in this series, Martin et al. (2015) manipulated friendship by having stranger dyads play a collaborative video game prior to the CPT; results again showed that these "friends" showed greater pain contagion than did strangers, thus providing evidence for the causal role of closeness/familiarity/friendship in facilitating contagion. Similar findings were reported in a recent study by Nahleen et al. (2019). Here, participants (n = 90; all female) first got to know another participant (in actuality, a research confederate) in order to build rapport. Following this, participants either did a version of the CPT (7 ℃, 45℉) for 1 min alone, or in the presence of the confederate. As with Martin et al., participants in the shared condition reported higher levels of sensory pain and higher levels of perceived stress, compared to those in the alone condition. Similar findings are also reported in a study by Tashjian et al. (2022). This study used a novel stress paradigm in which participants (n = 156) experience a haunted-house in small groups composed of a mix of friends and strangers. Here, results show that the presence of friends in the experience increased overall physiological arousal, or more specifically, a higher friends-to-stranger ratio was associated with an increase in tonic electrodermal activation. This indicates that the presence of friends can increase the overall stressful arousal and highlight the importance of contextual factors-in this case, the social dynamics of being with familiar others versus strangers. Historically, it has been thought that the greater empathy experienced with familiar others is evolutionarily adaptive as it should promote survival of in-group-and especially kin-over outgroup members. But an alternative interpretation is also possible in that we may feel more comfortable showing our distress in the presence of familiar others (because they can be trusted), whereas such vulnerability could be more risky-even dangerous-when we are in the presence of strangers, or alone.
Similarly-and in line with Martin et al. (2015)-Denk et al. (2021 found that a shared stressful experience can lead to a synchronized stress response. Here, participants either underwent a stressor as a group (n = 75; group version of the TSST) or a control condition (n = 63) (mean group size= 3.2). Importantly, the group version of the TSST tries to minimize contact between participants; for example, visual barriers prevent participants from having visual contact. Despite this limited interaction the authors observed that group peers' HPA activation as well as ANS activation predicted an individual's cortisol or alpha amylase response, indicating that stress contagion does not require actual (eye) contact between individuals, but might occur through other means as well (e.g., olfactory or audible information).
Notably, the discussed findings on shared stressful experiences are in line with the notion that shared experiences in general-both pleasant as well as unpleasant-are generally amplified and experienced as more intense (Boothby et al., 2014).

Effects of acute stress on affective empathy/contagion
Importantly, as noted at the outset, one caveat about the studies described above is that the perceiver's stress (or pain) is, or is part of, the measure of emotion contagion. Also, it is difficult to distinguish whether the perceiver's stress (or pain) response reflects pure emotion contagion (i.e., stress that is grounded in the target's experience) or includes feelings of personal distress, given that an overlap in biological activation (e.g., increased cortisol in both target and perceiver) indicates a coinciding stress response, but does not clarify whether the stress response is shared or egocentric (for a review: Lamm et al., 2016). Thus, this approach is not ideal for assessing the effects of acute stress, per se, on affective empathy; rather, what is needed is to elicit stress in the perceiver prior to the interaction with the target in order to test how acute stress itself influences affective empathy. We now turn to studies in which researchers have experimentally manipulated stress in perceivers to look at the causal effects of acute stress on affective empathy/emotion contagion.
In one study, Buruck et al. (2014) manipulated acute stress with the TSST (vs. a control condition; n = 104; 52 stress); participants then viewed pictures depicting neutral and painful situations and rated how painful the experience was for the person in the picture. Results showed that participants in the TSST condition had lower pain ratings (i.e., rated the pictures as less painful for the target), compared to those in the control condition. This study provides initial evidence that stress may undermine perceptions of others' pain.
In another study, Gonzalez-Liencres et al. (2016) also manipulated acute stress with the TSST (vs. a control condition; n = 52; 23 stress); similar to Buruck et al.'s study, participants then viewed pictures of others in pain but, instead of rating how much pain the person pictured was experiencing, Gonzalez-Liencres et al. had participants rate how unpleasant the pictures made them (i.e., the participants) feel. Here, results showed that participants in the TSST condition had higher unpleasantness ratings compared to those in the control condition.
At first blush, these two findings might seem inconsistent, but it may be that stress alters the mechanisms that modulate the automatic sharing of emotions, possibly by inducing a more ego-centric focus. Research by Tomova et al. (2017) is consistent with this idea. Tomova et al. investigated the link between acute stress, affect sharing (i.e., pain), and prosocial action (amount of money allocated to a future participant). The authors manipulated acute stress by having participants (n = 67; all male) undergo a modified version of the Montreal Imaging Stress Task (n = 35; Dedovic et al., 2005) or control task (n = 32) while undergoing functional imaging; they then showed participants a series of pictures that depicted targets undergoing a painful procedure (needle injection) or one of two control pictures (target's hand was "anesthetized", or a q-tip "prick"). Results showed that participants in the stress (versus control) condition showed greater activation in brain areas associated with pain-processing suggesting greater affective empathy/emotion contagion. Interestingly, stressed (versus control) participants also showed greater activation in the pain network when viewing the anesthetized hand; given that this condition should not result in increased activation of the pain network, it suggests that stress may predispose people to stronger emotional reactions in general. Consistent with this, results showed that stressed (versus control) participants increased engagement of brain areas implicated in emotion regulation and cognitive control during presentation of these stimuli. Moreover, this dysregulation was linked to difficulties differentiating self-and other-experienced negative affect, suggesting that acute stress, because of this blurring between self and other, can lead to more self-centered information processing (also see Krol and Bartz, 2021); this could explain the aforementioned research showing both reduced perceptions of the other's distress and heightened perceptions of one's own distress (Buruck et al., 2014;Gonzalez-Liencres et al., 2016). In addition, stressed participants showed an increase in prosocial behaviour, by allocating more of their own money to the next participants compared to controls. This effect was associated with increased anterior midcingulate cortex activation during the pain and anesthetized condition regardless of stress or control, suggesting that increased affect sharing-in general-is associated with increased prosociality.
In addition to these studies, other work indicates that acute stress can actually undermine emotion contagion. Returning to the aforementioned research by Martin et al. (2015), results indicate that stress, and specifically the stress hormone cortisol, impedes emotion contagion. The main objective of this research was to investigate why people (and mice) do not experience emotion contagion with strangers-it turns out that the culprit is stress: specifically, results showed that participants had significantly higher levels of cortisol in the stranger condition than in the friend condition and, moreover, cortisol levels in the stranger condition were negatively related to emotion contagion. Critically, Martin et al. were able to reverse this stranger-effect by administering the drug metyrapone, which blocks the synthesis of cortisol. In fact, stranger-dyads in the metyrapone condition showed comparable levels of emotion contagion to friend-dyads. These findings provide compelling evidence that stress can undermine emotion contagion, at least in some contexts (i.e., the presence of a stranger; of note, these findings are also consistent with Buruck et al.'s findings that stress attenuates ratings of other people's pain).
Up until this point we have focused on contagion of negative experiences-pain, stress, etc. But contagion is not limited to such experiences: positive stimuli-e.g., a smile-also readily elicits contagion in humans and non-human animals (Hecht et al., 2012;Paul et al., 2020); Importantly, such contagion also bodes well for social bonding and cohesion. Recently, Nitschke et al. (2020b) manipulated acute stress with the TSST and then measured automatic mimicry (also an index of contagion) with facial electromyography by having participants (n = 73; 50 women) view a series of emotional faces (either with unfolding smiles or frowns) during which their facial muscle movement was recorded. When participants were unstressed, they showed muscle activation in concordance with the emotions unfolding on screen--smiling to a smiling face, and frowning to a frowning face-indicative of the automaticity of emotion contagion in humans. However, when stressed, the same participants showed a significant reduction in reciprocal smiles, an effect that was driven by the amount of cortisol released as a result of the stress induction. (Frowning mimicry was unaffected by the experience of acute stress, but this may have been due to floor effects for the frowning trials.) This study is consistent with prior work showing that acute stress and specifically the stress hormone cortisol, can attenuate emotion contagion and further shows that the effects of stress can also influence contagion for positive emotional states.

Summary: acute stress and affective empathy/contagion
Taken together, research suggests that observing others in distress can lead to a distress response in the (initially) neutral perceiver; however, there are important moderators of this effect. First, vicarious emotional responses appear to be stronger when observing familiar others compared to strangers. This is in line with research that has shown that observing ostracism in familiar others is processed differently from observed ostracism in strangers, in that observing familiar others resulted in higher activation in brain areas implicated in the firsthand experience of social exclusion, compared to strangers (Beeney et al., 2011;Meyer et al., 2013). Second, vicarious emotional responses appear to be stronger when the other's experience is more palpable or salient.
Research also suggests that stressed individuals' emotional experience can be augmented in the presence of familiar others who are experiencing negative emotions (e.g., stress) as well, but attenuated in the presence of strangers. Precisely why this happens is unclear: it may be behavior that is grounded in our evolutionary history aimed at promoting kin and other in-group members, or it may be that it feels safer to express such emotions in the presence of trusted others.
Importantly, beyond these basic emotion contagion effects, research indicates that with increasing levels of self-experienced stress, perceivers can show a divergent response to that of the target. Notably, away from an allocentric experience (i.e., I am experiencing what you are experiencing), to a more egocentric perspective (i.e., my experience takes precedence over yours). This shift likely does not matter so much when perceivers experience something similar (i.e., all perceivers undergo the same stressful experience), but likely becomes important for affect sharing when there is a dissociation in affective experiences between a perceiver and a target. Further research is needed to delineate when stress leads to vicarious other-centric emotion sharing, and when it leads to experiences that are saliently grounded in the perceiver. This likely includes contextual factors (e.g., familiarity to target, settings that allow for interactions vs. purely observational settings), as well as traits of the perceivers-that is, anxiety (Hagenaars et al., 2014), sensitivity to vicarious affect sharing (Young et al., 2017), self-concept clarity (also see Krol and Bartz, 2021), as well as the ability to distinguish self-from other-related mental-representations and experiences (in particular during acute stress, cf. Tomova et al., 2014), and trait empathy (see Box 2 for the association of trait level empathy and affect sharing for stressed others). Indeed, such characteristics may lead perceivers to become overly stressed and/or less able to regulate their own distress. At this point, however, it is unclear whether or not a self-centered, or even a shared, stress response would be associated with increased or decreased feelings of closeness or levels of care.

Acute stress and cognitive empathy
We now turn to studies looking at the effects of stress on cognitive empathy-that is, the ability to identify and understand the thoughts, feelings, and intentions of another person. In contrast to affective empathy/emotion contagion, which is thought to be a relatively automatic response, cognitive empathy is generally thought to be less automatic, and to encompass a variety of skills ranging from simple processes such as emotion recognition, to more complex processes that require the integration of contextual information.
From what we know about the effects of stress on cognition, in general, we can speculate about the effects of stress on cognitive empathy. On the one hand, as noted, research suggests stress often leads to more habitual, automatic, rigid, and gist-like information processing (Dandolo and Schwabe, 2016;Hermans et al., 2014;Nitschke et al., 2020aNitschke et al., , 2019Satpute and Lieberman, 2006;Schwabe and Wolf, 2009). This suggests that simple tasks, such as emotion recognition, might be enhanced, at least in relevant contexts. However, these faster automatic processes also tend to lead to more rigid thinking. Given this, one might speculate that such facilitatory effects may not extend to more complex inferences, which require more effortful and flexible control (for a review: Contreras-Huerta et al., 2020), processes that are typically impaired following acute stress (Bogdanov et al., 2021;Nitschke et al., 2020a;Otto et al., 2013;Vogel et al., 2016). On the other hand, it has also been suggested that acute stress biases attention to contextually salient information (in order to deal with the aversive situation). In this regard, social information might be of particular importance (Oliveira and Faustino, 2017;Olsson et al., 2020), given its inherent salience. The notion that stress increases attention to salient information in the environment, suggests that acute stress may augment cognitive empathy.

Emotion recognition
In possibly the first study on this topic, Smeets et al. (2009) manipulated stress with the TSST and measured emotion recognition with the Reading the Mind in the Eyes Test (RMET; Baron-Cohen et al., 2001), in which participants (n = 64; 32 Stress) are presented with photos of faces cropped so that only the eye region can be seen; participants then must indicate the emotion from a list of four possible responses. Results showed no main effect of stress induction on RMET performance. Wolf

Box 2
Individual Differences in Empathic Abilities.
The individual difference factor that has received the most attention (for good reason) is trait empathy, which is commonly measured prior to stress induction using questionnaires such as the Interpersonal Reactivity Index (IRI; Davis, 1983), which assesses empathic concern, personal distress and perspective taking, or the Questionnaire of Cognitive and Affective Empathy (QCAE; Reniers et al., 2011). To date, the findings for trait empathy have been minimal. Several studies report no association between trait empathy and affective empathy. For example, Schury et al. (2020) measured trait empathy with the German version of the IRI and found no association between any of the IRI subscales and HPA axis activation in response to the TSST. Dimitroff et al. (2017) measured trait empathy with the QCAE. Again, results showed no difference between high and low (median split) self-reported affect sharing and changes in autonomic activation in the stress condition; however, for perceivers who watched targets recovering from stress, results showed that those high in self-reported affect sharing were quicker to tune into targets recovering from stress. Young et al. (2017) also used the QCAE. They found no association between affective empathy and ANS activation; however, they also split participants into high and low "vicarious pain responders" (as measured by their self-reports on the Empathy for Pain Scale); results showed (perhaps somewhat surprisingly) that high (vs. low) self-reported vicarious pain responders displayed less physiological co-activation (ANS; emotional respiratory behaviour).
Notwithstanding the lack of main effects, there is some indication that trait empathy might moderate stress resonance effects. Buchanan et al. (2012) found that perceivers who scored higher on the IRI's perspective taking and emotional concern scales showed significantly higher cortisol responses when observing a target/participant undergoing a stress task (TSST), possibly indicating that trait levels in empathy resulted in higher stress contagion. Engert et al. (2014) differentiated between "vicarious stress" (HPA activation when observing another person in distress) and "stress resonance" (the correlated occurrence of cortisol). They found no association between any of the IRI subscales and vicarious stress; however, emotional concern and perspective taking were positively associated with stress resonance. Finally, Brown et al. (2020) paired strangers in dyads (n = 140; all female). After a getting acquainted session (8 min), participants were randomly assigned to a stress or control condition. Those assigned to the stress condition watched 5 min of aversive videos followed by a 3-minute modified TSST; those in the control condition watched the stress participant undergo the modified TSST. In the unstressed perceiver, dispositional emotional empathy (measured with the Balanced Emotional Empathy Scale; Mehrabian and Epstein, 1972) was positively associated with negative affect in response to the distressed partner. There was no association between dispositional emotional empathy and physiological (co)activation (SNS reactivity); however, physiological (co)activation was linked to more accurate perception of the target's emotional state for perceivers high in emotional empathy. Finally, there was no association between cognitive empathy (IRI) and negative affect or physiological (co)activation. In sum, research on the relationship between trait empathy and emotion contagion is mixed. This could, however, be attributed to the range of different measures used (both trait measures as well as different measures of emotion contagion). Also, it is important to note that self-reported levels of empathy do not necessarily reflect actual aptitudes, as they are prone to the influence of social-cultural expectations, as well as self-presentation and social desirability concerns.
Notwithstanding these mixed findings, research looking at individual difference moderators can be useful in probing underlying mechanisms. For example, as noted in the main text, research has shown that individual differences in self-concept clarity (SCC; Campbell et al., n.d.) is associated with empathic responding (Krol and Bartz, 2021). Specifically, when confronted with another in distress (i.e., Batson's classic Katie Banks paradigm), those who reported lower levels of SCC reported more personal distress and less empathic concern than their higher SCC counterparts; these empathic reactions also were associated with subsequent prosocial helping behaviour (amount of money donated to Katie). Intriguingly, additional analyses indicate that self-other merging mediated the association between SCC and personal distress, suggesting that the reason why people low in SCC are more vulnerable to personal distress when confronted with another in distress is because they have difficulty distinguishing between the target's distress and their own distress. In addition to demonstrating who is vulnerable to more maladaptive empathic reactions, these findings also support a more general model of empathic responding that critically relies on the ability to distinguish between self and other. et al. (2015) conducted a similar study but used the Multifaceted Empathy Test (MET; Dziobek et al., 2008), 5 which is similar to the RMET, but participants (n = 99 men; 49 Stress) are shown photos depicting the whole face, and sometimes additional contextual information. As with Smeets et al., results showed no effect of stress on the cognitive component of the MET. Wingenfeld et al. (2018) also looked at the effects of stress induced with the TSST on MET performance in healthy women (n = 43; 23 Stress) and women diagnosed with borderline personality disorder (not discussed here). Similar to Smeets et al. and Wolf et al., they found no effect of stress induction on cognitive MET performance in healthy controls (Wingenfeld et al., 2018), although it is important to note that the authors administered the MET 65-minutes after stress induction, when levels of stress (including levels of cortisol) would likely have abated. While these studies suggest that there is no effect of stress on basic emotion recognition, it should be noted that both the RMET and MET are rather rudimentary; such simplified, static tasks, with forced-choice answers, can be problematic as they are vulnerable to range restrictions (Oakley et al., 2016;Quesque and Rossetti, 2020).
Other studies have focused on the ability to distinguish between different emotions. For example, Deckers et al. (2015) presented participants (n = 24 healthy controls, all female; note: this study also included borderline patients, and patients with Cluster C personality disorders) with video-vignettes of faces morphing from neutral to one of six emotions (anger, disgust, fear, happiness, sadness, and surprise). Participants had to classify the emotions displayed on two occasions, prior to stress induction (via the TSST), and then again after stress. Compared to baseline, stressed participants had a higher recognition rate for emotions of any valence. Although this study suggests that stress may indeed facilitate emotion recognition, learning effects cannot be ruled out since changes were compared to pre-stress levels of recognition.
This issue was resolved in a recent study by Domes and Zimmer (2019), who exposed one group of all male participants (n = 43; 23 Stress) to the TSST and the other to a control task. Participants then viewed a series of faces and were instructed to identify one of two emotions (happy, angry) at different degrees of morph with a neutral face (low, medium, or high). To quantify accuracy, Domes and Zimmer calculated a sensitivity score (d'). Results showed that participants in the TSST group had a higher sensitivity (d') for identifying emotions of any valence, compared to control participants. Participants in the TSST group also had lower response latencies for identifying negatively valenced emotions-an observation that is consistent with research showing that stress can increase attention to threat cues (Ali et al., 2020;Dandeneau et al., 2007;Roelofs et al., 2007). However, other research by this group, using a different stressor (group version of the TSST), reported somewhat contradictory results (von Dawans et al., 2020). In this study, stressed male (54; 28 Stress) participants had a higher sensitivity (d') for positive emotions, but a lower sensitivity for negative emotions of high intensity compared to control participants. Finally, Daudelin-Peltier et al. (2017) investigated the effects of stress on the ability to differentiate emotions from one another; or more precisely, the threshold of intensity needed to distinguish an emotion from another one. Here, the stimuli consisted of two different emotions presented simultaneously (e.g., fear and disgust) at different proportions (e.g., 86% fear and 14% anger; 50% fear and 50% anger etc). Participants (36; all men) were asked to make judgments on which prototypical expression the image most resembled. The authors found that when stressed individuals (group TSST vs. control; counterbalanced) were more likely to mis-categorize similar emotional facial expressions; specifically, faces expressing disgust needed to be shown at a higher intensity in order to be correctly identified. Conversely participants showed an increase in the ability to identify surprise even when displayed at lower intensity (i. e., lower threshold). This might suggest that under stress emotions are processed in a more gist-like manner, facilitating emotions that might be contextually more relevant (i.e., surprise vs disgust).
Not all studies, however, have found an effect of acute stress on emotion differentiation. Graumann et al. (2021) had healthy female participants (same sample as Wingenfeld et al., 2018) either undergo a stress (TSST) or control task before conducting a facial emotion recognition task (completed 65-minutes after the stressor). Here, participants had to correctly identify neutral as well as two negatively valenced (sadness and anger) faces at different levels of intensity (low: 40% intensity, high: 80% intensity). A subsequent sum-score was calculated for all correctly identified emotions at each intensity level. The authors did not find an effect of stress versus control on emotion recognition rate, for either low or high emotional intensity.

Beyond emotion recognition: mentalizing, theory of mind
While emotion recognition and differentiation are certainly important for cognitive empathy, in real-life we must also integrate various sources of information (face, voice, nature of the stressor itself, and the larger context); moreover, emotion processing in real life is dynamic and fleeting-we must continually update our representations as the situation unfolds. We now turn to research looking at the effects of stress on more complex and dynamic emotion processing.
In the aforementioned study by Smeets et al. (2009), the authors also looked at the effects of stress on emotion inference with the Movie for the Assessment of Social Cognition task (MASC; Dziobek et al., 2006). Specifically, 30-minutes after the TSST, participants (n = 64; 32 Stress) were instructed to watch a 15-minute movie of a dinner party involving four characters. At various time-points the video stopped, and participants were asked to make inferences about the feelings and intentions of the characters involved (e.g., "What is Betty feeling?"), from a selection of answer choices (e.g., happy vs. sad). Results showed that stress influenced performance on the MASC; however, the effect of stress depended on gender/sex, with stress improving MASC performance for men, but impairing MASC performance for women. It is important to note that in this study, there was no main effect of the stress condition; rather the effects were driven by the magnitude of the cortisol response in the TSST group. Specifically, for men in the TSST condition, high cortisol responders performed better than low cortisol responders on the MASC (median-split; n = 8 high; n = 8 low); however, high TSST cortisol responders did not perform better than the males in the control condition. For women in the TSST condition, the opposite effect was observed: here, low cortisol responders performed better on the MASC than high cortisol responders; again, though, the effect was specific to the TSST group as high cortisol responders did not differ from women in the control condition (although low cortisol responders did). Crenshaw et al. (2019) also observed that the effects of stress on cognitive empathy were moderated by gender/sex. The authors had romantic couples engage in an aversive discussion about a topic of disagreement in their relationship for 10 min while being video recorded. After the discussion, participants watched the recorded conversation and were prompted to write down the specific thoughts and feelings they were having at various time-points (i.e., 10 instances). Following this, participants either underwent the TSST (n = 48; equal gender split) 5 The MET contains both a cognitive component ("infer the mental states of another individual"), as well as an emotional concern component. For the latter, participants are instructed to rate their emotional reaction in response to the pictures ("how concerned are you for this person"). Research by Wolf et al. (2015) found that acute stress (TSST) increased the emotional concern felt for others in precarious situations in male participants (n = 49), compared to a control condition (n = 50). However, Wingenfeld et al. (2018) looked at the effects of acute stress on emotional concern in healthy women and women with borderline personality disorder (not discussed here) and did not find differences between stressed (TSST; n = 23) and unstressed healthy female participants (n = 24). Both groups showed the same levels of concern for others using the MET. or a control task in which they rated nature pictures (n = 48; equal gender split); immediately after they re-watched the video, but this time participants made ratings about how they thought their partner was feeling at each time-point (these ratings were compared to the partner's own ratings of how they were feeling as an index of "empathic accuracy"). Results showed that women were more accurate than men in the control condition; moreover, acute stress reduced empathic accuracy for women, but there was no effect of stress on empathic accuracy for the men. Although Crenshaw et al. did not measure cortisol, they partially corroborate Smeets et al.'s finding that high levels of stress (here psychological arousal) can be detrimental to women's empathic accuracy.
Recently, Nitschke et al. (2022b) also looked at the effects of stress on empathic accuracy. Here, empathic accuracy was assessed by having participants watch videos of targets discussing negative autobiographical events and rate how they thought the targets were feeling over the course of the videos (in contrast to Crenshaw et al. ratings were self-prompted and continuously recorded); these ratings were compared to the target's own ratings of how they were feeling to index empathic accuracy (cf. Zaki et al., 2008). Findings from two independent experiments (combined n = 268) indicate that the experience of psychosocial stress (TSST) increased men's empathic accuracy, an effect that was partially linked to the stress-induced cortisol response. Women, on the other hand, did not show such improvements following stress. Importantly, men and women did not differ in the empathic accuracy performance at baseline or in the no-stress condition, indicating that men actually got a boost from the stress manipulation. Nitschke et al.'s findings are consistent with broader theorizing about how acute stress produces a shift in two large scale brain networks, reallocating resources from the executive networks to the salience network (Hermans et al., 2014); moreover, and consistent with Hermans et al. (2014), Nitschke et al.'s (2022b) findings indicate that it is not cortisol per se, but rather stress-induced cortisol that appears to be important for affecting cognitive empathy. These findings also corroborate those reported by Smeets et al. (2009), but here the replicated effect is for men, not women (cf. Crenshaw et al., 2019). That said, methodological differences may be at play-whereas Nitschke et al. had participants track the emotions of strangers, Crenshaw et al. studied romantic partners discussing a real-life problem, and women may have been more affected by stress in this highly consequential interpersonal interaction.

Summary: acute stress and cognitive empathy
In summary, results from research looking at simple emotion recognition accuracy are mixed. There was no evidence that stress impacted the ability to identify discrete emotions from static images; however, it may be that simplistic tasks such as the RMET or MET are not sensitive enough to detect subtle differences, as they are prone to ceiling effects, at least for more neurotypical individuals. By contrast, there was some evidence that stress facilitated emotion recognition accuracy for tasks that required participants to distinguish one emotion from another, or from a neutral expression (face morphing tasks), although there were exceptions. Whether this differential effect is due to the fact that such tasks are more difficult (and, thus, not as prone to restricted range) or involve different computational processes is a question for future research. With regard to more complex emotion processing, the evidence to date suggests that acute stress does, indeed, have an impact; however, intriguingly, the effect of stress appears to differ for men and women. Although the strength of these effects vary across studies, the general pattern is similar, with men showing improvements in empathic accuracy performance following stress (Nitschke et al., 2022b;Smeets et al., 2009) or showing no effect (Crenshaw et al., 2019), and women showing declines in empathic accuracy performance following stress (Crenshaw et al., 2019;Smeets et al., 2009) or showing no effect (Nitschke et al., 2022b). Notably, three of these studies (Smeets et al.,and Nitschke et al. Study 1 and Study 2) linked the changes in empathic accuracy and social cognition with stress-induced glucocorticoids.
How might we understand these gender/sex differences? While none of these studies can definitively answer this question, there are clues in the data, and from prior work. We first address the differential effects for men and then turn to the women. As noted, both Smeets et al. and Nitschke et al. linked stress-induced changes in glucocorticoids with improvements in empathic accuracy, in men. Given that men typically show greater salivary cortisol reactivity in response to stress compared to women (Kudielka and Kirschbaum, 2005) (see Box 3 for gender/sex differences in the acute stress response)-an observation that was also observed in Nitschke et al.'s studies-it is possible that the increase glucocorticoids account (at least partly) for the beneficial effects of stress on cognitive empathy in men. Interestingly, for both Smeets et al. and Nitschke et al. the effects were specific to stress-induced glucocorticoids (i.e., there was no association between glucocorticoids and MASC or empathic accuracy performance in the no-stress conditions); this suggests that the broader stress context is critical for the effect to emerge. In fact, research that involves pharmacological manipulations of the HPA axis further support this notion, as exogenous administration of glucocorticoids, in the absence of a stressful context, typically shows no effect on cognitive empathy (see Box 4 for studies utilizing pharmacological manipulations of the HPA axis to test changes in empathic abilities).
Why did women show impairments in cognitive empathy following stress? As noted, to the extent that the facilitatory effects of stress on empathic accuracy are due to increases in glucocorticoids, it may be that women simply did not mount a strong enough glucocorticoid response. But that does not quite explain their performance declines (cf. Crenshaw et al., 2019;Smeets et al., 2009). Another possibility is that women may have experienced and/or coped with the TSST differently than men. For example, they may have been more likely to construe the TSST as a "threat" rather than a "challenge" (Moons et al., 2010), or they may have been more likely to cope with the TSST by ruminating (evidence suggests that women, in general, are more prone to rumination and emotion-focused coping; Nolen-Hoeksema, 2012); both of these factors could inhibit the attention and cognitive flexibility required to perform the empathic accuracy task well. That said, Nitschke et al. did not find detrimental effects of stress for women. Clearly future work is needed to better understand the complicated interplay between stress, gender/sex and cognitive empathy.

Discussion
Acute stress is widely acknowledged to be an adaptive response to a challenging situation (Selye, 1956). It is adaptive in the sense that acute stress mobilises resources and alters information processing to cope with an aversive situation. Changes in empathic behaviours-such as affect sharing and social cognitive abilities-due to acute stress should be considered within the framework of adaptiveness: how these changes might benefit an individual in an immediately challenging situation.
Traditionally, reactions to stress have often been attributed to a freeze or a fight-or-flight response (Cannon, 1939). As such, the adaptive function of acute stress is to highlight strategies that first-and-foremost help the stressed individual overcome that particular situation. In the context of experiencing another's distress this would likely result in, active and/or passive, withdrawal (trying to distance-physically and/or psychologically) of oneself from the aversive experience, or by focusing on the egocentric experience (i.e., what am I feeling?). The literature reviewed here, and in particular Section 2.3-Effects of acute stress on affective empathy/contagion-suggest that this might be the case during self-experienced stress (i.e., the experience of stress occurs prior to the engagement with others). Specifically, stressed individuals appear to focus more strongly on their own experience and are less likely to disentangle their own emotions or stress from that of another person. Importantly, these strategies of behaviour modification due to stress have been described in animal models of social behaviours under stress (Sandi and Haller, 2015), as well as in emotion regulation strategies in

Box 3
Gender/Sex Differences in Stress Reactivity.
When considering the effects of acute stress on empathic abilities, it is important to consider gender/sex differences in stress reactivity. The acute stress response in humans shows different biological profiles depending on gender/sex, with men typically showing a pronounced and robust cortisol response, at least when measured in saliva, whereas women do not tend to show as robust a response, although women do show a strong psychological stress response. Importantly, this effect extends beyond differences for men versus women (Kudielka and Kirschbaum, 2005) and include differences based on the presence of other hormones-such as across the female menstrual cycle and the use of hormonal contraceptives (Childs et al., 2010;Duchesne and Pruessner, 2013;Kudielka and Kirschbaum, 2005), and to differences based on perceived gender roles (Pruessner, 2018). It is therefore likely that all of these factors (i.e., gender/sex, hormonal levels) can differentially impact empathic abilities. It is important to point out that the perceived complexity and variability in females' biology has led to an under-representation of female participants in psychopathology and neuroscience research (Bale, 2019;Riecher-Rössler, 2017;Shansky, 2019), including stress research. As such, much less is known about how acute stress impacts female participants, and what role cyclical variations in ovarian hormones play in the female stress response, and subsequent cognition and behaviour.
As summarized in the main text, evidence indicates that the effects of stress on complex cognitive empathy appear to be moderated by gender/ sex. Smeets et al. (2009) found that higher levels of stress (defined as cortisol response) were beneficial to men, but detrimental to women's ability to make social inferences. Cranshaw et al. (2019) found that moderate levels of psychological arousal were beneficial for social cognitive abilities in both men and women, but that for women specifically acute stress led to a reduction in empathic abilities. Lastly, Nitschke et al. (2022b) found that acute stress benefitted men's ability to track the emotion of others over time, an effect that was tied to the cortisol response, but did not impact women's abilities. Interestingly, there was no evidence that the effects of stress on affective empathy/ contagion depended on gender/sex. Although only a limited number of studies have assessed potential gender/sex specific effects, those that do reported no such effects (Buruck et al., 2014;Martin et al., 2015;Nitschke et al., 2020b) (however see Gonzalez-Liencres et al., 2016 for non-behavioural differences between men and women). These findings highlight the distinction between cognitive and affective empathy and that it may be difficult to make uniform predictions about the effects of stress when considering that these systems work synergistically.

Box 4
Effects of Exogenous HPA Stimulation on Cognitive Empathy.
Centrally, the endocrine stress response involves two glucocorticoid binding receptors: glucocorticoid receptors (GR) and mineralocorticoid receptors (MR) (de Kloet, 2014). Acute stress triggers the release of the steroid hormone cortisol, which, through a feedback loop, centrally binds to one of these two receptors. This process can be short-circuited by the exogenous administration of corticosteroids, such as hydrocortisone, which has an affinity for GR and MR, and fludrocortisone, which has an affinity for MR. In addition to manipulating acute stress in the laboratory with psychosocial stress paradigms like the TSST, researchers have also used corticosteroids to simulate and test the effects of cortisol induced neuronal changes on social cognition. Schultebraucks et al. (2016) administered fludrocortisone (MR stimulation; 0.4 mg with 150 min delay to task) or placebo (n = 80; 40 women); they found no effect on emotion recognition (recognizing anger and sadness at varying levels of expressivity), although they did observe an increased attentional bias to negatively valenced faces (sad faces; dot-probe). In another study, Duesenberg et al. (2016) administered hydrocortisone (GR and MR stimulation; 10 mg with approximately 45 min delay to tasks) or placebo (n = 80; 40 women), and then assessed emotion recognition (anger, sadness, and neutral) in faces that were displayed for 1 s; participants also completed the MET. As with Schultebraucks et al., results showed no effect of drug administration on emotion recognition accuracy or on MET performance. Wingenfeld et al. (2014) also administered fludrocortisone, or placebo, after which participants (n = 35; all female) completed both the MET as well as the MASC. Again, there was no effect of fludrocortisone administration (0.4 mg with approximately 105 min delay to tasks) on either task; however, in a separate, larger, mixed gender study (n = 116; 91 women), the same research group did find an effect of fludrocortisone (4 mg with an approximately 180 min delay to task) on the MET (Nowacki et al., 2020) (Note: In addition to healthy controls discussed here, the researchers also looked at individuals diagnosed with major depressive disorder). Lastly, Chae et al. (2021) investigated the effects of hydrocortisone administration (10 mg) on facial emotion recognition using a puzzle task (Kliemann et al., 2013). Here, in two sets of tasks, participants (n = 104; all men) had to either match a selection of bottom-halves of faces (i.e., mouth region) to a top-half of a face (i.e., eye region)(implicit task); or, participants had to choose the correct emotion label out of four options for a displayed face (explicit task). While male participants generally performed better at the explicit task, compared to the implicit task, the administration of hydrocortisone (or noradrenergic stimulation (yohimbine), or combined) did not alter emotion recognition abilities.
In addition to studies investigating the effects of exogenous stimulation of the HPA-axis on empathic abilities, Wingenfeld et al. (2016) also investigated the effects of pharmacological MR blockade. Following administration of the MR antagonist spironolactone (300 mg), participants (n = 43, 23 women) completed the MET and the MASC. Here, the authors did not find an effect of pharmacological treatment, both the placebo and the spironolactone group performed equally well.
These results highlight the complexity of the human stress response and indicate that one biological stress system (in this case a partial stimulation of the HPA axis) alone likely does not account for the nuances in stress induced changes in empathic abilities. Indeed, research indicates that simultaneous ANS and HPA activation are necessary for an adaptive appraisal of information (cf. Ali et al., 2020;Hermans et al., 2011;Kukolja et al., 2008). They also suggest that the experience of an actual stressor may be important. As noted in the main text, research suggests that it is the stress-induced glucocorticoid response that is important for cognitive empathy; cortisol, in the absence of stress appears to be unimpactful (Hermans et al., 2014; for further discussion, see: Nitschke et al., 2022b). Going forward it will be important for research to parse these different aspects of the stress response, for example, by using pharmacological agents to augment and/or block them. humans (Gross, 1998).
Given that humans are a highly social species, and often have to deal with challenges that require interpersonal-problem solving-for example dealing with relationship conflicts-social strategies on how to deal with stressors are necessary (Chrousos and Gold, 1992). Furthermore, as a social species, humans often seek out others to overcome obstacles, or to socially modulate their emotional states (Gross et al., 2000;Zaki, 2020). In this regard, alternatives to the fight-or-flight hypothesis have been proposed; specifically, it has been suggested that behaviours that emphasize affiliation may increase under stressful circumstances (Depue and Morrone-Strupinsky, 2005;Taylor et al., 2000). As such, research suggests that we more readily share an experience with familiar others, as a shared experience might facilitate these affiliative processes (Echterhoff et al., 2009). Showing affiliative behaviours towards socially close others is likely more rewarding, and less risky, than investing time and effort in strangers (and by extension to abstract computer stimuli). To this end, stress has the power to draw individuals closer together, but it may only do so in situations where the benefits are obvious to the person engaging with the feelings of others. In this regard it has been argued that empathy is not necessarily automatic, but also motivational in nature (Weisz and Cikara, 2020;Zaki, 2014). Individuals can-at least to some degree-up-and down-regulate when to engage in, for example, affect sharing. When a perceiver is actively stressed themselves, a disengagement from their own emotional state is required in order to attend to another person's emotions. Here, it is beneficial to increase affect sharing when it comes to understanding and supporting a familiar other, while it will be beneficial to lower emotion sharing in situations where shared affect could be distracting (e. g., negotiations), or where social support provision (or tending and befriending more generally) seems untenable or impractical (e.g., with strangers, or computer stimuli). The findings discussed in Section 2.1-Evidence of affective empathy/stress contagion-and Section 2.2.-Evidence of affect sharing/contagion in simultaneously stressed perceivers-support this by showing that we more readily share the distress of others when we are at least somewhat familiar with them (i.e., romantic partner, but also individuals with whom we are teammates, or with whom we have had a previous interaction).
Lastly, it has been proposed that acute stress leads to a shift in information processing, with increased attention towards salient features in the environment (Hermans et al., 2014). As such, there is the notion that information conveyed by others carries special weight. It is therefore pertinent for individuals to track others carefully, including their emotions, as it might inform the perceiver about possible dangers, threats, and opportunities in the environment (Oliveira and Faustino, 2017;Olsson et al., 2020). As such social information serves to reduce uncertainties in the environment. In particular for emotion recognition and social cognitive abilities, increases due to stress might be best explained by this. For example, an increase in attention to faces, or the recognition of emotions in others can be explained by increased vigilance towards contextually salient information. In the context of the TSST, the identification of angry or neutral faces (emotions displayed by the confederates in order to elicit psychosocial stress) would therefore serve the function of understanding potential stressors in the environment, rather than being indicative of increased social motivation. As such, information others convey might be important clues for how to deal with a particular situation, and whether or not threats (or coping opportunities) are present (Ali et al., 2020;FeldmanHall and Shenhav, 2019;Olsson et al., 2020;Zaki, 2020). Similarly, shared emotions help individuals to better understand threats and opportunities in their environment. Sharing a stressful experience with a target prepares the individual to deal with the potential stressor themselves. These responses are adaptive and extend beyond the recruitment of social support or allies in general.
The findings discussed in Section 3-Acute stress and cognitive empathy-might be best explained this way. Acute stress can potentially facilitate emotion recognition abilities, but these effects (and the emotions involved) might be affected in a contextually dependent way.

Conclusions
Social connections are important for our health and well-being (Bzdok and Dunbar, 2020;Holt-Lunstad, 2018;Nitschke et al., 2021;Snyder-Mackler et al., 2020), identifying factors that can help and foster social relationships is thus important. One of these are abilities that can help us to understand others' emotions and intentions, commonly grouped under the umbrella term of empathy. In addition to these questions, there are several other outstanding issues in this research area. The following is not an exhaustive list but some questions we find to be especially compelling. First, how does acute stress affect the interplay between affective and cognitive empathy? Here, we, like others (e.g., Weisz and Cikara, 2020;Zaki and Ochsner, 2012), have treated affective and cognitive empathy as two (somewhat) distinct phenomena and, as outlined in the introduction, in many ways they are distinct from one another. But, at least in theory, there should be interplay between affective and cognitive empathy (cf. Schurz et al., 2020). For example, when engaging with another person's emotions in real life, we are likely drawing from a vast amount of information that involves several mechanisms to make inferences; ranging from shared emotions to idiosyncratic memories and experiences. Meta-analytic evidence from clustering of neuroimaging data by Schurz et al. (2020) further supports this: more "naturalistic" forms of empathic abilities rely on the co-occurrence of both cognitive as well as affective processes. More research is needed to understand this interplay and how stress might alter their typical patterns.
Second, is there a dose-response effect of acute stress? Relatedly, is the effect of acute stress linear or, like the effect of anxiety on performance, quadratic (cf. Joëls, 2006)? Indeed, one could imagine that moderate levels of stress could be beneficial in arousing attention but that extremely high levels of stress could be detrimental if, for example, the system becomes overwhelmed. This inverted-U pattern could explain why stress leads to sympathy and empathic concern in some situations (or for some individuals) but personal distress in other situations (or for other individuals).
A third question concerns delineating different aspects of the stress response. As noted at the outset, the acute stress response is highly complex, involving numerous biological systems including the fastacting autonomic nervous system and the slower acting hypothalamicpituitary-adrenal axis (Ulrich-Lai and Herman, 2009). Of course, there is also the psychological stress response, which can vary greatly across individuals and contexts-for example, some people are prone to experience acute stress as a challenge experience, whereas other people are prone to experience acute stress as a threat (e.g., Ali et al., 2017;Dickerson and Kemeny, 2004;Moons et al., 2010). What aspects of the stress response are important for affective and/or cognitive empathy? There is evidence that directly implicates the glucocorticoid cortisol (cf. Martin et al., 2015;Nitschke et al., 2022bNitschke et al., , 2020bPutman and Roelofs, 2011;Smeets et al., 2009), but what about other biological and psychological components of the stress response? Are some factors facilitatory but others inhibitory?
Fourth, are the effects of acute stress different from those deriving from chronic stress? One could imagine that they might be: in the context of acute stress, one is dealing with a momentary aberration to an otherwise healthy functioning system. By contrast, in the case of chronic stress, the stress system is strained and overburdened by chronic activation and may not function optimally (McEwen, 2012). This issue about acute and chronic stress also pertains to the question raised above about different aspects of the stress response. For example, if it is the stress-induced glucocorticoid cortisol that facilitates cognitive empathy (in men), then one might expect that chronic stress may not have facilitatory effects on cognitive empathy; this is because chronic stress can both blunt the HPA axis, in particular for severe and prolonged episodes (Herriot et al., 2020;Lam et al., 2019;Pruessner et al., 1999), and actuate the HPA axis response (for a review: Chrousos, 2009). As another example, people may be more likely to adopt a threat/anxiety response when experiencing chronic stress but a challenge response when experiencing acute stress; if these appraisal processes (and their downstream biomarkers; Moons et al., 2010) are important, we again might not expect the effects of acute stress to translate to conditions of chronic stress.
Fifth, how does an empathic understanding of others relate to prosocial actions under acute stress? It has been suggested that the experience of stress could translate to an increased motivation to affiliate with others (Taylor, 2006;Taylor et al., 2000;von Dawans et al., 2021), however, research in human subjects is inconclusive. As discussed in the current review, findings on the effects of acute stress on empathic abilities and behaviours are rather mixed and appear highly context dependent (cf. von Dawans et al., 2021). Relatedly, findings on the effects of acute stress on prosocial behaviours-such as charitable giving, reciprocity, but also retaliative behaviours-are similarly mixed. In a recent systematic review and meta-analysis, Nitschke et al. (2022a) showed that the effects of acute stress on prosocial behaviours were inconsistent with no mean effect of stress in either direction (i.e., stress leading to more or less prosocial actions) and importantly a high amount of inter-study heterogeneity. As with empathy, it is not that acute stress does not impact prosocial behaviours-as it clearly can, but it is still unclear under which conditions these changes can occur (cf. Faber and Häusser, 2021).
Lastly, moving forward, to instil more confidence in theoretical accounts, studies would benefit from pre-registered hypothesis, especially when gender/sex specific effects are predicted. Furthermore, to understand possible gender/sex specific effects of acute stress on empathy (broadly) it is important to have more adequately powered studies, and studies that systematically investigate gender/sex differences in the same sample to allow for adequate comparisons.
In conclusion, empathy is a multifaceted construct-spanning various cognitive, affective, and even behavioral responses-that often lacks an operational precise definition. Moving forward it will be important to clearly define the specific domain of interest (such as emotion recognition, affect sharing, stress contagion etc) in order to be able to more clearly delineate specific effects of acute stress. Acute stress does not uniformly change empathy, but rather we find domain (and even task specific) effects of acute stress. Over the last decade, there has been increasing interest in the effects of stress on empathy. This is an important question. Not only is stress an inescapable fact of existence in general, but it is, arguably, intimately tied to the phenomenon of empathy and social relations more broadly. Undoubtedly, being confronted with another in distress is stressful; how we navigate that stress will have important implications for our ability to respond to and care for the other. There are also other, less obvious, but equally relevant, contexts. For example, interpersonal and inter-group conflicts are inherently stressful; the ability to navigate those situations requires some degree of empathy and insight into the other's unique experience. A better understanding of the interplay between stress and empathy may thus shed light on issues related to interpersonal and inter-group conflict. We are excited about the research interest in this area and hope this review will inspire and aid others in this field going forward.

Data Availability
No data was used for the research described in the article.

Appendix A. Supporting information
Supplementary data associated with this article can be found in the online version at doi:10.1016/j.neubiorev.2022.105003.