(1) Behavioural and Clinical Neuroscience Institute, Department of Experimental Psychology, University of Cambridge, Cambridge, United Kingdom
(2) Institute for Empirical Research in Economics, Laboratory for Social and Neural Systems Research, University of Zurich, Zurich, Switzerland
Social behaviour is an integral part of the behavioural repertoire of mammals and includes intra-species aggression and social hierarchies; sexual, parental, affiliative and cooperative behaviours; social recognition; social play behaviour, and social development. Most animal studies have been dedicated to the investigation of intra-species aggression, and great advances have been made in the understanding of its neurochemical basis. Testosterone has emerged as a major player in driving social behaviour in humans and animals. Here we review the evidence on the role of testosterone in social interaction.
Animal models of testosterone’s role in social interaction
Early evidence on the role of testosterone in social behaviour suggested that it facilitates aggression.
For instance, an impressive finding in rodents is that castrated animals, which have virtually no circulating testosterone in their blood, show a near-complete absence of physical fights, but fights can be fully restored by providing testosterone supplementation to these animals [1-3]. Further studies established that in males, testosterone has to be present in a critical period of brain development to allow for endogenous or exogenous testosterone stimulation of aggression in the adult. This also applies to females, if they are exposed to experimentally raised levels of testosterone .
However, the role of testosterone in facilitating aggression appears to be specific to social forms of aggression. Examples are territorial and dominance aggression: the former refers to defending territory against intruders, which may result in fierce combat; dominance aggression serves the purpose of establishing a social hierarchy, in which the dominant animal displays supremacy over the submissive one in the context of mating or feeding [e.g. 5]. Territorial and dominance aggression share a common competitive element and are both modulated by testosterone . In contrast, it appears that testosterone is less involved in other forms of aggression such as e.g., predatory, anti-predatory aggression or irritable aggression .
While an animal’s ability to master aggressive social interactions is largely dependent on testosterone, it is well known that such interactions also influence testosterone levels in turn. This well-acknowledged fact is captured by the Challenge Hypothesis , originally postulated for the relationship between territory formation in birds , and has been extended to non-human primates  and humans . The hypothesis states that, perhaps due to the costs of chronically elevated testosterone levels  testosterone regulation mechanisms have evolved to elevate levels only in situations when the corresponding physical and behavioural changes are beneficial. Thus, levels are low during periods of social stability, while competitive challenges, mostly in the context of territory formation, dominance disputes, and mate-guarding, cause testosterone levels to rise, which further facilitates aggression as an adaptive mechanism to cope with the challenge . Research in rodents has shown that testosterone is a causal factor in this challenge effect, because short-lived testosterone preparations injected after a winning experience facilitated more distant future winning probability irrespective of basal levels of aggression . Therefore, it seems that winning aggressive social interactions might elevate testosterone levels which will then facilitate further aggressive interactions as a consequence, also referred to as the “winner effect”.
However, the role of testosterone in facilitating aggressive behaviour is not necessarily restricted to overt physical aggression: in primates, evidence suggests that testosterone instead drives a more general repertoire of behaviors, often subsumed under the concept of dominance behaviour. For instance, rhesus monkeys with high testosterone levels use stares, threats and displacements to maintain or ascertain dominance, rather than overt aggressive interactions . In humans, the way to assert dominance can potentially be even more subtle: for instance, increased staring duration is associated with dominant social status, while an averted gaze is generally a sign of submission . It is also in the context of dominance behaviour that one has to see testosterone driven aggressive behaviour in humans (see box 1).
Baseline testosterone as a biomarker of social interactions
The possibility of obtaining valid measures of bioactive steroid hormones from human saliva has led to an increase in the use of hormones in research of social behaviours (see Box 2).
In accordance with the biosocial status hypothesis , field and laboratory studies have shown that circulating testosterone is associated with constructs closely linked to dominance behavior. Measurements of testosterone at a single time-point have been shown to be positively linked to high social dominance in both adolescents [14, 15], and adults [16, 17]. In addition, salivary testosterone levels correlate with implicit measures of power motivation  and increased attention to status threats [19, 20]. Based on these relationships, and the stability of testosterone levels over time, some have suggested that baseline testosterone levels reflect a personality trait .
It is tempting to jump to the conclusion that the hormone testosterone is the causal driving factor of the observed behaviours . However, it cannot be emphasized often enough that correlation does not equate to causation . This fact is accentuated by research showing that testosterone secretion also responds to the social environment, which we discuss next.
Social modulation of testosterone
Although human testosterone levels seem to be relatively stable over time, short term fluctuations in testosterone secretion are also observed. These result partly from regulatory feed-forward and feed-back processes (see box 2), endogenous biological rhythms such as circadian and circannual variations, and from the female menstrual cycle; however, a number of studies have demonstrated that testosterone secretion also responds rapidly and strongly to social stimuli .
The most widely known and earliest reports of context effects on androgenic function is by , a researcher who spent time on an island where he was alone except when he travelled back to the mainland to visit his female romantic partner and engage in sexual activity. During his island stay, he collected his beard clippings and measured their weight. Androgens stimulate facial hair growth ; since the beard clippings were heavier on days before mainland visits, Anonymous concluded that his androgens were increased in anticipation of sexual activity.
In addition to sexual stimuli, social interactions outside a specifically reproductive context have also been shown to induce a testosterone response (see also box 2). For example, it is well documented that hormones change in a matter of minutes surrounding a competitive event (Booth A et al., 1989). In humans’ closest living relatives, chimpanzees (Pan troglodytes), anticipation of a dyadic food competition with a dominant male has been shown to induce an increase in salivary testosterone . In human males, anticipation of competition involving physical confrontation in e.g. [a tennis match 28, wrestling combat29] but also non-physical face-to-face competition such as chess and domino tournaments [30, 31] may induce a change in testosterone secretion as well.
Testosterone also reacts to contest outcomes, rather than only anticipation: for instance, in female soccer players the difference in testosterone levels before and after a match is positive in winners and negative in losers , but no such difference is observed when ties occur . Because physical exercise can itself lead to a testosterone response, an even stronger test of whether testosterone secretion is sensitive to social contexts is whether non-physical social interactions may elicit such a response as well. So far, relatively few studies have employed causal manipulations of social context to investigate its impact on the testosterone response in humans [e.g. 34, 35, 36, 37, Schultheiss, 2005 #3906]. Recently it was demonstrated that merely watching oneself engaged in a previous win of a competitive interaction on a video produces a significant surge in testosterone to approximately 140 % of baseline values whereas watching a neutral video did only result in a non-significant 6% increase .
Both aspects, i.e. testosterone’s predictive character for behaviour, and its responsiveness to the environment, are nicely shown in a study of London traders who work in a highly competitive environment . On days on which these traders made an above-average profit on the market at 11 am, they had significantly higher testosterone levels than average. Furthermore, the authors found that on days of high morning testosterone, the traders returned a significantly higher afternoon profit than on low (morning) testosterone days. Although correlative, these data suggest that changes in testosterone secretion might feed-back and adjust on-going or near-future behaviour in status-related contexts, i.e. induce a winner effect.
In summary, a number of studies have established that basal levels of testosterone correlate with status-related behaviours in males, and even more robustly in females. Conversely, sexual and status-related social contexts produce rapid fluctuations in testosterone levels in non-human animal and in humans. These rapid fluctuations give rise to an even more pronounced correlation-causation problem than in studies of social interaction. Thus, only exogenous administration of testosterone allows a clean investigation of its effects on social cognition and interaction. In the following section we review the effect of testosterone administration on social behaviour.
Testosterone administration effects on social cognition and interaction
Testosterone administration has long been used clinically in the treatment of the hypoandrogenic state in males, more recently for the treatment of female sexual dysfunction [39-41], and is being investigated as a male contraceptive . Furthermore, a number of studies document the causal impact of testosterone on motivation and emotion processing, e.g. gambling behaviour , fear processing , stress-related processes , and emotion perception . In-depth discussion of these studies is beyond the scope of the present review. Instead, we focus on studies that investigate the effects of testosterone on human social cognition and interaction, an area of research that has just begun.
That fact that significant relationships between testosterone and emotion processing have been shown to arise outside of awareness  emphasizes the importance of the investigation of actual behaviour or the assessment of implicit effects rather than self-report measures. This view is supported by the many null findings of exogenously administered testosterone on self-report measures [e.g. 44, 48], and by a study showing that administration of testosterone to healthy participants reduces unconscious fear but not consciously experienced anxiety . Because actual behaviour is influenced by both conscious and unconscious socioemotional processes, the study of actual social interaction in the laboratory is a way to simultaneously assess explicit as well as implicit effects of the hormone.
In a pioneering study on the effects of exogenously administered sex hormones on human economic decision making, the two sex hormones testosterone or estradiol were applied daily in the form of a topical gel to post-menopausal women over a period of a month. Results suggested that neither of the two hormones had an effect on a number of economic decision making paradigms, including risk taking, trust game, dictator game and ultimatum game . This null finding is interesting, because it emphasizes the importance of performing acute testosterone administration procedures, rather than administration over an extended period of time as done by Zethraeus et al. Acute administration shows greater reliability in the production of behavioural and neurophysiological effects compared to repeated administration [43-46, 49, 51-55].
There is much controversy around testosterone’s effect on human behavior, which reaches a boiling point when it comes to pinning down testosterone’s role in social interaction (Kenrick & Barr in ). Folk wisdom holds that testosterone causes antisocial, egoistic, or even aggressive behaviours in humans. These stereotypes likely derive from prevalent gender stereotypes about male behavior, in combination with the widely held, but wrong belief that testosterone occurs exclusively in males. However, the correlative testosterone studies mentioned above already suggest that this simple folk view is likely to require revision [8, 34, 56]. A recent testosterone administration study from our laboratory has confirmed this view: contrary to popular belief, we found that one acute dose of testosterone in women actually increased fair bargaining offers in proposers of an ultimatum game. As it is difficult to explain these higher proposer offers with selfish motives, these results suggest that the traditional folk view of testosterone’s role in social interaction should be revised. In an interesting twist, those individuals given placebo who believed they had been given testosterone also showed a change in bargaining behaviour, but in the opposite direction: bargaining offers were less fair, in line with people’s stereotypes about the effects of testosterone . As the beliefs about the treatment assignment were measured after the ultimatum game by self-report, the belief formation might have been endogenous (i.e. the behaviour affected beliefs and not vice-versa). Thus, future studies should test whether a causal manipulation of beliefs about testosterone treatment allocation confirms this result or not.
Previous research also suggests that behavioral effects of testosterone administration may extend to males, as within-sex variation in basal testosterone levels associates with similar behavioural tendencies in both sexes (i.e. status seeking). Despite this fact, another recent study showed that application of testosterone to a small sample of healthy males might induce proposers to be less generous, in line with the folk-hypothesis . These results have to be interpreted with caution, however, as the authors did not account for dependencies between multiple observations in each subject, which tends to inflate statistical significance.
Beliefs about testosterone might not only affect social behaviour, but the hormone might also affect beliefs about others. One study investigated the perceived ratings of trustworthiness of facial photographs after providing an acute dose of the hormone and showed that testosterone led to a significant reduction of rating of trust in others . However, this trust-reducing effect was only observed in a subset of participants. When the subject sample was divided into two groups – those who were more vs. less trusting when given placebo – it was only the more trusting women who showed any response to testosterone. These results suggest that testosterone makes individuals more pessimistic about others’ prosocial attitude. If we apply this reasoning to the ultimatum game, we might surmise that proposers on testosterone have a more pessimistic belief about the opponent’s prosocial attitude, i.e. about their rejection behaviour. Therefore, if testosterone had increased proposer’s pessimistic beliefs about the responder behaviour, this might also have induced an increase in proposer offers . Thus, future studies should also include measures of beliefs about rejection behaviour in the ultimatum game.
What are the possible mechanisms by which testosterone influences social behaviour?
Given the breadth of the results described above, it would be surprising if the effects of testosterone administration were linked to a particular action at one single receptor or within a single neurobiological circuitry. However, there are an increasing number of studies that delineate a distinctive set of psychological, neuroendocrinological and neurochemical processes relevant to status-related social behaviours, which are modulated by testosterone.
It is intuitively understandable that signals of potential impending status challenge might demand heightened awareness in individuals who are inclined to put high value on status. In macaques, it was shown that testosterone administration may increase selective attention to threatening social stimuli . Furthermore, a number of studies among healthy young adults have shown that individuals with high basal levels of testosterone show an attentional bias to pictures portraying facial expressions of anger, which are considered signals of an impending dominance challenge, in an emotional Stroop task. This relationship was more pronounced in women than in men  and findings were replicated in a more recent study . Furthermore, exogenous administration of the hormone leads to a cardiac acceleration in response to angry faces, but not to neutral or happy faces, a finding interpreted as willingness to face challenge .
The above effects could be mediated by the amygdala, a brain region rich in androgen receptors  and affected by circulating androgens [62, 63]. There is evidence linking testosterone levels to amygdala activity during the processing of threat-related stimuli; for instance, among healthy young men amygdala responses to stimuli depicting similarly fearful and angry faces co-vary significantly with individual differences in serum testosterone concentrations [64-66]. A recent testosterone challenge study has shown that the strongest increase in BOLD signal in response to the presentation of angry facial expressions is observed in the amygdala of young women . This finding was conceptually replicated among middle-aged women . In summary, findings suggest that testosterone may impact the salience of stimuli signalling a potential status challenge by a modulation of amygdala activity.
However, an intriguing twist to this account is the possibility that the testosterone effects in the amygdala are mediated not by testosterone itself, but its metabolite, the aromatic steroid estradiol. In this case, testosterone would act as a pro-hormone. This metabolisation is mediated by the enzyme aromatase, a protein that has attracted much interest recently. Aromatase is expressed in the brain of all mammals and is found in areas that are implicated in the regulation of social behaviours . In rats it has been detected in several sites in the hypothalamus including the preoptic nucleus, the sexually dimorphic nucleus and also in the bed nucleus of the striata terminalis, as well as in the medial amygdala [69, 70]. Using the PET ligand C-vorozole, high binding affinity to aromatase in vivo was detected in the amygdala of rhesus monkeys  and humans . A post mortem study detected high concentration of estrogen receptor alpha mRNA in this structure .
These findings suggest that the amygdala is both a likely candidate structure for threat detection, as well as a prominent site of action of aromatase; together this opens up the possibility that threat detection effects of testosterone operate via conversion to estradiol, rather than through testosterone itself.
Reward and motivation
Reward processing is considered to be crucial in social interactions  and has been shown to be influenced by testosterone. Animal research has shown that rodents self-administer testosterone both orally  and intracerebroventricularly [76, 77]. Furthermore, they exhibit conditioned place preference for locations associated with systemic testosterone administration , an effect which is blocked by peripherally administered mixed as well as selective dopamine D1 and D2 receptor antagonists . Further research has pointed to the shell of the nucleus accumbens (which corresponds to the human ventral striatum) as a neurobiological substrate of these conditioning effects . Human research has shown that testosterone administration can induce increased activity in the ventral striatum during a reward anticipation task in subjects with low appetitive motivation . Strikingly, in animals, effects on reward processing have been observed within short time periods (30 min) after systemic administration of testosterone , suggesting that naturally occurring fluctuations of testosterone may also modulate ongoing motivational aspects of human behaviour . More specifically, a testosterone surge following a social stimulus might reinforce any behaviour that led to that testosterone response in the first place (e.g. winning a fight in a resident intruder paradigm , or a contest in humans).
As described above for threat detection, one intriguing possibility is that the reward-related effects of testosterone are mediated through its metabolites. In particular, a large number of in vitro and in vivo studies investigating the neurophysiological role of estradiol revealed effects that could also explain some of the testosterone administration effects observed in humans.
For example, ovariectomy causes a decreased basal release of dopamine in the striatum of rats, which can be reversed by chronic estradiol replacement. In the same structure, estradiol potentiates amphetamine evoked dopamine release, and infusion of estradiol into the nucleus accumbens leads to an increase in K+ stimulated dopamine release and decreases K+ induced GABA release, while in the amygdala, estradiol dose-dependently enhances electrically evoked dopamine release [for review see 84]. Estradiol is also readily self-administered when applied intra-cerebroventricularly . Hence, testosterone administration might influence motivational and reward processes either directly via androgen receptors or indirectly via estradiol acting on estrogen receptors.
Aromatase activity is under neuronal control  and the concentrations of estradiol that are observed locally at the neuronal level may be very high, such that some have hypothesized that it fulfils the criterion of a neurotransmitter . Although speculative, these two aspects would allow rapid and situation specific conversion of testosterone into estradiol and by virtue of its modulatory effect on dopaminergic neurotransmission, estradiol could then modulate the acquisition of instrumental behaviors by modulating reinforcement learning processes.
While these findings might suggest that one has to perform estradiol administration studies to test the hypothesis whether testosterone or estradiol is causing the effects, one has to keep in mind that aromatase distribution in the human brain is concentrated in certain structures only. Hence, one cannot approach the aromatisation issue easily by estradiol administration; instead one has to use pharmacological blockade of aromatase after testosterone administration in humans.
Several studies have shown that testosterone and its metabolites reduce anxiety-like behavior in rodents in a number of behavioural paradigms including the defensive burying test  and the elevated plus maze . The anti-androgen flutamide was shown to block the anxiolytic-like effect of testosterone in castrated male rats. In humans, single acute doses of testosterone have been shown to be fear and startle reducing [44, 49]. Together, these results suggest that testosterone may take its effects on social cognition through decreasing anxiety and that increases in testosterone secretion in anticipation of a competitive interaction may help to facilitate engagement in the contest.
In contrast to the mechanisms described above for threat detection and reward processing, aromatisation to estradiol are unlikely to mediate the anxiolytic effects of testosterone. Instead, testosterone’s anxiolytic properties have been shown to be GABA-A receptor-dependent; aromatization to estradiol seems not to be necessary for the anxiolytic effects .
Box 1. Sources of testosterone and regulation of secretion
In women, testosterone is secreted by the adrenal zona fasciculata (25%) and the ovarian stroma (25%), the remaining 50% are produced from circulating androgens such as dehydroepiandrosterone (DHEA). DHEA is primarily an adrenal product and acts as a precursor for the peripheral synthesis of more potent androgens such as testosterone and dihydrotestosterone (DHT) . Daily production rate of testosterone is in the order of 0.1 – 0.4 mg and circulating levels are in the range of 0.2 – 0.7 ng/ml.
In the human male, more than 95% is secreted by the testes and daily production is approximately 6 – 7 mg . The metabolic steps required for the conversion of cholesterol into androgens take place in Leydig cells in the testes, which are of major importance for the generation of circulating androgenic hormones. Thus, the major stimulating factor for testosterone production in males is LH. However, as in females, the adrenal cortex also contributes to this production, but to a far lower extent.
Basal testosterone levels as well as rapid testosterone responses are tightly regulated via a cascade of hormonal feed-forward and feed-back processes. At the top of the hierarchy of the sex hormone cascade stands the hypothalamus, which receives converging inputs from other brain regions with socially relevant information (e.g. amygdala, XXX, ZIT). As part of the reproductive axis, the hypothalamus releases Gonadotropin-Releasing Hormone (GnRH) in a pulsatile fashion, which stimulates the release of Luteinizing Hormone (LH) and Follicle-Stimulating Hormone (FSH) from the anterior pituitary into the blood-stream. The stress axis is worth noting here, because of its inhibitory action on the reproductive axis in males  and potential facilitatory actions in females. In response to stress, Corticotropin-Releasing hormone (CRH) and Arginine Vasopressin (AVP) are released in the hypothalamus. These initiate the production and secretion of adrenocorticotropic hormone (ACTH) in the pituitary gland, which facilitates the production of glucocorticoids, most importantly cortisol, in the adrenal gland. In women, short term activation of ACTH after acute stress may increase the secretion of DHEA from the adrenals and therefore cause increased levels of circulating testosterone. Vice-versa, in males, short term activation of the HPA axis is assumed to suppress LH and hence, testosterone release from the testes.
Testosterone constitutes the main hormonal difference between the sexes, with average levels being approximately 10-fold higher in males than females. In both men and women, testosterone levels show pronounced circadian variation, peak levels being seen in the early morning hours , which may drop about 50% from morning to evening for both sexes . Furthermore, levels show seasonal variation with an autumn peak . Within the female menstrual cycle, testosterone is at its lowest concentrations in the early follicular phase of the cycle and rises to a mid-cycle peak. The following luteal phase concentrations are higher than those in the early follicular phase .
In males, a well-known determinant of basal testosterone levels is the body mass index (BMI) . The lower testosterone levels in obese men have been attributed to the lower levels of SHBG associated with higher BMI . Furthermore, because estradiol is known to potently inhibit the HPG-axis in males at the hypothalamic level , baseline testosterone levels are reduced in obese males as a result of increased conversion of testosterone to estradiol by the enzyme aromatase, which is found in high concentration in adipose tissue . Furthermore, testosterone level declines linearly with age, and approximately one fourth of elderly men have mild-to-moderate testosterone deficiency . In women, serum levels of both DHEA and its sulphate, DHEA-S decline with age .
However, if these sources of variability are controlled for, baseline testosterone levels are moderately stable over time. In both adult men and women, this has been shown for a period of 2 and 5 days , 2 weeks , 8 weeks , or 1 year in boys and girls . Furthermore, these levels have substantially high heritability with the majority of heritability estimates being between 50 percent and 70 percent, with a mean of 60 percent [e.g. 104]. The mechanisms underlying the interindividual variation in basal testosterone levels are poorly understood, but the moderate heritability suggests genetic effects. For males, it has been suggested that differences in feedback loop set point contribute to this variation, with part of the between-subject differences in serum testosterone levels reflecting genetically determined, subtle differences in androgen sensitivity . The set-point hypothesis is based on the rationale that a given serum testosterone level will result in reduced feedback suppression of serum LH in men with relatively lower androgen sensitivity, which in turn will result in increased LH levels and a consequent increase of serum testosterone levels. Indeed, testosterone levels were shown to be positively related to LH concentration in blood and to decreasing androgen sensitivity, as determined by CAG repeat polymorphism in exon 1 of the androgen receptor .
Rapid testosterone responses
Animal research has shown that rapid testosterone surges may be observed as a result of mating interactions . Such relatively fast responses have also been shown in humans: a brief social interaction with attractive females induces a testosterone response in young males , and the presentation of videos featuring attractive men induces such a response in females . Testosterone secretion changes in men have been observed after 15 min, but not at later time-points after a significant social interaction [106, 108]. Furthermore, because of the time testosterone takes to enter from the blood stream into saliva , significant changes of testosterone serum levels could have been expected to occur as early as 10 min after interaction.
In males, these rapid testosterone responses are thought to results from a pituitary independent mechanism, as testosterone release is triggered by pulses of luteinizing hormone (LH) occurring every 1 to 3 hours  and are therefore unlikely to be the cause of these fast testosterone responses. However, research on social status and reproductive physiology in baboons has shown that the catecholamines adrenaline and noradrenaline, which are released from the adrenals as part of the fast sympathetic stress response, have a stimulatory effect on testosterone secretion within minutes . Moreover recent research in male rats has revealed a neuronal connection between the brain and the testes, which seems to regulate Leydig cell sensitivity towards LH [111, 112] allowing fast neuronal control over testosterone release.
Box 2. Testosterone and human aggression
The growing interest on the role of testosterone in social interaction also brings up its controversial role in human aggression. However, due to ethical reasons, the experimental investigation of physical forms of aggression is difficult to investigate in humans. Therefore, studies in humans are lim