A highly controlled design, featuring a modified version of the TSST and using 76 healthy adults (38 women in the late luteal phase of their menstrual cycle, and 38 men), tested the following predictions: (a) overall WM performance, as measured by the n-back, is faster and more accurate on a 1-back than on a 3-back task, regardless of biological sex or elevated circulating cortisol; (b) cortisol elevations enhance performance on a 3-back task in men; but (c) impair performance on a 3-back task in women.
The literature commonly reports considerable variability in response to laboratory-based stress induction methods (e.g., Buchanan and Tranel, 2008; van den Bos et al., 2009; Zandara et al., 2016). Our findings are similar: 11 participants who completed the study procedures did not respond to the TSST with measurable increases in cortisol levels. Moreover, and again as is common in this area, more women (n = 8) than men (n = 3) did not respond to the TSST with such increases (e.g., Elzinga and Roelofs, 2005; Stephens et al., 2016).
Confirming the first hypothesis, the present analyses demonstrated that, overall, participants completed the 1-back trials (presumably, representing a lighter cognitive load) more quickly and accurately than the 3-back trials (presumably, representing a heavier cognitive load). This data pattern characterized mean performance in both men and women, regardless of whether exposure to the experimental manipulation elevated cortisol. This pattern replicates data showing that people are faster and more accurate on lighter- than heavier-load WM tasks, regardless of biological sex or cortisol status (Callicott et al., 1999; Speck et al., 2000; Pelegrina et al., 2015; Rac-Lubashevsky and Kessler, 2016). This performance difference likely arises because 1-back tasks, unlike 2-back and higher tasks, require no content manipulation, are not vulnerable to interference and do not require accessing information outside of immediate attentional focus (Verhaeghen and Basak, 2005; Oberauer, 2006; Schleepen and Jonkman, 2010).
The present analyses did not confirm the second and third hypotheses, detecting no between-group differences in 3-back task performance. Hence, elevated cortisol levels did not enhance performance in men, or impair performance in women, on a heavier-load WM task.
This data pattern does not systematically replicate those described elsewhere. For example, Cornelisse et al. (2011) reported that TSST exposure facilitated faster reaction time on a 2-back (but not a 3-back) task in men. Schoofs et al. (2013), using only a 2-back task, replicated that result and found that stress exposure was associated with slower reaction times in women. Our findings are, however, consistent with those described in other studies reporting that stress-induced cortisol elevations do not affect WM performance in men (Kuhlmann et al., 2005; Oei et al., 2009; Weerda et al., 2010; Stauble et al., 2013) or women (Cornelisse et al., 2011; Zandara et al., 2016).
On Failures to Systematically Replicate
These failures to systematically replicate data patterns under apparently similar conditions may be due to a number of between-study methodological differences including, but not limited to, the task used to represent working memory (e.g., n-back vs. digit span); the characterization of light and heavy cognitive loads (e.g., whereas some use 1-back vs. 2-back, others use 2-back vs. 3-back, or variations of the Sternberg WM paradigm); the outcome measures reported (e.g., accuracy vs. reaction time); the extent of practice on the tasks representing working memory (e.g., whereas some give a 0-back task and a series of practice trials at the beginning of each n-back block, others give two practice blocks preceding 2- or 3-back testing); if tasks representing working memory occur in isolation or as part of a task battery; the procedures used to elevate cortisol levels (e.g., hydrocortisone vs. psychosocial manipulations); the use of manipulation checks to exclude cortisol non-responders from at least some analyses; the stage of female participants’ menstrual cycle (e.g., whereas some include only female participants in the luteal phase, others include all women and do not use phase of cycle as a covariate in analyses); or if women using oral contraceptives are permitted to participate or not (Elzinga and Roelofs, 2005; Kuhlmann et al., 2005; Schoofs et al., 2008, 2013; Luethi et al., 2009; Oei et al., 2009; Weerda et al., 2010; Cornelisse et al., 2011; see Brown et al., 2014; for a detailed discussion and checklist).
In the present study, we considered the methodological choices that allowed us to best answer our questions of interest. So, for instance, we chose to use an n-back task because it appears to represent core WM processes more accurately than, for instance, digit span backwards. At the theoretical level, the n-back requires manipulation and organization of information, especially at heavier task loads (Kirchner, 1958; Veltman et al., 2003). At the level of experimental control, the n-back offers the option of including both accuracy and reaction time measures and can be of longer duration than digit span, where each difficulty level is typically tested with only two trials (Jarrold and Towse, 2006; Lynn and Irwing, 2008; Schoofs et al., 2008). Finally, at the empirical level, imaging studies indicate that n-back tasks activate the PFC reliably, and that this activation increases as cognitive load increases (Speck et al., 2000; Owen et al., 2005; Tomasi et al., 2007). Hence, even though some, based on tradition (Kane et al., 2007), or authority (Miller, 1956), justify digit span as a gold-standard measure of WM, performance on that task does not necessarily reflect the characteristics of WM as well as the n-back task does.
We also chose to control, by design rather than statistically, factors related to female hormonal states. Specifically, we excluded women using oral contraceptives, and we tested naturally-cycling women only in the late luteal phase of the menstrual cycle. As noted earlier, stressor-induced salivary cortisol levels obtained from men and from women in the late luteal phase are similar. Moreover, compared to men and to women in other phases of the cycle, women using oral contraception or who are in the follicular phase exhibit smaller stressor-induced cortisol increases (Kirschbaum et al., 1995, 1999).
Finally regarding methodological choices, we reconstituted our participant groups into those with and those without cortisol elevations. Doing so allowed us to focus exclusively on the major question of interest (i.e., relations between elevated cortisol and WM performance), and to maximize statistical power to detect differences relating to cortisol levels, rather than to stress exposure per se.
Regardless of our group reconstitution, the present pattern of cortisol data differs markedly from those reported in many other studies. The average baseline cortisol level in our sample was 1.92 (± 2.20) nmol/l. Post-manipulation, the average cortisol level in the EC groups was 5.65 (± 4.14) nmol/l. These numbers are notably smaller than those reported by most other studies using the same stress-induction method (Liu et al., 2017). Typically, studies in this field report baseline levels of over 5 nmol/l (at least one reported levels above 20 nmol/l), and post-TSST peak levels of over 10 nmol/l (at least one reported levels above 50 nmol/l; see, e.g., Kirschbaum et al., 1996; Wolf et al., 2001; Domes et al., 2004; Elzinga and Roelofs, 2005; Kuhlmann et al., 2005; Oei et al., 2006; Nater et al., 2007; Schoofs et al., 2008, 2013; Luethi et al., 2009; Cornelisse et al., 2011; Zandara et al., 2016). In the present study, the average increase in cortisol post-manipulation for the EC groups was 4.10 (± 3.47) nmol/l, a figure at the lower end of the range of increases reported by other studies, some of which describe average values of over 9 nmol/l (Kirschbaum et al., 1996; Wolf et al., 2001; Domes et al., 2004; Oei et al., 2006, 2009; Nater et al., 2007; Luethi et al., 2009; Cornelisse et al., 2011; Schoofs et al., 2013; Zandara et al., 2016). Indeed, based on the classification criteria suggested by Miller et al. (2013), 25% of participants in our EC groups (9 of 36) were “cortisol non-responders.”
The source(s) of differences between the pattern of cortisol data described in the present study and those described in some previously published studies cannot be easily explained. As in most other studies, our participants were undergraduate volunteers and we enforced standard exclusion criteria strictly. We were faithful to the TSST methodology, and an accredited and experienced laboratory staff analyzed our saliva samples (Pillay et al., 2008). Furthermore, previously published studies (Human et al., 2013; du Plooy et al., 2014) and unpublished theses and dissertations using similar samples consistently describe similar trends toward relatively low baseline cortisol levels and relatively small magnitudes of cortisol response.
We therefore propose a set of testable hypotheses, the core of which is this: “These between-study cortisol differences can be traced to the fact that our samples are drawn from the southern African population, a more racially, ethnically, culturally and genetically heterogeneous population than those from which Western European and North American laboratories sample” (Manica et al., 2007). For instance, the final sample in the present study comprised 32 white, 22 black African, and 22 colored (mixed ancestry) or Indian/Asian individuals (we use these racial terms following the descriptive nomenclature present in South African census publications; Statistics South Africa, 2015). Although previously published studies do not provide detailed racial/ethnic classification data, it is improbable that their samples were as diverse as ours.
The relatively small literature on individual racial/ethnic/cultural differences in baseline cortisol, and in magnitude of cortisol response to psychosocial stressors, may shed light on the observed discrepancies between our physiological data and those reported elsewhere. Chong et al. (2008), for example, found that, after TSST exposure, white American participants showed a significantly greater increase in cortisol levels than African-American participants. Other non-experimental studies conducted in North America report that black children have both lower basal cortisol levels and flatter cortisol curves than white children. Importantly, differences in socioeconomic status (SES) appear to strongly moderate these group differences, with black participants in those studies typically sampled from lower-SES neighborhoods or families (Chen and Paterson, 2006; Desantis et al., 2007; Dulin-Keita et al., 2012; but see Lupien et al., 2001). Although none of those studies were conducted in low- or middle-income countries (LAMICs), it is possible that low-SES individuals in LAMICs (a characterization that fits many of the participants in our sample) show diurnal cortisol rhythms (i.e., lower basal levels and flatter curves) similar to those observed in low-SES individuals in high-income countries (HICs) of North America and western Europe.
In addition, it is plausible that high-SES individuals in LAMICs have greater exposure to prenatal stress, lifetime stress and racial discrimination than high-SES individuals in HICs. Each of these adverse experiences bear a relation to both basal cortisol levels and magnitude of cortisol response to psychosocial stressors (Williams et al., 2008, 2012; Jorm and Ryan, 2014).
Furthermore, the ancestral life history (LH) strategies of individuals residing in LAMICS may differ from those residing in HICs. LH theory is an evolutionary biological framework that accounts for individual differences in reproductive strategies, suggesting that organisms inhabiting unstable environments with high mortality rates tend to produce many offspring and invest little in them (i.e., adopt faster LH strategies; see, e.g., Promislow and Harvey, 1990). Applied to humans, LH theory has guided studies seeking to understand individual differences in, for instance, the onset of puberty, age of first sexual experience and number of sexual partners (see, e.g., Belsky et al., 1991; Patch and Figueredo, 2017). Recent studies suggest cross-cultural and cross-national variations in LH strategy are associated with, for instance, variations in: (a) the parenting strategies to which individuals are exposed (Sotomayor-Peterson et al., 2012), and (b) the national prevalence of polymorphisms in the androgen receptor gene AR, the dopamine receptor gene DRD4, and the 5-HTTLPR VNTR of the serotonin transporter gene (Minkov and Bond, 2015). Of particular relevance here is that the latter study showed that nations with higher levels of socioeconomic inequality (as measured by the Gini coefficient) contained more individuals with those genetic polymorphisms, which are associated with a greater propensity toward risk-taking, dysfunctional impulsivity and unrestricted sociosexual attitudes (i.e., traits that covary with faster LH strategies). According to the World Bank (2017), South Africa has a Gini coefficient of 0.63, making it one of the most economically unequal countries in the world.
Animal studies demonstrate that variations in LH strategies are related to variations in glucocorticoid concentrations and in the strength of the HPA axis-driven physiological stress response. Although some evidence suggests that a faster LH strategy is present when basal hormone levels are lower and when stress responsiveness is dampened, there remains some debate about the precise direction and nature of this relationship (Crespi et al., 2013; Crossin et al., 2016). No published human studies address this question.
In summary, we suggest that the racial, socioeconomic and other variability we observe in our laboratory’s samples serve as proxies for long-standing cultural differences that have produced remarkable inter-individual (and perhaps genetically expressed) differences in human responses to environmental challenges (Cohen et al., 1996; Sapolsky, 2017). We could disconfirm these and related hypotheses by using a quasi-experimental design that: (a) assigns individuals to groups based on factors such as measured region of origin, individual and country SES, exposure to prenatal and lifetime stress, exposure to instances of racial/ethnic/cultural discrimination, fast vs. slow LH strategy and other variables hypothesized to be critical, and then (b) examines trends toward relatively low baseline cortisol levels and relatively small magnitudes of cortisol response in specified groups.
We have highlighted potential cultural, socioeconomic and ancestral life-history differences between participants in our study and those in previously published studies. We speculate that these differences contribute to the observed between-study differences in patterns of cortisol secretion (Oberlander et al., 2008; Huynh et al., 2016; Lovallo et al., 2017), and explain, at least partially, why the outcomes we present here are inconsistent with those obtained from less heterogeneous samples (e.g., Cornelisse et al., 2011; Schoofs et al., 2013). Specifically, if basal cortisol levels in our participants are relatively low, and the magnitude of cortisol response is also relatively small, then cortisol levels are unlikely to elevate to the supraphysiological levels known to affect cognitive performance significantly (Henry et al., 2014; Schultebraucks et al., 2015).
Limitations
Although the control procedure we used is similar to that described by others (e.g., Elzinga and Roelofs, 2005; Cornelisse et al., 2011), it did not require participants to complete cognitive and physical tasks formally equivalent to those completed by participants in the TSST condition. The literature describes other, perhaps more intuitively appealing, control procedures (e.g., Het et al., 2009; Wiemers et al., 2013). There are, however, no empirical demonstrations or accepted standards providing guidance in these matters (i.e., there are no direct comparisons of different types of control procedures in studies with aims similar to ours).
Unlike several other studies in this literature (e.g., Cornelisse et al., 2011; Schoofs et al., 2013), we did not collect biomarkers of the sympathetic nervous system’s response to the TSST (e.g., salivary alpha-amylase; Petrakova et al., 2015). However, our focus here was on possible sequelae of cortisol elevation, rather than on those of an overall stress response (i.e., both sympathetic and HPA-axis activation), on WM performance. Hence, the collection and analysis of additional biomarker data was superfluous. Moreover, we could not use statistical analyses more sensitive to detecting sex differences in cortisol response to the psychosocial stressor (e.g., growth curve modeling; Lopez-Duran et al., 2014) because post-manipulation cortisol sampling was not dense enough. Such analyses require samples at 5–10-min intervals in the hour immediately following stressor offset.
Finally, our group of participants was too small to run meaningful analyses on cortisol samples drawn from individuals with different cultural, racial, ethnic and/or SES backgrounds. Hence, we could not give proper statistical consideration to relations among, for instance, adverse life experiences, LH strategies, cortisol responses and WM performance. As noted above, such consideration is important when analyzing data from studies in this research area, particularly if those studies are conducted using samples drawn from LAMICs in the global south.