Adrenocortical and autonomic cross-system regulation in youth: A meta-analysis

Abstract

Childhood and adolescence are salient periods for the development of adrenocortical and autonomic arms of the stress response system (SRS), setting the stage for subsequent health and adaptive functioning. Although adrenocortical and autonomic systems theoretically function in highly coordinated ways, the strength of the relationship between these systems remains unclear. We leveraged a multivariate mixed effects meta-analytic approach to assess associations between adrenocortical, sympathetic, and parasympathetic functioning at rest and reactivity during stress-inducing tasks across 52 studies (N = 7,671; 5–20 years old). Results suggested a modest positive relation between adrenocortical and sympathetic systems as well as between adrenocortical and parasympathetic systems. Moderation analyses indicated the strength of associations varied as a function of several methodological and sociodemographic characteristics. Environmental effects on cross-system regulation were less clear, perhaps due to underrepresentation of adverse-exposed youth in the included studies. Collectively, our findings call for greater methodological attention to the dynamical, non-linear nature of cross-system functioning, as well as the role of experience in their organization across development.

1. Introduction

Individual differences in physiological regulation are posited to shape mental and physical health across the lifespan. The physiological stress response encompasses multiple co-activated biological systems that allow individuals to flexibly adapt to challenges and opportunities in the environment (McEwen & Wingfield, 2003). The human stress response system (SRS) is typically considered to comprise three anatomically distinct yet functionally coordinated neurobiological circuits: the sympathetic (SNS) and parasympathetic (PNS) branches of the autonomic nervous system (ANS) and the hypothalamic-pituitary-adrenal (HPA) axis. These systems support the regulation of energy metabolism and play a key role in attentional, emotional, and behavioral responses to fluctuating environmental demands (McEwen, 2008; Porges, 2001). Childhood and adolescence are thought to be particularly salient periods in the organization of these interwoven regulatory systems, setting the stage for subsequent health and adaptive functioning later in life (Gunnar, 2021).

Extensive research has demonstrated connections between physiological stress responsivity, early environments, and (mal)adaptive outcomes among youth (e.g., Lupien et al., 2009). However, most of this work has examined singular arms of the physiological SRS. Only recently has there been an increased focus on examining the coordination across systems, but findings have been inconclusive. In the current study, we leverage a meta-analytic approach to assess associations between the PNS, SNS, and HPA arms of the SRS and the extent to which the magnitudes of these associations vary systematically as a function of study methodological variables, sociodemographic characteristics, and environmental factors. Such meta-analytic findings have the potential to advance our understanding of multisystem physiological regulation by integrating across a complex and rapidly growing literature, offering theoretical and clinical implications that inform further research.

1.1. Multisystem physiological organization and coordination

The autonomic and adrenocortical arms of the stress response system are both biologically and functionally interdependent. Neuroanatomically, the ANS comprises dynamic neural networks that integrate activity across various brain regions, including the hypothalamus, brainstem, and spinal cord. During stress, the PNS arm of the ANS, with afferent nerves distributed throughout the body, releases tonic inhibitory influences on target organs, promoting activation of the SNS. Both arms of the ANS are partially regulated by the hypothalamus–the origin of HPA activation. Hypothalamic release of corticotropin releasing hormone triggers the pituitary gland to release adrenocorticotropic hormone, resulting in the release of the glucocorticoid hormone, cortisol (in humans), from the adrenal cortex, which then serves to regulate further hypothalamic activation (Herman, 2021). Given the closely linked neuroanatomical connections across the SNS, PNS, and HPA axis, activation in any arm of the stress system is inherently intertwined with the others through a complex, non-linear feed-forward and feedback set of excitatory and inhibitory signals (Ulrich-Lai & Herman, 2009). Research has begun to empirically evaluate this coordination, taking a “multisystem” approach that better acknowledges the self-organizing and synergistic nature of the stress response system. Importantly, however, the ‘true’ emergent dynamics of the stress response are difficult to capture with extant experimental paradigms and analytic tools, and thus the current study takes an important initial step in consolidating and clarifying linear associations across systems (Herman et al., 2003).

The interplay between systems promotes contextual adaptation, allowing individuals to maintain homeostasis at rest (e.g., body temperature regulation, digestion) or initiate allostasis in the face of challenge (e.g., mobilization of the stress response) (McEwen, 2004). During an acute stressor, such as threat, social evaluation, or noteworthy cognitive challenge, the stress response system re-organizes to flexibly adjust to changing environmental demands (Hamilton & Alloy, 2016). One framework for understanding real-time system coordination is the idea of ‘autonomic space’, which situates SNS and PNS functioning along two dimensions with varying degrees of reciprocity (i.e., how much one system increases in activation while the other decreases) and coactivity (i.e., how much the two systems increase or decrease together in activation) (Berntson et al., 1994). Broadly, when acute environmental demands are minimal, high levels of parasympathetic activation support the body’s ability to rest and repair and have been proposed to serve as a metaphorical ‘brake’ on the broader activity of the SNS and HPA axis (Porges, 2009). When contextual demands require greater physiological stress activation, autonomic balance tends to shift, lowering PNS activity, thereby removing the ‘brake’ and amplifying SNS activity to prepare the body and (indirectly) the brain to adapt to these changing contexts. When contextual changes require particularly pronounced activation (e.g., threat, injury), the comparatively slower HPA axis initiates its own cascade, culminating in the release of cortisol, which then binds to mineralocorticoid and glucocorticoid receptors distributed widely across the central and peripheral nervous system.

1.2. Physiological activation across childhood and adolescence

Early childhood and adolescence are characterized by pronounced developmental plasticity in the brain and related physiological stress response systems. Here we focus on the functional development of these systems, as indexed via activity during resting baseline (i.e., task baseline measures not intended to stimulate activation) and activation during instances intended to elicit an SRS response (i.e., acute reactivity) that are known to reflect adaptation to repeated stress exposures and can accumulate over time to influence wellbeing (Gunnar & Quevedo, 2007). Individual differences in developing physiological stress responses are captured by a variety of stress paradigms that generally take the form of a resting or baseline period, followed by a 5–10-minute stressor (e.g., social evaluative speech and/or math task, performance/challenge, startle), and a recovery period monitored for up to 40 minutes following peak stress. Biological indices of the SRS can be sampled throughout these tasks, with most studies collecting samples of salivary cortisol as an index of the HPA axis, salivary alpha amylase or pre-ejection period (PEP) as an index of the SNS, and respiratory sinus arrhythmia (RSA) or high frequency heart rate variability (HF-HRV) as an index of the PNS.

Previous meta-analytic findings examining these systems separately suggest a typical stress response involves decreases in PNS activation and increases in HPA axis and SNS activity, on average. While ANS responses occur immediately, the HPA axis response tends to peak approximately 20 minutes after stressor onset (Seddon et al., 2020; Wesarg et al., 2022). However, there is wide individual variation around this typical activation pattern, potentially driven by age, sex (Liu et al., 2017), and other contextual factors. Infants and children growing up in relatively safe, low stress environments tend to show more autonomically-driven responses to stress, characterized by lower cortisol reactivity and higher autonomic reactivity (Gunnar & Cheatham, 2003; Parent et al., 2019). As children transition into adolescence, biological processes (e.g., changes in gonadal hormones) associated with puberty begin to shift patterns of reactivity to more hyperresponsive patterns (see Joos et al., 2018 for a review). Given the variation in the developmental onset of puberty, assessing the role of puberty in addition to age is important to understanding associations across the SRS. Such pubertal changes are reflected by increases in cortisol reactivity and wider intra-individual variation in indices of autonomic reactivity that often depend on task context (Ellis et al., 2005; Gunnar & Cheatham, 2003). Youth exposed to early adversity, however, tend to demonstrate reactivity patterns marked by hypo- and hyper-responsivity across development (Gunnar & Vasquez, 2001; Hunter et al., 2011; Shirtcliff et al., 2021). These differences have been proposed to be reflective of the cumulative physiological toll, known as allostatic load, of prolonged elevations of the SRS in response to chronic stress—though theoretical models diverge somewhat in the extent to which these changes are typically framed as pathology (McEwen, 1998) or contextual adaptations (Boyce & Ellis, 2005; Del Giudice et al., 2011; also see section 1.3 below for further elaboration on this point).

In addition to the theoretical importance of examining cross-system physiological regulation, critical developmental competencies (e.g., self-regulation, psychosocial wellbeing) have been tied to the joint activation of the HPA axis and autonomic systems. Developmentally speaking, behavior and physiology are bi-directionally intertwined–thus, our understanding of core behavioral competencies likely require simultaneous consideration for how all three physiological systems operate to shape and predict observable behavior (Holochwost et al., 2021; Wadsworth et al., 2019). However, associations across arms of the SRS, particularly in relation to child outcomes, are largely unknown and potentially influenced by several moderators such as methodological characteristics of data collection (e.g., time of day reflecting natural circadian rhythms),sociodemographics, and early life experiences.

1.3. Multisystem co-activation patterns in developmental samples

Studies exploring patterns of multisystem physiological regulation among youth are limited. One of the earliest and most comprehensive of these studies in a developmental sample was from Quas and colleagues (2014), who used latent profile analyses to examine the extent to which there may be latent subgroups comprising varying combinations of autonomic and adrenocortical activation. Specifically, they found six cross-system profiles that included a range of high and low patterns of resting baseline and reactivity across systems. Recent evidence further suggests that meaningful clusters of physiological profiles may emerge as early as 18 months of age, with some profiles qualitatively re-organizing as children age (Rudd et al., 2021; Roubinov et al., 2021). Combined, these studies suggest moderate, average cross-system activation is the most common, followed by smaller proportions of children that fall into profiles characterized by anticipatory arousal and dominance of one system over another.

Complimenting this person-centered approach, related theoretical and empirical work has sought to specifically characterize the degree of symmetrical (i.e., positively correlated) versus asymmetrical (i.e., negatively correlated) physiological co-activation patterns across individuals (Bauer et al., 2002). Notably, most of this research has focused on co-activation between the HPA axis and SNS, with some work including PNS activation as a moderator. For example, in an adolescent sample, Glier and colleagues (2022) demonstrated greater positive (i.e., symmetrical) SNS-HPA co-activation in response to acute stress, but only when they also had increased PNS reactivity (i.e., greater decreases, indexing more withdrawal). In addition to theoretical relevance, one argument for this approach of analyzing patterns of stress system coactivation is that it may aid in better prediction of later stress-related mental health outcomes (Wadsworth et al., 2019). However, among the studies to explore the link between multisystem physiological co-activation and youths’ behavioral wellbeing, findings have been mixed. Internalizing and externalizing problems have been linked to both positively correlated (i.e., symmetrical) and negatively correlated (i.e., asymmetrical) co-activation of the HPA axis and SNS during resting baseline (El-Sheikh et al., 2008; Ursache & Blair, 2015) and during reactivity (Gordis et al., 2006; Allwood et al., 2011). However, asymmetrical patterns have also been associated with optimal developmental outcomes (Berry et al., 2012) and another study found no interactive effects in predicting disruptive behavior (de Vries-Bouw et al., 2012).

Adding to the complexity of this burgeoning multisystem literature, researchers note vast individual variation, with many of the demonstrated co-activation patterns varying across age (e.g., Rudd et al., 2021), family socioeconomic status (e.g., Quas et al., 2014), and adversity (e.g., Busso et al., 2017; Ellis et al., 2017). The interplay between multisystem stress physiology and environmental and sociodemographic factors remains unclear, but prominent theoretical models offer hypotheses that have growing support in the literature. One of the most notable theoretical frameworks is the Adaptive Calibration Model (ACM; Del Giudice et al., 2011). The ACM posits that the stress system is tuned to meet the demands of an individual’s environment despite its potential long-term costs on health and mental well-being. Specifically, the ACM proposes four SRS patterns (see Del Giudice et al., 2011 for a thorough review and visual depiction of profiles): sensitive (high PNS rest and reactivity, moderate SNS rest and reactivity, and moderate HPA rest and high reactivity), buffered (moderate-to-high PNS rest and reactivity, low-to-moderate SNS rest and reactivity, moderate HPA rest and reactivity), vigilant (low PNS rest and low-to-moderate PNS reactivity, high SNS rest and reactivity, and moderate-to-high HPA rest and reactivity), or unemotional (low rest and reactivity across all systems). Extant research adopting this model is limited, but recent empirical work has demonstrated some early support for the ACM, finding that typically developing youth from backgrounds characterized by lower stress tend to display buffered or sensitive profiles of SRS co-activation whereas youth from backgrounds characterized by higher stress tend to display more vigilant or unemotional profiles (Ellis et al., 2017; Quas et al., 2014; Glier et al., 2022; Hagan, et al., 2020; Rudd, et al., 2021). Clarifying these patterns empirically has implications for theory and practice aimed at supporting youth across a diverse array of contexts.

1.4. The current study

The overall strength of the relations between the three key arms of the stress response system is difficult to ascertain based on any single study due to variations in stress paradigms and sample characteristics. Thus, in the current meta-analysis, we aim to critically evaluate the literature on cross-system physiological regulation in youth ages 5–20 years old–a period in which acute stress task paradigms tend to be more qualitatively similar (Seddon et al., 2020). We use meta-analysis to (1) quantify the relation between HPA axis and SNS, HPA axis and PNS, and SNS and PNS systems during resting baseline, as well as in response to acute stressors (i.e., reactivity), and (2) determine if cross-system relations vary as a function of methodological characteristics, sample sociodemographics, or the presence of adversity. Given substantial variability across studies, we did not have specific directional hypotheses about average relations between arms of the SRS. The extent to which average cross-system relations vary as a function of other methodological and demographic characteristics were also exploratory. Investigating these relations meta-analytically, across childhood to late adolescence, may help inform this rapidly growing area of research.

2. Method

2.1. Literature search strategy and selection criteria

We followed PRISMA reporting guidelines when preparing the protocol, and all methods for this meta-analysis were preregistered via PROSPERO. We conducted searches through PsycINFO, Medline, Web of Science, as well as gray literature from Proquest (dissertations) and conference proceedings for studies that examined the adrenocortical arm of the SRS and at least one of the autonomic arms (i.e., SNS, PNS) in youth ages 5–20 years old. We focused on this age range to maximize similarity across stress tasks (i.e., stress tasks for youth under five years old tend to be more variable than tasks for older youth). All searches took place between March 1st 2021 to July 11th 2022.

Each study was required to meet the following criteria to be eligible for inclusion: (1) included a quantitative analysis of the statistical relations between HPA axis activation via salivary cortisol and at least one validated index of a specific branch of the ANS (e.g., salivary alpha-amylase, cardiac physiology, blood pressure), (2) assessed the SRS using an acute stress task among a youth population between the ages of 5–20 years old, (3) did not recruit participants with neurodevelopmental or cardiovascular disorders without inclusion of a healthy control group (i.e., we screened for studies that included participants with neurodevelopmental or cardiovascular disorders and if there was no healthy control group, we excluded it, and if they did, we included it).

Individual articles were uploaded to Rayyan (Ouzzani et al., 2016) where we deleted duplicate papers and screened abstracts for inclusion criteria. The full-texts of the remaining articles were then screened twice, independently, by three study team members. The agreement between raters was substantial (91.5% agreement). Disagreements regarding initial inclusion or exclusion were resolved by the lead author. Authors of articles that met the inclusion criteria but did not include the data required to calculate an effect size were contacted directly for additional information only for the SRS measures of interest. If no data were provided upon request, the study was excluded. Figure 1 details the flowchart of selection of studies into this meta-analysis.

Figure 1.

Preferred reporting items for systematic reviews and meta analyses (PRISMA) flow diagram for identification and inclusion of studies in the meta-analysis.

2.2. Data extraction and coding

Five study team members independently coded the included studies using a derived template. Extracted data included general article information such as authors, year of publication, and number of participants. A total of six effects were extracted to differentiate between systems and whether the effect corresponded to resting baseline levels or stress reactivity: HPA-SNS (1) resting baseline levels or (2) stress reactivity, HPA-PNS (3) resting baseline levels or (4) stress reactivity, and SNS-PNS (5) resting baseline levels, or (6) stress reactivity. We also coded to distinguish the method used to index stress reactivity for each system and provided descriptive labels indicating how stress reactivity was operationalized (e.g., AUC, difference/change score, mean score across stressor). When both AUCi and AUCg were given, we used AUCi to reflect the reactive increases or decreases in relation to the stress task (Pruessner et al., 2003). Difference scores in this literature are typically calculated as the difference between baseline and the peak at which a given system is theoretically posited to be highest. For example, studies identified that, cortisol peaks at around 20 minutes and alpha amylase peaks immediately or soon after post-stress task, and then sampled accordingly. Methodological and sociodemographic moderators tested included: age, sex composition (% female), racial/ethnic diversity (% White, % Black, % Hispanic), reactivity calculation method, time of day, and stress task type (e.g., TSST vs. Challenge/Emotion Task). Given our interest in developmental differences, we coded mean sample age as well as whether sample age consisted of primarily children (i.e., less than or equal to 11.9 years old) or adolescents (i.e., greater than or equal to 12 years old) and examined moderator effects of both age variables. We also coded for the time of day physiological measures were collected as well as children’s puberty stage, given known associations between these variables and SRS functioning (Jessop & Turner-Cobb, 2008).

Theoretically substantive variables related to child sociodemographics and environmental experiences were additionally coded and tested as moderators. These included family socioeconomic status (SES) and the presence of adversity or early life stress. Moderators were coded continuously where possible, and if insufficient information existed, we applied categories. This was the case for SES and adversity. Specifically, we dichotomized SES such that samples were identified as low SES if half or more of the sample included primary caregivers with a high school education or less, an income-to-needs ratio of two or below, and/or the sample was described as low-income. Similarly, adversity was dichotomized such that samples were identified as having the presence of adversity if at least half of the sample experienced maltreatment, neglect, poverty, or was described as a high adverse-exposed sample. Moderators with less than five effect sizes in a given group were not tested. Coding disagreements were resolved between individual coders and the first author, and overall rater agreement was 94.5%.

Reported Pearson correlations between the physiological variables of interest were the preferred method of effect size extraction; when not available, raters selected the beta weights from the model with the fewest covariates or contacted authors to obtain correlation matrices. For associations including PNS reactivity, we inverted effect sizes for studies that indexed reactivity as a difference score or a task mean score (i.e., during stressor), given that greater PNS withdrawal (i.e., decreases) means greater reactivity (as previous meta-analyses using similar measures have done; see Wesarg et al., 2022). After inverting these effects, a positive correlation reflected that higher levels of HPA or SNS reactivity was associated with higher PNS reactivity (i.e., greater decreases in PNS activation from baseline to stressor or greater PNS withdrawal).

The full dataset containing coded articles can be found on the associated OSF page (see section 2.4 below).

2.3. Analytic Strategy

In order to model heterogeneity and account for the non-independence of effects, all effect sizes reflecting associations (separated by resting baseline and reactivity per system) between HPA axis activity and indices of autonomic physiology were extracted and aggregated through random-effects meta-analysis. Specifically, we applied a multivariate mixed effects meta-regression model–a meta-analytic mixed model for the analysis of multiple correlated outcomes (Berkey et al., 1998; Cheung, 2019). Variance was separately estimated between subjects (sampling variance), within studies, and between outcomes using maximum likelihood estimation and unstructured random effects, which allows random effects to have different correlated variances for each outcome. This model also employed an estimation approach that accounted for studies that did not measure all of the outcomes. In other words, if a particular study had only measured one of the six outcomes, then it was still included in the analysis by adjusting the variance-covariance matrix to be a 1 × 1 matrix equaling the sampling variance of the observed outcome.

Models were implemented in R v. 4.0.2 using the metafor package (Viechtbauer, 2010) and adapted open-source syntax and tutorial provided by Moreau & Gamble (2022). An I2 statistic was calculated for each relation to estimate the proportion of the variance in study effect sizes that are due to between-study heterogeneity (i.e. not due to random sampling error; Higgins & Thompson, 2002). By convention, I2 values of 25% are typically considered to indicate low heterogeneity, 50% moderate heterogeneity, and 75% high heterogeneity (Migliavaca et al., 2022). Subsequent moderator analyses, again using a multivariate mixed effects meta-regression model, were performed to identify whether the aforementioned moderators explained variations in effect sizes across each SRS relation. Moderator analyses were performed based on all effect sizes for which moderator data were present. For continuous moderators, variables were grand mean centered and we report the slope indexing the linear association between a moderator and variations in effect size. For categorical moderators, we report whether the effect size for a given level of the moderator is statistically different from zero (see Moreau & Gamble, 2022).

Publication bias was assessed two ways. First, we applied an adapted multi-level Egger’s regression asymmetry test (Egger et al., 1997; Fernández-Castilla et al., 2019) using sampling variance as a moderator in our model. Second, we measured Cook’s distance to identify particularly influential study effect sizes. Study effects that were more than three times the mean were removed from the main meta-analytic model and compared against the full model as a sensitivity check. Examinations of funnel plots were not performed as these are limited to univariate meta-analyses; thus our multivariate and multilevel data structure was not appropriate for this method.

2.4. Openness and transparency

As mentioned above, study methods were pre-registered via PROSPERO and can be found here: https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=244929. We deviated slightly from our pre-registration in two ways. First, although we specified that we would use a mixed effects meta-regression, we determined later that the structure of our data called for a multivariate mixed effects model. Second, the decision to inverse-code effects that included PNS reactivity (see Data Extraction and Coding section above) was not pre-registered; the decision was made upon further inspection of the individually coded articles. All other methods used in the current study align with our pre-registration. Code, dataset containing coded articles, and supplementary materials presented here can be found on OSF: https://osf.io/yg7ws/?view_only=a5f1f0ea79824f5faaca13083b6c04c5.

3. Results

The current meta-analysis included a total of 52 studies (49 journal articles and three dissertations/preprints) with 133 effect sizes combined across N = 7671 participants. Eleven studies included all six physiological outcomes, with the remaining studies primarily limited to HPA-SNS activity (see Table 1). Most studies used the TSST as their stressor paradigm (78.85%), with some slight adaptations given the age range across the studies (age range = 5 to 20 years; M = 12.6; SD = 2.75). Table 1 provides descriptive information for each study (see Supplemental Table S1 on OSF page for additional article characteristics). Individual study and summary effect sizes are presented in Figures 2–4. Overall, average meta-analytic effect sizes for each SRS relation were small in magnitude and are expanded upon below. See Table 2 for meta-regression moderation effects for resting baseline outcomes and Table 3 for reactivity outcomes. Additional descriptive plots for select moderators can be found in supplementary materials on this article’s OSF page (see section 2.4 above).

Table 1.

Overview of Studies Included in Meta-Analysis.

Authors and year	Sample size	Mean age (years)	Percent female	Percent White	Percent Black	Percent Hispanic	Stress task	Physiological effects provided
								HPA-SNS		HPA-PNS		SNS-PNS
								Baseline	Reactivity	Baseline	Reactivity	Baseline	Reactivity
Allwood et al. (2011)	56	12	51.8%	79%	7%		TSST + Challenge Task	X	X
Bae et al. (2015)	169	11	41.4%	100%			TSST		X
Bendezu & Wadsworth (2018)	151	10.33	48.3%	93.4%			TSST	X	X
Busso et al. (2017)	169	14.9	55.6%	41.3%	18%	18%	TSST	X	X	X	X	X	X
Caster (2001)	43	14.69	51.1%	80%	4.4%	5.6%	TSST (speech only)		X
Chaplin et al. (2015)	193	14.69	50.3%		83.9%		TSST		X
Chen et al. (2020)	419	11.92	50%	12.6%	79.5%		TSST	X	X
de Veld et al. (2012)	158	10.6	52.5%	94.0%			TSST		X
DeJoseph et al. (2019) *	865	13.12	48.4%	54.9%	45.1%		TSST	X	X
Donley (2017)	60	13.98	0%	46.7%	6.7%	25%	TSST		X
Doom et al. (2020)	257	8	49%	52.9%	16.7%	10.1%	TSST + Challenge Task	X	X
Doussard-Roosevelt et al. (2003)	30	5.6	40%	43%	17%	23%	Challenge/Emotion Task				X
Duprey et al. (2021)	101	10.28	51%	11%	75%	9%	TSST	X	X	X	X	X	X
El-Sheikh et al. (2008)	64	8.72	56%	66%	34%		Challenge/Emotion Task	X	X
Ellis et al. (2017)	351	16.14	0%	100%			TSST	X	X	X	X	X	X
Evans et al. (2016)	327	17.3	44%	100%			TSST (speech only)	X	X	X	X	X	X
Glier et al. (2022)	72	12.5	45.8%	66.7%	27.7%		TSST	X	X	X	X	X	X
Gordis et al. (2008)	84	12.1	47.1%	10.7%	41.7%	39.3%	TSST	X	X
Gunnar, Frenn et al. (2009) *	42	11.25	55%				TSST	X	X	X	X	X	X
Gunnar, Wewerka et al. (2009) *	82	12	100%	88%	1.2%	2.4%	TSST	X	X	X	X	X	X
Gunnar et al. (2021)	68	15.84	39.7%	92%	1%		TSST	X	X
Han et al. (2020)	61	8.21	36%				TSST (speech only)			X	X
Hastings et al. (2011)	215	13.67	49.3%	71%	16%	2%	TSST	X	X
Katz & Peckins (2017)	54	16.6	74%				Group Public Speaking		X
Khoury et al. (2020)	114	8.88	44.3%	76.5%	2%		TSST	X	X
Kliewer et al. (2012)	228	14.1	55%		90%		Social Competence Interview	X	X
Lucas-Thompson et al. (2018)	153	12.92	52%	49%	17%	1%	TSST	X	X	X	X	X	X
Mücke et al. (2020) *	60	17.9	0%				TSST	X	X
Mücke et al. (2021)*	43	18	0%				TSST	X	X
Obradovic et al. (2010)	338	5.32	48.2%	43%	19%	4%	Challenge/Emotion Task			X	X
Platje et al. (2017)	197	17.31	41.6%	100%			Leiden Public Speaking		X
Quas et al. (2014) *	168	10.8		53%%	5%	8%	TSST	X	X	X	X	X	X
Quas et al. (2018)	94	11	53.2%	80%			TSST		X
Rabkin (2014)	48	10.6	52%	13%	38%	4%	TSST (math only)		X
Rahal et al. (2020)	91	18.39	57.5%	35.6%		64.4%	TSST			X	X
Rigterink (2013)	35	9.5	51.4%	54.5%	12.1%		TSST	X	X
Rudolph et al. (2011)	132	9.46	51.5%	71.2%			Challenge/Emotion Task	X	X
Schuurmans et al. (2021)	125	14.76	31%	100%			TSST	X	X	X	X	X	X
Shapero (2017)	127	15.28	49%	47%			TSST	X	X
Shenk et al. (2014)	110	17	100%	42%	51%	1%	Challenge/Emotion Task			X	X
Snoek et al. (2004)	70	10.02	15.7%	100%			Challenge/Emotion Task	X
Susman et al. (2010)	135	10.5	51.1%	89.6%	2.2%	3.7%	TSST		X
Teisen et al al. (2021)	98	9.6	52%	100%			Cold Pressor Task			X	X
Tsai et al. (2021)	95	11.63	48.1%	80.0%			TSST		X
Ungvary et al. (2018)	58	14.12	38.8%	45.1%	54.9%		Challenge/Emotion Task			X	X
van den Bos et al. (2014)	295	13.13	48.8%	100%			TSST (speech only)	X	X
Wade et al. (2020)	165	12	51%				TSST		X
Wadsworth et al. (2019)	149	10.31	48.9%	94.6%			TSST		X
Weyn et al. (2022)	101	11.61	54.6%				TSST	X	X	X	X	X	X
Wijnant et al. (2021)	137	14.2	49.6%				TSST		X		X		X
Wunsch et al. (2019) *	44	11.29	31.8%				TSST	X	X
Yim et al. (2015)	170	10.22	49%	50.6%	4.7%	8.8%	TSST				X

^*Denotes studies that included additional calculated indices or data than what was published. Received via email correspondence.

Figure 2.

Forest plots of individual study effect sizes from studies examining the association between HPA axis and SNS co-activation during resting baseline (left) and during reactivity (right). Horizontal lines indicate 95% CIs.

Figure 4.

Forest plots of individual study effect sizes from studies examining the association between SNS and PNS co-activation during resting baseline (left) and during reactivity (right). Horizontal lines indicate 95% CIs.

Table 2.

Meta-analyses of adrenocorticol and autonomic resting baseline co-activation.

	HPA-SNS restino baseline					HPA-PNS restino baseline					SNS-PNS restino baseline
	k	r	slope	SE	p	k	r	slope	SE	p	k	r	slope	SE	p
Total	30	0.066		0.020	0.001**	16	0.004		0.024	0.873	11	0.023		0.027	0.392
Methodological moderators
Time of day
Morning - Early Afternoon	5	0.129		0.050	0.010*	3	-	-	-	-	2	-	-	-	-
Late Afternoon - Evening	23	0.054		0.023	0.018*	11	0.019		0.032	0.550	7	−0.016		0.032	0.616
Stress task
TSST	25	0.058		0.029	0.045*	13	0.053		0.039	0.182	11	0.020		0.042	0.637
Other	5	0.119		0.063	0.062	3	-	-	-	-	0	-	-	-	-
Participant characteristics
Participant age (continuous)	30		−0.003	0.010	0.755	16		0.003	0.010	0.778	11		−0.002	0.017	0.891
Child	13	0.105		0.039	0.007**	6	0.017		0.057	0.77	4	-	-	-	-
Adolescent	17	0.039		0.032	0.223	10	0.030		0.042	0.473	7	0.014		0.048	0.763
Pubertal stage	5		−0.016	0.033	0.628	3	-	-	-	-	3	-	-	-	-
Percent female	29		−0.001	0.001	0.181	15		−0.001	0.001	0.568	10		−0.001	0.001	0.624
Percent White	24		−0.001	0.001	0.603	13		0.002	0.001	0.271	9		0.000	0.002	0.780
Percent Black	17		0.001	0.001	0.543	9		−0.002	0.002	0.307	6		0.002	0.002	0.361
Percent Hispanic	8		0.002	0.003	0.451	8		0.000	0.002	0.906	5		0.010	0.007	0.168
Socioeconomic status
High	12	0.033		0.039	0.403	10	0.000		0.042	0.993	6	0.009		0.052	0.862
Low	10	0.078		0.040	0.050	3	-	-	-	-	2	-	-	-	-
Adversity
Not present	20	0.083		0.031	0.008**	10	0.050		0.042	0.236	7	0.049		0.047	0.299
Present	8	0.043		0.048	0.377	5	0.054		0.064	0.399	4	-	-	-	-

Note. Moderators with less than 5 effect sizes in either group were not interpretted. HPA = hypothalamic adrenal axis, SNS = sympathetic nervous system, PNS = parasympathetic nervous system.

Table 3.

Meta-analyses of adrenocorticol and autonomic reactivity co-activation.

	HPA-SNS reactivity					HPA-PNS reactivity					SNS-PNS reactivity
	k	r	slope	SE	p	k	r	slope	SE	p	k	r	slope	SE	p
Total	43	0.092		0.025	<.001***	20	0.096		0.031	0.002**	13	0.004		0.050	0.939
Methodological moderators
Time of day
Morning - Early Afternoon	5	0.049		0.069	0.475	4	-	-	-	-	2	-	-	-	-
Late Afternoon - Evening	34	0.114		0.026	<0.001***	13	0.134		0.034	<0.001***	8	0.046		0.060	0.447
Stress task
TSST	37	0.085		0.024	<0.001***	15	0.084		0.036	0.020*	12	−0.062		0.040	0.120
Other	6	0.114		0.058	0.049*	5	0.152		0.068	0.025*	1	-	-	-	-
Reactivity index
HPA: Difference score	22	0.090		0.035	0.010**	9	0.143		0.048	0.003**	-	-	-	-	-
HPA: AUC	11	0.132		0.500	0.008**	2	-	-	-	-	-	-	-	-	-
HPA: Other index	9	0.043		0.054	0.424	9	0.046		0.045	0.311	-	-	-	-	-
SNS: Difference score	25	0.091		0.033	0.006**	-	-	-	-	-	6	−0.033		0.083	0.694
SNS: AUC	10	0.140		0.052	0.007**	-	-	-	-	-	2	-	-	-	-
SNS: Other index	7	0.050		0.062	0.414	-	-	-	-	-	5	−0.024		0.091	0.791
PNS: Difference score	-	-	-	-	-	11	0.072		0.044	0.106	8	−0.014		0.072	0.847
PNS: Mean score	-	-	-	-	-	9	0.117		0.048	0.014*	5	-	-	-	-
PNS: Other index	-	-	-	-	-	0	-	-	-	-	0	-	-	-	-
Participant characteristics
Participant age (continuous)	43		−0.006	0.009	0.521	20		−0.004	0.009	0.655	13		−0.023	0.016	0.142
Child	19	0.138		0.032	<0.001***	9	0.140		0.047	0.003**	5	−0.038		0.063	0.544
Adolescent	24	0.052		0.028	0.063	11	0.069		0.04	0.087	8	−0.058		0.045	0.191
Pubertal stage	9		−0.036	0.018	0.044*	3	-	-	-	-	3	-	-	-	-
Percent female	42		−0.003	0.001	0.077	19		−0.001	0.002	0.471	12		0.002	0.002	0.414
Percent White	33		0.001	0.001	0.229	16		−0.002	0.001	0.206	10		0.002	0.001	0.251
Percent Black	22		−0.000	0.001	0.738	11		0.003	0.002	0.233	6		0.002	0.004	0.616
Percent Hispanic	12		0.006	0.005	0.219	10		0.000	0.003	0.912	5		−0.005	0.016	0.769
Socioeconomic status
High	17	0.131		0.034	<0.001***	13	0.074		0.038	0.049*	7	−0.065		0.048	0.180
Low	12	0.038		0.037	0.311	3	-	-	-	-	2	-	-	-	-
Adversity
Not present	30	0.116		0.026	<0.001**	14	0.093		0.037	0.011*	9	0.020		0.042	0.638
Present	11	0.043		0.042	0.313	15	0.165		0.063	0.009**	4	-	-	-	-

Note. Moderators with less than 5 effect sizes in either group were not interpretted. Reactivity indices were only tested on relations that corresponded to the respective physiological index. HPA = hypothalamic adrenal axis, SNS = sympathetic nervous system, PNS = parasympathetic nervous system.

3.1. HPA – SNS co-activation

Among the available studies that examined the relations between HPA and SNS co-activation, Pearson’s r correlations ranged from −0.19 to 0.40 and −0.34 to 0.38 for resting baseline levels and reactivity, respectively (Figure 2). The multivariate mixed effects meta-regression model demonstrated that HPA and SNS co-activation showed a modest positive association for indices of resting baseline levels (k = 30, r = 0.07, SE = 0.02, p < 0.001) as well as reactivity (k = 43, r = 0.09, SE = 0.03, p < 0.001), on average.

Tests for heterogeneity using the I2 statistic indicated that across the included effect sizes, 43% of the variation in resting HPA-SNS levels and approximately 74% of the variation in HPA-SNS reactivity were due to meaningful (i.e., seemingly not due to sampling) variation between the studies. Moderation analyses revealed seven statistically significant moderators of the cumulative association between HPA and SNS activation (Table 2 and 3).

With respect to methodological moderators, time of day moderated both HPA-SNS resting baseline and reactivity. Specifically, effect sizes for HPA-SNS resting baseline were higher than average for studies that collected data in the morning to early afternoon hours (k = 5, r = 0.13, p = 0.01), and slightly lower for studies that collected data in the late afternoon to evening hours (k = 23, r = 0.05, p = 0.02). For HPA-SNS reactivity, effect sizes were slightly larger than average for studies that collected data in the late afternoon to evening hours (k = 34, r = 0.11, p < 0.001). Stress task type also moderated HPA-SNS resting baseline levels, such that effect sizes were smaller for studies using the TSST (k = 37, r = 0.06, p = 0.045). Stress task type also mattered for HPA-SNS reactivity, such that effect sizes were close to average for studies using the TSST (k = 37, r = 0.09, p < 0.001) and slightly higher for studies using other paradigms (k = 6, r = 0.11, p = 0.049). Finally, the way in which studies operationalized reactivity moderated HPA-SNS reactivity relations. Specifically, when HPA reactivity and/or SNS reactivity were indexed as area under the curve, effect sizes were significantly larger than average (k = 11, r = 0.13, p = 0.01 and k = 10, r = 0.14, p = 0.01, respectively). Effects were closer to average when HPA or SNS reactivity was indexed as a difference score (k = 22, r = 0.09, p = 0.01 and k = 25, r = 0.09, p = 0.01, respectively).

Additional moderator effects were evident for some participant demographic characteristics. Namely, samples consisting of more children than adolescents demonstrated larger than average associations between HPA-SNS during resting baseline (k = 13, r = 0.11, p = 0.01) and during reactivity (k = 19, r = 0.14, p < 0.001). Pubertal stage was also a significant moderator such that effect sizes for HPA-SNS reactivity tended to decrease in magnitude with increasing sample average pubertal stage (b = −0.04, SE = 0.02, p = 0.04). With respect to socioeconomic status, effect sizes for samples from higher socioeconomic backgrounds were significantly larger than the average effect size for reactivity (k = 17, r = 0.13, p < 0.001). Adversity was also a statistically significant moderator, but effect sizes only reached statistical significance for those not exposed to adversity. Specifically, effect sizes for HPA-SNS activation during resting baseline and during reactivity were slightly larger than average for resting baseline (k = 20, r = 0.08, p = 0.01) as well as for reactivity (k = 30, r = 0.12, p < 0.001). All other moderators were nonsignificant or unable to be interpreted given low cell counts (see Table 2).

3.2. HPA – PNS co-activation

Across included studies, associations between HPA and PNS co-activation (whereby increases in PNS reflect greater parasympathetic withdrawl) ranged from rs of −0.16 to 0.13 for resting baseline levels and −0.19 to 0.38 for reactivity (Figure 2). HPA and PNS co-activation showed a modest positive association for reactivity (k = 20, r = 0.10, SE = 0.03, p = 0.002), but no associations were found for resting baseline (k = 16, r = 0.004, SE = 0.02, p = 0.87).

Tests for heterogeneity using the I2 statistic indicated that approximately 23% of the variation in resting HPA-PNS levels and 65% of the variation in HPA-PNS reactivity were due to meaningful differences between the studies. Moderation analyses revealed six statistically significant moderators of the cumulative association between HPA and PNS co-activation during reactivity only (Table 3).

First, effects for HPA-PNS reactivity were more strongly correlated for studies that collected data in the late afternoon to evening hours (k = 13, r = 0.13, p <.0.001). Stress task type also moderated HPA-PNS reactivity, such that effects were slightly smaller for studies using the TSST (k = 15, r = 0.08, p = 0.02) but larger for those using other paradigms (k = 5, r = 0.15, p = 0.03). In addition, studies that indexed HPA reactivity via a difference score demonstrated greater correlations (k = 9, r = 0.14, p = 0.003). Studies using a mean score of PNS reactivity also demonstrated stronger associations (k = 9, r = 0.12, p = 0.01).

With respect to participant characteristics, child age was a significant moderator of HPA-PNS reactivity such that samples with more children than adolescents demonstrated stronger correlations (k = 9, r = 0.14, p = 0.003). With respect to socioeconomic status, effect sizes for samples from higher socioeconomic backgrounds were significantly smaller than the average effect size for reactivity (k = 13, r = 0.07, p = 0.049). Adversity was also a significant moderator. Specifically, effect sizes from samples not exposed to adversity showed slightly smaller effects (k = 14, r = 0.09, p = 0.01), whereas effect sizes from adversity-exposed samples showed higher associations (k = 15, r = 0.17, p = 0.01). All other moderators were nonsignificant or unable to be interpreted given low cell counts (see Table 2).

3.3. SNS – PNS co-activation

Across included studies, associations between SNS and PNS co-activation ranged from rs of −0.11 to 0.12 for resting baseline levels and −0.46 to 0.21 for reactivity (Figure 3). No overall associations between SNS and PNS co-activation were found for resting baseline k = 11, r = 0.02, SE = 0.03, p = 0.40) nor reactivity (k = 13, r = 0.004, SE = 0.05, p = 0.94).

Figure 3.

Forest plots of individual study effect sizes from studies examining the association between HPA axis and PNS co-activation during resting baseline (left) and during reactivity (right). Horizontal lines indicate 95% CIs.

Tests for heterogeneity using the I2 statistic indicated that approximately 27% of the variation in resting SNS-PNS levels and 83% of the variation in HPA-PNS reactivity were due to real differences between the studies. Moderation analyses revealed no statistically significant moderators for neither resting baseline nor reactivity.

3.4. Examination of publication bias

We used an adapted multilevel Egger’s test for asymmetry as well as Cook’s distance to examine potential publication bias. Results of the regression test indicated that HPA-SNS reactivity relations may be confounded by publication bias (z = 2.19, p = 0.03). An examination of Cook’s distance revealed three studies that had a disproportionate influence on the overall summary effect size estimates. Results from a sensitivity analysis with those outliers removed suggested nearly identical estimates and thus outliers were retained in our main models.

4. Discussion

The development of the stress response system across childhood and adolescence sets the stage for life-long mental and physical health (Nelson et al., 2020). Clarifying the interplay of the development of this coordinated system will likely play an important role in mitigating risk for developing stress-related diseases. In the present study, we meta-analytically quantified the effects from 52 studies comprising 7,671 participants ages 5–20 years old. This work was motivated by burgeoning research investigating multisystem coordination across the HPA axis and the sympathetic and parasympathetic branches of the autonomic nervous system. Our meta-analysis yielded several notable findings that help clarify inconsistencies in the literature. Broadly, overall patterns of associations were either null or small in magnitude but nonetheless offer insights into potential patterns of linear associations across the arms of the SRS and directions for future research. There was also significant heterogeneity in effects across studies and outcomes. Moderator analyses helped to inform the role of methodological decisions, sociodemographic factors, development, and experience in understanding cross-system coordination.

4.1. Overall associations across adrenocortical and autonomic systems

The null to modest effect sizes found across stress system relations is due in part to the noteworthy heterogeneity in the direction of effects across the individual studies. Significant average effects were observed for HPA axis and SNS co-activation, as well as HPA axis and PNS reactivity. Specifically, we found a modest positive association between the HPA axis and SNS, reflective of a symmetrical co-activation pattern during resting baseline and during reactivity such that higher adrenocortical activation generally corresponded with modestly higher sympathetic activation. A similarly modest positive correlation was found between the HPA axis and PNS during reactivity, reflective of higher adrenocortical activation corresponding with higher parasympathetic withdrawal. This association partially supports theory and some empirical work demonstrating intersystem balance necessary for maintaining homeostasis when faced with acute stress (Berntson et al., 1994; Del Giudice et al., 2011). Indeed, Quas and colleagues (2014) found that such a modulated, moderately activated pattern was the most common among youth ages 5–14 years of age, which they characterized as a ‘buffered’ response informed by the ACM (Del Giudice et al., 2011). Given the predominantly White and high-SES samples represented in the included studies and in Quas et al (2014), this buffering pattern may be indicative of a “typical” response specifically among children raised in environments with relatively low levels of hardship, structural inequalities, and daily stressors. In other words, this effect may be a consequence of sampling from primarily western (and White), educated, industrialized, democratic (WEIRD) populations (Henrich et al., 2010; Syed et al., 2018). Alternatively, it is also plausible that the HPA-SNS systems were only moderately correlated because we were unable to capture the more complex cross-system dynamics (see Limitations and Future Directions below).

No average associations were found for HPA-PNS resting baseline nor SNS-PNS co-activity at rest or during reactivity. There are at least three possibilities for these findings. First, far fewer studies included associations between these systems compared to associations we were able to extract for HPA-SNS patterns, and of the available associations, effects were highly mixed across studies. Thus, it is possible that there may be average patterns between the HPA axis and PNS at rest, and/or the SNS and PNS, but we were underpowered to detect the true population effects. Alternatively, the mixed findings we see in the literature reflect the possibility that there are no average patterns because the development and mobilization of the SRS occurs in context- and therefore person-specific ways, and examination of average linear associations in these systems may be missing the mark. A third possibility is that these null associations do in fact reflect meaningful general patterns of balanced and bi-directional coordination between these systems. For example, if youth maintain moderate parasympathetic activation, this may successfully dampen any similarly moderate sympathetic activation, and thus maintain homeostasis when environmental demands are minimal or nonthreatening (Bernston et al., 1994).

4.2. Moderator analyses: methodological considerations, sociodemographic correlates, and the role of environmental adversity

Our moderator findings suggest that relations across the SRS may vary as a function of several methodological and sample-level sociodemographic characteristics. First, the magnitude of associations between the HPA axis and both arms of the autonomic systems differed by the approximate time of day in which physiological samples were collected. Specifically, HPA-SNS co-activation at rest was more strongly associated than average in the morning to early afternoon hours, whereas the magnitude of this association decreased in the late afternoon to evening hours. This latter timing effect during resting baseline displayed the opposite pattern during reactivity, such that the HPA axis and SNS become more strongly positively associated when faced with an acute stressor in the later hours of the day than earlier hours. Time of day was also a significant moderator of HPA and PNS reactivity, with these systems demonstrating a more strongly positively associated pattern than average when measured later in the day. Taken together, these findings likely reflect normative differences in system-specific diurnal fluctuations occurring across the day (Sapolsky et al., 2000). For instance, diurnal salivary cortisol tends to demonstrate a peak in the morning and decrease throughout the day, whereas the opposite is true for salivary alpha amylase (Nater et al., 2007). These opposite diurnal trends are reflected in the opposite patterns of associations for resting baseline and reactivity for earlier versus later parts of the day.

Differences in reactivity relations were also found for the type of acute stress paradigm used as well as the computation method in which reactivity scores were indexed. First, HPA axis reactivity was less strongly correlated with SNS at rest for studies that used the TSST, potentially reflective of anticipatory stress previously seen in this task (e.g., Pulopulos et al., 2020). Reactivity to the TSST was similar to average HPA-SNS correlations, whereas correlations were stronger for studies that used other cognitive or emotionally challenging tasks. This latter pattern was also evident for HPA axis and PNS reactivity, paired with slightly smaller correlations for studies using the TSST. Although prior meta-analyses have shown the TSST to elicit reliable within-system HPA axis, SNS, and PNS responses, our findings suggest that reactivity during the TSST may rely on less coordination across these systems, on average. Alternatively, these systems may in fact be tightly coupled and unfold over time within individuals–as theory suggests–but we are unable to capture such dynamics with aggregate mean differences.

With respect to the way in which reactivity scores were operationalized, several significant effects emerged. HPA-SNS reactivity showed greater correlation when either the HPA axis or SNS was indexed by AUC, whereas this was not the case when either was indexed as a difference score. This pattern of findings supports the idea that AUC accounts for individual variation in the full profile of overall regulatory output compared to a difference score that may obscure this variation. The method of reactivity scoring also influenced relations between the HPA axis and PNS, but in qualitatively different ways. Specifically, HPA-PNS reactivity demonstrated greater co-variation when the HPA axis was indexed as a difference score or the PNS was indexed as a task-based average (e.g., average RSA value during the stress task). This was a notable finding given that the directionality of co-activation across these systems changes depending on computational decisions. Associations between the SNS and PNS did not appear to differ as a function of reactivity indices, but this may be because there was substantial variation in how the few studies included in our meta-analysis operationalized reactivity for these systems, possibly precluding the power to detect effects.

Developmental and demographic characteristics of the included study samples also moderated the effects of several SRS relations. Surprisingly, associations between systems did not change as the average age of the study samples increased. Descriptively, however, SRS relations exhibit more nonlinear patterns of change across continuous age (see Supplemental Figure S1). Additional exploratory analyses using age as a categorical variable differentiating predominately child (i.e., less than or equal to 11.9 years old) from adolescence (i.e., greater than or equal to 12 years old) samples, revealed several effects for HPA axis and SNS co-activation. For both resting baseline and reactivity, adrenocortical activation was more strongly associated with sympathetic activation for children. Relatedly, HPA-SNS reactivity associations were reduced with increasing pubertal stage. Studies including samples with more children than adolescents also exhibited greater HPA-PNS co-activation during reactivity. Combined, these developmental findings build upon our extant understanding regarding the developmental patterns observed within systems. Prior literature has shown that young children tend to show hyporesponsive patterns of SRS reactivity which slowly shifts towards increasing reactivity with age and puberty (Ellis et al., 2005; Gunnar & Cheatham, 2003; Parent et al., 2019). One interpretation of larger, or more positively correlated HPA-SNS and HPA-PNS associations in childhood than adolescence is that as development progresses, cross-system coordination becomes more organized and better able to return to homeostasis. In other words, although adrenocortical activation may increase across development, the autonomic system may become better equipped to counteract an overly reactive HPA response during acute stress–reflective of HPA-SNS and HPA-PNS cross-system effects that decrease in magnitude over time. The PNS may have a particularly salient role in regulating this developmental shift, as prior work has suggested (Glier et al., 2022; Quas et al., 2014), but our meta-analytic approach limited our ability to test this more complex coordination between all three arms of the system. Aside from developmental characteristics of the sample, we did not observe moderator effects of sex or racial group distributions.

Despite little research investigating the moderating role of environmental context on multisystem physiological regulation, our meta-analytic approach allowed us to explore emerging patterns as a function of SES and exposure to adversity (e.g., poverty, violence, maltreatment, neglect). In the current meta-analysis, very few moderation effects for the presence of poverty and adversity were observed (although see additional supplementary descriptive Figures S2 and S3). Among samples with youth who were not exposed to adversity, associations between the HPA axis and the sympathetic system during resting baseline were slightly more positively associated than average (i.e., more symmetrical). During reactivity to an acute stressor, youth from higher SES backgrounds as well as youth who were not exposed to adversity exhibited slightly greater positive associations (i.e., symmetrical reactivity) across the HPA axis and SNS. Different patterns were found for HPA-PNS reactivity. Specifically, youth from higher SES backgrounds youth as well as youth not exposed to adversity showed slightly less co-activation than average, whereas youth exposed to adversity showed greater co-activation. These patterns of findings do not align with any one ACM profile. Critically, lower representation of more disadvantaged samples obscure insight into a more comprehensive pattern of SRS activity that may exist between groups (see limitations below). Taken together, while our analytic approach does not afford rigorous tests of the ACM, the emerging sub-population-level patterns observed here calls for future work in this area.

4.3. Limitations and recommendations for future research

Several practical and substantive limitations of the current meta-analysis warrant additional consideration and inform our recommendations for future research on the development of multisystem physiological regulation. First, while many authors responded to data requests, we were unable to include 66 studies with potentially appropriate data because the relevant correlations were not reported or provided by authors upon request. This is a common limitation of many meta-analyses, but points to the importance of improving open data sharing practices and including comprehensive correlation matrices in main or supplementary texts. Second, we did not examine the role of health-related factors (e.g., smoking, exercise, prescription drugs) that are known to affect SRS functioning. This was because such factors were typically already screened for in participant recruitment, but could be an important moderator for future inquiry. Another key factor that we were unable to explore given the already large scope of the current meta-analysis was the role of behavioral outcomes–primarily internalizing and externalizing. Emerging work in this area has resulted in heterogeneous findings (e.g., Allwood et al., 2011; Gordis et al., 2006; Wadsworth et al., 2019), and future research should aim to clarify the extent to which positively associated (i.e., symmetrical) versus negatively associated (i.e., asymmetrical) patterns of SRS co-activation predicts various behavioral outcomes. Fifth, as illustrated in the participant characteristics of included studies (Table 1), our findings are primarily generalizable to White, upper middle-class youth growing up in relatively low-stress environments. As with most laboratory-based procedures, such homogeneity in sample sociodemographics often occurs because of accessibility constraints (e.g., proximity to a university) and a lack of trust for university research among historically marginalized groups due to historical patterns of injustice (Randolph et al., 2022). Recent innovations in adapted stress paradigms administered in the home (e.g., DeJoseph et al., 2019) or virtually (e.g., Gunnar et al., 2021) show particular promise for our collective goal of increasing representation in this area of research. Inclusion of more socioeconomically, racially, and gender diverse participants, combined with careful consideration of measures that index experiences of discrimination, is a particularly necessary future direction of this work.

Despite the many strengths of our meta-analytic approach for elucidating broad relations of multisystem regulation, a particularly notable limitation of this method is the inability to examine nonlinear patterns across all three of these arms of the SRS. Indeed, theory suggests that SRS functioning is a highly complex, dynamic, nonlinear process and thus may be why meta-analytic effects were small and variability in time of day as well as the computational method of reactivity across systems was such an influential moderator in our analyses. Although the specific cross-system dynamics remain somewhat unclear, evidence suggests that the PNS, SNS, and HPA-axis systems coordinate through complex, non-linear feed-forward and feedback processes that collectively support adaptation to and recovery from changing environmental demands. For example, catecholamines released by the SNS have been found to amplify multiple aspects of HPA-axis cascade (Dayas et al., 2001; Cecchi et al., 2002, Forray & Gysling, 2004), and glucocorticoid hormones are known to show both permissive and suppressive effects on different aspects of SNS functioning (Sapolsky et al., 2000). Collectively, this suggests that real-time and long-term adaptation to changing environmental demands is likely best supported by the dynamic functional organization of the ANS and HPA-axis that we were unable to capture in the current study.

Relatedly, this dynamic coordination of the SRS is highly dependent on individual experiences that differentially mobilize and canalize the SRS over time. According to the ACM, the SRS functions as an integrative mechanism that encodes environmental demands across development, which feed back on the longer-term calibration of the SRS. Such SRS calibration is thought to be finely-tuned to allow individuals to adapt to their unique context–ranging from safe and secure to harsh and unpredictable. Taken together, the complexity of SRS dynamics combined with its sensitivity to context calls for greater use of nonlinear and person-centered methods to fully capture such complexity. In addition to adopting within-person approaches such as latent profile analysis or growth mixture modeling, researchers can leverage recent innovations in time-series modeling of complex systems (e.g., Ihlen, 2012; Borsboom et al., 2021; Wallot & Leonardi, 2018). These latter methods are particularly well-suited for densely sampled measures of the SRS such as autonomic measures of cardiac activity. Indeed, recent work by our team has begun demonstrating the power of these methods for clarifying SRS functioning among youth (Berry et al., 2019; DeJoseph, 2023; Stallworthy, 2022). Application of such multisystem research would benefit not just how we understand early development of the SRS, but its nonlinear patterns of change across the lifespan and implications for health (Graham, Christian, & Kiecolt-Glasser, 2006).

4.4. Conclusion

In conclusion, the present meta-analysis aids in our burgeoning understanding of developing multisystem physiological regulation. We demonstrated that covariation between adrenocortical and sympathetic systems (at rest and during reactivity) as well as adrenocortical and parasympathetic (during reactivity) systems were modestly positively related, whereas no significant associations were found between the sympathetic and parasympathetic systems. Our methodological moderator findings underscore the importance of designing protocols that carefully consider diurnal rhythms, stressor paradigms, and the computation of reactivity indices that were shown to differentially influence the magnitude of associations across the HPA axis and both branches of the autonomic system. Developmentally, associations between adrenocortical and sympathetic activation during resting baseline and during reactivity were larger among children than adolescents. Similar developmental patterns were found for HPA axis and PNS reactivity. Environmental effects on cross-system regulation were less clear given underrepresentation of adverse-exposed youth in the included studies, but showed potential differences in HPA axis and SNS, as well as HPA axis and PNS co-activation. Collectively, our findings highlight the need for more person-centered and nonlinear dynamic system methods to capture the rich complexity of multisystem physiological regulation across development and how such developmental patterns adaptively calibrate across a diverse array of environmental contexts.

Highlights.

• Adrenocortical and sympathetic systems (at rest and during reactivity) as well as adrenocortical and parasympathetic (during reactivity) systems were modestly positively associated

• No average associations between other arms of the stress response system

• Associations varied as a function of methodological and sociodemographic characteristics

• Nonlinear dynamic system methods are needed to capture multisystem physiological regulation