The development of self-regulation is a primary task of early childhood (Calkins & Williford, Reference Calkins and Williford2009; Kopp, Reference Kopp1982). During the preschool period, roughly ages 3–5 years, self-regulation of negative emotions shows marked gains (Kopp, Reference Kopp1989). Because biological systems are believed to support the self-regulation of emotion (Levenson, Reference Levenson, Davidson, Scherer and Goldsmith2003, Reference Levenson2014; Thayer & Lane, Reference Thayer and Lane2000), empirical efforts to understand the early development of self-regulation frequently focus on associations between neurophysiological function and adaptive or maladaptive emotion propensities. Among the most frequently targeted are neural markers of self-monitoring, or the ability to effectively monitor one’s own behavior in order to make necessary adjustments (Brooker et al., Reference Brooker, Bates, Buss, Canen, Dennis-Tiwary, Gatzke-Kopp, Hoyniak, Klein, Kujawa, Lahat, Lamm, Moser, Petersen, Tang, Woltering and Schmidt2019; Brooker & Buss, Reference Brooker and Buss2014; Meyer et al., Reference Meyer, Carlton, Crisler and Kallen2018), and parasympathetic control, or the ability to effectively modulate metabolic output to meet shifting contextual demands (Porges, Reference Porges2003; Thayer & Lane, Reference Thayer and Lane2009). While systems of self-monitoring and parasympathetic control have overlapping neuroanatomical roots, we know surprisingly little about the degree to which they work together to support the development of self-regulation. That is, it is unknown whether these early systems of self-regulation work together or in a compensatory fashion to support the self-regulation of negative emotion in preschoolers. Understanding the nature of this association is important given that persistent psychophysiological dysregulation is linked to highly prevalent negative outcomes in young children, such as anxiety risk (Buss & McDoniel, Reference Buss and McDoniel2016; Kessler et al., Reference Kessler, Berglund, Demler, Jin, Merikangas and Walters2005; Van Hulle et al., Reference Van Hulle, Moore, Lemery-Chalfant, Goldsmith and Brooker2017). Thus, understanding associations between cortical and parasympathetic regulatory systems may be useful for more clearly elucidating dynamic psychophysiological correlates of self-regulation and identifying mechanistically-driven intervention targets for children at risk for negative outcomes.

Self-monitoring can be reliably indexed at the neural level in children as young as 3 years of age using ERN (Brooker et al., Reference Brooker, Bates, Buss, Canen, Dennis-Tiwary, Gatzke-Kopp, Hoyniak, Klein, Kujawa, Lahat, Lamm, Moser, Petersen, Tang, Woltering and Schmidt2019; Brooker, Reference Brooker2018; Grammer et al., Reference Grammer, Carrasco, Gehring and Morrison2014). ERN appears as a negative deflection in averaged EEG recordings approximately 50-100 ms following error commission (Falkenstein et al., Reference Falkenstein, Hohnsbein, Hoormann and Blanke1991). ERN is maximal at frontocentral electrode sites and is thought to be generated by the Anterior Cingulate Cortex (ACC; Dehaene et al., Reference Dehaene, Posner and Tucker1994; Van Veen & Carter, Reference Van Veen and Carter2002). The enhancement of ERN across development is associated with more adaptive learning and flexible response modulation (Tamnes et al., Reference Tamnes, Walhovd, Torstveit, Sells and Fjell2013). Notably, individual differences in ERN are linked to risk for anxiety problems in the form of fearful behavior in children. For example, both high and moderate (but not low) fear at age 2 predict frontal ERN activity at age 4.5 (Brooker & Buss, Reference Brooker and Buss2014) and fearfulness, but not shyness, is associated with reduced ERN in 6 year olds (Meyer & Klein, Reference Meyer and Klein2018). While a larger ERN is typically associated with elevations in generalized relative to specific (i.e., social) anxiety presentations in older samples (Meyer et al., Reference Meyer, Carlton, Crisler and Kallen2018), the nature of this relation appears reversed before age 9 (Meyer et al., Reference Meyer, Hajcak, Torpey-Newman, Kujawa, Olino, T., Dyson and Klein2018), whereby a smaller ERN predicts later anxiety and related disorders. Nonetheless, across ages, the ERN shows robust and reliable associations with anxiety risk (Moser, Reference Moser2017) and symptoms in young children (Meyer, Reference Meyer2017).

Another reliable marker of self-regulation in children is cardiac vagal flexibility, or the modulation of cardiac activity via the vagus nerve to fit a dynamic range of challenges (Porges, Reference Porges2009; Thayer & Lane, Reference Thayer and Lane2009; Wessel et al., Reference Wessel, Danielmeier and Ullsperger2011). Cardiac vagal flexibility can be indexed through Respiratory Sinus Arrhythmia (RSA). Suppression of RSA involves the withdrawal of parasympathetic control over autonomic activity to mobilize threat responsivity. Increased RSA suppression, linked to increased heart rate, facilitates efficient responses to real or perceived threats. Given that unique contexts call for different behaviors, there is not only one adaptive pattern of change in RSA (Hastings et al., Reference Hastings, Kahle and Han2014; Shakiba et al., Reference Shakiba, Ellis, Bush and Boyce2020). However, RSA suppression regularly predicts adaptive self-regulation in early life. In young children (2–4.5 years of age), reduced RSA suppression (or RSA augmentation) is associated with poorer self-regulation (Brooker et al., Reference Brooker, Buss, Lemery-Chalfant, Aksan, Davidson and Goldsmith2013; Brooker & Buss, Reference Brooker and Buss2010), increased fear and negative emotionality, and worse adjustment (Calkins & Keane, Reference Calkins and Keane2004). These social, emotional, and behavioral difficulties represent notable risk factors for the development and maintenance of anxiety and related disorders across the lifespan (Beauchaine, Reference Beauchaine2015; Cho & Buss, Reference Cho and Buss2017). However, some emotion-specific effects exist in the association between RSA and adaptation (Gatzke-Kopp et al., Reference Gatzke-Kopp, Willner, Jetha, Abenavoli, DuPuis and Segalowitz2015), underscoring the importance of interpreting effects in light of the incentives of the eliciting context.

According to the Neurovisceral Integration Model (Thayer & Lane, Reference Thayer and Lane2000), emotion–cognition interactions influence parasympathetic activity in order to facilitate goal-directed behavior. Therefore, the adaptive effects of cardiac vagal flexibility reflect the functional integrity of both neural and parasympathetic activity implicated in self-regulation. Polyvagal Theory (Porges, Reference Porges2003, Reference Porges2009) suggests that the vagus nerve specifically evolved to facilitate this brain-body feedback loop linking cortical with parasympathetic activity. In this way, individuals may adjust their physiological arousal to flexibly meet the demands of their environment, particularly social environments.

Expectations for coupling between cortical and parasympathetic systems in the service of self-regulation are grounded largely in theory, though a limited amount of empirical work with adults supports the plausibility of coupled systems. For example, consistent with both the Neurovisceral Integration and Polyvagal approaches, older samples evidence links between activation of the prefrontal cortex (PFC) and parasympathetic activation via bidirectional pathways mediated by the anterior cingulate cortex (ACC; Beauchaine, Reference Beauchaine2015; Thayer & Lane, Reference Thayer and Lane2009; Wessel et al., Reference Wessel, Danielmeier and Ullsperger2011). Further, both ERN and RSA suppression are directly associated with initial ACC activation and frontal brain activity (Wessel, Reference Wessel2012), suggesting a functional association between these cortical and cardiac indices might exist. Yet such a link has scarcely been tested directly, particularly in early childhood when coordinated regulatory processes are being calibrated (Hostinar & Gunnar, Reference Hostinar and Gunnar2013; Kopp, Reference Kopp1982) and critical brain development facilitates cognitive involvement in regulatory processes (Cachia et al., Reference Cachia, Borst, Vidal, Fischer, Pineau, Mangin and Houdé2014; Fjell et al., Reference Fjell, Walhovd, Brown, Kuperman, Chung, Hagler, Venkatraman, Roddey, Erhart, McCabe, Akshoomoff, Amaral, Bloss, Libiger, Darst, Schork, Casey, Chang, Ernst, Gruen, Kaufmann, Kenet, Frazier, Murray, Sowell, van Zijl, Mostofsky, Jernigan, Dale, Jernigan, McCabe, Chang, Akshoomoff, Newman, Dale, Ernst, Dale, Van Zijl, Kuperman, Murray, Bloss, Schork, Appelbaum, Gamst, Thompson, Bartsch, Jernigan, Dale, Akshoomoff, Chang, Ernst, Keating, Amaral, Sowell, Kaufmann, Van Zijl, Mostofsky, Casey, Ruberry, Powers, Rosen, Kenet, Frazier, Kennedy and Gruen2012). Indeed, developmental changes in the ACC during the preschool period are linked to concurrent age-related growth in observed behavioral monitoring (Tamnes et al., Reference Tamnes, Walhovd, Torstveit, Sells and Fjell2013) and cognitive control efficiency (Cachia et al., Reference Cachia, Borst, Vidal, Fischer, Pineau, Mangin and Houdé2014; Fjell et al., Reference Fjell, Walhovd, Brown, Kuperman, Chung, Hagler, Venkatraman, Roddey, Erhart, McCabe, Akshoomoff, Amaral, Bloss, Libiger, Darst, Schork, Casey, Chang, Ernst, Gruen, Kaufmann, Kenet, Frazier, Murray, Sowell, van Zijl, Mostofsky, Jernigan, Dale, Jernigan, McCabe, Chang, Akshoomoff, Newman, Dale, Ernst, Dale, Van Zijl, Kuperman, Murray, Bloss, Schork, Appelbaum, Gamst, Thompson, Bartsch, Jernigan, Dale, Akshoomoff, Chang, Ernst, Keating, Amaral, Sowell, Kaufmann, Van Zijl, Mostofsky, Casey, Ruberry, Powers, Rosen, Kenet, Frazier, Kennedy and Gruen2012); these associations are not present in late childhood or adolescence.

Slow frontal brain maturation in young children may facilitate central autonomic nervous system based response learning in childhood by allowing lower levels to “tune” priors (internal predictions based on experience) appropriately before top-down influences come fully online (Harteveld et al., Reference Harteveld, Nederend, ten Harkel, Schutte, de Rooij, Vrijkotte, Oldenhof, Popma, Jansen, Suurland, Swaab, de Geus, Prätzlich, Ackermann, Baker, Batchelor, Baumann, Bernhard, Clanton, Dikeos, Dochnal, Fehlbaum, Fernández‐Rivas, Gonzalez, González de Artaza‐Lavesa, Guijarro, Gundlach, Herpertz‐Dahlmann, Hervas, Kersten, Kohls, Konsta, Lazaratou, Kerexeta‐Lizeaga, Martinelli, van Nimwegen, Puzzo, Raschle, Rogers, Siklósi, Smaragdi, Steppan, De Brito, Fairchild, Kieser, Konrad, Freitag and Stadler2021; Porges & Furman, Reference Porges and Furman2011; Smith et al., Reference Smith, Thayer, Khalsa and Lane2017). Consistent with this possibility, individual differences in RSA would more likely stabilize before individual differences in ERN. However, efforts to establish normative developmental trajectories for both ERN (i.e., increases in amplitude [Meyer et al., Reference Meyer, Hajcak, Torpey-Newman, Kujawa, Olino, T., Dyson and Klein2018] and variability [Brooker, Reference Brooker2018] with age) and RSA suppression (i.e., increases from 2 months to 5 years, then decreases with age; Dollar et al., Reference Dollar, Calkins, Berry, Perry, Keane, Shanahan and Wideman2020), suggest considerable individual variability in both measures in early childhood. Such variability may reflect dynamic codevelopment across systems that could be leveraged for a more useful understanding of adaptive vs. maladaptive development of regulatory systems and individual differences in outcomes. For example, smaller childhood ERN in and of itself may be associated variably with maladaptive outcomes because it is only one of multiple influences on complex emotion behaviors. Thus, there is added utility in understanding the implications of a smaller ERN coupled with compensatory parasympathetic influence (i.e., greater RSA suppression) relative to less parasympathetic influence (i.e., less RSA suppression or even RSA augmentation) for more sensitively determining individual risk for negative outcomes. Therefore, one aim of this work was to identify the normative presence or absence of coupling in ERN and RSA during the preschool years.

Beyond normative function, a critical challenge in investigations of neurophysiological systems of regulation, including cross-system coupling, is that of interpreting adaptive and maladaptive patterns of function. Particularly in studies of childhood risk, before disorder manifests and in years when self-reports can be unreliable (Grills & Ollendick, Reference Grills and Ollendick2003), the implications of individual differences in physiological function can be unclear. However, consistent with the Developmental Psychopathology perspective emphasizing multiple levels of analysis (Cicchetti & Dawson, Reference Cicchetti and Dawson2002), longitudinal work that also includes observations of child behavior allows for the interpretation of individual differences in patterns of physiology alongside typical and atypical behavioral responses and/or childhood outcomes. For example, coupling between regulatory systems may be interpreted as adaptive if it occurs in the presence of typical emotion behaviors, and this argument is strengthened if it is also associated with more adaptive long-term function. In addition, it may be useful to interpret patterns of coupling in relation to other established biological correlates of regulation or dysregulation, embedding it into an overall picture of individual propensities for adaptive or maladaptive function.

The typical (versus atypical) nature of children’s regulatory behaviors is perhaps most interpretable in light of the eliciting context. Consistent with a functionalist perspective on emotion (Campos, Reference Campos1996), context-appropriate emotion behaviors suggest a response that is well-matched to the incentive properties of the eliciting context (i.e., displays of fear in a fear-eliciting context), while context-inappropriate emotion behaviors suggest a mismatch between the eliciting context and children’s emotion responses (i.e., displays of fear in a positivity-eliciting context). Context-inappropriate fear responses, in particular, are linked to dysregulated emotion tendencies (Brooker et al., Reference Brooker, Davidson and Goldsmith2016; Buss, Reference Buss2011) and enhanced risk for the development of anxiety problems (Buss et al., Reference Buss, Davis, Kiel, Brooker, Beekman and Early2013). In contrast, context-appropriate fear responses such as stranger fear are evolutionary in origin, putatively adaptive, and common in 2-3 year olds (Brooker et al., Reference Brooker, Buss, Lemery-Chalfant, Aksan, Davidson and Goldsmith2013; Marks, Reference Marks1987). Thus, patterns of coupling across biological systems in children with propensities for context-inappropriate fear responses may be more likely to reflect atypical regulation than patterns of coupling linked to context-appropriate emotion responses.

Similar to contextual affect, neuroendocrine reactivity has established associations with early emotion behaviors and maladaptive outcomes (Brooker et al., Reference Brooker, Buss, Lemery-Chalfant, Aksan, Davidson and Goldsmith2013; Hostinar & Gunnar, Reference Hostinar and Gunnar2013; Koss & Gunnar, Reference Koss and Gunnar2018; Van Hulle et al., Reference Van Hulle, Moore, Lemery-Chalfant, Goldsmith and Brooker2017). Neuroendocrine reactivity reflects activation of the hypothalamic–pituitary–adrenal (HPA) axis, the human end product of which is cortisol. Like other regulatory systems, the HPA axis functions to maintain the biological resources available for responding to environmental challenge, although it works on a much slower time scale relative to systems like RSA (Gunnar & Quevedo, Reference Gunnar and Quevedo2007). Even so, blunting of the HPA axis (i.e., hypo-cortisolism) is accordingly linked to chronic stress and threat reactivity in preschoolers (Badanes et al., Reference Badanes, Watamura and Hankin2011; Doom et al., Reference Doom, Cook, Sturza, Kaciroti, Gearhardt, Vazquez, Lumeng and Miller2018; Gunnar & Vazquez, Reference Gunnar and Vazquez2001), and potential chronic dysregulation across the lifespan. Indeed, cortisol output shares unique associations with fear expression and emerging internalizing behaviors in the preschool period (Buss et al., Reference Buss, Davis and Kiel2011). Therefore, activity in well-explicated systems like the HPA axis can help to interpret coupled ERN and RSA activity in relation to emotion propensities and anxiety risk in young children.

In sum, although supported by overlapping neurovisceral mechanisms, indices of multi-systemic regulatory function are not well understood, both broadly and in early childhood when self-regulatory abilities undergo critical development (Beauchaine, Reference Beauchaine2015; Bornstein & Suess, Reference Bornstein and Suess2000; Dawson et al., Reference Dawson, Panagiotides, Klinger and Hill1992; Hostinar & Gunnar, Reference Hostinar and Gunnar2013; Koenig, Reference Koenig2020; Kopp, Reference Kopp1982, Reference Kopp1989). Understanding how regulatory systems develop in relation to each other and in relation to emotion behaviors in early childhood is important for elucidating dynamic psychophysiological correlates of self-regulation and identifying mechanistically-driven correlates of adaptive relative to maladaptive developmental outcomes. We tested for distinct patterns of coupling between cortical (i.e., ERN) and parasympathetic (i.e., RSA) regulatory systems in the preschool period. We used concurrent and longitudinal observations of children’s emotion behaviors linked to child anxiety risk along with neuroendocrine reactivity (i.e., cortisol output) to characterize significant patterns of coupling.

We hypothesized that significant coupling, or correlation, would exist when predicting ERN from RSA during a cognitively challenging Go/No-go task (GNG; Brooker & Buss, Reference Brooker and Buss2014; Meyer, Reference Meyer2017). Specifically, we predicted a larger ERN would be coupled with more RSA suppression. We chose a standard GNG task given its effectiveness in eliciting ERN in children as young as age 3 (Brooker, Reference Brooker2018), and also because there is an established pattern of RSA response to cognitive challenge (Zeytinoglu et al., Reference Zeytinoglu, Calkins and Leerkes2020). Thus, we could be confident that null effects, if found, were not the result of an inability to elicit sufficient variability in either ERN or RSA response. However, we do not assert that any effect would be specific to a GNG challenge.

In line with both the Neurovisceral Integration Model and Polyvagal Theory, significant ERN-RSA coupling may be most expected when accounting for emotion behavior (Porges, Reference Porges2009; Thayer & Lane, Reference Thayer and Lane2009). Therefore, we further hypothesized that patterns of coupling characterized by a smaller ERN and less RSA suppression (or RSA augmentation) would be most visible in relation to markers of anxiety risk such as greater atypical fear (high fear in a low-threat context), lower typical fear (low fear in a high threat context), greater anxiety symptoms, and blunted neuroendocrine reactivity. In contrast, we expected coupling characterized by a larger ERN and more RSA suppression, reflecting coordinated regulation, to be most related to adaptive child outcomes such as less atypical fear, greater typical fear, less anxiety symptoms, and more neuroendocrine reactivity. Separate models were employed to allow for a conceptual replication of effects across different levels of analysis related to anxiety risk.

Methods

Participants

This research was approved by the Institutional Review Board at Montana State University [RB070213-FC]. The sample included 121 preschoolers (58% female) assessed longitudinally at ages 3 (M = 3.59, SD = 0.15, range = 3.13–3.99 years) and 5 (M = 5.52, SD = 0.12, range = 5.28–5.98 years) as part of a larger study between Spring 2014 and Winter 2017. Children were recruited from the general community via flyers, news advertisements, mailings based on local birth records, and word-of-mouth. Parents were compensated after each study visit. Sample characteristics are summarized in Table 1.

Table 1. Participant characteristics for children and their biological parent

Note. Data for those mothers (n = 89) and fathers (n = 68) that reported gross annual income are presented.

Procedures

Parents enrolled children in the study by contacting research staff. Upon arrival at the laboratory, parents were informed of study procedures and provided written consent for their child’s participation. At age 3, children participated in a modified Go/No-Go (GNG) task while continuous EEG and ECG data were collected. Children then participated in two standardized behavioral paradigms intended to elicit either positive (watching a puppet show) or negative (meeting a stranger) emotion (Buss & Goldsmith, Reference Buss and Goldsmith2000). At age 5, the primary caregiver completed the Anxiety and Related Disorders Interview Schedule for DSM-5 (ADIS-5 PV; Albano & Silverman, Reference Albano and Silverman2016) and salivary cortisol was collected from children at three time points: upon arrival, 30 minutes following peak arousal to the EEG procedure, and at the end of the visit.

Measures

ERN

ERN was elicited from children using a modified GNG task (Brooker, Reference Brooker2018) during which either a spaceship (no-go stimulus) or an asteroid (go stimulus) was presented vertically in the center of a 23 computer screen using Presentation stimulus delivery software (Neurobehavioral Systems, Inc.). Children were instructed to push the response button as quickly and accurately as possible to destroy the asteroids but be careful not to destroy other spaceships. Stimuli were presented for 1200 ms and were preceded (200 ms) and followed by (300–800 ms) a gray fixation cross. The full task comprised 2 blocks of 40 trials presented pseudo randomly (at least 60% go trials) for a maximum of 80 trials per child. Prior to the task, children practiced with a research assistant using laminated pictures and completed two computerized practice blocks of 10 trials each. Children received a sticker after completing each block. To equate task difficulty across participants, an error rate of roughly 50% was maintained through an automated procedure that decreased stimulus presentation time by 50 ms following 2 consecutive correct responses and increased stimulus presentation time by 50 ms following 2 consecutive incorrect responses.

Continuous EEG data were acquired during GNG using a BioSemi Active 2 recording system and a 64-channel cap with Ag-AgCl-tipped electrodes arranged according to the 10–20 labeling system. To collect eye movements, electrodes were also placed at the outer canthi of the left and right eye and at the supra and infra orbital sites of the left eye. Two electrodes were also placed on the mastoids for later re-referencing. During recording, data were referenced to the Common Mode Sense and Driven Right Leg electrodes and sampled at a rate of 2048Hz.

Data were processed offline using Brain Vision Analyzer Version 2.1. All electrodes were re-referenced to the average of the right and left mastoids, high-pass filtered at 0.1 Hz, and corrected for eye movement or blinks (Gratton et al., Reference Gratton, Coles and Donchin1983). Correct and incorrect trials were segmented (−200 – 600 ms) and baseline corrected for 200 ms prior to the response. Artifacts were marked in segmented data when one of the following criteria were met: a voltage step of more than 75 μV between data points, a difference of 150 μV within 200ms, amplitudes below 0.5 μV within a 50ms period, and activity that exceeded + 100μV or − 100 μV. Remaining segments were visually inspected for artifacts. Clean segments were averaged and low-pass filtered at 30 Hz. For participants with at least six trials of usable data (Olvet & Hajcak, Reference Olvet and Hajcak2009; Pontifex et al., Reference Pontifex, Scudder, Brown, O’Leary, Wu, Themanson and Hillman2010), peak negative amplitudes between 0 and 100 ms were marked as ERN at electrodes Cz and Fz, which were selected based on previously-reported ERN in these data (Canen & Brooker, Reference Canen and Brooker2017); see Brooker (Reference Brooker2018) for waveforms. Split-half reliabilities for ERN were moderate at age 3 (r = .62; ICC = .64). To isolate error-specific activity, a linear regression was run predicting incorrect trial amplitudes from correct trial amplitudes (Meyer & Klein, Reference Meyer and Klein2018). Unstandardized residuals from this analysis were saved, with more negative values indicating larger ERN and more positive values indicating smaller ERN (independent of correct trial activity).

RSA

Measures of cardiac output were collected during GNG using the same BioSemi Active 2 recording system by placing 2 exogenous electrodes on each wrist (i.e., either side of the midline). The ECG signal was sampled at a rate of 500 Hz and bandpass filtered at 40 and 250 Hz to reduce low (e.g., signal drift) and high-frequency artifacts (e.g., changes in muscle tension) that can distort the signal and reduce data quality (Pérez-Riera et al., Reference Pérez-Riera, Barbosa-Barros, Daminello-Raimundo and de Abreu2018).

RSA analyses were performed offline using the Mindware editing program Mindware HRV, Version 2.51 (Mindware Technologies, Ltd, Westerville, OH) and a validated algorithm to identify IBIs and detect physiologically improbable intervals based on the overall distribution (Berntson et al., Reference Berntson, Quigley, Jang and Boysen1990). All data were visually inspected for artifact identification and editing. Data were detrended using a first-order polynomial to remove the mean and any linear trends, cosine tapered, and submitted to fast Fourier transform. Child RSA was defined as the natural log integral of the .24–1.04 Hz power band and calculated in 30 sec epochs (minimum of 20 sec for final epochs). To protect against artificial changes in RSA due to changes in respiration (Grossman et al., Reference Grossman, Van Beek and Wientjes1990), children who cried or had hiccups (n = 3) were not scored. Thus, the final data set contained 92 children with complete heart rate data during GNG.

Similar to previous work (Brooker & Buss, Reference Brooker and Buss2010), estimates of RSA in response to challenge were extracted from two-level mixed growth curve models (Singer, Reference Singer1998). Data were centered at the beginning of the tasks (i.e., the first 30 s epoch). Then, growth curve models in which linear time predicted RSA values were used to estimate linear changes in RSA across the GNG task. Individual scores of linear change were extracted for each child using the SAS PROC REG procedure. Calculated in this way, positive and negative slope values reflect linear changes in RSA over time, resulting in positive values when RSA increases across the task (less suppression/more augmentation) and negative values when RSA decreases across the task (more suppression).

Context appropriate and inappropriate fear

Fear at age 3 was observed during two standardized behavioral paradigms derived from the preschool version of the Laboratory Temperament Assessment Battery (Lab-TAB- Buss & Goldsmith, Reference Buss and Goldsmith2000): Context-appropriate fear was assessed during a Stranger Approach episode designed to elicit fear and wariness. For this, an experimenter led the parent and child into a room where parents were asked to sit quietly and remain uninvolved. The child was given a set of age-appropriate toys to play with. After the child had played with the toys for 30 sec, a male experimenter entered the room. The stranger approached the child slowly and watched them play for up to 2 min while he asked several questions (e.g., ‘‘Are you having fun today?’’) before leaving the room.

Context-inappropriate fear was assessed during a Puppet Show episode designed to elicit positive affect. For this, an experimenter instructed the parent to sit with their preschooler on their lab opposite a puppet theater approximately 10 feet away. Two puppets played a series of three games together (catch, fishing, sticker), continually inviting the toddler to join them. Each game lasted 1 minute, after which the puppets declared the show over and said goodbye.

Trained coders scored both Stranger Approach and Puppet Show fear behaviors offline using video recordings of the episodes. Coders were required to achieve a minimum reliability of κ = 0.70 with a master coder before coding independently. The following behaviors were coded for across each episode: intensity of facial fear, bodily fear, distress vocalizations, verbal hesitancy, gaze aversion, and escape behaviors. Codes for fear behaviors were assigned every 5–10 sec, per Lab-TAB coding instructions. Roughly 20% of episodes were double coded to prevent coding drift (Stranger Approach mean κ = 0.80; Puppet Show mean κ = 0.78). Codes were averaged across the episode for each behavior. Average scores were then standardized within behavior and mean composited to create a single measure of fearfulness for each episode.

Anxiety risk

GAD symptoms

Generalized Anxiety Disorder (GAD) symptoms were evaluated when children were 5 years old. Parents completed the ADIS-5 (Albano & Silverman, Reference Albano and Silverman2016), a semi-structured interview derived from DSM-5 criterion for anxiety and other related conditions in children. Interviewers were psychology graduate students who established training reliability on administering and scoring the interview with a Master-level clinical psychologist. Interviews were tape recorded and double coded on 7% of cases and reliability was high (>90% agreement across all scales). For each DSM-5 diagnosis, interviewers rated the degree of distress and lifestyle impairment associated with characteristics of each disorder (0 = none, 8 = very severely disturbing/disabling). Interviewers inquired about each of the 4 worry behaviors proposed for GAD (i.e., avoidance, over-preparation, procrastination, reassurance seeking). After recording a description of the worry behavior, the behavior was rated on a 0–8 scale of frequency/severity (0 = never, 8 = always/very severe). Symptoms endorsed correspond to 6 criteria across the 4 worry behaviors. Therefore, possible sum scores range from 0-6, with 6 indicating clinical threshold. Three participants met the criteria for GAD disorder at age 5.

Neuroendocrine reactivity

To assess cortisol levels across the lab visit, three saliva samples were collected from each preschooler; a pre-task baseline collected upon arrival, a sample 30 minutes following peak arousal to the EEG procedure, and a sample at the end of the visit. Saliva samples were collected by holding a dental swab (Salimetrics, LLC part 5501.20) under the child’s tongue or in the corner of the mouth for approximately 90 s.

All samples were stored at −80°C until they were thawed and centrifuged. All samples were assayed in duplicate using a highly sensitive enzyme immunoassay with a detection range from 0.012μg/dL to 3μg/dL. Samples were included in analyses only if the coefficient of variation (CV) was less than 20%. Mean CV values for included data were 7.18 (SD = 5.15). Average intra- and inter-assay coefficients of variation were < 7%. Raw cortisol samples were regressed on time since arriving to the lab. The distributions of lab cortisol measures were skewed and so were log10 transformed. Samples were excluded from analyses if the child was ill, taking antibiotics at the time of the sample, and/or if the sample was taken within 1 hour of eating. A reactivity score indexing total cortisol output across the lab visit was calculated as the area under the curve with respect to ground using precise time of sample collection (Pruessner et al., Reference Pruessner, Kirschbaum, Meinlschmid and Hellhammer2003); higher values reflect greater neuroendocrine reactivity across the lab visit (Gunnar, Reference Gunnar1992).

Data analysis

All analyses were performed using Mplus software package version 8.4 (Muthén & Muthen, Reference Muthén and Muthen2017). Following preliminary analyses to examine data distributions and bivariate associations to exclude potential demographic covariates, the primary study aims were addressed. Electrode locations Cz and Fz were tested in separate models based on previous work suggesting ERN and child fear relations may show some electrode specificity (Brooker & Buss, Reference Brooker and Buss2014). Considering existing literature suggesting common sex-effects across measures of RSA (Koenig et al., Reference Koenig, Rash, Campbell, Thayer and Kaess2017; Rudd et al., Reference Rudd, Alkon and Yates2017), ERN (Grammer et al., Reference Grammer, Carrasco, Gehring and Morrison2014; Moser et al., Reference Moser, Moran, Kneip, Schroder and Larson2016), and anxious behaviors (Zhou et al., Reference Zhou, Trainer, Vallorani, Fu and Buss2022) in early childhood, sex and age were controlled for in primary analyses.

First, to assess whether coupling was present between ERN and RSA, we tested simple linear associations between these measures. Then, to assess whether the relation between ERN and RSA may be linked to concurrent and longitudinal risk for anxiety problems, we tested context-appropriate fear (concurrent), context-inappropriate fear (concurrent), GAD symptoms (longitudinal), and cortisol reactivity (longitudinal) in separate models as moderators of the linear association reflecting coupling (Hayes, Reference Hayes2022). The rationale for including longitudinal measures as moderators of earlier associations was to test whether children’s early patterns of coupling might be explained as adaptive or maladaptive based on observed outcomes.

Missing data

ERN was missing for 64 participants; RSA was missing for 27 participants; observed fear was missing for 25 (Puppet Show) and 28 (Stranger Approach) participants; anxiety symptom data were missing for 39 participants; cortisol data were missing for 51 participants. An analysis of patterns of missing data suggested that data were missing at random (Little’s MCAR χ ² (119) = 98.752, p = .912), and identified 11 cases missing all variables. Missing data were handled using a Full-Information Maximum Likelihood procedure that uses all available data to produce unbiased parameter estimates, thus allowing one to take advantage of the full sample from which at least some data are available (Enders, Reference Enders2010). Of the original sample (N = 121), the final analytic sample comprised 110 children.

Results

Descriptive statistics, t-tests, and bivariate correlations are reported in Table 2. Younger children showed more context-appropriate (r[109] = −.329, p < .001) and context-inappropriate fear (r[109] = −.234, p = .015). ERN amplitudes were highly correlated across electrodes (r[109] = .710, p < .001). Cortisol output at age 5 was negatively correlated with frontal ERN (r[109] = −.369, p = .034) and positively correlated with context-appropriate fear (r[109] = .351, p = .029) at age 3.

Table 2. Descriptive statistics and bivariate correlations for primary study variables

Note. RSA slope, GAD symptoms, and cortisol output were mean-centered prior to analyses. For child sex, male = 0 and female = 1. *** p < .001, ** p < .01, * p < .05; results presented in italics are t-tests with effect sizes measured using Cohen’s d.

ERN-RSA coupling

The bivariate relation between ERN and RSA at age 3 was significant at both electrode Cz (r[109] = −.241, p = .028) and Fz (r[109] = −.223, p = .050), whereby smaller ERN was associated with more RSA suppression or larger ERN was associated with less RSA suppression (hereafter “negative coupling”). We then tested whether coupling between cortical and parasympathetic activity exists when predicting ERN from RSA, controlling for child age and sex, using multiple linear regression. Results indicated no significant relation between ERN and RSA at either electrode Cz (B = −4.257, SE[B] = 2.213, 95% CI [−8.595, 0.081], p = .054) or Fz (B = −3.156, SE[B] = 1.732, 95% CI [−6.550, 0.238], p = .068).

ERN-RSA coupling and anxiety risk

Context-appropriate fear

Context-appropriate fear at age 3 moderated the relation between RSA and ERN at Cz (B = 16.010, p = .013) but not Fz (B = 1.976, p = .710) (Table 3; Model 1, 2). Probing this effect at high (+1 SD) and low (−1 SD; Aiken & West, Reference Aiken and West1991) levels of context-appropriate fear revealed no significant coupling at high levels of context-appropriate fear (B = 5.249, p = .247). However, at low levels of context-appropriate fear, preschoolers showed more negative coupling (i.e., smaller ERN, more RSA suppression or larger ERN, less RSA suppression; B = −8.803, p = .001) (Fig. 1).

Figure 1. Age 3 context-inappropriate fear as a moderator of ERN-RSA coupling.

Table 3. Age 3 anxiety risk as a moderator of ERN − RSA coupling

Note. CI = confidence interval; LL = lower limit; UL = upper limit. Dependent Variable: Error−Related Negativity (ERN). *p < .05, **p < .01, ***p < .001.

Context-inappropriate fear

Context-inappropriate fear at age 3 moderated the relation between RSA and ERN at Fz (B = −5.815, p = .031) but not Cz (B = −3.222, p = .373) (Table 3; Model 3, 4). Probing the interaction in the same manner as previously noted revealed that, when context-inappropriate fear was high, coupling was more negative (i.e., smaller ERN, more RSA suppression or larger ERN, less RSA suppression; B = −9.004, p = .004). There was no relation between RSA and ERN at low levels of context-inappropriate fear (B = 1.288, p = .627) (Fig. 2).

Figure 2. Age 3 context-appropriate fear as a moderator of ERN-RSA coupling.

GAD symptoms

Anxiety symptoms at age 5 significantly moderated the association between RSA predicting ERN at Fz (B = −7.710, p = .002) but not Cz (B = −5.459, p = .139) (Table 4; Model 1, 2). When age 5 GAD symptoms were high, preschoolers showed more negative coupling (i.e., smaller ERN, more RSA suppression or larger ERN, less RSA suppression; B = −11.290, p < .001) at age 3. However, when GAD symptoms were low, preschoolers showed more positive coupling (i.e., smaller ERN, less RSA suppression or larger ERN, more RSA suppression; B = 14.516, p = .014) (Fig. 3).

Figure 3. Age 5 GAD symptoms as a moderator of age 3 ERN-RSA coupling.

Table 4. Age 5 anxiety risk as a moderator of ERN − RSA coupling

Note. CI = confidence interval; LL = lower limit; UL = upper limit. Dependent Variable: Error−Related Negativity (ERN). *p < .05, **p < .01, ***p < .001.

Neuroendocrine reactivity

Neuroendocrine reactivity at age 5 significantly moderated the association between RSA and ERN at Fz (B = 1.357, p < .001) and Cz (B = 1.811, p < .001) (Table 4; Model 3, 4). When cortisol reactivity was high, preschoolers showed positive coupling (i.e., smaller ERN, less RSA suppression or larger ERN, more RSA suppression) at both Fz (B = 11.855, p = .003) and Cz (B = 16.744, p = .010). When cortisol reactivity was low, preschoolers showed more negative coupling (i.e., smaller ERN, more RSA suppression or larger ERN, less RSA suppression) at both Fz (B = −22.171, p < .001) and Cz (B = −28.643, p < .001) (Fig. 4).

Figure 4. Age 5 cortisol as a moderator of age 3 ERN-RSA coupling.