Context of childhood family SES

Childhood family SES is typically assessed through objective indicators such as parent educational attainment, occupational prestige, and household income, as well as subjective evaluations of social status by parents or children and adolescents themselves [4, 5]. Childhood family SES is related to a myriad of life experiences [2, 4, 6]. Lower family SES is linked to increased food scarcity, reduced access to food with high nutritious quality, and more limited health care [7, 8]. Children from lower SES backgrounds are more likely to live in lower-quality residential spaces, with elevated risk for exposure to neurotoxins like lead and indoor air pollutants, residential crowding, and poor housing maintenance [8–10]. Children from lower SES backgrounds are also more likely to experience adverse neighborhood conditions, such as excessive noise and exposure to hazardous wastes [11, 12]. Furthermore, children from lower SES backgrounds are less likely to have access to well-maintained child-care facilities, schools, and recreational amenities [10, 13]. Nutrition, health resources, and the physical quality of settings where children live and learn have been linked to child outcomes directly, as well as indirectly through their associations with the well-being of family members, teachers, and peers [10, 14, 15].

In addition to the material and physical characteristics of children’s daily environments, specific family SES indicators have also been associated with the psychosocial dynamics in children’s lives, especially through parenting behaviors [16–18]. Low parental income and increased parental financial stress have been linked to disadvantageous child outcomes through increased parental distress and disrupted parenting [19]. In addition, parent educational attainment has been linked to parental beliefs and expectations about child development and what children would need to be successful, which in turn predict the extent to which parents seek cognitively stimulating activities inside and outside of the home environment [20]. It has been proposed that childhood SES may relate to brain development through the unique, cumulative, and interactive effects of these psychosocial factors, especially via changes in the neurobiology of stress regulation and via neurodevelopmental changes afforded by cognitive stimulation in children’s daily lives [2, 8, 21].

Neurodevelopment in the context of childhood family SES

There is converging evidence that links family SES to neurodevelopment in childhood and adolescence [1, 22, 23]. SES-related differences have been reported for the development of broad brain structures, including brain volume, gray matter density, total surface area, mean cortical thickness, and white matter volume in children and adolescents [24–27]. SES-related differences have also been observed in the structure and functioning within specific brain regions supporting cognitive control, memory, language, and reading acquisition [23, 28–30], as well as the functional connectivity within and between brain networks, including the frontoparietal, sensorimotor, and the default mode networks [31–33].

Childhood family SES has also been linked to the electrophysiology of the developing brain [34, 35]. Lower parental educational attainment and lower family income-to-needs ratio have been associated with alterations in neural responses in children, such as larger neural responses to sounds appearing in a distracting story [36, 37], smaller neural responses to novel stimuli [38], and attenuated neural responses to targets in tasks of cognitive control [39, 40]. Together, these findings provide corroborating evidence that childhood family SES is associated with neurodevelopment in childhood.

A few longitudinal neuroimaging studies raised the possibility that the links between childhood family SES and brain structures and functions may persist into adulthood. For example, lower family income-to-needs ratio in childhood was associated with less white matter organization in the frontolimbic and white matter association tracts [41] and altered neural responses to stimuli with negative emotional valence [42]. Furthermore, lower average family income-to-needs ratio assessed across multiple time points from childhood to adolescence was linked to less cortical thickness and surface area in brain regions involved in a wide array of functions in adulthood [43]. However, it remains unclear whether other indicators of childhood family SES, especially parent educational attainment and subjective social status, which have distinct links with developmental outcomes [5, 20], relate to brain functions in adulthood.

Capturing childhood family SES retrospectively

Undoubtedly, the most accurate way of investigating the links between childhood family SES and adult brain functioning is a longitudinal design that follows individuals from childhood through adulthood. Although some researchers have successfully done so [41–43], longitudinal studies spanning a decade, or more, are rare due to the extensive resources required. In the absence of longitudinal studies, and despite its limitations, an alternative approach is to assess childhood family SES by retrospectively asking adults about their childhood experiences [44, 45].

Family SES is most commonly assessed in childhood and adolescence by parent reports of parent educational attainment, occupation, and family income [4]. However, it may be harder for adults to accurately report on their childhood family income or wealth. To address this issue, similar to what is done by developmental researchers when access to parent reports is limited and youth reports of household income and wealth may not be accurate [4], childhood parent educational attainment may be used as an objective proxy for childhood family SES.

Another approach for assessing family SES is asking individuals about their subjective social standing in a particular sociohistorical context [4, 46]. Subjective social status is most commonly measured by having respondents rank themselves on a ladder of status, which places people who are the worst off, those who have the least money, the least education, and the worst jobs or no jobs at the bottom [47]. Subjective SES generally has weak to moderate associations with objective SES indicators [48] and may capture not only economic circumstances, but also relative judgments of social status based on power, control, social influence, and standing in a community and society [4, 49].

It has been argued that lower subjective social status may be linked to poorer physical and mental health outcomes, not only because it is related to deprivation from economic resources, but also because it may be related to recurring negative emotions that are elicited by upward social comparisons and relative deprivation and associated alterations in stress regulation systems [46, 49]. Several studies linked concurrent subjective family SES to physical and mental health outcomes and school achievement in adolescence [5, 50, 51]. In adults, both retrospective reports of childhood and adolescence family SES and concurrent subjective social status were associated with neural responses in situations with overt socioemotional and interpersonal salience, including when adults were asked to make social comparisons, were presented with stimuli with negative emotional valence such as threatening faces, and were given negative feedback [44, 52–54]. It remains to be investigated whether the links between subjective family SES in childhood and brain functioning in adulthood would extend to fundamental brain functions supporting perception and cognition in daily life, even in situations without overt socioemotional valence.

Present study

The main goal of the present study was to investigate the links between objective and subjective indicators of childhood family SES and fundamental brain functions supporting perception and cognition in young adults. We used the highest parent educational attainment in childhood as a proxy for objective childhood family SES because we reasoned that parent educational attainment may be more robust to inaccuracies in recall compared to other objective indicators such as family income or wealth in childhood. In addition, adults reported on their subjective evaluations of their childhood family SES. To index brain functions supporting perception and cognition, we recorded ERPs from young adults (N = 86) during a passive auditory oddball and an active visual oddball task, which were optimized to capture two well-established and widely used event-related potentials (ERP) components—the Mismatch Negativity (MMN) and P3b respectively [55].

MMN is a neural index of change detection that occurs after an echoic memory trace is formed and is elicited once deviance in the auditory context is detected, even in the absence of attention [56, 57]. In neurotypical adults, MMN is observed as a greater negative deflection for rare versus frequent stimuli over the frontocentral electrode sites between 125 and 250 ms post-stimulus onset [55, 58]. Previous research on the links between childhood family SES and MMN is scarce. However, a limited number of studies demonstrated that MMN could be altered with training among neurotypical children and adults [59, 60]. These findings suggest that MMN can be modified by experiences. Given that childhood family SES is related to childhood auditory experiences at and outside of home [10, 20] and that supportive experiences, such as trainings have been linked to larger MMN responses [59, 60], we reasoned that higher objective childhood family SES would be associated with larger (more negative) MMN responses.

P3b, which is a subcomponent of the P3 ERP component, is considered to index brain functions that support cognitive control and are involved in the detection of rare targets, allocation of attentional resources to the target, stimulus evaluation, and updating neural representations associated with memory operations [61, 62]. In neurotypical adults, P3b is observed as a robust positive voltage deflection for rare targets compared to frequent stimuli over parietal electrode sites between 300–600 ms [55]. Specific indicators of childhood family SES, in particular, higher maternal education [40] and higher family income [63] have been associated with larger P3 amplitudes in children. Building on these findings, we expected higher objective childhood family SES to be related to larger (more positive) P3b responses. As subjective evaluations of childhood family SES may capture SES-related life experiences above and beyond those related to parent educational attainment, such as stress resulting from limited financial resources or social comparisons [4, 49], we also hypothesized that higher childhood subjective family SES would be associated with larger neural responses for MMN and P3b, even after taking childhood parent educational attainment into account.

Participants

Participants were recruited via flyers posted on a university campus and at community centers in the Western United States. The age range for participant recruitment was set to be 18 to 30 years to make our sample comparable in age with the sample in the ERP CORE study from which we adopted our tasks, data processing pipeline, and ERP scoring methods [55]. The inclusion criteria for recruitment were as follows: normal or corrected-to-normal vision, no hearing issues, no history of brain injuries or neurological disorders, and no current use of medications that could alter brain functioning. The study protocol was approved by the Institutional Review Board of the University of California Merced. All individuals gave written consent to participate in the study. All participants received cash for their participation. The data collection for the study started on February 23^rd, 2023, and ended on June 16^th, 2023.

An a priori power analysis was conducted in G*Power to determine an adequate sample size before data collection had started. With two predictors (objective and subjective childhood family SES), an alpha level of .05, an effect size of .15, and a minimum power established at .80, we determined that a minimum sample size of 68 would be needed per ERP component. To account for a 5% data loss rate in adult studies, we aimed for a minimum sample size of 72.

Participants were excluded from the analytical sample if they completed the sociodemographic questionnaire incorrectly (i.e., reported on the educational attainment of adults with whom they currently live with, such as roommates, instead of their parents/legal guardians, n = 5) or had poor ERP data quality for both ERP tasks (n = 1). Due to experimenter error, participant performance, or ERP data quality, we had usable data for only one of the two ERP tasks for some participants. To maximize our sample size, we included all participants who had at least one usable ERP task in the present study. The participants who had data from only one ERP task (n = 24) did not differ from participants who had data from both tasks (n = 62) in terms of age, gender, education level, highest childhood parent educational attainment, and subjective childhood family SES (all ps > .17). Table A in S1 Text includes the descriptive statistics for these sociodemographic variables for each group. Table B in S1 Text provides a summary of the independent samples t-tests comparing the sociodemographic characteristics of the participants who had data from only one ERP task to participants who had data from both tasks.

The final analytical sample consisted of 86 adults between the ages of 18 and 30 years (M = 21.85, SD = 2.74). When asked to report their gender, 55% of the participants chose female, 45% chose male, and no other genders were reported. Participants reported their race/ethnicity as follows: 57% Hispanic or Latino/x, 16.3% Asian/Asian American, 16.3% selected multiple categories, 9.3% White/European American, and 1.1% not reported. All participants reported having completed at least high school, and 12% had an associate degree, 21% had a bachelor’s degree, and 3.5% had a master’s degree (see Fig A in S1 Text for participant education frequencies). The childhood parent educational attainment levels of the participants were as follows: 28% less than a high school diploma, 20% high school diploma, 14% some college but no degree completed, 6% associate degree, 14% bachelor’s degree, 12% master’s degree, and 6% professional or doctorate degree.

Childhood family SES

The SES questionnaires were administered after the completion of the ERP tasks to avoid prompting the participants to think about their childhood SES before or during the ERP tasks. To obtain retrospective reports of objective childhood SES backgrounds and subjective social status, participants were asked to think back to when they were 10 years old to answer the questions.

Oddball tasks

The passive auditory oddball and the active visual oddball tasks were adapted from ERP CORE, a set of paradigms designed to optimally capture widely studied ERP components [55]. The tasks were presented using the Presentation (Neurobehavioral Systems) experiment software. The visual stimuli were presented on a Dell LCD monitor with a resolution of 1280 x 1024, a refresh rate of 60 Hz, and a viewing distance of 95 cm. Participants were offered to take a break in between tasks.

Electroencephalogram (EEG) recording

EEG was recorded with Brain Products actiCHamp Plus, using BrainVision Recorder Version 1.25.0101, and collected with an actiCAP slim active electrode system, mounted on elastic snap caps [65]. The ground electrode was placed at FPz. From the 32-channel electrode bundle, 2 electrodes were repurposed and placed on the mastoid bones behind the left and right ears for offline re-referencing. Three additional electrodes were repurposed to record electrooculogram (EOG). The vertical EOG (VEOG) electrode was placed below the right eye and the horizontal EOG (HEOG) electrodes were placed lateral to each eye’s external canthus. The remaining 27 electrodes were used as scalp electrodes, mounted in accordance with the international 10/20 system (see Fig 1 for the electrode configuration). EEG was sampled at 500 Hz and referenced to Cz. The stimulus presentation delays in the monitor and the headphones were assessed with StimTrak [65]. Both the visual and auditory stimuli were delayed by approximately 20 ms.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Configuration of the scalp electrodes.

Channels shaded in orange are included in the frontocentral cluster used for MMN analyses. Channels shaded in blue are included in the posterior cluster used for P3b analyses.

https://doi.org/10.1371/journal.pone.0307406.g001

EEG signal processing and averaging

EEG signal processing and averaging were conducted in MATLAB using customized EEGLAB [66] and ERPLAB [67] scripts, as well as scripts adapted from ERP CORE [55] and ICLabel [68]. The data processing and analysis modifications were made to have a pipeline that can be easily adapted for use with a wider age group and was previously used with young children [40, 69]. All scripts can be found on Open Science Framework (OSF): https://osf.io/43h75/.

EEG data were re-referenced to the arithmetic average of the left and right mastoids. To detect ocular artifacts more easily, bipolar eye channels were created as follows: the vertical bipolar eye channel was computed by subtracting FP2 from the right VEOG, and the horizontal bipolar eye channel was computed by subtracting the left HEOG from the right HEOG. EEG data were band-pass filtered with the EEGLAB default finite impulse response (FIR) filter between 0.1 to 40 Hz (-6 dB cutoff frequency). This filter was chosen to be consistent with previous and ongoing research with children [40, 69].

EEG artifacts were removed using a combination of independent component analysis (ICA) and the rejection of trials with residual artifacts. To prepare data for ICA, we removed recording periods with no event codes (defined as no event codes for 6000 ms or longer, with a buffer of 3000 ms before and after any event codes). To detect bad channels that would cause unnecessary data loss, we applied the ERPLAB moving window peak-to-peak threshold algorithm across a 500 ms window, moving in 50 ms increments, with a +/- 300 μV threshold to all the scalp channels. Except for FP1 and FP2, any scalp channel that resulted in excessive data loss (defined as more than 10% overall data loss and ≥ 3.29 SD data loss compared to other scalp channels) was excluded from pre-ICA artifact rejection. Then, data segments with extreme artifacts were rejected from the continuous data with the ERPLAB moving window peak-to-peak threshold algorithm (across a 500 ms window, moving at 50 ms increments, with a +/- 300 μV threshold) applied to all remaining scalp channels, excluding FP1 and FP2.

ICA was applied to all channels, except the bipolar eye channels. The computed ICA weights were applied to the preprocessed data files (i.e., before pre-ICA cleaning). Artifact correction was conducted by removing components that were selected as “eye” components if ICLabel classified them as “eye” with at least 80% confidence and as “brain” with less than 5% confidence. These criteria were selected based on a systematic comparison of different confidence thresholds for young adults.

Before epoching, to account for the equipment presentation delays described above, the stimulus event codes were shifted in time for 20 ms. Consistent with the ERP CORE parameters, ICA-corrected data were epoched -200 to 800 ms relative to stimulus onset, and the -200 to 0 ms pre-stimulus onset was used for baseline correction. To remove remaining eye artifacts, ICA-corrected bipolar HEOG and VEOG channels were computed. To avoid unnecessary data loss due to a bad channel, first, a simple voltage threshold algorithm with a +/- 200 μV threshold was applied to all the scalp channels of analytic interest. The channel exclusion criteria were the loss of more than 10% of trials and a data loss rate of at least 3.29 SD above the other channels. In the auditory oddball task, 4 participants had 1 “bad channel” and these scalp channels were excluded from further artifact rejection and analyses. In the visual oddball task, no channels of interest were detected to be excluded.

The final artifact rejection included the following steps: To detect artifacts in scalp channels of interest, we applied a simple voltage threshold algorithm with a +/- 200 μV threshold and a moving peak-to-peak window algorithm moving at 100 ms increments, with a 125 μV threshold to each epoch. To remove residual blinks and saccades, a moving peak-to-peak window algorithm was applied to the ICA-corrected VEOG (across a 200 ms window, moving at 50 ms increments, with a 150 μV threshold), and a step-like algorithm was applied to the ICA-corrected HEOG (across a 100 ms window, moving at 10 ms increments, with a 64 μV threshold). The following steps were added for the visual oddball task: To reject trials where a blink occurred too close to when the visual stimuli were presented, we applied a moving peak-to-peak window algorithm to the uncorrected VEOG (between -25 to 225 ms with respect to stimulus onset, moving at 10 ms increments, with a 150 μV threshold). To reject trials where a saccade occurred too close to when the visual stimuli were presented, we applied a step-like algorithm to the uncorrected HEOG (between -50 to 250 ms with respect to stimulus onset, moving at 10 ms increments, with a 32 μV threshold). Consistent with ERP CORE, in the visual oddball task, trials with incorrect behavioral responses and trials with responses faster than 200 ms and slower than 1000 ms were also excluded from analyses.

After artifact rejection, individual ERP plots were visually inspected for data quality. In addition, analytic standardized measurement error (aSME) was computed for observed waveforms as an objective metric of data quality [70] and used to check for data quality in channels of interest across participants. In the auditory oddball, 1 participant was excluded for being an aSME outlier and not showing discernable auditory evoked potentials. In the visual oddball, 2 participants were excluded for being aSME outliers and not showing discernable visual evoked potentials.

ERP amplitude and data quality scores

In ERP CORE [55], single channels were selected for each component of interest (e.g., FCz for MMN and Pz for P3b). However, in a follow-up study, analyses conducted with the ERP CORE data, as well as Monte Carlo simulations, revealed that the ERPs obtained from multi-site clusters had as good or better data quality compared to ERPs obtained from a single channel of interest for all ERP components, including MMN and P3b [71]. Therefore, we inspected the grand average plots and selected multiple channels in which ERP components of interest were similarly observable. We created multi-site channel clusters for ERP analyses as follows: Fz, F3, F4, FC1, FC2, C3, C4 for MMN; Pz, P3, and P4 for P3b. Figs F and H in S1 Text show the grand average plots for the MMN and P3b channel clusters respectively, and Figs G and I in S1 Text Fig show the grand average plots for MMN and P3b over representative scalp channels respectively.

ERP time-window mean amplitude scores (hereafter referred to as mean amplitude) were extracted for MMN between 125 and 225 ms in the auditory oddball task, and for P3b between 300 and 600 ms in the visual oddball task, consistent with the time windows recommended in ERP CORE [55]. Given that difference waves eliminate concurrent neural processes across conditions [72], we computed difference waves (rare minus frequent) to isolate the experimental effects in the auditory and visual oddball tasks. To provide an objective data quality metric for the MMN and P3b difference waves, bootstrapped standardized measurement error (bSME) scores [70] were computed.

Behavioral performance

A discriminability index (d′) was calculated as a measure of behavioral performance in the active visual oddball task as follows: d′ = Z(Correct/Hit)—Z (Incorrect/False Alarm). Higher values of d′ indicate a greater ability to distinguish signals from noise, corresponding to better overall task performance [73].

Data analytic plan

Hierarchical regression analyses were conducted separately for MMN and P3b mean amplitudes. In each regression analysis, the highest parent education level in childhood was entered in Step 1. The subjective childhood family SES was added to the model in Step 2. All artifact-free trials were used in MMN analyses since there were no response demands. Only correct trials were included in P3b analyses. All analyses were performed in Jamovi (Version 2.3.28). The data file used in the analyses and the accompanying data dictionary are available on OSF: https://osf.io/43h75/.