Prefrontal activation during executive tasks emerges over early childhood: evidence from functional Near Infrared Spectroscopy

Abstract

Functional Near Infrared Spectroscopy (fNIRS) is a brain imaging technique that is well-suited for use in young children, making it particularly useful for investigating the neural bases of the development of executive functions. In the present study, children (ages 4–10) underwent fNIRS while completing response inhibition and working memory tasks. While both tasks were associated with increases in oxyhemoglobin and decreases in deoxyhemoglobin, we found that strength of activation increased with age and with improvements in task performance. These findings support the relation between emerging executive functions and maturation of the prefrontal cortex.

Introduction

Executive functions are a set of cognitive abilities that promote goal attainment, including working memory, response inhibition and set shifting. Deficits in executive functions are associated with a range of childhood disorders including attention deficit hyperactivity disorder and autism spectrum disorders (Barkley, 2006; Rosenthal et al., 2013). As such, the typical developmental trajectory for executive task performance and associated neural activity plays an important role in understanding executive function in childhood disorders. Executive functioning skills begin to develop in infancy (Diamond, 1990), and accuracy in executive tasks improves across childhood and into adulthood, although this varies by task (Huizinga, Dolan, & van der Molen, 2006). This variance is due both to underlying differences in the developmental trajectory of specific executive abilities and also to differences in task demands (e.g., degree of difficulty within task). It can therefore be difficult to disentangle differences in executive function that are due to developmental change from those due to task effects (Koppenol-Gonzalez, Bouwmeester, & Vermunt, 2012). In addition, less is known about the relation between brain development and specific executive functions, particularly in early childhood, which limits understanding of how deficits in executive functions emerge in neurodevelopmental disorders. Here, we use functional Near Infrared Spectroscopy (fNIRS) to measure oxygenation changes associated with two executive functions (response inhibition and working memory) via two commonly studied paradigms (the Go-No/Go and N-back tasks, respectively).

While the neural substrates for both working memory and response inhibition have been studied in adults and children with functional Magnetic Resonance Imaging (fMRI), fMRI study participants are limited to children who are old enough to stay still within an MRI machine. This has limited our ability to understand the neural substrates of age-related changes in executive function in preschool and younger school-age children. fNIRS, which uses differential light absorption patterns of oxy- and deoxyhemoglobin to measure task-related changes in cerebral hemodynamics, is an imaging method that is generally tolerated by even very young children and is less sensitive to subject motion than other neuroimaging methods. fNIRS has been successfully used to measure neural activation during executive tasks in adults (Heilbronner & Münte, 2013; Sato et al., 2013), with results generally reflecting those found with fMRI. Specifically, response inhibition is associated with the inferior frontal gyrus (IFG) and anterior cingulate (AC) in adults (Aron, Robbins, & Poldrack, 2014) and the network between the IFG and supplementary motor cortex (Zandbelt, Bloemendaal, Hoogendam, Kahn, & Vink, 2012), while working memory is associated with more diffuse activation, including activity in the frontal and parietal lobes, anterior cingulate, and cerebellum (Dima, Jogia, & Frangou, 2014).

NIRS has also been used in children across a range of executive tasks, including response inhibition (Inoue et al., 2012; Mehnert et al., 2013), working memory (Buss, Fox, Boas, & Spencer, 2014; Perlman, Huppert, & Luna, 2016; Tsujii, Yamamoto, Masuda, & Watanabe, 2009; Tsujimoto, Yamamoto, Kawaguchi, Koizumi, & Sawaguchi, 2004), fluency (Kawakubo et al., 2011; Tando et al., 2014), and set shifting (Moriguchi & Hiraki, 2009, 2011). Studies have focused on both preschoolers (Buss et al., 2014; Mehnert et al., 2013; Moriguchi & Hiraki, 2009, 2011; Perlman et al., 2016) and school-age children (Inoue et al., 2012; Kawakubo et al., 2011; Tando et al., 2014; Tsujii et al., 2009; Tsujimoto et al., 2004), and almost all have focused on activation in the frontal lobe (except for Buss et al., 2014; Mehnert et al., 2013). Results have been inconsistent in this literature, with some studies finding increasing task-related activation with age (Buss et al., 2014; Kawakubo et al., 2011; Moriguchi & Hiraki, 2011; Perlman et al., 2016) while others do not (Inoue et al., 2012; Mehnert et al., 2013). This may be related to variability in study design, including treatment of age as a categorical variable, use of either longitudinal or cross sectional design (e.g., studies comparing “children” vs. “adults”), and to small age ranges within studies. Investigating changes in fNIRS across age measured continuously, especially within a broad age-range of preschool and school-age children has the potential to clarify age-related changes in the fNIRS signal related to executive function.

In addition to inconsistent results that are potentially driven by restricted range of ages and modeling age as a categorical rather than dimensional variable, many fNIRS studies lack control conditions. Instead, cerebral oxygenation during the active condition or task is compared to a resting baseline. Comparing a task to a resting baseline could result in oxygenation differences that are associated with completing any challenging task, rather than due to the construct of interest (Aslin, 2012). Therefore, a control task is used to control for any other processes (e.g., attention, motor requirements, visual stimulation) that are not of interest. Therefore, while fNIRS has been used in combination with executive tasks in children, treatment of age as a categorical variable and lack of an fNIRS control condition leaves a significant gap in the literature on development of the neural processes underlying executive functions in children.

In the present study, we assessed two executive functions—working memory and response inhibition—within the same fNIRS session, allowing us to investigate both shared and unique patterns of activation. We investigated the role of participant age as a continuous variable across a wide range (4–10 years) of the understudied preschool to young school age period. Some children also completed these tasks at a separate visit approximately two years later; their data from both visits was included with age modelled as both a within and between subjects factor, thus increasing power to detect potential effects of subject age. In addition, we maximized analyses in both response inhibition and working memory tasks by comparing changes in oxy- and deoxyhemoglobin between the two executive function tasks (No-Go, 1-back) and their respective control conditions (Go, 0-back), rather than comparing these conditions to a resting baseline. Finally, we separately analyzed contributions of both age and task performance- which can have related and potentially confounding effects- on brain activity. In this way, we were able to investigate the relation between changes in oxy- and deoxyhemoglobin and both age and ability across the frontal lobe during executive tasks in children.

Materials and Methods

This study was approved by an Institutional Review Board at the National Institutes of Health under NIMH protocol 2013-M-0007. Legal guardians of all participants provided informed consent prior to study participation. Participants provided informed verbal or written assent, depending on age.

Participants

Participants were 26 right-handed 4–10 year old children. Exclusion criterion were (1) IQ lower than 80 as determeined by the Differential Ability Scales, 2^nd edition (Elliott, 2007) or the Wechsler Preschool and Primary Scales of Intelligence, 4^th edition (Wechsler, 2012), and (2) any neurological disorder or DSM-IV-TR Axis I disorder, as determined by parent interview in combination with the Child Behavior Checklist (Achenbach et al., 2001), and medical history that contraindicated fNIRS (e.g. head injury).

Participants completed both an N-back working memory task and a Go/No-Go response inhibition task (see Table 1). Data from five children were excluded due to excessive motion (n=2) and equipment failure (n=3). While the study design is cross-sectional in nature, 10 participants returned for a second visit approximately two years later as part of a larger longitudinal study, during which they completed the same NIRS tasks. Here, we include data from both timepoints to aid in analyses of effects of age, but do not evaluate longitudinal hypotheses.

Table 1.

Participant characteristics and behavioral performance for the Go/No-Go and N-back tasks. Continuous variables are presented with means, standard deviations in parentheses, and range in brackets.

Demographics	Go/No-Go	N-back
N	21	15a
Number of repeats	10	9
Total NIRS data sets	31	24
Sex (m:f)	17:14	13:11
Age (years)	6.6 (1.43) [4.07–10.41]	6.73 (1.5) [4.07–10.4
4–5 years	n=11	n=8
6–7 years	n=13	n=10
8–10 years	n=7	n=6
Time between repeat visits (years)	2.06 (.59) [1.18–2.76]	1.98 (.57) [1.18–2.69]
FSIQ	117.48 (12.43) [94–143]	117.4 (13.48) [94–143]
Performance	Go/No-Go	N-back
Reaction time (ms)	459.1 (96.0) [291.9–693.0]	675.0 (102.1) [505.3–849.8]
Reaction time variability (ms)	122.4 (39.2) [53.1–217.5]	160.5 (35.5) [98.1–222.5]
dprime	2.61(.95) [.46–4.65]	3.72(1.13) [1.2–4.65]

^aWe chose to alter the design of the N-back task after running 6 subjects, such that only 15 participants completed the modified N-back task at their first visit and 9 at their second visit.

Task

Go/No-Go:

A series of red and green spaceships appeared sequentially on a laptop screen, and children were instructed to push a button on a peripherally attached button box when they saw a green spaceship, and not when they saw a red spaceship (see Supplmental Figure 1). Participants completed a practice block with both red and green spaceships, which was repeated until the child attained a score of 70% correct or above. Each child was then presented with four Go blocks alternating with four No-Go blocks with 10 seconds of rest between each block. The Go blocks (control condition) included green spaceships only to account for basic visual processing and motor response. The No-Go blocks included both green and red spaceships at a ratio of 10:4. For all blocks, 10 spaceships appeared for 500 milliseconds with an interstimulus interval of 1000 ms, for a total of 15 seconds per block, with 10 seconds of rest between blocks.

N-back:

For the 1-back task, participants viewed black shapes (e.g., star, circle) presented on a white screen and were told to push the button if they saw the same shape twice in a row (see Supplemental Figure 1). Stimuli were presented for 1500 ms each with an interstimulus interval of 500 ms. Participants completed practice blocks until the child attained a score of 70% correct or above. Then, the participant completed four blocks of 16 seconds each with a target/nontarget ratio of 3:5. For the 0-back blocks, participants were told to press the button whenever they saw a circle and were shown a circle on the screen as an example. Participants completed four 0-back blocks, each of which was 16 seconds long and had a target/nontarget ratio of 3:5. There was a 10 second rest period between all blocks. Stimuli were presented on a 13” laptop screen placed on a child-sized table in front of the seated participants.

fNIRS

Optical imaging was performed with a continuous wave fNIRS device (fNIRS Devices LLC, Potomac, USA).This device uses a silicone headband that contains four light sources and 10 detectors. Near infrared light at two frequencies (730 and 850 nm, 25 mW), was emitted at each source and backscattred light was received at the detectors. The source-detector distance was fixed at 2.5 cm, while the intersource distance was fixed at 3.5 cm. The headband was placed across the participant’s forehead prior to completing computer tasks. The band was secured at the back of the head with adjustable fabric ties followed by application of athletic wrap to keep the band secure and block ambient light from exterior sources. Four fiber optic cables (two sides by two rows of detectors) extended from the back of the band to the fNIR control box, which was positioned to the side of each participant. The band was positioned on each child’s forehead such that the sources and detectors were centered horizontally at FPZ based on the international 10–20 coordinate system. The bottom middle detector was carefully placed at location Fpz, but given the fixed interoptode distance, all other locations are approximate and depend on the size and shape of each child’s head. Therefore, the source-detector pairs approximately cover Brodmann areas 10 and 46 (i.e., right and left ventrolateral prefrontal cortex (VLPFC) and dorsolateral prefrontal cortex (DLPFC)).

Once the band was placed, an experimentor seated next to the child explained the task and provided appropriate feedback on practice trials. A second experimentor began data acquisition while the first experimentor initiated the Go/No-Go task using using EPrime (v. 2.0). During the task, EPrime recorded button presses and reaction time, and sent analog pulses to the fNIRS control box to mark the onset of each block, stimulus, and button press.

fNIRS data were collected at a rate of 2 Hz through the COBI Studio acquisition interface, which shows real-time data for both oxy- and deoxyhemoglobin and allows the user to mark events (e.g., subject motion) in real time. Prior to data analysis, NIR light intensities at both 730 and 850nm wavelengths were converted to changes in oxy- and deoxyhemoglobin using fNIRSOFT software (H Ayaz, 2010). In short, changes in chromophore concentration from baseline were calculated using the modified Beer-Lambert Law (Baker et al., 2014). We then applied a low pass filter (FIR 0.1 Hz) to the signal to remove frequencies related to heartbeat and respiration and removed blocks containing motion artifact (i.e., with unidirectional changes in oxy- and deoxyhemoglobin or visible sharp signal spikes). Using the SMAR filter (H. Ayaz, Izzetoglu, Shewokis, & Onaral, 2010), we removed channels that were consistently saturated (i.e., had a high coefficient of variance), which can be caused by motion artifact or poor skin-sensor coupling. As a result, we report data from the two medial emitters and their corresponding detectors, resulting in 8 source detector pairs, as the lateral source-detector pairs frequently were positioned over the hair for smaller children, which led to poorer skin contact in younger and more active children.

Analyses

Given our sample size, all analyses of NIRS data were completed after removing outliers for measurements of both oxyhemoglobin (37 out of 897 measurements) and deoxyhemoglobin (39 out of 897 measurements) in order to reduce the potential for significant effects driven by isolated cases. We defined outliers as being greater than 2 standard deviations from the group mean for either oxy- or deoxyhemoglobin values. In order to normalize the effect of individual differences in optical path length, changes from baseline were converted to z-scores relative to each subject’s measurements (Moriguchi & Hiraki, 2013).

Behavioral analyses for the Go/No-Go task included computing average reaction time (RT) and reaction time variability (RTV, the standard deviation of the reaction time variable for each participant) for correct responses. Trials were excluded from each subject’s data if reaction times were below 100 ms or above 1000 ms. We then calculated omission error rate (OER;not pushing when seeing a green spaceship) and commission error rate (CER; pushing when seeing a red spaceship). We determined overall accuracy using d’prime, where d’prime represents accuracy on this task while accounting for different response styles (Macmillan & Creelman, 2004):

$$\mathrm{d}’\mathrm{prime}\left({\mathrm{p}}_{\mathrm{i}}\right)=\mathrm{z}\left(\mathrm{omission}\mathrm{error}{\mathrm{rate}}_{\mathrm{i}}\right)-\mathrm{z}\left(\mathrm{commission}\mathrm{error}{\mathrm{rate}}_{\mathrm{i}}\right)$$

where z is the z score for the i^th participant.

We analyzed data in R using the lme4 package for linear mixed effects analyses (http://CRAN.R-project.org/package=lme4, R Core R Core Team, 2014). All continuous, non-normalized variables were assessed for fit to a normal distribution prior to data analysis via the Shapiro Wilk test. All variables were associated with a non-significant finding. (For age, W=.97, p=.52; For Go/No-Go RTV W=.96, p=.28; RT W=.96, p=.23. For N-back, RT W=.96, p=.40; RTV W=.96, p=.45).

We used a multilevel linear random effect approach to model effects of individual characteristics (i.e., age), condition (e.g,Go versus No-Go), and source-detector location (i.e., medial vs. lateral, ventral vs. dorsal, left vs. right) on hemodynamic change, and adjusted for main effects prior to investigating interactions between terms.

Results

Behavioral Performance

Means, standard deviations, and ranges for behavioral measures for all participants at all study visits are reported in Table 1.

Go/No-Go.

For the No-Go task, as expected, RT and RTV were correlated with age, with increasing age associated with quicker (β=−53.8, SE=7.1, t(28)=−7.54, p<.001) and more consistent responses (β=−17.3, SE=3.9, t(23.8)=−4.49, p<.001). Neither omission error rates, comission error rates nor d’prime were correlated with age (β=.02, SE=.01, t(29)=1.7, p=.1; β=.0009, SE=.02, t(26.7)=.05, p=.96; β=.15, SE=.12, t(29)=1.3, p=.2).

N-back.

For the N-back task, neither RT nor RTV varied with age (β=−13.8,SE=13.2, t(19.5)=−1.0, p=.3; β=−2.9, SE=4.9, t(22)= −.6, p=.6). However, omission error rates were significantly correlated with age (β=−.08, SE=.02, t(20.6)=−3.3, p<.005) and the correlation between comission rates and age approached significance (β=−.006, SE=.003,t(22)=−1.99, p=.06), with increasing age associated with fewer of both types of errors. For the Nback task, d’prime was also associated with age (β=.37, SE=.13,t(29)=2.87, p<.01).

FNIRS results

Go/No-Go.

For the No-Go task, the general linear model showed an interaction between condition (i.e., No-Go versus Go) and side (left, right) on changes in oxygenated hemoglobin (β=.48, SE=.18, t=2.73, p<.01, see Figure 1). Specifically, oxyhemoglobin increased more from the control condition to the No-Go condition at right hemisphere channels versus left hemisphere channels, indicating right lateralization of activation. There were no interactions between condition and row (β=.03, SE=.18, t=.14, p=.89) or condition and medial/lateral location (β=−.025, SE=.18, t=−.14, p=.89). Similarly, for deoxyhemoglobin, there was a significant interaction between condition and source-detector side, with a relative increase in deoxyhemoglobin between the control and No-Go task on the left and a decrease in deoxyhemoglobin on the right (β=−.65, SE=.18, t=−3.61, p<.001).

Figure 1.

Effects of task and age. Heat maps show statistical contrast for oxy- and deoxyhemoglobin change between executive function tasks and their control tasks. For Go/No Go, oxyhemoglobin increased more from the control condition to the No-Go condition at right hemisphere channels versus left hemisphere channels, while deoxyhemoglobin increased more at left hemisphere channels versus right hemisphere channels. For N-back, increases in oxyhemoglobin for the 1-back compared to 0-back task were greater for the ventral versus dorsal row of optodes, while decreases in deoxyhemoglobin between the 1-back and 0-back conditions did not depend on optode location. Line graphs show Z-scores for changes in oxy- and deoxyhemoglobin for the Go/No-Go and N-back tasks by age. For all models, age was represented as a continuous variable, but here we use a median split (e.g., older, younger) to aid in visualization of the interaction between age and task. Error bars represent standard error of the mean.

N-back.

For the Nback task, the general linear model predicting oxyhemoglobin change showed a significant interaction between condition (i.e., 1-back versus 0-back) and source-detector row, with a higher contrast between the task (1-back) and control condition (0-back) in the ventral versus dorsal row of sensors (β=−.52, SE=.20, t=−2.54, p=.01, see Figure 1). There was no effect of either hemisphere (β=−.037, SE=.20, t=−.18, p=.85) or medial versus lateral location (β=.34, SE=.20, t=1.69, p=.09). There was an effect of condition on deoxyhemoglobin levels, with decreases in deoxyhemoglobin for the 1-back compared to 0-back task (β=−.26, SE=.10, t=−2.55, p=.01), and this did not vary by source-detector position.

fNIRS Results by Age

We next addressed the question of whether participant age affected changes in oxy- and deoxyhemoglobin. For the Go/No-Go task, increasing age was associated with a greater change in oxyhemoglobin levels from the Go to the No-Go conditions (β=.14, SE=.07, t=2.1, p=.04, see Figure 1). This effect of age was not mitigated by optode position. Age did not predict deoxyhemoglobin change between Go and No-Go (β=.08, SE=.07, t=1.22, p=.22). Increasing age was also associated with a bigger increase in oxyhemoglobin levels for the N-back task, (i.e., higher levels of oxyhemoglobin for the 1-back versus 0-back condition, β=.21, SE=.07, t=2.9, p=.005). This effect did not vary by optode position. For the N-back task, changes in deoxyhemoglobin (i.e., reduction in deoxyhemoblin from the 0-back to the 1-back conditions) also increased with age (β=−.25, SE=.07, t=−3.4, p<.001).

fNIRS in Relation to Task Performance

Finally, we determined whether changes in oxy- and deoxyhemoglobin across all optode locations were correlated with task performance after accounting for variance related to age. For the Go/No-Go task, both oxy- and deoxyhemoglobin change were associated with RTV (β=−.006, SE=.003, t=−2.18, p=.03; β=.008, SE=.003, t=2.63, p=.009). For the N-back task, deoxyhemoglobin change was associated with dprime (β=−.29, SE=.09, t=−3.35, p<.001) as well as with RTV (β=.012, SE=.003, t=3.44, p<.001). In all cases of significant correlations, higher oxyhemoglobin was associated with better performance, while higher deoxyhemoglobin was associated with worse performance. See Figure 2 for the visualization of these effects by optode position, showing diffuse effects across the frontal lobe.

Figure 2.

Partial correlations between oxy- and deoxyhemoglobin levels and measures of task performance (after accounting for the effects of age) for each location measured. The effect across regions showed that better performance was associated with higher levels of oxyhemoglobin change and lower levels of deoxyhemoglobin change.

Discussion

We analyzed changes in oxy- and deoxyhemoglobin in the frontal cortex across two executive function tasks in preschool and school age children. We first characterized patterns of activation across all participants for both the Go/No-Go and N-back tasks, which index response inhibition and working memory, respectively. In comparison to the control “Go” condition, the No-Go task was associated with increased oxyhemoglobin and decreased deoxyhemoglobin, with greater changes over the right versus left hemisphere. In comparison to the control “0-back” task, the 1-back task resulted in increases in oxyhemoglobin and decreases in deoxyhemoglobin, with the contrast stronger in the ventral versus dorsal row of optodes for oxy- but not deoxyhemoglobin. Further, these effects varied by participant age. Specifically, increasing age was associated with a larger effect of condition (No-Go versus Go, 1-back versus 0-back) that was reflected in change in oxyhemoglobin for the Go/No-Go task as well as changes in both oxy- and deoxyhemoglobin for the N-back task. Finally, changes in both oxy- and deoxyhemoglobin were related to task performance, with higher levels of oxyhemoglobin related to better task performance and higher levels of deoxyhemoglobin related to worse task performance.

Pattern of Activation: Go/No-Go

Our finding of right-lateralized activation in children for both oxy- and deoxyhemoglobin replicates the extant literature and extends it by showing evidence for a differential effect on increasing activation for left versus right hemispheres in children. Specifically, the Go/No-Go task has previously been associated with generalized frontal activation (Heilbronner & Münte, 2013; Steele et al., 2013) including in children with typical development and attention deficit hyperactivity disorder (Inoue et al., 2012). However, a localized right IFG activation has also been reported as central to response inhibition in adults (Aron et al., 2014; Zandbelt et al., 2012), although some research using deoxyhemoglobin suggests that this lateralized response emerges with age (Bunge, Dudukovic, Thomason, Vaidya, & Gabrieli, 2002; Mehnert et al., 2013). Here, we show not only significant activation near the right IFG, but also a differentially larger response on the right versus left hemisphere.

Pattern of Activation: N-back

Other NIRS studies of working memory in children have shown widespread activation of bilateral prefrontal cortex as indicated by changes in both oxy- and deoxyhemoglobin, with some lateralization effects depending on task and age (Buss et al., 2014; Perlman et al., 2016; Tsujii et al., 2009; Tsujimoto et al., 2004). While we replicated this effect across a wide age range in children by showing significant increases in oxy- and decreases in deoxyhemoglobin for the 1-back vs. 0-back task, we also found that for the N-back task the changes in both oxy- and deoxyhemoglobin were greater in the ventral row of optodes. This pattern likely corresponds to activation in the orbitofrontal cortex. Notably, all other fNIRS studies of working memory in children have used visuospatial working memory tasks. While it is possible that the regional pattern seen in the present study is related to the nature of the task (i.e., remembering what something “was” vs. remembering where it “is”), it is also possible that this interaction with location is spurious, given the limited sample used here (n=15, with 9 returning for a second visit, for a total of 24 data points), and replication is a necessary next step in characterizing localization of this task in children.

Relation to Age

For the Go/No-Go task, we found that increasing age was associated with greater changes in both oxy- and deoxyhemoglobin. The two NIRS studies of Go/No-Go in children have not described an effect of age on activation change in children (Inoue et al., 2012; Mehnert et al., 2013), although Mehnert and colleagues do report an increase in activation in adults compared to children. Differences in findings could be attributable to a smaller and younger age range in that study (i.e., 3–4 year old children versus 4–10 year old children in present study). The study by Inoue and colleages was focused on group differences between typical development and ADHD, and therefore a lack of age effect in the full sample may have been related to the sample characteristics. This is therefore the first study to report an effect of age on frontal activation during response inhibition in children, although this pattern has certainly been reported in other executive tasks in children (e.g., Kawakubo et al., 2011).

For the N-back task, we found that increasing age was associated with an increase in oxy- and a decrease in deoxyhemoglobin. The extant literature on the effect of age on NIRS signal for working memory has been mixed, with studies showing decreases in activation with age (Tsujii et al., 2009), no effect of age (Buss et al., 2014), increases in activation with age (Perlman et al., 2016), and mixed increases and decreases varying by region in 8–20 year olds (Tamm, Menon, & Reiss, 2002). One potential reason for mixed findings involves the role of task difficulty and its interaction with age. Specifically, while activation related to maturation may increase with age, activation related to change in the relative level of difficulty may decrease, with the resulting effect depending on the balance of those two effects. Here, the effect of age on changes in oxy- and deoxyhemoglobin may be related to the wide range of ages included in our sample paired with tasks of limited difficulty for children.

Relation to task performance

In the present study we report that improved task performance is associated with increases in oxyhemoglobin (RTV for the Go/No-Go task) and decreases in deoxyhemoglobin (RTV for the Go/No-Go task, RTV for the N-back task, and dprime for the N-back task) after accounting for the effects of age. Very few studies of brain activity during executive function in children have reported relations between levels of activation and task performance. However, there are multiple examples from the fMRI literature of an increase in activation related to better performance across multiple ages, including children (Bunge et al., 2002; Rubia, Smith, Taylor, & Brammer, 2007). One study from the fNIRS adult literature shows a positive correlation between working memory performance and oxyhemoglobin increases (Ogawa, Kotani, & Jimbo, 2014). In children, increased fNIRS activation with better task performance has been noted only for cognitive flexibility tasks (Li, Grabell, Wakschlag, Huppert, & Perlman; Mehnert et al., 2013) and no correlation has been reported for response inhibition (Inoue et al., 2012; Mehnert et al., 2013), letter fluency (Kawakubo et al., 2011), or working memory (Perlman et al., 2016). In previous studies, lack of a significant positive relation between neural activity and task performance in children may be due to the interfering role of increased cognitive load, which is also associated with increased activity in adults and children in both fNIRS and fMRI studies (Fishburn, Norr, Medvedev, & Vaidya, 2014; Kharitonova, Winter, & Sheridan, 2015; Perlman et al., 2016). Specifically, while performance increases with age while relative cognitive load for a given task level decreases with age, one effect could potentially suppress the other. Here, we account for this effect by measuring the relation between performance and activity after accounting for changes related to age.

Limitations

This study contributes to the growing literature on NIRS studies of executive function and children by looking across a range of ages, using both response inhibition and working memory tasks, and provides novel findings regarding the role of age and task performance on NIRS signal that augment the literature. However, there are limitations that are worth noting. First, we used a limited number of optodes here, as we started with sixteen but used only eight in our analyses due to reduced data quality along the hairline. This means that we are limited to analyzing activity within Brodmann areas 10 and 46 (and limited within those as well), and therefore could not capture all activity associated with these tasks or even all frontal activity associated with these tasks. In addition, while fNIRS captures local changes in oxygenation, its spatial resolution, including both precision of localization on the surface of the cortex and interrogation depth, is limited in comparison to other imaging modalities (i.e., fMRI, MEG). This data is therefore useful in describing patterns of relation between age, performance, and some frontal activity during executive functions, but cannot fully characterize activation related to these tasks across the frontal cortex.

Second, we used tasks associated with executive functions (No-Go, 1-back) as well as control tasks (Go, 0-back), but did not include any further variations on difficulty. Evidence shows that task difficulty (particularly, task difficulty relative to an individual’s ability) affects activation levels (Amyot et al., 2012; Fishburn et al., 2014). Therefore, future studies of fNIRS activation for executive function tasks should use a range of task difficulty to elicit individual differences at different ages. It should also be noted that both the Go/No-Go and N-back tasks were presented as a block design, a common method used when measuring the BOLD response, which may result in more diluted effects when compared to an event-related design.

Third, we acknowledge that given the wide age range, the limited sample size hinders statistical power and potentially limits interpretability of the findings, and that having repeated measurements at different ages for only part of the sample precludes any ability to address longitudinal patterns. Any lack of findings here (e.g., localization of the effect of task performance) could be due to a lack of power, and as such should be interpreted carefully. In addition, in a small sample size, extreme data values in a few participants can spuriously drive significant findings. We addressed this concern by stringently excluding all data points with z-scores over 2 relative to data from the full sample.

Conclusions

This study shows frontal activation changes for both response inhibition and working memory in children and relates those changes to participant age and task performance. It replicates work from both the fMRI and fNIRS literature by revealing a lateralized activation pattern for response inhibition in children. It also supports continued investigation of the interaction between age, ability, and task difficulty and their effects on neural activation in children.