History of concussion impacts electrophysiological correlates of working memory

Abstract

Sports-related concussions occur in approximately 21% of college athletes with implications for long-term cognitive impairments in working memory. Working memory involves the capacity to maintain short-term information and integrate with higher-order cognitive processing for planning and behavior execution, critical skills for optimal cognitive and athletic performance. This study quantified working memory impairments in 36 American football college athletes (18–23 years old) using event-related potentials (ERPs). Despite performing similarly in a standard 2-back working memory task, athletes with history of concussion exhibited larger P1 and P3 amplitudes compared to Controls. Concussion History group latencies were slower for the P1 and faster for the N2. Source estimation analyses indicated that previously concussed athletes engaged different brain regions compared to athletes with no concussion history. These findings suggest that ERPs may be a sensitive and objective measure to detect long-term cognitive consequences of concussion.

1. Introduction

Sports-related concussions are a major public health concern as approximately 1.6–3.8 million occur every year (Langlois, Rutland-Brown, & Wald, 2006). Concussion is considered a mild traumatic brain injury and is defined as a complex pathophysiological brain injury that is caused by biomechanical forces (McCrory et al., 2013). Athletes with a history of head injury are believed to be at an elevated risk for future brain injury, as well as possible residual effects (Kontos, Kotwal, Elbin, Lutz, Forsten, Benson, & Guskiewicz, 2013), such as sleep disorders (Bramoweth & Germain, 2013; Gosselin, Lassonde, Petit, Leclerc, Mongrain, Collie, & Montplaisir, 2009) and cognitive deficits (Bleiberg et al., 2004; Guskiewicz, Marshall, Bailes, McCrea, Cantu, Randolph, & Jordan, 2005; Konrad et al., 2011; Larson, Farrer, & Clayson, 2011). Working memory is one such critical aspect of cognition known to be impacted by concussion that describes the maintenance and integration of information for higher cognitive functions (Baddeley & Hitch, 1994; Just & Carpenter, 1992) that is commonly measured as a part of short-term post-concussion evaluations (Allen & Gfeller, 2011; Lovell & Getz, 2006). Following a concussion, working memory deficits appear to resolve within 7–10 days (Belanger & Vanderploeg, 2005; McCrea et al., 2003). Yet, some research suggests that subtle cognitive deficits including working memory impairment may persist long after somatic concussion symptoms (e.g., headaches, nausea) have dissipated (Iverson, Brooks, Collins, & Lovell, 2006; Sandel, Lovell, Kegel, Collins, & Kontos, 2013).

Numerous behavioral investigations have challenged the idea that concussion results in a transient change in cognitive function that resolves quickly (e.g. Broglio, Moore, & Hillman, 2011; Collins, Grindel, Lovell, et al., 1999; Matser, Kessels, & Lezak, 1999). For instance, a large-scale study of approximately 600 youth athletes by Covassin and colleagues (2013) found that athletes with three or more concussions experience impaired verbal memory for longer periods of time compared to athletes with one or no previous concussions. Specifically, on the Immediate Post-Concussions Assessment and Cognitive Testing (ImPACT) neurocognitive battery subtest for verbal working memory, athletes with a history of one or two concussions had scores that resolved to baseline by eight days post-injury unlike athletes with three or more concussions. In addition, longer-term follow-up studies demonstrate working memory impairments in athletes later in their career or after retirement. For one, Guskiewicz and colleagues (2005) had approximately 2550 retired professional football athletes (≥ 50 years old) complete a general health questionnaire and followed up with ~750 retirees to more specifically target memory and cognition. The study found that retirees with a history of multiple concussions (three or more) are five times as likely to be diagnosed with mild cognitive impairment and three times as likely to have more self-reported memory impairments, relative to retirees without a history of concussion. A second long-term study indicates that deficits in working memory tasks, as well as other cognitive domains, may increase as athletes age, signifying that the effects of concussion amplify over time (Siechepine et al., 2013).

It is critical to understand how subtle deficits in working memory may carry negative consequences for lifelong challenges (Hux & Hacksley, 1996; Sholberg & Ledbetter, 2016). However, traditional measures, including behavioral performance, may not fully capture underlying deficiencies in the working memory system that may be evident at a neural level (Hammeke et al., 2014; Mayer et al., 2012). Neuroimaging methods may be helpful to better understand the long-term cognitive sequelae affecting working memory (Bigler & Omison, 2004; Sharp, Scott, & Leech, 2014). Specifically, it may be useful to include neuroimaging tools, such as event-related potentials (ERPs), source estimation, and functional magnetic resonance imaging (fMRI), in non-symptomatic, young adults to assess the short and long-term consequences of concussion. After behavioral indicators of concussion such as reaction time and accuracy on cognitive tasks appear to have recovered, differences in brain function between concussion and control groups are still evident (Hammeke et al., 2013; Ledwidge & Molfese, 2016).

ERP research on working memory highlights mid-latency components including the N2 and P3 (Duncan, Kosmidis, & Mirsky, 2005; Gosselin et al., 2012). The N2 is elicited at approximately 150–300 ms post-stimulus onset at fronto-central or prefrontal electrode sites, particularly during tasks involving mismatched visual stimuli (Folstein & Van Petten, 2008). The N2 is sensitive to deviant or mismatching information (Eimer, Goschke, Schlaghecken, & Stürmer, 1996; Folstein & Van Petten, 2008) and precedes the P3 component, a marker of attention allocation during the updating of working memory (Donchin, 1981; Donchin & Coles, 1988). The P3 is a positive component of the ongoing EEG waveform that is elicited between 300–800 ms post-stimulus onset and maximal at centro-parietal or central electrode channels (Donchin, Karis, & Bashore, 1986).

Although research using attentional tasks highlights attenuated amplitudes following concussion (Broglio, Pontifex, O’Connor, & Hillman, 2009; De Beaumont, Brisson, Lassonde, & Jolicoeur, 2007; De Beaumont et al., 2009; Dupuis, Johnston, Lavoie, Lepore, & Lassonde, 2000; Moore, Hillman, & Broglio, 2014), there is less evidence about N2 and P3 working memory responses (Gosselin et al., 2012). There is evidence that working memory impairments persist more than six months following remote concussion (Ozen, Itier, Preston, & Fernandes, 2013), as reflected by a reduced P300 amplitude, suggesting that concussion has a long-term impact on the ability to allocate attention appropriately during working memory. Another working memory study by Gosselin and colleagues (2012) found a similar P300 reduction for patients with mild traumatic brain injury and sports concussion; yet, there were no amplitude or latency N2 differences between groups, aligned with two other studies (Gosselin, Bottari, Chen, Petrides, Tinawi, de Guise, & Ptito, 2011; Duncan et al., 2005). The lack of group N2 differences may indicate that the attention orienting phase of working memory is not affected. However, across four other studies, tasks that target other aspects of cognition (i.e., attention, response inhibition, switching) have found increased N2 amplitudes (Moore et al., 2014; Moore, Pindus, Drolette, Scudder, Raine, & Hillman, 2015; Ledwidge & Molfese, 2016; Rugg, Cowan, Nagy, Milner, Jacobson, & Brooks, 1988) and slower latencies (Moore et al., 2015; Rugg et al., 1988). One study found an N2 reduction on a task related to working memory (Broglio et al., 2009). These discrepancies in the literature warrant further investigation and the present study aimed to address these issues.

The objective of this study was to examine long-term effects on working memory brain function following a history of concussion in young adults. Specifically, this study set out to address three key issues within the existing literature. First, our objective was to evaluate working memory, specifically, to assess whether known discrepancies in the literature related to attention (e.g., opposing N2 and P3 amplitude effects, Moore et al., 2014; Rugg et al., 1988) extend to working memory, as well. For instance, there are mixed findings related to attention, such that some studies suggest delayed P3 latencies associated with concussion (De Beaumont et al., 2009; Ledwidge & Molfese, 2016), while others find no differences (Broglio et al., 2009). Second, despite the fact that these mid-latency components (i.e., between 150–800 ms) are known to characterize group differences between concussion and control groups, there is a potential bias in using a priori defined temporal windows for ERP components (see Dien, 2010; Dien, Beal, & Berg, 2005; Foti, Hajcak, & Dien, 2009 for a review of possible common misinterpretations). For instance, if a concussion event causes long-term alterations to the organization and/or efficiency of the working memory system, then it is less appropriate to constrain analyses to classic temporal components. Thus, our objective was to determine the temporal window of the ERP most associated with group differences that we anticipated would identify a prefrontal N2 and central P3, yet would not bias possible morphological shifts (Dien & Santuzzi, 2005). Third, no prior studies evaluating the long-term effects of concussion on working memory have combined the use of ERP with source analysis. By exploring the underlying source generators of ERPs, our objective was to better describe the neural mechanisms involved with long-term deficits.

We measured electrophysiological correlates of working memory in varsity collegiate athletes with and without a history of concussion using a standard 2-back task (i.e. Braver, Cohen, Nystrom, Jonides, Smith, & Noll, 1997; Cohen, Perlstein, Braver, Nystrom, Noll, Jonides, & Smith, 1997). In order to clarify the long-term consequences of concussion, we proposed two hypotheses. First, in line with prior work describing a long-term return to normative cognitive task performance following concussion (Hillary et al., 2011; McAllister, Sparling, Flashman, Guerin, Mamourian, & Saykin, 2001), we predicted that the Concussion History and Control groups would have equivalent accuracy and reaction times on behavioral cognitive tasks. Second, we anticipated that the working memory task would elicit the N2 associated with attention orienting, but would not be affected by concussion history. We predicted that the P3 would exhibit decreased activation (i.e., smaller amplitudes, slower latencies, less source engagement) in participants with a history of concussion. Decreased P3 activation for concussed individuals may indicate reduced working memory capacity (Broglio et al., 2009; Ozen et al., 2013) while increased activation may indicate increased resource demands for working memory (Ledwidge & Molfese, 2016).

2. Methods and materials

2.1. Participants and characterization

Collegiate American football athletes were invited to participate in a research study that consisted of testing prior to the onset of the respective sport season, including cognitive assessment and an electrophysiological battery. The present investigation included male college American football athletes (N = 36) ranging in age from 18 to 23 years (M = 20.84 years, SD = 1.58). Participant characterization is provided in Table I. The University’s institutional review board and the Department of Athletics approved all study procedures. All athletes provided written informed consent to participate. Data reported in this study was collected over a period of two years. Athletes participating in the second year of testing received financial compensation of $25.

Table I. Group demographics and characterization.

Participant demographics and characterization of working memory, inhibition, and processing speed are reported for each group as Mean (Standard Deviation).

	Concussion	Control	Group differences
Domain / Measure	History Group	Group	t (df)	p
Number of subjects	17	19
Previously diagnosed concussions	1.47 (.94)	0
Years since concussion	2.94 (3.31)	n/a
Number of Match trials	36.88 (5.79)	42.16 (6.93)	−2.46 (34)	0.019
Number of Non-match trials	42.35 (6.42)	42.37 (5.78)	−0.008 (34)	0.994
Age (years)	20.9 (1.5)	20.79 (1.69)	0.2 (34)	0.84
Working memory (scaled score)	10.18 (2.07)	9.47 (1.9)	1.06 (34)	0.30
Inhibition - errors (#)	10.59 (2.81)	10.59 (2.81)a	1.10 (33)	0.30
Inhibition - speed (ms)	12.35 (1.8)	10.94 (3)a	1.67 (33)	0.10
Processing speed (ms)	18.31 (4.64)	19.2 (4.39)	−0.59 (34)	0.56
Switching speed (ms)	45.38 (19.57)	43.45 (15.46)	.33 (34)	0.74

^aData from one participant removed due to color-blindness.

Asterisk (*) highlights significant differences between groups.

Participants were divided into two groups based on their concussion history prior to enrolling in this study, similar to other studies (Broglio et al., 2009; De Beaumont et al., 2009; Moore et al., 2014; Ozen et al., 2013). The Concussion History group (n = 17) included participants who reported at least one prior medically diagnosed concussion on the ImPACT (Immediate Post-Concussion Assessment and Cognitive Testing; ImPACT Applications, Pittsburgh, PA), which was confirmed by Department of Athletics staff. Athletes in the Concussion History group reported 1–4 previously diagnosed concussions (M = 1.47, SD = 0.94). For this group, an average of approximately three years elapsed since their most recent concussion (M = 2.94, SD = 3.31). The Control group (n = 19) reported no prior concussion history. Power analyses indicated that a minimum of 13 participants per group were needed to achieve 80% power for effects between r = .65-.88 based upon a power analysis conducted for cognitive function post-concussion (Moore, Lepine, & Ellemberg, 2017), indicating that our groups were adequately powered.

Participants completed a battery of neuropsychological measures to evaluate overall group differences in working memory and other cognitive features closely associated with working memory. In addition to the ERP task, traditional working memory performance was assessed via pencil-paper testing on the Letter-Numbering Sequencing subtest of the Wechsler Adult Intelligence Scale–IV (Wechsler, 2008). The Color-Word Interference subtest of the Delis-Kaplan Executive Function System (D-KEFS; Delis, Kaplan, & Kramer, 2001) assessed inhibition (Dimoska-Di Marco, McDonald, Kelly, Tate, & Johnstone, 2011). Lastly, the Trail-Making Test Forms A and B assess processing and switching speed, respectively (Collie, Makdissi, Maruff, Bennell, & McCrory, 2006; Sánchez-Cubillo, Periáñez, Adrover-Roig, Rodríguez-Sánchez, Ríos-Lago, Tirapu, & Barceló, 2009). As shown in Table I, there were no between-group differences in working memory, inhibition, or processing speed.

2.2. Working memory 2-back task

A standard visual 2-back task assessed working memory (Shelton, Elliott, Hill, Calamia, & Gouvier, 2009). Participants were seated facing a standard Dell 15.5” (39.5 cm) laptop screen. The researcher instructed the participant to attend to uppercase English language letters presented in the center of the screen with a visual angle of 1.16° × 1.16° viewed from a distance of 1 m. Each letter was displayed for 1000 ms followed by a blank screen that appeared for a varied period of 1600–2200 ms. Stimuli were displayed using E-Prime 2.0 software (Psychology Software Tools, Inc, Pittsburgh, PA). Participants used a handheld response pad with two buttons to indicate whether the letter matched or did not match the letter presented two stimuli prior (e.g., 2-back from the current stimuli). Participants were instructed to respond as quickly and accurately as possible. Fifty Match and 50 Non-match trials were presented in random order. Analysis was restricted to correct trials only (overall: 81% of Match trials; 93% of Non-match trials). The Concussion History group did have fewer correct Match trials relative to the Control group (see Table I), which we acknowledge as a limitation. However, both groups contributed a minimum number of trials for behavioral and ERP analysis consistent with other work (e.g., > 30 trials for each condition per individual; Duncan, Kosmidis, & Mirsky, 2003; Ledwidge & Molfese, 2016), even within a single participant (Kota, Kelsey, Rigoni, & Molfese, 2013). Average reaction time (RT; from stimulus onset) and accuracy (percentage of correct trials) were computed for each participant for both the Match and Non-match conditions.

2.3. Electrophysiological recording and preprocessing

Ongoing electroencephalogram (EEG) was recorded from a high-density 256-channel silver/silver chloride electrode array using Net Station 4.4.2 software and high-impedance amplifiers (Electric Geodesics Inc., EGI, Eugene, OR). Following laboratory standard procedures, electrode impedances recorded before and after the task measured below 60 kohms to maximize signal-to-noise ratio. The unfiltered EEG at all electrode sites was recorded with a gain of 10 k and referenced to C_z. EEG signals were filtered offline from 0.3 to 30 Hz per Rossini, Cracco, and Cracco (1981). Correct trials were segmented with a 200 ms baseline period prior to stimulus onset through 800 ms post-stimulus onset, adjusting for computer timing offsets and digital finite impulse response filters (fixed). Channels with voltage shifts greater than 150 µV due to eye blinks or bad signal or greater than 50 µV due to eye movements were automatically classified as artifacts. Following artifact detection, trials were reviewed by manual visual inspection by a researcher blind to concussion status to confirm automatic, algorithmic decisions before being deleted and replaced with signals via spline interpolation from immediately adjacent electrode sites per EGI commercial software algorithm (Ferree, 2000). The entire trial was excluded if it contained more than 40 bad channels. Trials were then baseline-corrected, re-referenced to the average reference, and averaged for each condition. Equivalent numbers of trials per condition for each participant were included in the analyses. The averaged ERPs were clustered into prefrontal and central scalp regions (see Supplemental Table I) (Ledwidge & Molfese, 2016; Kota et al., 2013). This procedure increased statistical power by reducing electrode locations from 256 individual sites into a smaller set of scalp regions (Curran, 1999).

In order to derive temporal windows associated with the N2 and P3 that are unbiased towards group status, we implemented a temporal principal components analysis (tPCA) approach (Kayser & Tenke, 2003; Molfese, Nunez, Seibert, & Ramanaiah, 1976) with complete details available within Supplemental Materials. The tPCA identified a prefrontal window from 132 to 304 ms associated with P1 and N2 components, a central P3 window from 248 to 620 ms, and a prefrontal late positive component (LPC) from 468 to 800 ms. Preliminary analyses indicated that the LPC was not related to group status; thus, analyses focused only the P1, N2 and P3.

2.4. Behavioral, ERP amplitude and latency analysis

Statistical analyses were performed using SPSS version 21 (IBM, Chicago, IL). First, One-Way ANOVAs compared the Concussion History and Control groups on reaction time and response accuracy on the 2-back task. Next, for each individual, the P1, N2, and P3 peak amplitude and peak latency values at each scalp region were extracted for Match and Non-match conditions. N2 and P3 values were extracted according the tPCA-derived temporal windows. To avoid overlap between the positive peak of the P1 and a second positive-going prefrontal component, P1 values were extracted from the beginning of the tPCA-derived time window (132 ms) to the average latency of the N2 across all subjects and condition (224 ms). Group differences were tested via univariate ANOVA with group, condition, and the subsequent interaction included as fixed effects. Post-hoc statistics for each condition were computed using independent samples t-tests.

2.5. ERP brain source analysis

Source estimation analysis examined group differences in the underlying brain source localization for each experimental condition. A finite difference model (FDM) was applied using a forward modeling approach to compute the electrode locations in relation to brain tissues (Vanrumste, Van Hoey, Van de Walle, D’Havé, Lemahieu, & Boon, 2001). This brain source analysis was conducted using the standardized low-resolution brain electromagnetic tomography (sLORETA) method (Pascual-Marqui, 2002) within the NetStation GeoSource 2.0 software package (EGI, Eugene, OR). FDM estimates were constrained by the Montreal Neurological Institute (MNI) average adult MRI database. Tissue volumes were parceled using 7-mm voxels, each serving as a dipole source location with three orthogonal orientations (in x-, y-, and z-orientations). The FDM applied estimations across a total of 2,447 source dipole triplets. Conductivity values used in the FDM model included 0.25 S/m for brain, 1.8 S/m for cerebral spinal fluid, 0.018 S/m for skull, and 0.44 S/m for scalp (Ferree, Eriksen, & Tucker, 2000). Weighting was placed equally across locations with regularization carried out via Tikhonov regularization (1×10⁻²), consistent with recommendations by Congedo (2006) and other recent work using sLORETA (Chung et al., 2007; Hudac, Kota, Nedrow, & Molfese, 2012; Ledwidge & Molfese, 2016). Source estimation was conducted within 33 GeoSource-defined brain gyri for both left and right hemispheres (66 brain regions total). Athletes’ mean source estimation was computed for the two tPCA-derived time windows for the P1 and N2 (132–304 ms) and P3 (248–620 ms) components for each condition across the 66 gyri locations. Group differences were tested via One-Way ANOVAs.

3. Results

3.1. Behavioral analysis

Figure 1 illustrates group-average task accuracy and reaction time for each condition. As predicted, there was no significant difference between groups in accuracy or reaction time, F(1, 34) < 2.39, p’s > .13. Means and standard deviation for each group are reported in Supplemental Table II.

Figure 1. Working memory task performance by group.

Group-averaged task accuracy (top, percent correct) and reaction time (bottom, milliseconds) are plotted for each condition. Error bars represent +/−1 SD. Control group = grey; Concussion history group = white.

3.2. ERP amplitude analysis

Figure 2 illustrates group grand-averaged waveforms for each condition across prefrontal and central electrodes. Group means and post-hoc differences for each condition are reported in Table II. Both P1 and N2 exhibited a main effect of group (P1: F(1, 68) = 4.44, p = .039; N2: F(1, 68) = 6.11, p = .016), such that the Concussion History group exhibited larger P1 amplitudes and smaller N2 amplitudes than the Control group. There were no main effects or interactions with condition for the P1 or N2, and pairwise comparisons did not indicate significant group P1 or N2 amplitude differences for either condition. For the P3 component, there was a main effect of group, F(1, 68) = 4.38, p = .040, such that the Concussion History group had a larger P3 amplitude overall relative to the Control group. Pairwise comparisons (reported in Table II) indicated a larger Match condition P3 amplitude for the Concussion History relative to the Control group. There was also a marginal effect of condition, F(1, 68) = 3.12, p = .082, such that the P3 amplitude was larger for Match than Non-match condition, which is in the anticipated direction for this task.

Figure 2. Grand-averaged ERP waveforms for conditions by history of concussion.

ERP waveforms are plotted over time for athletes with (red) and without (black) a history of concussion for both Match (left) and Non-match (right) conditions. Waveforms are presented as a grand-average across prefrontal electrodes (top row, P1 and N2) and parietal electrodes (bottom row, P3). The time window derived via tPCA is noted as a yellow bar along the x-axis.

Table II. Group amplitude differences in microvolts.

Group means and standard deviation are reported for P1, N2, P1-N2 peak-to-peak amplitude difference, and P3 amplitude in microvolts (µV).

		Concussion History	Control	Group differences
Component	Condition	M (SD)	M (SD)	t (df)	p
P1	Match	3.53 (2.29)	2.57 (1.71)	−1.44 (34)	0.159
	Non-match	3.53 (1.86)	2.49 (2.17)	−1.54 (34)	0.133
N2	Match	−0.7 (1.99)	−1.93 (1.74)	−1.98 (34)	0.056
	Non-match	−0.95 (2.02)	−1.99 (2.05)	−1.54 (34)	0.133
P3	Match	6.94 (1.81)	5.3 (2.39)	−2.3 (34)	0.028
	Non-match	5.46 (1.51)	5.1 (2.17)	−0.56 (34)	0.58

Asterisk (*) highlights significant differences between groups.

3.3. ERP latency analysis

Group means and post-hoc differences for each condition are reported in Table III. There was a significant P1 latency difference, F(1, 68) = 11.19, p = .001, such that the Concussion History group had a longer P1 latency. In contrast, the Concussion History group had a faster N2 latency, F(1, 68) = 5.03, p = .028, but no group difference in P3 latency, F(1, 68) = .16, p = .70. There were no main effects or interactions with condition for the P1 or N2. There was an anticipated P3 main effect of condition, F(1, 68) = 28.07, p < .0001, such that the P3 was faster for Match relative to Non-match.

Table III. Group latency differences in millisecond.

Group means and standard deviation are reported for P1, N2, and P3 latency in milliseconds (ms).

		Concussion History	Control	Group differences
Component	Condition	M (SD)	M (SD)	t (df)	p
P1	Match	171.61 (20.39)	155.46 (19.6)	−2.42 (34)	0.021
	Non-match	171.25 (17.39)	157.78 (17.53)	−2.31 (34)	0.027
N2	Match	209 (51.02)	231.37 (44.97)	1.4 (34)	0.171
	Non-match	212.88 (57.02)	241.56 (39.45)	1.74 (28.06)	0.094
P3	Match	399.69 (43.66)	426.06 (63.35)	1.47 (32.03)	0.152
	Non-match	489.11 (44.07)	472.85 (61.39)	−0.9 (34)	0.373

Asterisk (*) highlights significant differences between groups.

3.4. ERP source analysis

Results and group means are reported in Table IV. For the P1 and N2 time window, the Concussion History group had less Non-match condition source activation than the Control group within the left lingual and superior frontal gyri, F’s (1,34) > 4.56, p’s < 0.040. For the P3 Non-match condition, the Concussion History group generated less source activation than the Control group within the right fusiform, sub-gyral, and parahippocampal gyri, F’s > 4.60, p’s < 0.039 (see Figure 3). For the P3 Match condition, the Concussion History group generated more source activation than the Control group within the left superior parietal lobule, F (1,34) = 4.25, p= 0.024.

Table IV. Significant group source estimation differences.

Significant P1-N2 and P3 source estimation differences between groups are reported separately by condition. Group means and standard deviations also presented.

				Concussion History		Control		Group differences
Component	Condition	Hemi	Region	M	SD	M	SD	F(1,34)	p
P1 and N2	Non-match	L	Lingual Gyrus	0.114	0.035	0.164	0.081	5.56	0.024
		L	Superior Frontal Gyrus	0.029	0.012	0.039	0.016	4.56	0.04
P3	Match	L	Superior Parietal Lobule	0.067	0.037	0.048	0.014	4.25	0.047
	Non-match	R	Fusiform Gyrus	0.096	0.034	0.152	0.102	4.6	0.039
		R	Parahippocampal Gyrus	0.118	0.046	0.184	0.104	5.83	0.021
		R	Sub-Gyral	0.075	0.027	0.117	0.073	5.02	0.032

Figure 3. P3 brain source activation between concussion symptom groups.

Whole-brain source estimation maps are plotted as average values within 50 ms moving windows from 250–600 to illustration source activation of the P3 component (248–620 ms). As a guide, yellow lines intersect the right parahippocampal gyrus. Amount of source activation is presented on a scale from 0.05 (dark red colors) to 0.1 (yellow) nA.

4. Discussion

This study explored how a history of concussion relates to behavioral and electrophysiological correlates of working memory. Collegiate football athletes with and without a history of concussion completed a visual 2-back working memory task during ERP recording. Group differences were evident in ERP and source estimation, but not in behavioral performance. The use of tPCA to derive temporal windows ensured that the component time windows were not biased against concussion history status. Instead, the time windows were associated with the most variance, which unexpectedly revealed a P1-N2 complex (132–304 ms) as well as the anticipated P3 (248–620 ms). The Concussion History group exhibited increased P1 and P3 amplitudes, slower P1 latencies, and faster N2 latencies compared to the Control group. These results lend partial support to our hypothesis that despite a lack of clear behavioral problems, there may be long-term cognitive consequences of concussion, which might support ERPs as a sensitive measurement tool to detect cognitive changes following concussion.

Few ERP studies investigating concussion symptoms or long-term effects have targeted early perceptual components, such as the P1. P1 responses have been associated with sensory selection (Finnigan, O’Connell, Cummins, Broughton, & Robertson, 2011; Heinze, Luck, Mangun, & Hillyard, 1990; Hillyard, Vogel, & Luck, 1998) and are thought to reflect the earliest ERP index of attentional control (Klimesch, Sauseng, and Hanslmayr, 2007; Klimesch, Sauseng, Hanslmayr, Gruber, & Freunberger, 2007; Klimesch, Schack, Schabus, Doppelmayr, Gruber, & Sauseng, 2004; Natale, Marzi, Girelli, Pavone, & Pollmann, 2006). The Concussion History group had slower P1 latencies suggesting delayed perceptual processing and together with the increased P1 amplitudes may indicate an increase of working memory resources tied to this early perceptual processing. Aligned with other work (i.e., Missonnier et al., 2005), this finding emphasizes the importance of examining early components, as these early group differences may implicate deficits within pre-cognitive aspects of working memory.

However, the N2 latency was faster for athletes within the Concussion History group, which is in contrast to results from other cognitive tasks that found increased N2 latencies (Moore et al., 2015; Rugg et al., 1988). These findings indicate that it may be helpful to conceptualize group differences in N2 latencies in the context of earlier perceptual components to better differentiate deficient aspects of working memory. In addition, it is important to consider the task specificity when comparing across studies, such that group differences may not be universally evident across all cognitive tasks or populations.

Multiple studies present P3 amplitude attenuation related to concussion history (Broglio et al., 2009; De Beaumont et al., 2007; De Beaumont et al., 2009; Dupuis et al., 2000; Moore et al., 2014), though others find no group differences (e.g., Gosselin et al., 2011). Our results identify increased P3 amplitude in previously concussed athletes, which may be consistent with the interpretation of increased demands on resources for other cognitive processes (e.g., selective attention) in athletes with a history of concussion (Ledwidge & Molfese, 2016). Alternatively, the increased amplitudes may also be tied more specifically to long-term structural changes that might be unique to the population in our study. For instance, prior work by Penkam and Mateer (2004) found increased P3 amplitudes in patients with a brain injury to the orbitofrontal region of the brain. Unlike other work demonstrating delayed P3 latencies (De Beaumont et al., 2009; Di Russo & Spinelli, 2010; Gaetz, Goodman & Weinberg, 2000; Gaetz & Weinberg, 2000; Gosselin, Thériault, Leclerc, Montplaisir, & Lassonde, 2006; Ledwidge & Molfese, 2016), we did not find any P3 latency group differences. One possible explanation is that the results of the present study are derived from a bilateral cluster of electrodes, as opposed to a single medial electrode (e.g., F_z, C_z, or P_z), and that this P3 latency effect may be evident when including more lateral electrodes.

Finally, to explore the group differences in neural mechanisms associated with the P1-N2 and P3, source analyses tested whole-brain regional group differences across gyri. The dense array EEG recording permits the estimation of brain-tissue source estimation for the ERP components we observed. Source analysis identified differences in the underlying brain mechanisms employed to complete the 2-back task for each group. Specifically, the Concussion History group exhibited reduced P1-N2 brain source activation compared to the Control group in gyri that may be related to working memory such as the left lingual and superior frontal gyri (Cohen et al., 1997; Ungerleider, Courtney, & Haxby, 1998; Wild-Wall, Falkenstein, & Gajewski, 2011). P3 source activation results were mixed. Areas implicated in working memory demonstrated reduced activation for the Concussion History group for Non-match trials, including the right parahippocampal, sub-gyral, and fusiform gyri (Courtney, Ungerleider, Keil, & Haxby, 1997; Ungerleider et al., 1998; Yoo, Paralkar, & Panych, 2004). Both the parahippocampal and fusiform gyri have been associated with memory encoding and retrieval (Ranganath, Cohen, Dam, & D’Esposito, 2004). However, the Concussion History group exhibited increased activation within the left superior parietal lobule for the P3 Match condition, a region that has been associated with the manipulation of information within working memory (Koenigs, Barbey, Postle, & Grafman, 2009).

Taking the ERP and source results together, these findings indicate important group differences in underlying brain activity, even though behavioral performance was equivalent across groups. During the early ERP components P1 and N2, the Concussion History group showed larger, slower ERP activity related to perceptual processing (P1), and smaller, earlier ERP activity associated with attention orienting and mismatch detection (N2), both related to less source activation in working memory brain regions compared to controls. The larger, later P1 amplitude for those in the Concussion History group may indicate less efficient perceptual processing, with less task-relevant brain activation. For those in the Concussion History group, it is possible that the larger positive amplitude of the P1 component affected the amplitude of the N2, resulting in a smaller (less negative) N2 relative to controls. N2 results are somewhat difficult to interpret given the mixed findings in previous literature, but the Concussion History group still demonstrated less task-relevant source activation. For the P3 window, the Concussion History group exhibited larger ERP amplitudes. Similar to the P1, increased amplitude related to less source activation for brain regions associated with working memory encoding and retrieval for the easier Non-match trials (as indicated by overall higher accuracy on Non-match compared to Match trials). This is consistent with our interpretation of less task-relevant brain source processing alongside increased electrophysiological activation for the Concussion History group. However, for the more difficult Match condition, during the P3 window we observed increased activation in brain regions associated with the manipulation of information in working memory for the Concussion History group. This may indicate that when working memory is taxed, such as during the Match condition of our 2-back task, increased electrophysiological response and increased activation in task-relevant brain regions may indicate the engagement of greater neural resources to achieve typical behavioral accuracy.

These results may be related to findings from functional magnetic resonance imaging (fMRI) that indicates that individuals with more severe symptoms following concussion engaged more working memory resources, including frontal (Pardini, Pardini, Becker, Dunfee, Eddy, Lovell, & Welling, 2010) and parietal networks (Pardini et al., 2010; Smits, Dippel, Houston, Wielopolski, Koudstaal, Hunink, & van der Lugt, 2009). In other words, symptomatic individuals exhibit increased activation within working memory regions, suggesting a reduction in the overall efficiency of the working memory system. It is possible that athletes with a history of concussion recruit additional attentional resources during tasks that tax working memory in order to compensate for deficits in the working memory system, even after acute symptoms of concussion have resolved (Hammeke et al., 2013).

Alternatively, it is possible that concussion results in more diffuse activation, perhaps as a compensatory mechanism following injury, that results in decreased levels of focal activity compared to controls. For example, some fMRI investigations report decreased neural activity in concussion groups compared to controls in regions related to working memory including the dorsolateral and medial prefrontal cortex (Chen, Johnston, Frey, Petrides, Worsley, & Ptito, 2004; Chen, Johnston, Petrides, & Ptito, 2008; Keightley, Saluja, Chen, Gagnon, Leonard, Petrides, & Ptito, 2014; van der Horn, Liemburg, Scheenen, de Koning, Spikman, & van der Naalt, 2015; Witt, Lovejoy, Pearlson, & Stevens, 2010) and the left superior parietal lobule (Keightley et al., 2014). Our results would suggest that this might be more likely to occur during less-taxing tasks. In several studies, however, regions of the brain outside the task-relevant regions of interest tend to exhibit increased neural activity in concussion groups compared to controls (Chen et al., 2004; Slobounov, Zhang, Pennell, Ray, Johnson, & Sebastianelli, 2010; Witt et al., 2010), consistent with our results during more taxing trials of increased P3 source activation within the left superior parietal lobule.

Our results suggest that athletes with a history of concussion demonstrate different electrophysiological and brain source responses during a working memory task. Specifically, athletes with a history of concussion exhibited consistently greater P1 amplitude and latency compared to controls that, in conjunction with reduced task-relevant source activation, may highlight an overall inefficiency in early perceptual aspects of working memory. In addition, compared to controls, the Concussion History group allocated inefficient P3 neural resources (larger amplitude, increased source activation compared to controls) for the difficult condition (Match) and less task-relevant neural resources (smaller amplitudes, reduced source activation) for the easier condition (Nonmatch). This indicates that there are some long-term changes in the allocation of neural resources needed to update working memory following concussion (Donchin, 1981; Donchin & Coles, 1988).

These findings emphasize the ongoing vulnerability and neural changes associated with a concussive event, even long after somatic symptoms have resolved, and athletes returned to athletic activities and academics. Our findings contribute to the growing body of research that reports the long-term sensitivity of electrophysiological correlates to detect changes in cognition related to concussion (Folmer, Billings, Diedesch-Rouse, Gallun, & Lew, 2011; Gosselin et al., 2012; Ozen et al., 2013). Resolution of concussion symptoms and improved neuropsychological test performance have been thought to indicate short-term recovery within 7–10 days (e.g., McCrea et al., 2003). However, our results show persistent behavioral, electrophysiological, and neural organizational differences between Control and Concussion History groups, approximately 3 years on average since last injury. Tracking recovery trajectories using more sensitive methods such as ERPs, and following athletes for a longer period of time post-injury may be necessary to fully appreciate the consequences of concussion to cognitive health.

In the research setting, the 2-back task isolates working memory demands. However, as part of daily routine, individuals are expected to simultaneously utilize an integrative network of neurocognitive resources to navigate a more complex and dynamic world. The recruitment of inefficient neural resources in the laboratory working memory task represents a cognitive disadvantage for the athletes with a history of concussion. Real-world situations involve more complex tasks, such as encoding memory across multiple domains (e.g., integrating audio and visual information regarding advancing opponents). Thus, athletes with a history of concussion may be at an even greater cognitive disadvantage in ecological settings that require rapid working memory encoding.

It is critical to consider how these results impact the development of concussion assessment tools. Neuropsychological tests can be unreliable as a measure of impairment due to factors such as repeat exposure to tests, learning effects, and psychological interference (e.g., reduced effort on baseline tests, or ‘sandbagging’). Neuroimaging techniques such as ERPs can elucidate the brain mechanisms and networks involved following head injury (Broglio et al., 2011; Duncan, Summers, Perla, Coburn, & Mirsky, 2011). These strategies are highly specific and provide ongoing information about neurocognitive mechanisms and efficiency. Recent work by Kota and colleagues (2013) demonstrated the consistency between ERPs following periods of physical activity and rest, highlighting possible use of ERPs as an immediate sideline measure. In addition to identifying aberrant function, ERPs can indicate multimodal impairments that persist beyond neurocognitive processing (Folmer et al., 2011). For instance, concussions and mild traumatic brain injuries commonly produce sensory problems, including visual problems, such as post-trauma vision syndrome (Hellerstein et al., 1995; Padula et al., 1994).

We acknowledge several limitations to this study. First, we acknowledge that there is substantial heterogeneity within the Concussion History group, which may diminish the ability to assess other key factors. For instance, although many of the college athletes in this study reported multiple head injuries, preliminary post-hoc analyses did not indicate that number of injuries or time since injury correlated with ERP amplitude or source activation (p’s > .05). This may be due to limited power and the variability across participants, yet these results are similar to other studies (e.g., Broglio et al., 2009). In addition, the current results may not represent all sports-related concussions. Concussions are extremely heterogeneous, and endophenotypes, such as patterns of ERP responses, may vary depending on the mechanism of initial injury, secondary cerebral responses, or environmental factors during the acute phase of recovery (i.e., sleep loss, return to play/academics, stress) (Grady, 2010). Relatedly, we did not confirm via independent evaluation that athletes within the Control group did not have an undiagnosed or unreported concussion. Second, we made several methodological and analytical decisions that should be considered. Although we based our 2-back task from prior work (Shelton et al., 2009), it may be the case that bigger group differences would have been evident with more trials, as suggested by Kappenman and Luck (2016), especially considering that the Concussion History group contributed fewer Match trials in our study. Additional trials might also have permitted the assessment of dynamic, ongoing processing of working memory (i.e., habituation or intensification of the signal over the course of the experiment), which has been used to specify cognitive mechanisms in other special populations, such as autism (Hudac et al., 2015; Hudac et al., 2017). Other recent work emphasizes the need for standardized procedures (e.g., filter settings, artifact removal, visual angle) in order to permit better comparisons across studies (Keil et al., 2014; Webb et al., 2015).

Another consideration is our use of peak amplitude and latency measurements in order to isolate the early components of interest (P1, N2) instead of mean amplitude measurement, which may better map onto the source estimation results. In particular, the susceptibility of peak measurements to high-frequency noise and the variability of timing across different electrodes within the regional clusters have been noted as potential problems in drawing conclusions regarding the underlying cognitive processes (Handy, 2005; Otten & Rugg, 2005). Additionally, we selected a priori spatial regions of interest for our analyses, which may have introduced bias. Third, as illustrated by the grand-average waveforms in Figure 2, there are potential baseline differences at the onset of each stimulus that may be related to the anticipation of the stimulus and/or ocular artifacts. Although the temporal PCA did not highlight this early period of the ERP (0–40 ms) as contributing a significant amount of variance, future work should better investigate this effect to determine if there are indeed group differences related to pre-attentional sensory processing. Lastly, it will be important to continue to look more closely at the organization and efficiency of the working memory system, particularly at other ERP and source components. For instance, here we targeted the P3 component, though it may be beneficial to look specifically at the P3a and P3b subcomponents, as they related to attention and memory processing, respectively (Polich, 2007).

5. Conclusions

This study provides evidence of long-term electrophysiological differences following concussion, highlighting the need to better understand the underlying cognitive and neural mechanisms implicated in concussive injuries. Despite a lack of behavioral task performance differences between groups electrophysiological methods demonstrated group differences in the amplitude and latency of ERP components, and differences in neural source activations. In particular, those with a history of concussion exhibited increased P1 and P3 amplitudes, decreased P1 latencies, increased N2 latencies, as well as different neural source activation for brain source regions critical for working memory. These results may implicate an inefficient use of the working memory neural system for athletes with a history of concussion. This highlights the necessity of replications and additional research in this area. Continued emphasis within neuroimaging research on long-term recovery will help develop biomarkers that might be more sensitive for the identification and ongoing tracking of concussive symptoms to support athlete recovery and improve overall long-term prognosis.

Highlights.

1. Increased P1 and P3 ERP working memory amplitudes found following concussion.

2. Decreased P1 and increased N2 latencies highlight possible perceptual differences.

3. Atypical neural source engagement suggests inefficient working memory demands.