Episodic Memory in Former Professional Football Players with a History of Concussion: An Event-Related Functional Neuroimaging Study

Participants

Participants in this study were 27 former professional NFL players who played a minimum of two seasons of professional football (M age=63.4 years; standard deviation [SD]=5.9; all male) and 14 healthy age and education-matched older adults with no history of concussion or memory impairment (M=62.2 years; SD=6.3; all male; Table 1). An additional five former football players (two from the low concussion group and three from the high concussion group) underwent functional MRI, but their data were excluded because of poor performance or difficulty following task instructions. Participants were all right-handed native English speakers without a history of psychiatric illness or neurological disorder. Before participating in the study, all persons gave written informed consent in accordance with the requirements of the Institutional Review Board at the University of North Carolina (UNC) at Chapel Hill.

Former football players in the current study were divided into two groups based on self-reported concussion history. Previous research suggests that likelihood of future cognitive impairments increases for persons with three or more reported concussions. ¹⁰ As such, the “low concussion” group included those participants with zero, one, or two reported concussions sustained during their professional playing years (n=15) and the “high concussion” group included those persons with three or more (n=12). The two groups of former football players were balanced in terms of primary playing position (linemen and skill positions). In addition, the two groups were matched with the third group, as well as with one another, in terms of age and education (Table 1). Former athletes with zero reported concussions were included in the low concussion group, because the tendency to underreport concussions makes it highly likely that these persons have sustained at least one concussion during their careers. ³⁵

Recruitment of former professional football players (Fig. 1)

Potential members of the former football player sample were recruited from the Center for the Study of Retired Athletes at the UNC at Chapel Hill. In 2001, a General Health Survey was sent to all living members of the retired players' section of the NFL Players Association (n=3729), and 2536 (68%) of the surveys were completed and returned. Each person was asked a variety of questions about musculoskeletal, cardiovascular, and neurological conditions that the retired player may have experienced during or after his football career.

OPEN IN VIEWER

FIG. 1.

Visual depiction of the process of selecting former National Football League (NFL) players for the current study. UNC, University of North Carolina.

In 2008, an additional questionnaire focused on memory and issues related to MCI was sent to a subset of 905 retirees. The subset comprised all respondents from the General Health Survey who were 50 years and older in 2008 and had valid mailing addresses. The survey also included the AD-8 Dementia Screening Interview. In addition, duplicate forms were provided for a close family member to provide their opinions of the former player's symptoms.

A criterion score was used to identify subjects experiencing some level of memory impairment, especially those experiencing increasing memory problems over the past year. An algorithm was created that identified higher (worse) scores for selected questions on the Likert-style questionnaire for symptoms that were experienced “sometimes,” “quite a lot,” and “all the time.” Results from these memory impairment scores were combined with an indicated change in memory over the past year on the AD8. The composite survey scores ranged from 0–2, with 0 indicating no memory impairment and 2 indicating severe memory impairment.

The participants selected for this study had also agreed to participate in our clinical trial studying the effects of an omega-3 fatty acid supplementation as a potential treatment intervention. As such, of those who returned these surveys (n=633), we disqualified participants with no memory complaints (i.e., a score less than 0.13; n=224), or those who had a diagnosis of Alzheimer's disease or clinical dementia (n=36). In addition, persons who regularly ate more than two servings of fish per week or who were currently taking dietary supplements such as fish oil, docosahexaenoic acid, flaxseed, or food fortified with omega-3 fatty acids were disqualified from the study (n=211).

Potential participants were also disqualified because of disinterest in the study (n=18), being over 75 years of age (n=23), claustrophobia (n=4), contraindications for MRI research (e.g., metal implant or pacemaker; n=2), or being deceased (n=10). Of the 633 former professional football players who returned the survey, a total of 105 potential participants were called and screened. While multiple measures of neuropsychological function, behavior, and neuroimaging were taken over the course of a 9-month period, the focus of this article is on the baseline fMRI measures of those persons selected for participation in our clinical trial.

Recruitment of age-and-education–matched participants

Age-and-education–matched men were recruited from the UNC Cognitive Neuroscience of Memory Laboratory's volunteer database. Additional age-matched older adults were recruited via information e-mails in the UNC system. Participants were selected based on age, sex, and education, and were screened for psychiatric disorders, psychotropic medications, and contraindications for fMRI.

Neuropsychological assessment

Former football players were given neuropsychological measures to determine their level of cognitive status (see Table 1 for a summary).

Design and procedure

The memory paradigm involved encoding and retrieval of word pairs to examine item and relational memory. ^{31 –34} Stimuli included 288 one-to-three-syllable unrelated nouns (M Freq=56.3; SD=63.5). Participants engaged in extensive practice with the behavioral paradigm before placement in the scanner. During the practice session, participants were verbally provided with instructions for the entire task.

The experimental task consisted of four study lists and four test lists. Although only the test portion of the task was scanned, the entire task took place in the MRI scanner because of the alternating study-test-study-test pattern (Fig. 2). Each study list consisted of 24 unrelated word pairs (e.g., SURGEON ARROW) for a total of 96 studied word pairs across the entire experiment. Participants viewed word pairs for 8 sec each and were asked to covertly create a sentence that incorporated both words, ensuring that the word on the left side of the screen was used in the sentence before the word on the right side of the screen (e.g., “The surgeon removed the arrow”). Participants indicated via button press that they had created and encoded a sentence on each trial. This same encoding task was used for both memory tasks (i.e., participants encoded pairs tested in the item task and pairs tested in the relational task in the same manner), and the words used for the item and relational tasks were counterbalanced across subjects.

OPEN IN VIEWER

FIG. 2.

Each retrieval phase consisted of an event-related task design with alternating blocked task periods of relational retrieval (“together previously?”) and item retrieval (“both old?”). There were four study/retrieval phases. Imaging data were acquired during the retrieval phase only.

Immediately after each study task, participants engaged in a scanned retrieval task. Each retrieval session included an event-related design constructed of test-type blocks (4 blocks for each of the four sessions for 16 retrieval blocks total) that alternated between item and relational tasks (Fig. 2). During each of the 16 retrieval blocks, participants viewed nine word pairs and three control trials.

In each item retrieval block, participants saw three pairs of words previously seen, but not together (Recombined Items), three pairs consisting of one old word and one new word (Old/New Items), and three pairs consisting of two new words (New Items; for 6 pairs of each type in each retrieval session and 24 pairs of each type total). During the presentation of each pair, participants were asked to indicate whether both words were previously seen at any point during the previous study session (but not necessarily together; “Both Old?” 1=yes; 2=no). The pair appeared on the screen until the participant made a response. The pair was immediately followed by a fixation cross so that the total amount of time spent on a trial (i.e., viewing the pair plus viewing the fixation cross) equaled 6 sec.

In each relational retrieval block, participants were presented with three pairs of words previously seen together (Intact Pair), three pairs of words previously seen, but not together (Recombined Pair), and three pairs of new words (New Pair; for 6 pairs of each type in each retrieval session and 24 pairs of each type total). Each pair was presented for a total of 6 sec, during which participants were asked to indicate via button press whether the two words were previously seen together (“Together Previously?” 1=yes; 2=no).

Each retrieval block also included 3 control trials (for 48 control trials total). During the control task, participants viewed ampersands and number signs and were instructed to indicate via button press which side of the screen the ampersands appeared. Control trials were used to introduce jitter (3 to 12 sec) during each scanner run and were included to control for neural activity associated with visual activity, reading instructions, and executing a motor response. All trials were randomized within each task block.

fMRI data acquisition

MRIs were acquired using a Siemens Allegra 3-T scanner. Participants' heads were held in place using cushions and a headrest. An initial localizing scan was followed by a high resolution T1-weighted structural scan for anatomical visualization (160 1-mm slices, TR=1750 msec, TE=4.38 msec). Next, functional scans were collected during memory retrieval. Whole brain, gradient-echo, echo planar images (50 interleaved 3-mm slices, TR=3 sec, TE=30 msec, Flip angle=90 degrees, 3×3×3 mm) were acquired at an angle parallel to the long axis of the hippocampus, identified during the T1 scan.

fMRI data analysis

The purpose of this study was to use event-related fMRI to examine long-term neural changes associated with multiple sport-related concussions. To this end, whole-brain images were acquired every 3 sec during item and relational memory tasks, capturing the average hemodynamic response for each trial individually. ⁴³ The use of event-based fMRI enabled us to examine brain activity during item memory and relational memory trials under conditions of successful retrieval (i.e., correct responses only).

Images were preprocessed and analyzed using Statistical Parametic Mapping (SPM)8 (Wellcome Department of Cognitive Neurology, London, UK) software implemented as a suite of commands in MATLAB (MathWorks, Natick, MA). Images were co-registered with each person's anatomical scan, slice-time corrected, realigned, normalized, and smoothed using a Gaussian 8-mm kernel. Only trials in which participants responded correctly were modeled and analyzed. In the current task, a participant could respond correctly by indicating “old” to re-combinations of old items in the item task (recombined item hit), “new” to a combination of an old and new item in the item task (old/new item correct rejection), “new” to new items in the item task (new item correct rejection), “old” to intact pairs in the relational task (intact pair hit), “new” to re-combinations of old items in the relational task (recombined pair correct rejection), or “new” to new pairs in the relational task (new pair correct rejection).

Recombined item hits provide a measure of an person's ability to correctly identify two words that had been studied together, which may rely exclusively on item memory retrieval. New item correct rejections provide a measure of a person's ability to recognize that two words were never studied before, again relying on item memory retrieval. Importantly, in both cases, participants are required to make decisions using both words, equating memory load between this item memory task and the relational memory task. In the old/new item correct rejections, participants are presented one word that had previously been studied and one that had not been studied. Participants may correctly respond “new” if they remember one word as new, or if they mistakenly remember both words as new. Thus, it is not possible to pinpoint how participants make their responses in this condition, and as such, the condition was not analyzed and will not be discussed further.

Intact pair hits are a measure of a person's ability to remember that two words had been presented together previously, a decision that requires relational memory processes. Similarly, recombined pair correct rejections provide a measure of a person's ability to use relational memory processes to remember that words had not been presented together, even though both words had been presented at some point during the study phase. For new pair correct rejections, persons can use either item or relational memory processes to remember that the words had not been presented during the study. This condition is included during the task to better match the item and relational tasks, but was not included in the analysis because of this mixture of item and relational processes.

In the current analysis, we were interested in the behavioral and neural responses to relational and item memory conditions. Item memory is assessed as the difference in responses between recombined item and new items, whereas relational memory is the difference in responses between intact pairs and recombined pairs. Thus, in the current study, we examined responses to recombined item hits (i.e., correct responses to recombined items), new item correct rejections (i.e., correct responses to new items), intact pair hits (i.e., correct response to intact pairs), and recombined pair correct rejections (i.e., correct response to recombined pairs). As has been previously noted, old/new item correct rejections and new pair correct rejections were included in the study for design reasons, but are not meaningful conditions to examine at the neural level. As such, although all six response conditions were included in the model at the individual subject level, only the four conditions of interest were analyzed and will be discussed.

Each of the four conditions of interest was compared to the control condition using t tests (e.g., intact hits>control trials). The results of these four t tests were entered into group t tests at the group level: recombined item hits, new item correct rejections, intact pair hits, and recombined pair correct rejections. Activity during each of these four contrasts was examined for all three groups (i.e., the age-matched older adults, the low concussion group, and the high concussion group). In addition, two-group t tests were used to examine activity in a given condition that was greater for one group than another (e.g., low concussion intact hits>high concussion intact hits).

Conjunction analyses can be used to identify regions showing significant activation in multiple groups during a particular memory condition (i.e., low concussion intact hits and high concussion intact hits). In the current study, these analyses were performed using the masking function of SPM8 to identify neural overlap in mnemonic processing across concussion groups. To create a conjunction, once contrast of interest (e.g., low concussion intact hits) was used to create an inclusive mask at p<0.025 (i.e., a mask including all voxels reaching significance at p<0.025). This mask was applied to the second contrast of interest (e.g., high concussion intact hits) at p<0.025 with an extent of at least 10 contiguously activated voxels. Using the Fisher method of estimation, ^44,45 the probability that a voxel would be activated in both contrasts (i.e., the conjoint probability) is p<0.005.

Multiple regression analyses were conducted in SPM8 at the group level to identify regions in which the raw number of reported concussions in former football players significantly predicted neural activity in our four conditions of interest. All former players were included in these contrasts, with number of concussions included as a regressor of interest. In a final analysis, accuracy was included as a covariate of interest in the two-group t test comparing persons in the low and high concussion groups. This analysis identified regions preferentially engaged by persons in each group who performed well on the memory task. All activations were considered significant at p<0.005, k≥10, because this threshold has recently been shown to be an acceptable balance between Type I and Type II error in neuroimaging studies. ⁴⁶ For all analyses, the peak Montreal Neurological Institute coordinates of active regions (reported here) were converted to Talairach space and localized in reference to the Talairach and Tournoux atlas. ⁴⁷

Behavioral results

Recombined item hits, new item correct rejections, intact pair hits, and recombined pair correct rejections were compared across the three groups. On average, participants had 14.81 recombined item hits (out of a possible 24; SD=4.84), 18.89 new item correct rejections (SD=3.72), 18.63 intact pair hits (SD=3.70), and 14.26 recombined correct rejections (SD=5.07). The two concussion groups did not differ in any behavioral measures of item or relational memory (p>0.3 for all contrasts). In addition, the two concussion groups did not differ from the third group in either their proportions of recombined item hits or new item correct rejections (i.e., item memory measures; p>0.1 for all contrasts). Both groups, however, demonstrated relational memory impairments. The high concussion group was significantly worse at recognizing intact pairs as “old” (intact pair hits) relative to the age-matched group (t(24)=2.129, p<0.05), and both the low and high concussion groups had significantly fewer recombined pair correct rejections than the age-matched group (t(27)=2.871, p<0.05 and t(24)=2.916, p<0.05 for low and high concussion groups, respectively (Fig. 3). In addition, the item or relational memory measures were not correlated with the raw number of reported concussions in former football players (p>0.2 for all contrasts).

OPEN IN VIEWER

FIG. 3.

Behavioral performance presented by concussion group (i.e., control, low, and high) and memory condition.

Imaging results

The current analysis examined activity associated with recombined item hits, new item correct rejections, intact pair hits, and recombined pair correct rejections compared with control trials. Activity during each of these four contrasts was examined for all three groups (i.e., the age-matched group, the low concussion group, and the high concussion group). In addition, two-group t tests were used to examine activity in a given condition that was greater for one group than another (e.g., low concussion intact hits>high concussion intact hits). Finally, a conjunction analysis was performed to identify regions that were commonly activated by two groups during a particular memory condition (i.e., low concussion intact hits and high concussion intact hits).

Neural activity during recombined item hits

Neural activity associated with item hits, compared with activity associated with control trials, was examined in the age-matched group, low concussion group, and high concussion group. During item hits, the age-matched group recruited bilateral parietal and frontal regions. The low concussion group recruited bilateral parietal and frontal regions, as well as a more extensive network that included bilateral lateral and medial temporal lobes (left amygdala extending into the hippocampus and right parahippocampal gyrus). The high concussion group also recruited bilateral lateral and medial temporal lobes (specifically bilateral hippocampus), as well as parietal and frontal regions (Table 2).

To test the hypothesis that concussion history was associated with alterations in memory networks, a comparison of activity was performed between groups. This analysis was conducted for the low relative to high concussion groups, eliminating the age-matched group. We did so for three reasons. First, the members of our age-matched group in the current study had highly heterogeneous activation patterns and did not recruit many episodic memory regions, even when the threshold was lowered to p<0.05. In addition, these age-and-education–matched participants did not activate any additional regions beyond those activated by the low and high concussion groups. Such results suggest that this group may be atypical in their neural recruitment. Second, the interpretation of neural differences is more straightforward when behavioral performance is equivalent, because this eliminates the possibility that observed differences are from the disparity in success rate. The behavioral memory measures and neuropsychological scores between the two concussion groups were equivalent, isolating concussion history as the only difference contributing to distinct neural recruitment. Although the age-matched older adults were significantly better than both concussion groups on some behavioral measures, the two concussion groups were equivalent in all respects. Finally, we were primarily interested in examining how sustaining multiple sport-related concussions might contribute to underlying differences in the neural architecture supporting episodic memory; the low versus high concussion group contrasts address that question directly.

The low versus high concussion group analysis examined regions commonly recruited by both groups, as well as regions preferentially recruited by each group compared with the other. The conjunction analysis revealed that the low and high concussion groups recruited many of the same regions in this analysis, including left lateral and medial temporal regions, and bilateral frontal and parietal lobes. With such a high degree of overlap, only a few regions were preferentially engaged by the low and high concussion groups. Regression analyses were also used to examine the effect of concussion history on activity. All regression analyses were also conducted with response time and accuracy as covariates, with highly similar results.

Relative to the high concussion group, the low concussion group showed greater activation in the right thalamus and the left middle frontal gyrus (Table 2 and Fig. 4). Similarly, the regression analysis identified the right thalamus and inferior frontal gyrus as regions in which activity was negatively associated with concussion history (i.e., increased concussion history was associated with decreased activity). In addition, these persons preferentially recruit bilateral lateral temporal lobes. Importantly, activity in the left middle frontal gyrus was associated with increased accuracy in this group, demonstrating that increased activity in this region in the low concussion group contributes to memory accuracy (Table 2 and Fig. 5).

OPEN IN VIEWER

FIG. 4.

Regions of activation (at p<0.005, k≥10) preferentially recruited by the low concussion group relative to the high concussion group (blue), preferentially recruited by the high concussion group relative to the low concussion group (red), and commonly recruited by both low and high concussion groups (green). In recombined item hits (a), both groups engaged medial temporal and lateral frontal regions, but differentially engaged medial frontal regions. In new item correct rejections (b), the low concussion group preferentially recruited a number of parietal regions including BA40. In intact pair hits (c), the low concussion group preferentially engaged important relational memory regions, including right parahippocampal gyrus and inferior parietal gyrus, while the high concussion group recruited frontal regions. Finally, in recombined pair correct rejections (d), the high concussion group preferentially recruited a number of parietal, frontal, and temporal regions. Color image is available online at www.liebertpub.com/neu

OPEN IN VIEWER

FIG. 5.

Regions of activation (at p<0.005, k≥10) preferentially recruited by persons with increased memory accuracy in the low concussion group relative to the high concussion group (blue) and high concussion group relative to the low concussion group (red). In recombined item hits (a) and intact pair hits (c), accuracy was associated with increased activity in frontal regions in the high concussion group to a greater extent than the low concussion group. Conversely, in new item (b) and recombined pair (d) correct rejections, accuracy was associated with increased activity in frontal and parietal regions in the low concussion group to a greater extent than the high concussion group. Accuracy was associated with temporal regions in the high concussion group relative to the high concussion group in all four memory conditions. Color image is available online at www.liebertpub.com/neu

The high concussion group showed greater activation in the bilateral anterior cingulate and left middle temporal gyrus and precentral gyrus (Table 2 and Figure 4). As in the low concussion group, regions preferentially recruited by the high concussion group, such as left middle temporal gyrus and anterior cingulate, were associated with increased accuracy (Table 2 and Figure 5). These results suggest that the two concussion groups may rely on distinct neural regions to support memory performance.

Neural activity during new item correct rejections

Neural activity associated with item correct rejections, compared with activity associated with control trials, was examined in the age-matched group, low concussion group, and high concussion group. Only the low concussion group showed significant clusters at the threshold of p<0.005. The low concussion group engaged a large bilateral network of regions, including a number of frontal, parietal, and temporal lobe regions. Of interest, this network included bilateral medial temporal lobe activity (specifically, left hippocampus and right amygdala). The low and high concussion groups commonly engaged bilateral lateral temporal lobes in this condition (Table 3 and Figure 4).

Relative to the high concussion group, the low concussion group showed greater activation in bilateral middle temporal gyrus and inferior parietal lobe (Table 3 and Figure 4). Interestingly, these regions were not preferentially associated with accuracy in the low concussion group, suggesting that increased activation of these regions is not related to memory performance (Table 3 and Figure 5). The high concussion group did not show increased activation relative to the low concussion group in any regions (Table 3 and Figure 4).

Neural activity during intact pair hits

Neural activity during correct old responses to intact pairs (relational hits) was examined for all three groups (i.e., age-matched group, low concussion group, and high concussion group) separately. Of note, this contrast for age-matched older adults revealed no significant clusters. The same contrast in the low concussion group revealed extensive bilateral activation in frontal, temporal, and parietal regions. Activity in the high concussion group was primarily left lateralized along midline structures, in addition to activity in the left lateral temporal lobe. The conjunction analysis identified regions that were commonly activated by both concussion groups. Both groups rely heavily on medial frontal regions, including the anterior cingulate.

In the comparison between groups, the low concussion group recruited more posterior regions, including temporal and parietal regions, relative to the high concussion group. Importantly, the low concussion group engaged the right parahippocampal gyrus during retrieval of intact pairs to a greater extent than the high concussion group (Table 4 and Figure 4). The only region associated with accuracy in this group, however, was the posterior cingulate (Table 4 and Figure 5). The high concussion group recruited medial prefrontal regions to a greater extent than the low concussion group (Table 4 and Figure 4), and these regions were associated with increased memory performance. In addition, activity in the left parahippocampal gyrus was associated with increased accuracy in the high concussion group (Table 4 and Figure 5).

Neural activity during recombined pair correct rejections

During correct “new” responses to recombined pairs, the age-matched group recruited left caudate, thalamus, and middle frontal gyrus, as well as right precentral gyrus. The conjunction analysis revealed that the right inferior parietal lobe and middle temporal gyrus were commonly recruited by the low and high concussion groups in this task (Table 5 and Figure 4).

The low concussion group recruited right parietal regions, with the supramarginal gyrus as the only region showing greater activation for the low concussion group relative to the high concussion group (Table 5 and Figure 4). A wide range of frontal and parietal regions were associated with increased accuracy in the low concussion group to the high concussion group (Table 5 and Figure 5). The high concussion group engaged a much more extensive bilateral network of regions, including frontal, temporal, and parietal regions. Relative to the low concussion group, persons with a high number of previous concussions showed greater activation in a number of midline regions, as well as a region in the inferior parietal lobe and lateral temporal lobe (Table 5 and Figure 4). Notably, only the right lateral temporal lobe was associated with increased accuracy in the high relative to the low concussion group (Table 5 and Figure 5).

Limitations

A secondary goal of the current study was to compare former NFL players with a group of healthy, age-matched adults. The paradigm used in the current study has been used successfully in a number of previous experiments with healthy older adults ^31,34 and young adults, ³³ revealing a number of regions recruited during memory tasks. As such, this task has been shown as a reliable method of activating both item and relational memory networks in both young and older adults. Specifically, older adults recruit right inferior parietal lobe and right medial temporal lobe during relational relative to item memory, and frontal regions in item relative to relational memory. ³⁴

We hypothesized that our healthy older adults would also activate these regions and to a greater extent than our former football players. The age-matched adults in the current study, however, did not activate these regions, suggesting that the age-matched older adults in our study were atypical in some way. As such, any direct comparisons between the age-matched group and either concussion group are uninformative. Results from such contrasts could be caused by functional changes associated with concussion history, or by an unexpected and unexplained difference between our age-matched group and the true healthy older adult population. Because our data cannot distinguish between these options, our analysis was limited to identifying differences between concussion groups (i.e., our first research question).

Investigating the long-term effects of sport-related concussion is necessarily accompanied by the loss of experimental control. For instance, our study relies on self-report of concussion history and subsequent symptoms. Although it is true that most athletes will underreport exposure to concussion, it is likely that this pattern is systematic across individuals. Our study also does not capture the subconcussive insults sustained by this group of former players, although it is very likely that those with a higher number of concussions are most prone to a higher exposure of impacts to the head, leading to a greater number of subconcussive insults. The current study divides participants into two distinct groups that are likely to be equivalent in their tendency to underreport concussions and subconcussive insults. This has been a consistent limitation of studies involving self-reported concussion history, but all of these studies have accepted the assumption that this cohort is clearly more at risk for having sustained repetitive subconcussive head traumas compared with their age matched counterparts in the general population.

Because of the highly specialized sample (i.e., former NFL players who are physically able and willing to participate in a 1 h MRI scan), our sample size was small in this study. Such small sample sizes can lead to low power in MR studies and, consequentially, an inability to identify activity in some conditions. Future studies are needed to replicate and extend our findings, but the results of the current study introduce an important first step in understanding the long-term function effects of multiple sport-related concussion.

Finally, the sample of former players in our study was biased by selection criteria because the study was interested in examining persons who expressed some degree of recent memory problems. Specifically, former players were recruited based on their self-reported mild memory impairments in their daily lives. As such, differences between athletes and healthy age-matched adults may not be surprising; however, the current study was primarily interested in differences between groups of former NFL players. Importantly, we found important neural differences between the low and high concussion groups who were equivalent on all measures, including selection criteria. Identifying differences in the underlying neural correlates of relational memory, even when behavioral performance was equivalent, suggests that significant functional reorganization may be occurring because of repeated sport-related concussion. In addition, we disqualified persons who reported taking supplements before the study began, potentially excluding a subset of persons who were already concerned about memory changes. Although we do not believe that including this subset would have altered our results, it is possible that our findings would not generalize to persons who have taken action to reduce memory impairment.