Disrupted flexible use of context-dependent relational memory in adults following moderate-severe traumatic brain injury

Abstract

Learning associative information and extracting regularities from that remembered information to adaptively meet goals is a hallmark of navigating life. Adaptive goal-directive behavior has been historically attributed to prefrontal functions, and more recently to hippocampal relational memory. Disruptions in either of these systems, both frequently seen in Traumatic Brain Injury (TBI), have far reaching consequences in everyday life. In the current study, we investigate the impact of chronic, moderate-to-severe TBI on both relational memory processes as well as the ability to use regularities or rules extracted from that remembered information to guide behavior via both overt responses and eye-tracking. Individuals with and without TBI completed a context-dependent relational memory task designed to assess both 1) the formation and organization of overlapping relational associations (hippocampal-dependent); and 2) the acquisition and flexible use of learned, context-dependent rules (ventromedial prefrontal-dependent). Behavioral measures revealed that relative to neurotypical matched comparison participants, participants with TBI were significantly impaired on context-dependent relational memory measures, but showed spared memory guided rule-use. Eye-tracking data indicated largely intact information gathering at study for participants with TBI, but impaired flexibility at test leading to poor behavioral outcomes. Critically, these data suggest that relational memory impairment is a significant source of behavioral dysfunction in TBI, which likely contributes to poor outcomes in both laboratory testing and real-life, long-term trajectories following injury. Furthermore, this study highlights the feasibility and strength of incorporating eye-tracking into studies of TBI to gain novel insights into information gathering and use across time.

1. Introduction

As information is accumulated over time, we not only remember specific learned associations (e.g., my new co-worker’s name is Jade and her office is on the 3^rd floor), but also more global or general regularities of our environments (i.e., the office tends to be colder after the weekend so I should bring a sweater), both of which can be used to adaptively influence our future behavior. The learning of specific associations relies on relational memory, which is critically supported by the hippocampus and related structures in the medial temporal lobe, and is characterized as the ability to bind together information across domains and delays into coherent memory representations that can be used both in the moment and into the future (Cohen, 2015; Eichenbaum & Cohen, 2014; Maguire & Mullally, 2013; Moscovitch et al., 2016; Verschure et al., 2014). These processes are, of course, essential as we navigate life, allowing us to encode new information, update representations in the presence of new information, remember learned information about the world around us, and use that information to shift behavior based on situational and contextual demands.

Traumatic brain injury results in diffuse neural pathology and heterogeneous patterns of cognitive deficit (Covington & Duff, 2021). Impairments in memory, however, are the most reported and frequently targeted in rehabilitation (Cicerone et al., 2011; Vakil, 2005; Wilson, 1998). Memory deficits are present deep into chronic phase of injury (Draper & Ponsford, 2008) and are associated with poor long-term psychosocial outcomes (i.e., navigating life; Kekes-Szabo et al., 2023). That memory deficits are highly prevalent following TBI is unsurprising given the vulnerability of the larger memory network to injury mechanisms. For example, several frequently occurring pathophysiological consequences of TBI (e.g., hypoxia, seizure activity) disproportionately affect the structure and function of the hippocampus, making hippocampal damage one of the most likely consequences of injury (Palacios et al., 2013; Sharp et al., 2014; D. F. Tate & Bigler, 2000).

While memory complaints are ubiquitous among individuals with a history of TBI, it is now well understood that memory is not a unitary construct, and thus characterizing the types or aspects of memory affected, and the extent of the impairment, in TBI is an active area of study (Rigon et al., 2019; Vakil, 2005). A growing body of evidence (e.g., Dulas et al., 2022; Morrow et al., 2020; Rigon et al., 2019, 2020) suggests that TBI may impact relational memory abilities disproportionately. Relational memory is negatively affected in chronic TBI, and impairments can manifest increasingly in the years and decades after even mild brain injury. Monti et al. (2013) reported that young adults who had experienced a mild TBI in adulthood showed neither deficits on an item-context binding relational memory task nor changes in hippocampal volume compared to those without a history of TBI. However, middle aged adults who had sustained a mild TBI as young adults showed a significant deficit in relational memory task performance and also had significantly smaller hippocampi than matched neurotypical comparison participants. This study documented the long-term impact of mild TBI on relational memory decades after injury emphasized that importance of assessing memory abilities in the chronic period.

Several recent studies have documented impairment from moderate-severe TBI on relational memory performance across a variety of tasks for different domains (i.e., spatial vs. temporal relational memory) and across varying delays (i.e., immediate vs. delayed test). Morrow and colleagues (2020) showed that individuals with chronic, moderate-severe TBI were impaired on a continuous version (Hannula et al., 2006) of the item-context task used by Monti et al. (2013), where study trials and test trials were interleaved so that participants were either tested immediately or after a moderate delay. Compared to a matched neurotypical comparison group, memory for the TBI group was significantly impaired both immediately and after a delay, demonstrating relational memory deficits on both long-term and working memory timescales.

Rigon and colleagues (2020) compared performance between individuals with and without a history of moderate-severe TBI on a spatial reconstruction task in which they had to learn the spatial locations of 5 objects and, after a brief (4-second) delay, reconstruct the spatial layout that they had just studied. The TBI group was significantly impaired on putting the right objects back in the right locations, and also made significantly greater errors in how far they placed each object from its original studied location. Dulas and colleagues (2022) demonstrated a TBI-related impairment in temporal relational memory. They compared performance between individuals with and without a history of TBI on a temporal order task in which objects were presented one at a time in groups of 3, 5, 7, or 9. After an object set was presented, participants were immediately asked to report back the order in which the items had appeared. Compared to demographically matched non-injured comparison participants, the TBI group showed memory impairment at all set sizes >3. Taken together, these studies demonstrate the impairment produced by TBI on both spatial and temporal relational memory.

While relational memory theory has long identified the hippocampus as a major hub of relational processing (Cohen & Eichenbaum, 1993; Eichenbaum & Cohen, 2001), recent expansions of this theory (e.g., Rubin et al., 2017; Wang et al., 2015) have highlighted a complementary role for prefrontal regions. In particular, the ventromedial prefrontal cortex regions has been shown to support the extraction of regularities from individual memories in order to abstract underlying rules or schemas that can be used to guide behavior (Badre & Nee, 2018; Kroes & Fernández, 2012; Miller, 1999a, 1999b). Indeed, ventromedial prefrontal processes support strategic control of memory retrieval – this type of control is engaged when, for example, contextual information determines what specific information one should retrieve (Preston & Eichenbaum, 2013; Schwarb et al., 2019; Wang et al., 2015). To return to an earlier example, while the hippocampus is essential for learning that Jane’s office is on the 3^rd floor, ventromedial prefrontal cortex is essential for me to extract information about office temperature on multiple Mondays compared to multiple Wednesdays so that I learn to bring a sweater after the weekend. We have previously referred to this ability to extract regularities from distinct remembered events and use that information to flexibly guide behavior based on situational demands as “memory guided rule-use” (Schwarb et al., 2019). Importantly, like the medial temporal lobe, prefrontal deficits have also long been emphasized in the literature on traumatic brain injury as being particularly vulnerable to the pathophysiological mechanisms of injury (e.g., Adams et al., 1985; Bigler, 2013; Hayes et al., 2016; Levin & Kraus, 1994) Accordingly, here we approach TBI as a possible window for exploring the critical shared and unique contributions of prefrontal and medial temporal structures in various aspects of relational memory and memory guided rule-use.

While the literature on the effects of TBI on relational memory continues to grow, the literature on memory guided rule-use in TBI remains scant. However, there are several clues from related work that provide insights about how we might expect individuals with a history of chronic, moderate-severe TBI to use memory guided rule-information to shape performance. Participants with a history of moderate-severe TBI are impaired on learning regularities in artificial grammars, even when they have learned individual grammatical chunks (Pothos & Wood, 2009). Similarly, TBI groups are impaired on implicit sequence learning tasks, demonstrating that they are less able to learn a repeating sequence of stimuli over time (Schwizer Ashkenazi et al., 2021; Vakil et al., 2002). Together, these data suggest that participants with a history of TBI may be impaired at identifying task regularities extracted from remembered information. Additionally, participants with a history of moderate-severe TBI are typically impaired on tasks of source memory or context-dependent memory when asked to explicitly retrieve source/context information (for review, see Vakil, 2005). Therefore, we expect that participants with chronic, moderate-severe TBI may struggle to use contextual information to arbitrate between competing alternatives learned over time. Furthermore, it remains unclear whether moderate-severe TBI results in deficits to item-context memory, integration of information across contexts, or both.

The goal of the current research was to characterize the impact of chronic moderate-to-severe TBI on distinct facets of memory performance. To this end, we implemented a context-dependent relational memory task (Schwarb et al., 2015) that we have shown in our previous work to have robust behavioral indices of the distinct, complementary contributions from hippocampal- and prefrontal-dependent systems (Schwarb et al., 2015, 2019). In this task (Figure 1), participants learn that “people” (i.e., pictures of unique faces) belong in unique “rooms” in two distinct “buildings,” differentiated only by color. Each face is associated with one room in each of the two building contexts (e.g., face 1 belongs in room A of the blue apartment building context, but room 11 of the purple office building context), thus requiring learning of two partially overlapping face-room associations for each face. Unbeknownst to participants, the assignment of faces to rooms has a consistent underlying structure, with female faces always on the left in one building context, and always on the right in the other (and vice versa for males). While participants are never informed of this underlying structure, over the course of the experiment they may extract information about these more general, context-dependent structural regularities or rules and use that information to make appropriate responses at test. At test, participants must put each studied face back into its associated room in the context-appropriate building. Accurately placing a face in its appropriate studied room is a measure of relational memory. Previous work using eye-tracking has demonstrated dynamic interactions between which rooms participants consider in the presence and absence of relevant contextual information (i.e., building color; Schwarb et al., 2015) demonstrating participants’ ability to flexibly consider multiple overlapping relational representations before selecting the context-appropriate room. Additionally, sometimes participants are asked to place novel faces into an appropriate room. Participants’ ability to place a novel face in a room consistent with the underlying context-dependent rule structure is a measure of memory guided rule-use. This behavior demonstrates that participants can use context-dependent rule information flexibly when presented with new, never before encountered information.

Figure 1.

Example (A) study and (B) test trials for the context-dependent relational memory task used in this study. Slides outlined in red represent when eye-movements were recorded.

Our previous work with this task combined behavioral measures with converging neuroimaging measures to show that memory for specific face-room associations engages different neural structures than memory for context-dependent regularities and memory guided rule-use. In an fMRI study of non-injured adults, remembering specific studied face-room associations uniquely engaged the hippocampus, while applying structural regularities to guide performance in novel (i.e., unstudied) situations engaged the ventromedial prefrontal cortex (Schwarb et al., 2019). These findings were replicated in a structural imaging study where face-room memory performance correlated with magnetic resonance elastography (MRE)-derived measures of hippocampal integrity as well as diffusion tensor imaging (DTI) derived measures of fornix (i.e., white matter tract that connects the hippocampus) integrity. Similarly, using information about global regularities in task structure to guide performance when presented with novel information correlated with MRE-derived measures of ventromedial prefrontal integrity and DTI-derived measures of uncinate fasciculus (i.e., a white matter tract connecting medial temporal and prefrontal structures) integrity (Schwarb et al., 2019). These data document the dissociable roles of hippocampus and prefrontal cortex in integrating information across multiple episodes, extracting regularities, and flexibly guiding behavior in the face of novel information. Thus, while neuroimaging data were not collected as part of the current study, these prior data, dissociating behavioral indices of hippocampal and ventromedial prefrontal contributions to performance, allow us to make inferences about the potential underlying deficits in these distinct regions in TBI using the same task.

In the current study, we tested individuals with chronic, moderate-severe TBI using this context-dependent relational memory task. As both medial temporal and ventromedial prefrontal neural systems are vulnerable to impairments in TBI, the goal of this study was to characterize behavioral performance for individuals with a history of moderate-to-severe TBI on these dissociable behavioral measures that depend more heavily on either hippocampal or ventromedial prefrontal processes. Both behavioral choice responses and eye-tracking behaviors were evaluated. Eye-tracking is a powerful tool for investigating memory phenomena because while behavior responses are binary (remember or don’t remember), eye-tracking can provide information about the ways in which information was gathered during encoding and also which choice competitors were considered at test. Further, eye-tracking patterns in non-injured individuals have been previously characterized in this task specifically, with findings converging with the behavioral and imaging results (Schwarb et al., 2015, 2019). The inclusion of both behavioral and eye movement outcome measures here allowed us to evaluate task performance at multiple scales to characterize the scope of potential dysfunction in TBI for both information gathering and information use. We hypothesized that we would see deficits in both relational memory and memory guided rule-use for individuals with chronic, moderate-severe TBI.

2. Method

2.1. Participants

Participants were 36 adults with moderate-severe TBI (18 females; 18 males) and 36 demographically matched neurotypical comparison (NC) participants (18 females; 18 males). They were recruited from the Vanderbilt Brain Injury Patient Registry (Duff et al., 2022) and the community using social media ads and flyers. All participants were 18 or older to limit the effects of developmental changes and were younger than 55 to conservatively limit the effects of expected age-related cognitive decline. All participants had normal or corrected-to-normal hearing and vision per lab screening protocols or self-report. All participants had to have stable eye-tracking calibrations to be included in the final data set. Four additional individuals with TBI (1 male, 3 female) and one additional NC individual (female) participated but were excluded because stable eye-tracking calibrations could not be attained. All study activities were approved by the Vanderbilt University Medical Center Institutional Review Board (IRB #201647; approved November 25, 2020) and all participants gave their informed consent.

All 36 participants with TBI (Table 1) sustained a moderate-severe TBI as determined using the Mayo Classification System (Malec et al., 2007). All met at least one of the following criteria: (1) Glasgow Coma Scale (GCS) <13 within the first 24 hours of acute care admission, (2) positive neuroimaging finding (acute CT findings, or lesions visible on chronic MRI), (3) loss of consciousness (LOC) >30 minutes, or (4) post-traumatic amnesia (PTA) >24 hours. Injury information was determined from available medical records and a semi-structured participant interview. Loss of consciousness information was available for 34 participants. Post-traumatic amnesia information was available for 35 participants. Acute imaging information was available for 35 participants (34 with positive findings). GSC was available for 32 participants (median = 10.2, ranging from 3 to 15). See Table 1 for demographic and injury information for participants with TBI.

Table 1.

Demographics and injury information for participants with TBI

ID	Age	Edu	Etiology	TSO	LOC	Neuroimaging	GCS	PTA
5003	25–30	18	Ped vs. Auto	15	N/A	SDH	11	>24 hours
5005	30–35	16	MVA	22	LOC >30 minutes	SAH; SAH; IVH	14	>24 hours
5011	39–44	12	Fall from height	48	N/A	SAH; frontotemporal contusion; EDH	15	>24 hours
5013	30–35	18	Ped vs. Auto	26	No LOC	SAH	15	< 24 hours
5014	48–53	16	MVA	184	LOC >30 minutes	N/A	N/A	>24 hours
5016	20–25	16	MVA	20	LOC >30 minutes	SAH	13	>24 hours
5018	36–41	18	MVA	156	LOC >30 minutes	SAH	3	>24 hours
5020	49–54	16	MCC	68	LOC >30 minutes	Right frontal SAH	N/A	>24 hours
5027	27–32	16	Ground-level fall	11	LOC >30 minutes	SAH	9	>24 hours
5029	20–25	14	Non-motorized vehicle	11	LOC < 30 minutes	SDH; IPH; SAH	14	< 24 hours
5031	50–55	14	Struck by object	7	No LOC	SDH; SAH; IPH	13	N/A
5038	38–43	16	Ground-level fall	20	LOC >30 minutes	SDH; Multifocal hemorrhages; Post-traumatic hemorrhagic contusions	N/A	>24 hours
5040	39–44	12	MVA	77	LOC >30 minutes	SDH; SAH; Uncal herniation	3	>24 hours
5041	32–37	16	MVA	93	No LOC	No acute intracranial findings	10	>24 hours
5044	23–28	12	Non-motorized vehicle	75	LOC < 30 minutes	SDH; IPH	15	>24 hours
5046	45–50	18	Non-motorized vehicle	51	LOC < 30 minutes	SAH	14	>24 hours
5050	29–34	18	Ground-level fall	20	LOC >30 minutes	IPH	15	< 24 hours
5051	51–56	16	MVA	35	LOC < 30 minutes	SAH; SDH	14	< 24 hours
5052	29–34	14	MVA	37	LOC < 30 minutes	SDH; SAH	9	>24 hours
5056	23–28	12	Non-motorized vehicle	33	LOC >30 minutes	Hemorrhagic shear injury	11	>24 hours
5057	24–29	12	MVA	18	No LOC	SDH	N/A	No
5058	32–37	12	MCC	109	LOC < 30 minutes	SAH; SDH; PCH	8	>24 hours
5068	25–30	16	Fall from height	69	LOC < 30 minutes	ICH	3	>24 hours
5079	37–42	18	MVA	114	LOC >30 minutes	PCH; SAH	5	>24 hours
5086	33–39	16	Ped vs. Auto	126	LOC >30 minutes	SAH	15	< 24 hours
5104	36–41	20	Struck by object	37	LOC < 30 minutes	SDH; scattered SAH; right temporal hemorrhage	15	< 24 hours
5109	25–30	14	MVA	115	LOC >30 minutes	SDH; IPH; IVH	5	>24 hours
5126	44–49	12	MVA	35	LOC >30 minutes	SDH	3	>24 hours
5129	52–57	12	Other	18	LOC < 30 minutes	SDH; SAH	12	< 24 hours
5131	40–45	12	MVA	21	LOC >30 minutes	SDH	12	>24 hours
5133	24–29	12	MCC	32	LOC < 30 minutes	Contusions; SDH; IVH	15	< 24 hours
5137	25–30	16	Ped vs. Auto	17	LOC >30 minutes	EDH; SDH: SAH	3	>24 hours
5141	26–31	12	MVA	18	LOC >30 minutes	SDH	13	< 24 hours
5161	23–28	12	MVA	8	LOC >30 minutes	SDH; PCH; Diffuse axonal injury	10	>24 hours
5164	46–51	16	Fall from height	13	LOC >30 minutes	SDH	3	>24 hours
5166	32–37	12	MVA	8	LOC >30 minutes	SAH	7	>24 hours

ID = participant ID number. Age is presented as a five year range to protect participants’ identities. Education (edu) reflects years of highest degree obtained. MVA = motor vehicle accident. MCC = both motorcycle and snowmobile accidents. Ped vs. auto = participant was hit by car while walking or running. Time since onset (TSO) is presented in months. Loss of consciousness (LOC) is presented in minutes. SDH = subdural hematoma. SAH = subarachnoid hemorrhage. IPH = intraparenchymal hemorrhage. IVH = intraventricular hemorrhage. PCH = parenchymal hemorrhage. EDH= epidural hematoma. ICH= intracerebral hemorrhage. Glasgow Coma Scale (GCS) is total score at time of first post-injury measurement. PTA = post-traumatic amnesia. N/A = information was unknown or not available.

Participants with TBI were in the chronic phase of injury (>6 months post-injury, mean time since injury = 4.1 years (SD: 3.8 years)). Thus, participants’ neuropsychological profiles were stable (Salmond et al., 2006). All participants sustained their injuries in adulthood (i.e., after age 18), so no participants sustained developmental TBIs. No participant had a history of neurological or cognitive disability prior to sustaining their brain injury. Injury etiologies included motor vehicle accidents (n = 16), motorcycle/snowmobile accidents (n = 3), ground-level falls (n = 3), non-motorized vehicle accidents (e.g., bicycle; n = 4), falls from height (n = 3), being hit by a car while walking (n = 4), being struck by a moving object (n = 2), and other (n = 1). Aphasia was ruled out via clinical assessment by a certified speech-language pathologist.

NCs had no history of neurological or cognitive disability. To determine eligibility, NC participants completed a medical history screening to rule out diagnoses and medications that can interfere with cognition (e.g., neurological or psychiatric conditions, developmental or learning disorders, untreated diabetes, or sleep apnea). NCs and were matched to the TBI group on age, sex, and education. The groups did not differ statistically on age (TBI mean = 35.6 years (SD: 9.7), NC mean = 32.2 years (SD: 9.0; t(70) = 1.57, p = 0.121) or years of educational attainment (TBI mean = 14.8 years, (SD: 2.5); NC mean = 15.7 years (SD: 2.1); t(70) = −1.74, p = 0.086).

2.2. Neuropsychological Testing

To characterize the cognitive abilities of the sample, participants completed six Cognition Measures from the NIH Toolbox battery (Gershon et al., 2013) to assess episodic memory (task: picture sequence memory), executive functioning (tasks: Flanker and dimensional change card sort), language (task: picture vocabulary), working memory (task: list sorting), and processing speed (task: pattern comparison). See Weintraub et al. (2013, 2014) for task descriptions. The NIH Toolbox Cognitive Battery has established validity for individuals with TBI (Carlozzi et al., 2017; Tulsky et al., 2017) and is recommended for TBI research (e.g., NIH Common Data Elements).

Tasks were completed on an iPad on a separate day from the main context-dependent relational memory task (average time between the battery session and context-dependent relational memory task session = 36.3 weeks). The total battery took approximately 45 minutes. All participants in the TBI group and 32 participants in the NC group completed the neuropsychological battery. NC participants who did not complete the NIH Toolbox tasks did so because they were either unable to return to the lab to complete the assessment or had moved away. Age-corrected standard scores (National Institutes of Health, 2021) were derived for each of the six tasks. Independent samples t-tests were calculated to identify TBI vs. NC group differences.

Consistent with the well-documented long-term cognitive impairments following moderate-severe TBI (Ponsford et al., 2014; Rabinowitz & Levin, 2014), the participants with TBI performed significantly poorer than the NC group on all subscales: picture sequence memory (TBI: 105.25, NC: 117.25, t(66) = −2.91, p = 0.005, d = −0.71), Flanker (TBI: 79.39, NC: 88.66, t(66) = −3.04, p = 0.003, d = −0.74), dimensional change card sort (TBI: 93.33, NC: 112.47, t(66) = −4.74, p < 0.001, d = −1.15), picture vocabulary (TBI: 106.61, NC: 115.81, t(63.2) = −3.65, p < 0.001, d = −0.87), list sorting (TBI: 101.83, NC: 114.62, t(66) = −3.91, p < 0.001, d = −0.95), and pattern comparison (TBI: 97.89, NC: 113.67, t(66) = −3.36, p = 0.001, d = −0.82).

2.3. Context-Dependent Relational Memory Task

2.3.1. Apparatus.

Eye-tracking data were collected at 1000hz using a Eyelink1000 table mounted eye-tracking system (SR Research, Ontario, Canada). Participants were positioned in a chin rest 65cm from the computer monitor while completing the experiment. All stimuli were presented on an 18”×14” monitor. The experiment was administered using Presentation software (Neurobehavioral Systems, http://nbs.neuro-bs.com) on a Windows computer.

2.3.2. Stimuli and Design.

The task has been described in detail elsewhere (Schwarb et al., 2019). Context stimuli included three 21-room buildings (3 floors each with 7 rooms) sized to 22.9° × 13.2° of visual angle. The three buildings were blue, purple, and gray in color. The gray building had consistent shading throughout; for the blue and purple buildings there was a color gradation radiating out from the middle column (Figure 1). Face stimuli included 64 male and 64 female faces, all faces were selected from our face database (Althoff & Cohen, 1999) and presented in full color. Four male and four female faces were used during the practice block, all other faces were used during the eight experiment blocks. Each face appeared in two sizes; a small version that was sized to 1.3° × 1.3° visual angle used in the study phase, and a large version that was sized to 3.3° × 3.3° of visual angle that was used in the test phase. Six unique male and six unique female faces were included in each of the eight experimental blocks.

In each experimental block, 16 unique rooms were used. Thus, on a given block, each room was only occupied by one face. To ensure that there were no intrinsic spatial biases that could impact performance, we sought to sample each individual room an equal number of times. However, to account for all 128 study trials, two rooms (i.e., the center room on each side of the building) were sampled one additional time across the whole experiment.

2.3.3. Procedure.

Each block was divided into a study phase and a test phase (Figure 1). Eye-tracking calibration and validation was performed prior to each block to ensure accurate eye movement measurements. On each study phase trial, a face was presented in one room in one of the two buildings. The color of the building served as the context for distinguishing in which room a given face belonged. Participants were instructed to treat the blue building as an apartment building and the purple building as an office building (see instructions below). A total of 4 male and 4 female faces were included in each study phase. After all eight face-room pairs were studied in the first building context, those same faces were studied again in different rooms in the second building context (e.g., first Face 1 was studied in Room A of the blue apartment building context, then later, Face 1 was studied in Room F of the purple office building context). After all face-room associations had been studied once, study was repeated two more times such that each face-room pair was studied a total of three times before test. On each study trial, the face remained on the screen for 3000ms and participants clicked the left mouse button to indicate that the face was on the left side of the building and the right button to indicate that the face was on the right. Trials were separated by a 250ms intertrial interval.

Face assignment was determined by an underlying gender-by-side rule structure that differed between the two building contexts but ensured that male and female faces were always on opposite sides of each building. Participants were not told about the underlying rule. For example, male faces always appeared on the left and female faces always appeared on the right in one of the building contexts. Then, the rule was switched in the other building contexts such that male faces always appeared on the right and female faces always appeared on the left. Thus, if Face 1 was a male face, it was always on the left in one building context and the right in the other building context.

Each test phase trial was divided up into four phases. First a face was centrally presented for 3000ms. Next the gray building was presented for 3000ms, followed by one of the two color buildings for 3000ms. Finally, the face reappeared with the colored building and participants were asked to use the mouse to put the face back in the room in which it had been studied. The trial ended when the participant made a response. During the testing phase each of the eight studied faces was tested in one of the two buildings (half in the blue building and half in the purple building). In addition to the eight studied faces, four novel faces were also included. Participants were instructed that if they did not recognize a face, they should put it where they thought it fit best. Eye-tracking data were collected when the gray and colored houses were presented.

All participants completed a single practice block prior to starting the experimental blocks. The practice block was identical to the experimental blocks except no novel faces were included at test. Face stimuli from the practice block were not repeated in the experimental blocks.

2.3.4. Task instructions.

Prior to the study phase and study phase practice, participants were provided with the following instructions: “In this experiment there are two buildings: A blue apartment building and a purple office building. During the learning period, you will see different people appear in different rooms of each building. Every person belongs in one room of the blue apartment building and a different room of the purple office building. For example, this woman lives in this room of the apartment building [woman shown in one room of the blue apartment building]. But she works in this room of the office building [same woman shown in a different room of the purple office building]. You will see every person 3 times in each building. When you see a person appear, your job is to try to learn where the person belongs. Also, if the person is on the left side of the building, click the left mouse button; if the person is on the right side of the building, click the right mouse button. Do you have any questions?”

Prior to the test phase and test phase practice, participants were provided with the following instructions: “Next we will ask you to try to put each person back into the room where they live in the blue apartment building and where the work in the purple office building. If you are not sure, make a guess! We are very interested in the types of errors that people make on this task.

During the test, first you will see a person [show example face]. Next you will see a gray building, feel free to look around the building and try to remember in which room you have seen that person [show example gray display]. Finally, either the blue apartment building or the purple office building will appear, continue to look around [show example display]. After a few seconds, the person will appear again and you can use the mouse to put the person back in the room where you think he/she belongs [show example display]. Sometimes you will see a face that you have never seen before. When this happens, just put that person in the room where you think they fit the best. Do you have any questions?”

2.4. Outcome measures and statistical analysis

Behavioral data were analyzed as follows:

Relational memory performance was assessed by computing the proportion of Accurate Studied Face Placements a participant made (i.e., frequency with which a studied face was placed in the correct “Target-Room”). Face placements on inaccurate trials were also evaluated. Placements were grouped into three regions of interest: Target-Room Adjacent responses (i.e., a room directly next to the studied room), Other Context-Correct responses (i.e., a room on the appropriate side of the building given the gender of the face and the building context, but not adjacent to the studied room), and Context-Incorrect responses (i.e., a room on the side of the building opposite to the studied location). Group differences in responding patterns on inaccurate trials were assessed with a Placement Type (proportion of responses that were to Target-Adjacent vs. Other Context-Correct vs. Context-Incorrect rooms) × Group (TBI vs. NC) Repeated measures ANOVA and follow up post hoc t-tests.

Memory Guided Rule-Use was assessed by computing the proportion of rule-consistent placements made for novel faces that had never been seen. Group differences were determined using independent samples t-tests.

To characterize potential differences in first-level eye-tracking measures, total number of fixations made and the duration of those fixations across the whole study were calculated for each group separately and independent samples t-tests were performed to assess group differences.

Eye movements were evaluated separately for study and test phases of the study. At study, when a face was visible on the screen, two particular patterns of eye movements were considered. The first was the proportion of viewing to that face (i.e., the Target-Room) compared to all other fixations. The second was the proportion of viewing to the Other-Studied-Room on the context-incorrect side of the building (i.e., the room where the visible face had been previously studied in the other building context) compared to all fixations that were not to the Target-Room. Study phase data were analyzed only for trials from the third study repetition when participants had maximum knowledge of individual face-room associations. Group differences were again evaluated using group independent samples t-tests. Correlations were also computed comparing behavioral accuracy with proportion of viewing to the face and proportion of viewing to the Other-Studied-Room for each group separately.

At test, for studied face trials when participants went on to be correct, proportion of viewing to the context-correct studied room and the context-incorrect Other-Studied-Room were computed separately for the gray building period and the color building period. Group (NC vs TBI) × Room (Target-Room vs. Other-Studied-Room) × Context (color vs. gray building) repeated measures ANOVA was performed on the proportion of viewing measure. Significant interactions were followed up with independent-samples t-tests. For novel face trials, proportion of viewing to the rule-consistent side of the building was calculated, and independent samples t-tests were used to identify group differences.

At test, for studied face trials when participants went on to be wrong, proportion of viewing to the context-correct studied room, the context-incorrect Other-Studied-Room, as well as the chosen room (i.e., the room that the participant erroneously selected) were computed separately for the gray building period and the color building period. Only participants who made 10 or more incorrect responses were included to provide sufficient power for stable eye movement data. As such, three participants were excluded from each group. A Group (NC vs TBI) × Room (Target-Room vs. Other-Studied-Room vs. Chosen-Room) × Context (color vs. gray building) repeated measures ANOVA was performed on the proportion of viewing measure. Significant interactions were followed up with independent-samples t-tests. Levene’s test for equality of variance was considered and adjustments made where appropriate. Effect sizes are reported throughout. Statistical significance is reported at p < 0.05.

3. Results

3.1. Task Behavior

3.1.1. Accurate Studied Face Placement:

Accuracy for putting the correct face in the Target-Room was 45.6% (SD: 23.1%) for the TBI group and 59.8% (SD: 19.2%) for the NC group (chance = 5.6%); both groups performed significantly above chance, t > 10.0, p < 0.001 (Figure 2A). Accuracy was significantly worse for the TBI group compared to the NC group, t(70) = −2.83, p = 0.006, d = −0.67. These data suggest that relational memory was significantly impaired in the TBI group.

Figure 2.

(A) Bars represent mean accurate studied face placements with 95% confidence intervals shown. Each dot represents an individual participant’s performance. (B) Percentage of Rule-Consistent Face Placements for novel faces. Each dot represents an individual participant’s performance. Asterisk indicates statistical difference less than p = 0.05. Error bars = 95% confidence intervals.

3.1.2. Inaccurate Studied Face Placement:

When a studied face was not put in the Target-Room, it was put on the rule-consistent side of the building 67.9% (SD: 23.0%) of the time for the TBI group and 78.2% (SD: 24.5%) of the time for the NC group (chance = 50%); t > 4.0, p < 0.001, d > 0.75 for both groups (Figure 2B). The group difference was not statistically significant, t(70) = −1.85, p = 0.069, d = −0.44. These data suggest that when participants are unsure of the appropriate face-room pair, they may be using the underlying structural regularities to guide their performance and therefore chose a rule-consistent room most of the time. An alternate explanation is that participants may have low resolution relational representations of specific face-room pairs and rather than selecting the appropriate studied room, they instead select one of the rooms adjacent to the studied room. To arbitrate between these possibilities, incorrect trails were sorted into Target-Room Adjacent responses (i.e., a room directly next to the studied room), Other Context-Correct responses (i.e., a room on the appropriate side of the building given the gender of the face and the building context, but not adjacent to the studied room), and Context-Incorrect responses (i.e., a room on the side of the building opposite to the studied location). For the TBI group, 35.6% (SD: 16.5%) of placements were Target-Room Adjacent, 33.8% (SD: 14.0%) of placements were to an Other Context-Correct room, and the remaining 30.6% (SD: 22.8%) of placements were to a Context-Incorrect room. For the NC group, 49.3% (SD: 20.8%) of placements were Target-Room Adjacent, 20.9% (SD: 18.9%) of placements were to an Other Context-Correct room, and the remaining 21.8% (SD: 24.5%) of placements were to a Context-Incorrect room (Figure 3).

Figure 3.

Proportion of studied face placements to either a Target-Room adjacent room, a non-adjacent context-correct room, or a context-incorrect room on incorrect trials. Each dot represents an individual participant’s performance. Asterisks indicate statistical difference less than p = 0.05. Error bars = 95% confidence intervals.

A Placement Type (Target-Adjacent vs. Other Context-Correct vs. Context-Incorrect) × Group (TBI vs. NC) Repeated measures ANOVA revealed a significant main effect of Placement Type, F(1.8,122.9) = 8.4, p < 0.001, η_p² = 0.11, as well as a significant Placement Type × Group interaction, F(1.8,122.9) = 4.4, p = 0.018, η_p² = 0.06. Post hoc comparisons revealed that the NC group made significantly more Target-Room Adjacent placements than either Other Context-Correct, t(35) = 3.9, p < 0.001, d = 0.65, or Context-Incorrect placements, t(35) = 4.0, p < 0.001, d = 0.67. These data suggest that about half of the incorrect responses made by the NC group can be explained by a less precise relational representation leading them to make a placement right next to the studied room. There was no difference between Other Context-Correct and Context-Incorrect placements, t(35) = 1.1, p = 0.273, d = 0.19. There were no significant differences between placement types for the TBI group (t < 1.0, p > 0.40 in all cases). Only the between group simple effect of Target-Adjacent placements was significant, t(70) = 3.1, p = 0.003, d = 0.73, the groups did not differ on either Other Context-Correct or Context-Incorrect placements (t < 1.6, p > 0.10 in both cases)

3.1.3. Accurate Novel Face Placement:

As an assessment of whether participants were able to use underlying structural regularities of the task to guide behavior, novel face trails were considered. Proportion of Novel Face placements to the rule-consistent side for the given building context was 68.8% (SD: 20.0%) for the TBI group and 76.2% (SD: 22.0%) for the NC group (chance = 50%); t > 5.0, p < 0.001, d > 0.90 for both groups (Figure 2B). The group difference was not significant, t(70) = −1.51, p = 0.136, d = −0.36. These data suggest that the TBI group was not impaired relative to the NC group and that both groups were able to learn and use the underlying structural regularities to guide performance.

3.2. Eye-tracking

3.2.1. General.

We first examined general eye movement allocation to the displays including the number of fixations and viewing time. This was to ensure there were no observed group differences in eye movement behaviors that may have been the result of brain injury and that were independent of the task manipulations. No group differences were identified¹. Therefore, difference in eye movement patterns as a function of task were further investigated.

3.2.2. Study Phase.

On each trial of the study phase, a face appeared on the screen in the Target-Room; participants, unsurprisingly, directed most of their viewing to that face². Indeed, by the third study repetition, the proportion of viewing to the face was 81.7% (SD: 9.6%) and 77.8% (SD: 11.4%) for the TBI and NC groups respectively. There was no significant group difference, t(69) = 1.59, p = 0.059, d = 0.38.

Previously we have shown that during study, as participants learn information about the two different rooms in which a face belongs, they shift their eyes away from the face on the screen and toward the Other-Studied-Room on the other side of the building (Schwarb et al., 2015). To evaluate if this pattern exists in the current sample, we calculated the proportion of viewing to the Other-Studied-Room for all fixations that were not to the Target-Room. The proportion of viewing to the Other-Studied-Room was 15.5% (SD: 16.6%) for the TBI group and 22.4% (SD: 17.1%) for the NC group³. Proportion of viewing was greater than chance (5.9%) for both groups, ts>3, ps <.001, and the group difference was significant, t(69) = −1.72, p = 0.045, d = 0.35.

To assess the relationship between study phase looking behavior and relational memory outcomes, two correlations were performed for each group (Figure 4). Data from one NC participant was excluded from the analyses as their proportion of viewing to the Target-Room was greater than three standard deviations below the mean. The first correlation compared Accurate Studied Face Placements and proportion of viewing to the visible face in the Target-Room. The correlation was significant for both the TBI group, r = −0.46, p = 0.005, and the NC group, r = −0.40, p = 0.018. Thus, when participants in both groups were willing to move their eyes away from the face on the screen and sample other rooms in the building, they had better memory at test. The second correlation compared Accurate Studied Face Placements and the proportion of viewing to Other-Studied-Room. The correlation was significant for the TBI group, r = 0.59, p < 0.001, but not the NC group, r = 0.33, p = 0.051. These data suggest that individuals who spent less time looking at the visible face and/or more time looking at the Other-Studied-Room during study performed better on the memory task at test. Indeed, the proportion of viewing to the visible face in the Target-Room and proportion of viewing to the Other-Studied-Room was indeed negatively correlated for both the TBI group, r = −0.76, p < 0.001, and the NC group, r = −0.36, p = 0.034.

Figure 4.

(A) Correlation between Accurate Studied Face Placements (i.e., faces that were correctly placed back in their studied location) and proportion of viewing to the face on the screen and (B) correlation between Accurate Studied Face Placements and proportion of viewing to the Other-Studied-Room compared to all fixations that were not to the Target-Room for TBI group (red/light symbols) and NC group (blue/dark symbols) participants. The eye movement data include all study trials from the third study repetition when participants have had the full opportunity to learn each individual face-room association. Asterisks indicate statistically significant correlations at p < 0.05.

3.2.3. Test phase viewing duration, studied faces, correct trials.

During the gray building period, for the TBI group the proportion of viewing to the context-correct studied room (i.e., Target-Room) and proportion of viewing to the context-incorrect studied room (i.e., the Other-Studied-Room) was 20.8% (SD: 13.7%) and 13.4% (SD: 9.4%) respectively. For the NC group, Target-Room proportion of viewing was 21.2% (SD: 11.2%) and proportion of viewing to the Other-Studied-Room was 19.0% (SD: 10.3%). During the color building period, for the TBI group Target-Room proportion of viewing was 33.8% (SD: 14.9%) and Other-Studied-Room proportion of viewing was 7.1% (SD: 4.2%). For the NC group, Target-Room proportion of viewing was 40.3% (SD: 19.0%) and Other-Studied-Room proportion of viewing was 7.7% (SD: 4.6%; Figure 5).

Figure 5.

Proportion of viewing to the context-correct studied room (Target-Room; solid lines) and the context-incorrect studied room (Other-Studied-Room; dotted lines) during both the gray building and color building periods of the test phase. Error bars = 95% confidence intervals. Asterisks indicate statistically significant post hoc comparisons at p < 0.05.

A Group (NC vs TBI) × Room (Target-Room vs. Other-Studied-Room) × Context (color vs. gray building) repeated measures ANOVA was performed. The main effect of Group, Group × Room, and Group × Context interactions were not significant, F < 3.0, p > 0.14 in all cases. The main effect of Context was significant, F(1,70) = 25.2, p < 0.001, η_p² = 0.27, with greater proportion of viewing to studied rooms during the color building period. The main effect of Room was also significant, F(1,70) = 133.3, p < 0.001, η_p² = 0.66, with greater proportion of viewing to the context-correct compared to context-incorrect studied room. Notably, the 3-way interaction was also significant, F(1,70) = 6.6, p = 0.012, η_p² = 0.09, indicating that viewing patterns during the test period differed between the two groups. Post hoc paired samples t-tests were performed to characterize the source of the interaction. For the NC group, during the gray building period, proportion of viewing to the two rooms did not differ, t(35) = 1.34, p = 0.263, d = 0.19, but this difference was significant during the color building period, t(35) = 10.20, p < 0.001, d = 1.70. These data suggest that both rooms were similarly considered during the gray building period and viewing shifted to the correct Target-Room and away from the Other-Studied-Room once contextual information was available. For the TBI group, proportion of viewing to the two rooms did significantly differ for both the gray building period, t(35) = 2.80, p = 0.009, d = 0.46, and the color building period, t(35) = 10.98, p < 0.001, d = 1.83. These data suggest that the TBI group already preferred one of the two rooms during the gray building period when they went on to be correct and were less likely than participants in the NC group to consider both rooms associated with a given face at test.

3.2.4. Test phase viewing duration, studied faces, incorrect trials.

For incorrect trial viewing patterns, three regions of interest were considered: The context-correct studied room (i.e., Target-Room), the context-incorrect studied room (i.e., the Other-Studied-Room), and the chosen (incorrect) room (i.e., the room that was selected at the end of the trial; Chosen-Room).

During the gray building period, for the TBI group the proportion of viewing to the Target-Room was 5.8% (SD: 4.1%), proportion of viewing to the Other-Studied-Room was 10.5% (SD: 7.4%), and proportion of viewing to the Chosen-Room was 15.5% (SD: 9.4%). For the NC group, the proportion of viewing to Target-Room was 6.7% (SD: 5.0%), proportion of viewing to the Other-Studied-Room was 14.1% (SD: 9.1%), and proportion of viewing to the Chosen-Room was 15.3% (SD: 11.8%). During the color building period, for the TBI group Target-Room proportion of viewing was 6.9% (SD: 3.8%), Other-Studied-Room proportion of viewing was 5.7% (SD: 4.0%), and Chosen-Room proportion of viewing was 26.8% (SD: 9.0%). For the NC group, Target-Room proportion of viewing was 8.8% (SD: 4.7%), Other-Studied-Room proportion of viewing was 8.8% (SD: 6.8%), and Chosen-Room proportion of viewing was 29.6% (SD: 12.6%); Figure 6).

Figure 6.

Proportion of viewing to the Chosen-Room (long dash lines), the context-correct studied room (Target-Room; solid lines) and the context-incorrect studied room (Other-Studied-Room; dotted lines) during both the gray building and color building periods of the test phase for Incorrect trials. Error bars = 95% confidence intervals. Asterisks indicate statistically significant post hoc comparisons at p < 0.05.

A Group (NC vs TBI) × Room (Target-Room vs. Other-Studied-Room vs. Chosen-Room) × Context (color vs. gray building) repeated measures ANOVA was performed. The main effects of Group, F(1,63) = 5.9, p = 0.018, η_p² = 0.09, Room, F(1.5,98.2) = 80.4, p < 0.001, η_p² = 0.56, and Context, F(1,126) = 59.4, p < 0.001, η_p² = 0.49, were all significant as was the Context × Room interaction, F(1.6,98.2) = 59.3, p < 0.001, η_p² = 0.49. No other interactions were significant (F<2.5, p>0.10 in all cases). To interpret the significant Context × Room interaction, data were collapsed across Group for post hoc analyses (Figure 6). In the gray building period, participants looked to the Target-Room less than either the Chosen-Room or the Other-Studied-Room, t > 4.5, p < 0.001, d > 0.60, in both cases. Viewing to the Chosen-Room and Other-Studied-Room did not significantly differ, t(64) = 1.6, p = 0.055, d = 0.20. During the color building period, participants looked to the Chosen-Room more than either the Target-Room or the Other-Studied-Room, t >12.0, p < 0.001, d > 0.15, in both cases. Viewing to the Target-Room and Other-Studied-Room did not significantly differ t(64) = −0.69, p = 0.245, d = 0.09.

3.2.5. Test phase viewing duration, novel faces.

Proportion of viewing to the rule-consistent side of the building was computed for each group again for correct trials only (Figure 7). NC participants distributed 60.4% (SD: 17.4%) of their viewing to the future rule-consistent side during the gray building period and 81.2% (SD: 8.9%) during the color building period. TBI participants distributed 60.8% (SD: 17.5%) of their viewing to the rule-consistent side during the gray building period and 76.4% (SD: 10.1%) during the color building period. A Group (NC vs TBI) × Context (color vs. gray building) repeated measures ANOVA was performed. The main effect of Context was significant, F(1,70) = 93.5, p < 0.001, η_p² = 0.57. Neither the main effect of Group nor the Group × Context interaction was significant, F<2, p>.16 in both cases. These data indicate that viewing patterns did not differ between the groups for novel faces suggesting that context-dependent viewing did not differ in the absence of relational memory information. Finally, while it may seem surprising that rule-consistent viewing was above chance for both groups in the Gray Building (when no context information was available), this pattern was specific to this analysis that included only correct trials. Indeed, when all trials were included, proportion of rule-consistent viewing was 46.9% (SD: 7.4%) for the TBI group and 50.4% (SD: 9.8%) for the NC group during the Gray Building period. While additional research is necessary, it is possible that with only 4 novel face trials per block, participants were able to predict the likely upcoming context on some trials leading to increased rule-consistent viewing. Previous work with a non-TBI sample has shown that this predictive looking is common in a variant of this context-dependent relational memory task (Schwarb et al., 2015).

Figure 7.

Proportion of viewing to the rule-consistent side of the building on trials with a novel face during the gray building period (left) and color building period (right). Means are depicted with diamond icon and 95% confidence intervals. Each dot represents an individual participant’s proportion of viewing.

3.2.6. Context-dependent relational memory task outcomes compared to NIH Toolbox tasks.

Because many studies investigating the relationship between TBI history and cognitive outcomes have used measures from the NIH Toolbox (e.g., Holdnack et al., 2017; Picon et al., 2022; Tulsky et al., 2017), in an exploratory analysis we compared performance on the context-dependent relational memory task with standard outcome measures from the NIH Toolbox. The goal of these analyses was to determine if the context-dependent relational memory task captures overlapping or separable aspects of cognitive functioning. We first correlated NIH Toolbox measures with Accurate Studied Face Placements, our measure of relational memory. For the NC group, Accurate Studied Face Placements significantly correlated with the picture sequence task (r = 0.57, p < .0001) and the picture vocabulary task (r = 0.42, p = 0.02), but not any of the other tasks (p>0.10 in all cases). For the TBI group, the relational memory measure significantly correlated only with the picture sequence task (r = 0.48, p = 0.003), but not any of the other tasks (p > 0.10 in all cases). The picture sequence task measures episodic memory, so it is unsurprising that that task correlates with relational memory accuracy in the current study. Our measure of Memory Guided Rule-Use (i.e., the proportion of rule-consistent placements made for novel faces) did not correlate with any of the NIH Toolbox measures for either the NC or TBI groups (p > 0.20 in all cases). The Memory Guided Rule-Use measure in the current task appears to capture cognitive processes and variability in performance that are not measured by the NIH Toolbox tasks.

4. Discussion

Individuals with moderate-severe TBI can exhibit a range of cognitive deficits, often resulting in difficulty navigating life. Such individuals report difficulty integrating back into the community and independently participating in vocational, leisure, interpersonal, and daily living activities (Kekes-Szabo et al., 2023; Pugh et al., 2018; R. L. Tate & Broe, 1999). A variety of cognitive processes (and the associated neural regions) that are critical for supporting aspects of flexible cognition like relational memory and memory guided rule-use are particularly vulnerable to the pathophysiological mechanisms of brain injury. The current work applied a context-dependent relational memory task (Schwarb et al., 2015) that provides robust behavioral indices of distinct and complementary contributions from medial temporal- and ventromedial prefrontal-dependent systems (Schwarb et al., 2015, 2019) to characterize patterns of spared and impaired performance relevant to altering behavior depending on the context cues among individuals with a history of moderate-to-severe TBI.

Findings here showed that for individuals with chronic moderate-severe TBI, relational memory was significantly impaired, and participants were less able to put studied faces back in their studied (target) rooms (i.e., Accurate Studied Face Placement) than individuals without a history of brain injury. The TBI group also showed differences compared to the NC group in how they directed their eye movements to the Target-Room at test, even on trials where they went on to be correct, further details of which will be discussed below. Together these findings demonstrate an impairment in relational memory abilities (here, for specific face-room associations) in TBI, in line with a growing body of work demonstrating that deficits in relational memory are prevalent and persistent in the chronic phase of TBI (e.g., Dulas et al., 2022; Monti et al., 2013; Morrow et al., 2020; Rigon et al., 2020).

This is the first study to our knowledge to examine relational memory performance in a task that combines behavioral and eye-tracking measures in TBI. Eye-tracking has become a valuable tool in measuring memory processes in both neurotypical and clinical populations to gain novel insights into information gathering and use across time (e.g., Hannula et al., 2010; Nickel et al., 2015; Ryan & Shen, 2020). We note that while the cranial nerves supporting a range of oculomotor functions can be susceptible to the pathophysiological impacts of TBI, we successfully collected eye movement data from 36 individuals with moderate-severe TBI who passed our calibration procedures (four TBI participants and one NC participant failed our calibration procedures and were excluded from the study). Findings from the completed study revealed no differences between the TBI and NC group in such first-order measures of viewing as number of fixations to and amount of time spent fixating on the experimental display. These data show that eye-tracking methods for studying memory processes are tractable for many individuals with moderate-severe TBI. Further, the group similarities and differences in eye movement behaviors highlighted below suggest that such an approach provides a novel window into the nature, scope, and time course of the observed deficit in relational memory in TBI.

During the study phase, group differences observed in the eye-tracking data were minimal. Participants in both groups unsurprisingly spent the great majority of their viewing time (greater than 80%) looking at each study face presented singly on the screen. When participants shifted their eyes away from the face, for both groups viewing time was largely allocated to fixating on the Other-Studied-Room (i.e., the room in which that particular face had been studied in the other building), with the NC group engaging in this viewing behavior significantly more than TBI group. Interestingly, this latter viewing behavior correlated significantly with later memory outcomes for the TBI group, although the relationship was similar and approached significance for the NC group. TBI participants who spent more time looking at the currently empty Other-Studied-Room had higher accurate studied face placement (i.e., had better memory for individual face-room associations) at subsequent test. Thus, it appears that this viewing behavior may benefit subsequently-tested relational memory for individuals with and without a history of TBI.

Previous work with this task has suggested that as participants learn overlapping relational memory representations (i.e., Face 1 belongs in Room A in the blue apartment building context and Face 1 also belongs in Room F in the purple office building context) they form triadic memory representations (i.e., Face 1 belongs in both Rooms A and F; Schwarb et al., 2015). We believe that this behavior is reflected in the study time eye movements in the current study. Because each face-room association was studied three times, with additional repetitions, participants spent less time looking at the face on the screen and also shifted their eyes to look at the other room where they remember that the face belonged. Participants who engaged in this verification behavior (i.e., this face also belongs in this other room) appeared to have more precise representations about the two appropriate rooms associated with a given face and therefore can successfully put the face back in its studied room when presented with a specific building context.

Group differences in eye movements also emerged during the test phase, revealing group differences in how participants approached the task. While memory performance is often binary (i.e., either a correct or incorrect response was made), eye movement data provide novel insights as to which competitors were considered by participants before making their selection. During the gray building period, when no contextual information is available, NC participants tended to distribute their viewing time equally between both rooms in which a given face had been studied. Then, after contextual information was revealed with presentation of the colored building, they strongly shifted their viewing to the context-correct studied room. By contrast, participants with a history of TBI tended to favor a single studied room during the gray period building, which they would go on to endorse once the contextual information became available. These data, we suggest, represent further evidence of relational memory deficit in participants with a history of TBI whereby they were either unable to access both learned room associations for a given face or were unable to flexibly consider both relevant rooms and therefore preferentially looked to only one room even in the absence of the discriminative contextual information.

While relational memory impairments have been shown on a variety of other tasks (Dulas et al., 2022; Morrow et al., 2020; Rigon et al., 2020) very few studies have investigated the impact of TBI on memory guided rule-use. Extracting regularities from remembered episodes to inform future decisions is a hallmark of navigating life. Memory guided rule-use in the current study was largely intact among individuals with a history of chronic, moderate-severe TBI. The TBI group showed no difference in their ability to place novel faces in a manner consistent with the underlying rule structure of the task and their pattern of eye movements were also similar when making decisions about novel faces. This was surprising given that participants with a history of TBI are impaired on measures of implicit learning (e.g., Cotrena et al., 2014; Pothos & Wood, 2009; Skidmore, 2015) and also struggle to incorporate contextual information into their memory representations (cf. Vakil, 2005). Future work will need to investigate this point more thoroughly to better understand, and predict, patterns of spared and impaired rule learning and use in individuals with TBI. Characterizing the specific anatomical regions impacted by injury in the study sample may provide important insights. While our previous work associated memory guided rule-use with the ventromedial prefrontal cortex (Schwarb et al., 2019) and the ventromedial prefrontal regions are vulnerable to TBI (e.g., Adams et al., 1985), it is possible that this region was relatively spared in the current sample. These results, however, are also intriguing in that perhaps individuals with chronic moderate-severe TBI, or subgroups based on anatomical characters, could benefit from interventional programs that take advantage of these intact cognitive processes (see Wilson, 1998; Ylvisaker et al., 2003; Ylvisaker & Feeney, 1998).

Linking relational memory impairments to disruptions in flexible, adaptive, goal-directed behavior has significant implications for rehabilitation. For example, the hippocampal relational memory system benefits from a range of lifestyle and non-invasive treatments such as neuromodulation (e.g., Wang et al., 2014), nutrition (e.g., Monti et al., 2014), sleep (Stickgold & Walker, 2005), and physical activity interventions (e.g., Erickson et al., 2011). While there has been a historical emphasis in the TBI literature between prefrontal structure impairment and behavioral dysfunction in flexible cognition, a broader approach that also includes targeted intervention of the hippocampal relational memory system, offered alone or in combination with other treatments, holds tremendous promise for improving long-term trajectories after TBI. Exploring the relative contributions of medial temporal cortex and prefrontal cortex to relational memory deficits, and their impacts on flexible cognition, in TBI has, to date, been under-investigated. The current study begins to fill this gap.

An interesting future direction would be to examine performance on our context-dependent relational memory task, and the fate of different types of memory representations, over time. Most lab-based memory experiments, including ours here, asks participants to encode and explicitly retrieve or use newly acquired information in a single session. Yet, much of the information we use to flexibly guide our behavior in the world is accumulated, updated, and consolidated over time. Indeed, previous work has shown that performance deficits significantly increase following days and weeks for individuals with hippocampal-dependent relational memory impairments (Morrow et al., 2023; Schapiro et al., 2019). These findings raise the possibility that a range of both hippocampal-dependent and hippocampal-independent memory processes may require the hippocampus for consolidation over time. This suggests that future investigations may show that performance on aspects of our task that relied on more general task regularities, and that were not impaired in TBI relative to NCs, may become significantly different from NCs if tested over time.

Finally, a second important way in which the current work goes beyond many previous studies investigating the impact of TBI on cognition, is that the intentional design of the current task requires the contribution of multiple processes and systems for successful performance. Many experimental and neuropsychological approaches to understanding behavioral dysfunction in TBI attempt to isolate a single behavior, cognitive domain, or neural system to investigate. The context-dependent relational memory task used here, however, differs from that approach in that it affords, within a single task, assessment of multiple behavioral outcomes (memory for specific learned associations vs. global regularities) that have been linked in prior work to distinct neural systems (hippocampal vs. ventromedial prefrontal) at multiple scales. Using this task, we found here that, relative to the NC group, individuals with moderate-severe TBI were impaired on aspects of task performance that required relational memory ability. However, they were not impacted on aspects of task performance that relied on memory guided rule-use derived from task regularities to guide behavior. The success of this task in revealing impaired and spared ability is promising for future work that aims to identify the relative contribution of a distinct system to a complex behavior, and for its subsequent treatment, even in the context of diffuse disruption as is the case in TBI.

To tie these findings to specific brain systems, we can turn to previous neuroimaging work in non-injured adults with this task. We have demonstrated dissociable roles of hippocampus and prefrontal cortex, such that the ability to remember specific studied face-room associations uniquely engaged the hippocampus, whereas extracting and applying structural regularities to guide performance uniquely engaged the ventromedial prefrontal cortex and the connections between medial temporal and prefrontal systems (Schwarb et al., 2019). While neuroimaging data were not collected as part of the current study, impairments here in TBI selectively on measures linked to relational memory ability and hippocampal function are consistent with a growing body of work demonstrating deficits in relational memory in moderate-severe TBI (Dulas et al., 2022; Morrow et al., 2020; Rigon et al., 2020) and with well-documented disruptions in hippocampal structure and function after TBI (Palacios et al., 2013; Sharp et al., 2014; D. F. Tate & Bigler, 2000). While impaired relational memory was pervasive here, the TBI group fared well on aspects of the task that were less dependent on learned associations and more reliant on global or general regularities of the task design that, as noted above, are linked more to prefrontal cortex than to the hippocampus. Together these data support the idea that relational memory ability is critical to flexible, adaptive, goal-directed behavior (i.e., Cohen, 2015; Hannula & Duff, 2017; Rubin et al., 2017) and we propose relational memory impairment is a significant source, and predictor, of behavioral dysfunction and poor outcomes not just in laboratory testing but in real-life long-term trajectories after TBI (e.g., affecting vocational, interpersonal, and daily living outcomes; Kekes-Szabo et al., 2023).

Highlights.

• Chronic moderate-to-severe severe TBI impairs relational memory

• Ability to extract and use global task regularities relatively spared in TBI

• Eye movements are an effective tool for investigating memory deficits with TBI

• Memory impairment is a major source/predictor of behavioral dysfunction in TBI