Virtual Reality Water Maze Navigation in Children with Neurofibromatosis Type 1 and Reading Disability: an Exploratory Study

Abstract

Deficits in visuospatial skills have long been considered a hallmark characteristic of the neurofibromatosis type 1 (NF1) cognitive profile. Yet, whether previously used measures reveal the full nature of these visuospatial deficits in NF1 remains unclear. An exploratory study was conducted using a virtual reality water maze developed to serve as a human analog of the Morris water maze animal model. Children with NF1 were compared to children who did not have NF1 (typically developing; TD), with the two groups matched on reading skill due to the common occurrence of reading difficulties (RD) in NF1. Metrics of virtual navigation revealed that the TD group (n = 14) generally outperformed NF1 (n = 17) on learning trials of the water maze on multiple measures (path length, latency, Gallagher cumulative search error). Results remained significant even when controlling for IQ and working memory. However, the final (probe) trial showed no significant difference between the groups. These preliminary results suggest that though NF1 initially performs poorly on virtual navigation, with additional practice this group resembles TD children. These findings prompt future avenues of research to further explore potential unique relations between visuospatial skill and reading skill in NF1 as well as to further delineate individuals with NF1 from their typically developing peers.

Neurofibromatosis Type 1 (NF1) is a genetic disorder with estimates of occurrence in approximately 1 in 3000 individuals (Friedman, 1999). This autosomal dominant disorder results from an inherited affected allele in roughly half of cases, while the other half are a result of new mutations (Littler & Morton, 1990; North, 1993). Mutations are located on chromosome 17 at 17q11.2 (Barker et al., 1987; Seizinger et al., 1987) and result in disrupted function of the protein neurofibromin. Within the Ras signaling cascade, neurofibromin is involved in negatively regulating cellular proliferation and differentiation, with dysregulation of the pathway leading to a broad range of neurocutaneous manifestation (Monroe et al., 2017). Additionally, disruption of neurofibromin can result in increased GABA transmission, potentially engendering deficits in long-term potentiation (LTP) that may impact learning and cognition (Anastasaki et al., 2022). Although many of the genetic aspects of NF1 are understood, determination of NF1 is generally diagnosed based on clinical criteria that include café au lait spots, axillary or inguinal freckling, Lisch nodules, neurofibromas, optic pathway glioma, bone deformation, and having a parent with an NF1 diagnosis (Legius et al., 2021). These physical characteristics of NF1 vary in terms of extent and expression, with significant heterogeneity present (Monroe et al., 2017). Despite considerable variation in NF1 phenotype, learning deficits and cognitive impairments are common, occurring in about half of all individuals with NF1 (Hyman et al., 2005; Levine et al., 2006). However, similar to the heterogeneity of physical manifestations in NF1, the cognitive/academic profile of those with NF1 is also variable, with no clear deficit profile.

Despite the variable presentation of cognitive/academic deficits in NF1, there are some consistencies; notably, the majority of individuals with NF1 struggle with academic performance (approximately 70%; Brewer et al., 1997; Doser et al., 2022; Lehtonen et al., 2013). Executive function deficits, particularly in working memory and planning/problem solving, as well as visuospatial deficits (Clements-Stephens et al., 2008; Hyman et al., 2005; Krab et al., 2008; North, 2000; Schrimsher et al., 2003), are considered to be consistent weaknesses in NF1 (Beaussart et al., 2018). Indeed, visuospatial deficits have been long considered as a hallmark characteristic of NF1 (Hyman et al., 2005). Poor academic performance has been shown to co-occur with visuospatial deficits in NF1 (Brewer et al., 1997; Cutting & Levine, 2010), with our prior research having shown that children with NF1 with reading difficulties (RD) had significantly poorer visuospatial skills than TD children who showed similar weaknesses in reading (Barquero et al., 2015). Yet, it remains unclear whether the reading deficits in NF are intrinsically linked to visuospatial deficits.

While visuospatial deficits are a relatively consistent deficit in NF1, few studies have fully explored the visuospatial deficits in NF1. Most commonly, visuospatial impairment in NF1 has been evidenced by lower performance on visuospatial tasks presented as two-dimensional (on a printed page) stimuli, such as the Judgement of Line Orientation (JLO; Benton et al., 1983) and the Developmental Test of Visual Perception (DTVP; Hammill et al., 1993) assessments (Barquero et. al., 2015; Clements-Stephens et al., 2008; Cutting et al., 2000; Hyman et al., 2006). Interestingly, while children with NF1 appear to have intact visual form discrimination abilities, they may be experiencing impaired top-down integration of visual information (Van Eylen et al., 2017), a somewhat nebulous construct that is difficult to tap through simplistic means.

Further exploration of visuospatial skills with more complex tasks that include navigation and learning may provide additional insight into the profile of visuospatial skills in NF1, which may be important for linking mouse models with human findings. Previously, a mouse model of NF1 demonstrated deficits in visuospatial learning during the Morris Water Maze task (Costa et al., 2002), a classic animal (murine) spatial learning task. The Morris Water Maze task is performed by placing a rodent in a tank of opaque water in which it learns to locate a hidden platform and climb out of the water. Once the animal has learned where the platform is over a series of training trials, the platform is removed and retention of the platform’s location is inferred from the animal’s swim path (Morris, 1981; Morris et al., 1982). Additional work has shown that spatial learning deficits in a mouse model of NF1 can be rescued through treatment with lovastatin (Li et al., 2005), which downregulates the Ras signaling pathway and is thought to counteract the excessive GABA-mediated inhibition of synaptic plasticity and LTP.

In an effort to demonstrate the same effect in humans, visuospatial learning in NF1 has been examined in two studies using a standard computer monitor to display task graphics for an Arena Maze task (Ullrich et al., 2010, 2020). First, in a pilot study comparing children with NF1 (n = 10) with siblings without NF1 (n=6), Ullrich et al. (2010) showed that some (but not all) children with NF1 had more difficulty in early search trials relative to their unaffected (without NF1) siblings. Further, Ullrich and colleagues (2010) found that during the probe trial (with the target removed), unaffected siblings of those with NF1 spent more time in the target quadrant, an indication of remembering target location. A subsequent study compared youth with NF1 who were assigned to either the lovastatin treatment or placebo group (Ullrich et al., 2020). Findings indicated that the treatment group showed no improvement in maze navigation metrics relative to the placebo group. Though useful in depicting the complexities of spatial problem solving, such human maze simulations are limited in their representation of physical navigation as they are typically presented on a computer screen (Thornberry et al., 2021). Thus, it remains unclear whether a navigational environment that more closely approximates the three-dimensional nature of the water maze could provide clearer information. Further, neither of the studies by Ullrich and colleagues took into consideration whether the NF1 group also had learning difficulties.

Of particular interest for the current study, reading difficulties are the most common form of learning difficulty in the general population, and this is also true among children with NF1 (Orraca-Castillo et al., 2014). RD has been associated with NF1 across multiple studies (Arnold et al., 2018; Chaix et al., 2018; Cutting & Levine, 2010; Hyman et al., 2005). Children with NF1 and RD have been shown to resemble TD children with RD including deficits in performance on phonological awareness, rapid naming, and reading comprehension measures (Cutting & Levine, 2010). Yet, children with NF and RD showed marked visuospatial deficits when compared to TD individuals (with and without RD), as evidenced by lower performance on visuospatial tasks presented as two-dimensional (on a printed page) stimuli, such as the Judgement of Line Orientation (JLO) and the Developmental Test of Visual Perception (DTVP; Hammill et al., 1993) (Barquero et al., 2015; Clements-Stephens et al., 2008; Cutting et al., 2000; Hyman et al., 2006), even when controlling for IQ (Cutting & Levine, 2010). What remains unclear is how closely the underlying causes of RD in children with NF1 resemble those of TD, and whether RD in NF1 may be attributable to features of NF1, such as visuospatial deficits.

Traditionally, visuospatial weaknesses are not thought to be a central causal deficit in RD (White et al., 2019). However, this remains controversial (White et al., 2019), with some studies suggesting that the visuospatial skills of individuals with RD may not be equivalent to individuals who are TD. For example, a recent study suggested that virtual maze navigation may be impaired in RD (Gabel et al., 2021). Indeed, our own work has shown that while not statistically significant, TD with RD tend to have somewhat lower two-dimensional visuospatial skills, albeit still within the average range, when compared to TD individuals (Barquero et al., 2015). Nevertheless, studies comparing NF1, have shown differences. Of specific relevance, those with NF1 and RD have significantly poorer two-dimensional visuospatial skills than TD individuals with RD (Barquero et al., 2015).

In sum, while prior studies indicate that individuals with NF1 have visuospatial deficits compared to TD individuals (with and without reading difficulties), a more granular understanding is necessary to effectively understand the visuospatial functioning in those with NF1, particularly in terms of attempting to link NF1 mouse model findings with cognitive profiles in humans with NF1. Therefore, the present exploratory study aimed to investigate visuospatial processing skills, as measured by performance on a virtual three-dimensional Morris Water Maze task, while also targeting individuals with NF1 who have reading difficulties. To our knowledge, no studies have investigated the visuospatial skills of those with NF1 using a virtual three-dimensional navigation task while also considering reading performance levels in both the NF1 and TD comparison group. Given (limited) evidence that children with NF1 underperform on computer water maze tasks compared to TD siblings (Ullrich et al., 2010), and children with NF1 underperform on two-dimensional visuospatial tasks compared to TD with RD (Barquero et al., 2015), we hypothesized that the NF1 group would underperform relative to TD in the virtual reality water maze. Further, though not consistent in the literature, some studies have shown that NF1 samples exhibit visuospatial deficits even when accounting for IQ (Cutting & Levine, 2010; Descheemaeker et al., 2013; Hyman et al., 2005; Krab et al., 2008); accordingly, we hypothesized that the underperformance would be largely independent of overall intellectual functioning (IQ). Finally, given that recent studies have suggested that the visuospatial deficits in NF1 could, in part, stem from deficits in cognitive control skills, including working memory (Beaussart et al., 2018), we hypothesized that observed differences would be due, in part, to working memory.

Methods

Participants

A total of 31 (11 female) children and adolescents, average age 11.72 years (SD = 3.23), with either NF1 (n = 17) between the ages of 8 and 17 years or TD (n = 14) between the ages of 8 and 16 years, all with evidence of reading difficulty, participated in the current study. Reading difficulty (for both NF1 and TD) was defined by standard scores equal to or less than 90 (25th percentile) on measures of basic word reading skills. The current study was part of a larger, ongoing longitudinal study of NF1 and academic difficulties, which explores academic (reading) intervention in conjunction with (and without) lovastatin treatment for the NF1 group in effort to elucidate the most efficacious intervention methods for these unique poor readers. Accordingly, the study design provides comparison of participants with NF1 to a TD group with comparable reading skills, effectively controlling for reading skill. All participants in the NF1 group had a clinical diagnosis of NF1, as confirmed by medical records reviewed by a medical geneticist provided prior to study enrollment. All participants were native English speakers, and all testing took place at a research university in the southeastern US. Due to the targeted population for recruitment (children and adolescents with NF1), participants in the NF1 group were recruited nationally in the USA, whereas TD participants were recruited locally. Participants were recruited through educational clinics, neuropsychology clinics, neurofibromatosis advocacy and support organizations, word of mouth, and the clinicaltrials.gov website. Participants were enrolled in the study if they met the inclusion criteria of MRI compatibility, normal or corrected-to-normal vision, normal hearing, no known history of spina bifida, cerebral palsy, and/or traumatic brain injury, and received no treatment with any psychotropic medication except for stimulant medications for ADHD. Participants were not excluded for ADHD, autism symptomology, or mild to moderate intellectual disability, which are known to commonly co-occur with NF1. All participants reported their race and ethnicity. Of those, 81% reported as White, 13% as Black/African American, 6% self-identified as more than one race, and 100% reported ethnicity as Not Hispanic/Latino. During this experiment, participants continued medications as usual, including medications for ADHD and depression. One NF1 participant took medication due to a history of seizures. All participants were compensated as part of the larger study. For NF1 participants, travel expenses were partially or fully covered. The Institutional Review Board of Vanderbilt University approved all procedures.

Measures

All currently reported measures were taken at visit 1, prior to intervention. As part of the larger study, a battery of behavioral measures was administered. IQ, word-level reading skill, reading comprehension, oral language skills, working memory, and visuospatial measures were selected for inclusion in the present study, as well as a parent rating of executive function. Individual measures are described below.

WASI-II

The four-subject version of the Wechsler Abbreviated Scale of Intelligence, Second Edition (WASI-II; Wechsler, 2011), including the Block Design, Similarities, Matrix Reasoning, and Vocabulary subtests, was administered as a metric of full-scale IQ. The Verbal Comprehension Index includes Vocabulary and Similarities. The Perceptual Reasoning Index includes Matrix Reasoning and Block Design. Standard scores are reported.

WRMT-III

The Woodcock Reading Mastery Test, Third Edition (WRMT-III; Woodcock, 2011) measures reading readiness and reading skills. Participants received the Word Identification and Word Attack subtests which comprise the WRMT Basic Reading composite score. The Word Identification subtest instructs participants to read aloud English words of increasing difficulty. The Word Attack subtest instructs participants to read aloud decodable pseudowords of increasing difficulty. Standard scores (age normed) are reported. A standard score ≤90 (25^th percentile) was used for study inclusion. An additional subtest, Passage Comprehension, was administered. This subtest assesses the ability to read and understand short passages of increasing difficulty by directing participants to supply a missing word.

WISC-V Digit Span

The Digit Span subtest of the Wechsler Intelligence Scale for Children, Fifth Edition (WISC-V; Wechsler, 2014) measures auditory working memory and is one of two subtests that comprise the Working Memory Index. Digit Span involves listening to sequences of numbers and then recalling the numbers 1) in the order in which they were heard, 2) in reverse order, and 3) in ascending sequence. Scaled scores are reported.

CELF-5

The Clinical Evaluation of Language Fundamentals, Fifth Edition (CELF-5; Wiig, et al., 2013) is a comprehensive standardized assessment of oral language skills. The Core Language Score is comprised of four subtests, combinations of which differ depending on participant age range. All participants completed Recalling Sentences and Formulated Sentences. In accordance with CELF-5 guidelines, 8-year-old participants received Word Structure and Sentence Comprehension; 9- to 12-year-old participants received Word Classes and Semantic Relationships; 13-to 17-year-old participants received Semantic Relationships and Understanding Spoken Paragraphs. Standard scores are reported.

CTOPP-2

The Comprehensive Test of Phonological Processing, Second Edition (CTOPP-2; Wagner et al., 2013) Phonological Awareness Composite is comprised of the Elision, Blending Words, and Phoneme Isolation subtests. Elision assesses the ability to remove phonemes or phonological segments from spoken words. Blending words requires synthesizing sounds to form words. Phoneme Isolation assesses the ability to isolate individual sounds in words. Standard scores are reported.

TOWRE-2

The Test of Word Reading Efficiency, Second Edition (TOWRE-2; Torgesen, et al., 2012) has two subtests that comprise Total Word Reading Efficiency, a measure of single word reading fluency. The Sight Word Efficiency subtest measures the number of real words that can be read correctly in 45 s. The Phonetic Decoding Efficiency subtest measures the number of decodable pseudowords that can be read correctly in 45 s. Standard scores are reported.

TOSCRF-2

The Test of Silent Contextual Reading Fluency, Second Edition (TOSCRF-2; Hammill et al., 2014) is a normed, standardized measure of the ability to use morphological and syntactical cues to decipher text presented in unspaced word strings of increasing difficulty. Participants are instructed to delineate individual words by drawing lines between words during a 3-min period. Scores are reported as standard scores.

BRIEF-2

The Behavior Rating Inventory of Executive Function, Second Edition (BRIEF-2; Gioia, et al., 2015) was completed by parents of the participants. The Global Executive Composite is comprised of nine scales (inhibit, self-monitor, shift, emotional control, initiate, working memory, plan/organize, task-monitor, and organization of materials). Scores are reported as T-scores, with higher scores denoting executive function difficulties.

Judgement of Line Orientation

The Judgement of Line Orientation (JLO; Benton et al., 1983) assesses spatial discrimination by having the participant correctly match lines at different orientations on a page to a reference. The reference is a fan-shaped display of lines each numbered and each oriented in a direction 18-degrees different from those adjacent; the reference is essentially a protractor that the participant uses to indicate which reference line matches the orientation of a line presented as a test stimulus. Raw scores (with and without controlling for age) are reported.

Virtual Reality Water Maze

The HVS Water Maze (Version 2017.7; HVS Image Software, 2016), which is a virtual reality program for PC designed as a human analog of the Morris Water Maze task (Morris, 1981), was used for the current study. The task is administered using an Oculus Rift Virtual Reality headset (https://www.oculus.com/rift/). Like the original murine Water Maze Task (Morris, 1981), participants are virtually dropped into one of four quadrants of a large tank of water (Fig. 1) spanning 180 virtual cm in diameter. The participant uses either Oculus handheld controllers or an Xbox controller (depending on individual preference) to swim in the tank of water. Much like the original murine Water Maze Task, the participant is tasked with finding a static platform. Unbeknownst to the participant, the static platform is located in Zone B of the Northwest quadrant. While in the virtual tank, participants are able to see landmarks outside the water tank, which are designed to mimic a lab environment (Fig. 2). For each trial, the session statistics are recorded, including path length and latency. Speed is fixed at 0.14 m/sec. Gallagher cumulative search error is recorded for learning trials, and Gallagher proximity measure is recorded for the probe trial. Water maze metrics, which parallel the murine water maze task, are explained in detail below.

Fig. 1.

Map of HVS Morris water maze quadrants, zones, and platform

Fig. 2.

Screenshot from a practice trial with visible platform

Path Length

Path length is the distance traveled from the origin to the end of the path in meters. For learning trials, path length is measured as the entire path travelled from the point of entry to the point of reaching the platform. Floating (time spent remaining motionless) does not extend the length of the path as no distance is traveled. For probe trial, path length is measured as the entire path travelled from the point of entry to the final point reached at 90s, measured in meters per second.

Latency

Latency is defined as the time it takes participants to find the counter (40 cm circle encompassing the 20 cm platform) from the time they enter the water tank. Counter latency was the measure of latency used for both learning (platform hidden) and probe (platform absent) trials. Participants who never reached the platform counter were recorded as having used the maximum time (60s for learning trials and 90s for probe trial).

Gallagher Cumulative Search Error

The Gallagher cumulative search error measure is the sum of the distances between the participant and the goal taken at 1 second averages, corrected for starting point. Hence, lower scores indicate better performance. This measure is often used to gauge performance during the learning trials (Gallagher et al., 1993; Hawthorne & Baker, 2017; Pereira & Burwell, 2015).

Gallagher Proximity Measure

The Gallagher proximity measure is the average distance of the participant from the platform, after correcting for starting position. This performance measurement is often used during probe trials. Lower scores indicate greater learned knowledge of where the platform should be (Gallagher et al., 1993; Hawthorne & Baker, 2017). See Fig. 3 for example.

Fig. 3.

Example probe trial paths for better navigation (left) and poorer navigation (right)

Procedure

The virtual reality Water Maze session was comprised of nine consecutive trials: two 20-s practice trials with a visible platform, followed by six learning trials (with an invisible platform) lasting 60 s each, and ending with one probe trial (with no platform) lasting 90 seconds. Participants were not told that the probe trial contained no platform. Each participant’s starting point was within a randomly assigned tank quadrant. Participants sat at a desktop computer and put on the headset. Participants performed the task seated in a stationary office chair (without wheels to minimize motion sickness) and were acclimated to the virtual environment by briefly viewing a virtual scene.

Once acclimated, participants were held in a temporary “holding tank” to practice the controls before starting the task. After establishing understanding of the controls, the two learning trials were initiated. Participants were “dropped” into one of four quadrants of a large water tank and swam (speed cannot vary when the control is pressed for any direction; motion stops if the control is released) to attempt to find a static platform located in Zone B of the Northwest (NW) quadrant (Figs. 1 and 2) The start zone for each trial was systematically randomized, but the target zone and platform remained static throughout the experiment.

Data Analysis

First, the two groups were compared on practice trials to establish similarity of understanding task directions and operation of controls. Next, to determine potential differences between NF1 and TD groups across learning trials, mixed linear regression was performed using lmer (Bates et al., 2015) in R (version 3.5.3; R Core Team, 2019), with maximum likelihood fit and t-tests computed using Satterthwaite’s (Satterthwaite, 1946) method. Outcome measures (path length, latency, Gallagher cumulative search error) were predicted with group (coded as NF1=1, TD=0) and age entered as fixed effects. Because learning had multiple trials, an additional fixed effect of trial number (L1-L6) was included in the learning model. Individual participant intercept and slope across learning trials were included as random effects. For the probe trial, linear regression was run using lm in R with outcome measures (path length, latency, Gallagher proximity measure) predicted by group (NF1, TD). Age was entered as a covariate in all models. Multiple comparison correction of p-values was performed using the Benjamini-Hochberg procedure (Benjamini & Hochberg, 1995) to control the false discovery rate (FDR). Group by trial interactions were explored. Additionally, to explore a potential role of working memory, models were rerun with WISC Digit Span as a covariate; further, all models were rerun with IQ as a covariate to be assured that any variability in global cognitive abilities did not account for any group differences.

Results

Behavioral

The two groups, NF1 (n=17) and TD (n=14), differed on age and IQ, with NF1 being older and scoring lower on IQ than TD (Table 1). The two groups did not differ on JLO raw scores; yet, when controlling for age, the difference approached significance (p = 0.091). As anticipated by the study design, groups did not differ on Basic Reading scores, with both groups showing markedly poor reading scores compared to the normative population.

Table 1.

Descriptive statistics and reading scores for groups

	NF1		TD
	(n = 17)		(n = 14)
Measure	M	SD	M	SD	p
Age (years)	12.86	3.27	10.33	2.66	0.02
Sex	M:12, F:5		M:8, F:6
JLO, raw	6.25	4.80	6.93	3.41	0.54
JLO, raw, controlling for age					0.091
WRMT-III Basic Reading, ss	70.94	15.01	77.38	14.86	0.25
WRMT-III Passage Comp., ss	79.54	14.09	83.79	15.54	0.45
WASI-II FSIQ, ss	79.69	9.66	89.57	11.05	0.01
WASI-II Verbal Comprehension Index, ss	83.07	14.31	91.46	10.39	0.09
WASI-II Perceptual Reasoning Index, ss	78.07	10.38	89.34	10.38	0.02
CELF-5 Core Language Score, ss	78.36	13.29	85.36	15.44	0.21
CTOPP-2 Phonological Awareness, ss	78	13.63	85.21	16.22	0.20
TOWRE-2 Total WRE, ss	76.41	18.31	75	14.03	0.83
TOSCRF-2 Index Score, ss	68.5	13.01	72.57	11.33	0.39
BRIEF-2 Global Executive, T-score	67.71	8.66	56.5	9.36	0.003
WISC-V Digit Span, scaled score	5.64	1.86	7.85	2.64	0.029

JLO Judgement of Line Orientation; JLO is not normed; only raw scores are available. WRMT Woodcock Reading Mastery Test; WASI Wechsler Abbreviated Scale of Intelligence; FSIQ full scale IQ; CELF Clinical Evaluation of Language Fundamentals; CTOPP Comprehensive Test of Phonological Processing; TOWRE Test of Word Reading Efficiency; WRE Word Reading Efficiency; TOSCRF Test of Silent Contextual Reading Fluency; BRIEF Behavior Rating Inventory of Executive Function; WISC Wechsler Intelligence Scale for Children; ss standard score

Virtual Reality Water Maze

The virtual reality water maze task had nine consecutive trials: two practice trials (platform visible), six learning trials (platform hidden), and one probe trial (platform absent). Latency and path length were explored for all trials. For learning trials, we report Gallagher cumulative score, and for the probe trial, we report Gallagher proximity score (Table 2). All significant values reported were significant beyond the critical value created by the Benjamini-Hochberg procedure (p = 0.01788). Outcomes for each type of trial are described below.

Table 2.

Water maze measures by group

Trial	Measure	NF1 (n = 17) M	SD	TD (n = 14) M	SD	p
Learning
	Counter Latency (sec)	34.6	19.75	31.7	19.93	0.00894**
	Path Length (m)	4.7	2.41	3.7	2.02	0.00853**
	Gallagher Cumulative	29.5	14.46	25.6	14.08	0.00548**
Probe
	Counter Latency (sec)	37.8	27.98	34.2	30.13	0.42137
	Path Length (m)	10.1	2.20	8.6	2.37	0.0456
	Gallagher Proximity	0.6	0.17	0.5	0.142	0.2442

p-values were adjusted for multiple comparisons using the Benjamini-Hochberg Procedure. Values were significant beyond the critical value of 0.01788

Practice Trials

Prior to the learning and probe trials, all participants performed two practice trials. These trials were provided to establish understanding of the goal of the task and understanding of how to navigate within the water maze virtual environment.

All participants indicated understanding of the task and were able to operate the controls. There were no significant differences between NF1 and TD in practice trials for latency (p = 0.516) or path length (p = 0.185), controlling for age.

Learning Trials

Across six learning trials, mixed effect models were employed to examine the three outcome measures: path length, latency, and the Gallagher cumulative search error (Table 2). Models were then rerun with IQ and working memory as fixed covariates to determine if global cognitive abilities accounted for any of the resulting group differences.

Path Length

Group differences emerged for path length (p = .00853), controlling for age, with NF1 exhibiting poorer performance than TD as evidenced by longer path length (Fig. 4).

Fig. 4.

Path length across learning trials

Latency

Significant differences were found between groups on counter latency (p = 0.00894), with NF1 spending significantly more time than TD to locate the platform across learning trials. (Fig. 5).

Fig. 5.

Latency across learning trials

Gallagher Cumulative Search Error

Controlling for age, significant differences were found between NF1 and TD on Gallagher cumulative search error across the learning trials (p = 0.00548), with TD outperforming NF1 (Fig. 6).

Fig. 6.

Gallagher cumulative search error across learning trials

Interactions

Group × trial interactions were not significant.

Additional Analyses

All significant results in learning trials analyses remained significant when controlling for working memory, as well as when controlling for IQ, prior to correction. All of these maintained significance upon applying Benjamini-Hochberg correction to each set of analyses (working memory and IQ), with the exception of latency which did not survive correction when controlling for IQ (adjusted p=0.05405 exceeded critical value of 0.03081)

Probe Trials

Linear regression models, with group and age as predictors, examined the three outcome measures: path length, latency, and the Gallagher proximity measure. Models were then rerun with IQ and working memory as covariates to determine if global cognitive abilities were associated with any group differences.

Path Length, Latency, and Gallagher Proximity Measure

Linear models of latency and Gallagher proximity measure were highly nonsignificant per omnibus testing (Table 2). The difference in path length, with longer path length for NF1, did not survive multiple comparison correction. As such, the nonsignificant overall models precluded any meaningful reporting of group differences for path length, counter latency, or Gallagher proximity measure. Hence, no group differences emerged for any of the probe trial metrics.

Additional Analyses

All regressions of probe trial metrics remained nonsignificant when additionally controlling for working memory and IQ.

Discussion

As the Morris water maze task was originally designed to assess spatial learning in rodents, the current study explored a virtual reality version of the task developed to emulate the water maze environment for humans. This is of particular interest for refining the understanding of NF1, a disorder with known associated visuospatial impairment. This exploratory study compared a group with NF1 and reading difficulties to those without NF1, but who also had reading difficulties (TD). Across learning trials, the NF1 group exhibited impaired spatial learning as evidenced by greater latency, longer path length, and greater cumulative search error relative to the TD group. The learning trial findings remained significant even when controlling for working memory (and IQ), though the difference in latency did not survive multiple comparison correction when controlling for IQ. There were no significant group × trial differences, indicating that the rate of learning did not differ between groups. Nevertheless, the slopes over the trials generally shifted from a discrepancy between the two groups towards similarity at the final learning trial. Lack of significant interaction (or, conversely, the observed trends) may be attributable to the small sample size as these preliminary interaction analyses are underpowered. As definitive conclusions are precluded, the trends observed warrant investigation on a larger scale. In the probe trial (the ninth and final trial), no significant differences were observed. As such, though the NF1 group continued to have greater difficulty in arriving at the target location during the learning trials, differences in most metrics were no longer apparent in the probe trial. This may suggest that with additional practice, the visuospatial deficits of NF1 can be tempered. That is, these preliminary findings implicate that although individuals with NF1 initially perform worse that their typically developing peers, subsequent learning sessions may allow them to “catch up.” Hence, these findings may have implications for instruction and warrant further investigation.

Given the limited work that has been done in virtual maze navigation in both TD and NF1, this study provides a step toward elucidating the role of visual and spatial skills that are overlapping and unique for the two populations. For NF1, previous work has suggested that maze navigation may be distinct from other visual tasks as demonstrated by the lack of correlation with other measures of spatial skills and learning (Ullrich et al., 2020). Comparisons on traditional measures of visuospatial skills have shown that NF1 differs from TD, even though both groups are matched on academic (reading) performance, perhaps indicating a unique relation between visuospatial skill and reading skill in NF1 (Barquero et al., 2015; Levine et al., 2006). The current study points to further delineation of NF1 from TD, especially in the context of controlling for learning difficulties, in terms of severity of visuospatial deficit.

Future directions should focus on further teasing apart how visuospatial learning and memory relate to variability in reading and other academic skills in NF1. In the current study, NF1 and TD with RD had similar profiles across multiple reading/language measures, but the two groups were discrepant in intellectual functioning, findings that are consistent with the literature. It remains unclear whether visuospatial deficits are specifically causal to reading and learning difficulties in NF1, or if these are co-occurring deficits, perhaps tied to an underlying (and yet to be pinpointed) domain general deficit. Further exploration is needed to understand if the reading deficits in NF1 are a functional proxy for greater brain/cognitive involvement and are substantively different from the reading deficits in TD that are more confined to domain-specific networks of reading and language. A future step in understanding the visuospatial skills in NF1 should include investigation with neuroimaging along with behavioral assessment. Expanding the understanding of the neurobiology of navigation and exploring how that neurobiology differs for NF1, especially while controlling for academic performance in comparison groups will provide needed insight. To illustrate, “place cells,” found to fire depending on route traveled by rats, were discovered in the hippocampus decades ago (O’Keefe & Dostrovsky, 1971). However, more recently, place cells and other spatial encoding cells have been found beyond the hippocampus (e.g., thalamus), suggesting a complex spatial network is involved in navigation (O’Mara & Aggleton, 2019). Given the multifaceted differences in the brain associated with NF1, which include hyperintensities on T2-weighted images often located in the thalamus among other sites, it is possible that these spatial encoding cells are affected. However, it is also possible that differences in visual areas (e.g., visual cortex) play a greater role. Although there is considerable variability in NF1 phenotype, it may be useful to consider some of the characteristic NF1 neurobiological manifestations when explicating possible causes underlying visuospatial deficits.

This study has some limitations. Clearly, humans differ from rodents in terms of how the task is experienced and what approaches are employed in solving the task, including the significant level of fear that mice have of water, which in a virtual situation (or real) can be quite distinct from humans. The stress that mice experience when confronted with the task, and thus the drive to escape the water, may facilitate learning (Schoenfeld et al., 2017), although higher corticosterone has been associated with poorer water maze performance in mice (Harrison et al., 2009). Additionally, mice are likely to be much more reliant on olfactory sense for navigation than are humans (Lam et al., 2018). As such, any conclusions that attempt linkages to animal models must be made with caution. Further, as with all virtual analogs of the water maze, the task does not involve the demands of physical locomotion or the motivation (and some of the sensory information) inherent to finding a platform when swimming in a tank of water (Devan & Hendricks, 2018; Thornberry et al., 2021). Of note, however, comparison of rodents in a real physical water maze and humans using a computer maze have shown similar spatial performance (Schoenfeld et al., 2017), with the task in the current study additionally having the advantage of head mounted virtual reality display rather than two-dimensional computer monitor display. Another consideration for future studies is that the current study did not broadly explore the role of cognitive control/executive function on Morris Water Maze performance, but rather was limited to working memory (which had a minimal impact on findings). A recent meta-analysis has revealed that working memory, planning, and problem-solving are affected in children with NF1, more so than other cognitive constructs such as mental flexibility and inhibition (Beaussart et al., 2018). It has even been suggested that visuospatial deficits in NF1 may be related to cognitive control/executive function deficits (Van Eylen et al., 2017). Future work could further explore a fuller range of the impact of these skills on visuospatial functioning. Further, the effect of autism should be considered given the greater occurrence of autism symptomology in individuals with NF1 relative to TD.

Additionally, we recognize that this study would greatly benefit from a larger sample size. This study was undertaken as to explore feasibility of using virtual reality to investigate maze navigation and learning in NF1, including gathering a preliminary dataset. To our knowledge, no other study has investigated the use of virtual reality in maze navigation with an NF1 sample. The current study, albeit underpowered, is larger than the two studies of NF1 and maze navigation on computer screen (Ullrich et al., 2010, 2020). These small sample sizes reflect the early stage of exploration in this type of work, and expectations are for expanded studies in the future to allow for adequately powered analyses.

In conclusion, the current exploratory study extends understanding of the differences in visuospatial skills between NF1 and TD and lays groundwork for future studies. Current results are to be interpreted with caution due to the small sample size, but nonetheless appear worthy of further investigation. Visuospatial deficits in NF1 were evident across multiple metrics and largely maintain significance even when controlling for working memory and IQ. While previous studies have established with pencil/paper visuospatial tasks that NF1 differs from TD, this is the first study of which we are aware that demonstrates differences for navigation within a virtual reality environment while controlling for concomitant learning (reading) difficulties in NF1.