Age-related impairment of navigation and strategy in virtual reality star maze

Background Although it is well known that aging impairs navigation performance, the underlying mechanisms remain largely unknown. Previous study suggested that the egocentric strategy was relatively unimpaired during aging. In this study, we aimed to examine strategy use and the underlying cognitive supporting mechanisms in older adults with the virtual reality star maze. Thirty young adults and thirty-one older adults participated in the study. During the learning trials, participants were required to reach a fixed destination in a virtual reality star maze task. In an additional probe trial, the distal landmarks around the maze were removed, and the strategy using was classified into egocentric and non-egocentric according to whether participants could reach the destination directly. were selectively


Background
Navigation means finding and maintaining a route in a familiar or unfamiliar environment [1].Itis one of the fundamental cognitive functionsthat decline the most with increasing age [2]. Empirical studieshave found thatnavigation is vulnerable to aging [3].Compared with young adults, older adults tend to commit more errors and take longer in solving navigation tasks [4].
Many studies havefound that there are two mainstrategies during navigation: egocentric strategy and allocentric strategy [5][6][7].Allocentric strategy relies on a world-centred representation, whereas egocentric strategy is based on a self-centered (i.e., body-centred, a series of left and right turns) representation [7,8]. Specifically, allocentric strategy is based on a "map-like" representation, which enables the navigators to find a detour or reach a destination from different starting point [9,10]. In contrast, egocentric strategy is based onthe association of stimulus and responses (idiotheticinformation such as body turns and vestibular sense) and allows individuals to navigate in a fixed route [7,8].This strategyenables individualsto navigate from a fixed origin to a fixed destinationthrough the same route [9].Spatial memory task involving egocentric and allocentric strategy may be a sensitive tool to detect the cognitive decline during aging [11]. In age-related studies, previousstudies have found the egocentric strategy is relatively preserved in older adults, whileallocentric strategy and other strategies experience decline [2,9,12,13]. According to the retrogenesis theory, abilities acquired later in life are more vulnerable to aging and are the first to deteriorate [14]. As a more fundamental representation,egocentric strategy is acquired 5 earlier, andis more rooted and well-preserved in older adults [15].Several studies reported that the visuo-spatial deficitsmight help discriminate mild cognitive impairment (MCI) from dementia in elderly population [16][17][18]. Using different 2-dimensional visuoconstructive tasks, previous study found that the real world navigation was selectively related with the visuo-spatial memory [19].Though the visuo-spatial working memory correlated with the navigation performance both in young and older adults, visuo-spatial working memory performance decreased with age [20]. Therefore,visuo-spatial ability may supportthe strategyuse in older adults.
Recent studies revealed that a successful navigation primarily relies on the capacity to flexibly choose the appropriate strategy depending on the task demands rather than using one specific strategy [12,21]. Therefore, though some evidence suggests that the egocentric strategy is well preserved in older adults, the navigation performance is still impaired if they fail to choose the appropriate strategy. A virtual reality study has already confirmed the capacity to choose both allocentric and egocentric strategy in young adults [7]. In contrast, older adults appear to have a deficit in strategy choosing and are more likely to choose the inappropriate strategy across different studies [22,23].
Visuo-spatial ability is typically assessed using paper-and-pencil tasks such ascomplex figure task and clock drawingtest [3]. These tasks were designed to evaluate the basis spatial abilities, for instance mental rotation and spatial memory, which are moderately or weakly correlated with spatial memory or navigation in the real world[3, 24, 6 25].Recently, a study compared walking and non-walking space task in an equivalent virtual reality [26]. It shows that walking space task is much easier for participants to search different viewpoints to get themselves familiar with the environment and locate the targets, whereas the non-walking space task may demand extra cognitive effort.
Therefore, an environment in which participants can move and rotate freely should be more compatible for older adults and sensitive to the effect of aging [27]. Given that the real world study was relatively complex, costly and difficult to control, virtual reality becomes an attractive approach to investigate navigation [28][29][30]. Recently,the virtual reality technique with elaborately designed scenes has been widely used in navigation tasks, including the Morris Water maze [31], the Eight-Arm maze [21,32,33], and the star maze [7,34]. A recent study reported a strong correlation (r 2 = .73) between virtual reality and real-world navigation performance [35], indicating the applicability of virtual reality in navigation studies. However, it is worth noting that avirtual reality navigation paradigm may be good at distinguishing navigation strategies but weak in evaluating navigation performance (e.g., navigational speed, distance travelled), or vice versa [33,36]. The virtual reality star maze may be a trade-off for evaluating both navigation performance and navigation strategy [37]. In the virtual reality star maze task, participants navigate freely in a pentagram-shaped virtual reality environment and the task is always to find a fixed destination. The latency,total degree rotated and distance travelled are recorded as navigation performance. In addition, landmarks may be removed or changed to detect the strategy [7,37,38]. Previous studies have already applied the virtual reality star maze in healthy children [6] and young adults [7]. 7 Moreover,virtual reality star maze might provide a selective behavioural marker for Alzheimer's disease(AD) [38]. Therefore, in healthy aging, the virtual reality star maze is considered as an ideal paradigm to evaluate age-related changes both in navigation performance and strategy.
In thecurrent study, weaimed to use the virtual reality technique to examine strategy use and the supporting cognitive mechanism in older adults. We expected a preserve of egocentric strategy in older adults, andthe success of using egocentric strategy in older adults was related with visuo-spatial ability.

Participants
Thirty-one young adults (15 males and 16 females; mean age: 22.68 ± 3.07) and thirty-two older adults (14 males and 18 females; mean age: 66.91 ± 5.27) were recruited using notices distributed at local universities or in the community or by word of mouth. Inclusion criteria of participants were as follows: 1) no history of neurological or psychiatric diseases base on self-report; 2) 6 years or more of education; 3) a score of 24 or higher on the Mini-Mental State Examination (MMSE) [39]; 4) for young adult group, the age should be within the range from 18 to 30, for older adult group, the age should be over 55. The study was approved by the local ethics committee of theInstitute of Psychology, Chinese Academy of Sciences (IPCAS).Written informed consent was obtained from all participants inaccordance with the Declaration of Helsinki prior to the study. 8 Participants were asked about their frequency of computer and virtual reality use. For instance, 'How often do you use computers? 1 for never, 2 for occasionally, 3 for often'.

Neuropsychological tests
All participants completed neuropsychological tests, including the Multifactorial Memory Questionnaire (MMQ) [40], the digit span tests and block design. MMQ was used for general memory status estimation; it included three subtests: satisfaction of memory, memory ability and memory strategy. The digit span tests were selected from the Wechsler Adult Intelligence Scale (WAIS) [41], including the forward and backward version. The block design test, also selected from the WAIS-IV, mainly evaluates visuo-spatial ability. The order of neuropsychological tests was counterbalanced among participants.

The virtual reality star maze paradigm
During the entire duration of the following virtual reality, participants were comfortably seated in front of the computer screen and moved the joystick (Sony DualShock 4) freely in a first-person view. Prior to the virtual reality star maze task, participants went through a joystick-familiarity task. Moreover, participants were asked to move along the boundary of a yellow square different from the star maze task to help familiarize with the joystick. Movement in virtual reality was achieved by pushing the joystick forward and pulling it backward, and horizontal rotation was achieved by pushing the joystick left and right. Young adults spent an average of 5 minutes and 9 older adults an average of 15 minutes in this joystick training. After participants declared their confidence to complete the task, the joystick training ended and the star maze task began.
The human virtual star maze is similar to the well-defined rodent navigation paradigm.
Current study adapted it from previous virtual reality star maze research in older adults and made a few changes [38].  Figure 1). Participants used a joystick to freely move around at a constant speed (8 vm/s) and rotate (95°/s) in all the alleys. To avoid guidance strategy, all distal environment cues were presented twice, but the exemplar differed (e.g. two different forests).

[PLEASE INSERT FIGURE 1 HERE]
The virtual reality star maze task included two phases: the learning phase and the probe phase. Prior to the learning phase, all participants went through at least two practice trials to ensure they learned the task and remembered the destination. If participants failed to find the destination within 90 seconds, they would be teleported to it, a process 10 aimed at helping participants familiarize themselves with the task. If participants thought they were able to locate the destination after two practice trials, then the formal learning phase would begin; otherwise, participants were provided with additional practice trials.

Learning phase
The learning phase was designed to evaluate navigation performance and learning effect. Prior to the formal trials, participants were informed that the goal of the virtual reality task was to find and maintain a location, which would always be located at the same place during the following tests. Once reached, sparking fireworks would be shown to provide a reward and signal the beginning of the next trial. If they learnt the destination, they should always take the optimal route to the destination. In each trial, participants would always start from the same place with same facing direction to the same destination (Figure 1a, indicated by the blue arrow on the left-hand pictures).
Instructions were designed to avoid any words involving view, route, circumstance, etc., which excluded bias from any strategy. Participants were provided with 9 learning trials. Each learning trial had a 90-second time limitation; if participants failed to reach the destination within 90 seconds, the trial terminated immediately and the next trial began. In the learning phase, participants could reach the destination by remembering the sequence of body turns of each Y-shaped intersection (egocentric strategy) or other strategies. 11 A computer automatically collected the location every 20 milliseconds, as well as the facing direction related to the start point. The total length and time of navigation were also recorded for each learning trial. The total length travelled was expressed in vm and the time in milliseconds. To evaluate the navigation performance of each participant, the following parameters were calculated for the 9 learning trials: speed, distance error, rotation, percentage of successful people and percentage of successful trials. Notably, in current study, speed was considered as measurement of navigation efficiency, whereas the distance error, rotation, percentage of successful of people and percentage of successful of trials were considered as measurement of the navigation accuracy.
Speed was calculated by dividing total length by total time and was expressed in virtual length units per second (Formula 1). Speed here included stopping time and therefore was considered a factor in navigation efficiency. Rotation(°)= ∑ |facing direction tn -facing direction tn-1 | (Formula 2) Distance error was calculated by subtracting the optimal path distance from the total length. To compare distance error between individuals, we further divided this result by 12 the optimum path distance (Formula 3). Participants who travelled the optimal path exactly scored 0% in distance error, whereas a detour would increase distance error.

Probe phase
After the 9 learning trials, participants had to finish one probe trial within 90 seconds.
The probe trial shared the same maze structure as the learning phase but with no salient landmarks (i.e., all distal cues were removed). For the volunteers, the probe trial just like the tenth learning phase, the existence of the probe trial was not mentioned in advance. Even the total number of trials were not informed before the experiment, volunteers were only been told the experiments consists of dozens of trials, the program will stop automatically when it ends. 13 The probe trial was designed to distinguish the strategy used and the navigation performance in probe trial was not analysed to avoid circular reasoning. The absence of distal landmarks made strategies relying on the landmarks invalid, including allocentric or guidance strategy. In contrast, egocentric strategy by remembering a sequence of body turns was unaffected. In consequence, participants were further divided into two strategy groups according to the probe trial. Those participants who reached the destination directly in the probe trial (did not enter any irrelevant alleys other than optimal path) were classified as egocentric strategy users (by remembering a series of body turns), whereas the rest of the participants who failed to perform a direct trial were classified as non-egocentric strategy users. It is worth noting that in this study, the two strategy groups we defined did not means the participants kept using the same strategy during the entire study. The major difference between two strategy groups is if the participants could use the appropriate and efficient strategy when the environment and task quest changed. And then we looked back the learning phase, whether those older adults who were able to use an egocentric strategy in the probe trial also navigated much more efficient. Therefore, in the currents study, there were four subgroups: egocentric older adults, non-egocentric older adults, egocentric young adults and non-egocentric young adults.

Data analysis
Demographics were analysed using independent sample t tests for continuous variables and chi-square tests for categorical variables.
14 Three-way repeated-measure ANOVA was carried out for speed, rotation, distance error and percentage of successful trials separately, with age (young adults, older adults) and strategy (egocentric, non-egocentric) as between factors, and learning trials (9 learning trials) as within factor. Because of large rotation, 3 young adults (2 males and 1 female) exceeded three-standard deviations (SDs) from their group in the 6th and 7th learning trials, and were excluded from the repeated-measure ANOVAs and subsequent simple effect tests.
Pearson correlations among navigation performance during the learning phase and neuropsychological tests were applied in young and older adults separately. Moreover, to further explore whether the two strategy groups in older adults differed in WAIS block design and digit span, independent sample t tests were applied.
Furthermore, to validate the discrimination of the probe trial, we performed an additional analysis. We administered paired sample t tests between the 9th learning trial and the probe trial in allocentric strategy users and egocentric strategy users, respectively.

Demographics
Two older adults were excluded for not completing the entire virtual reality task (one male and one female older adult  16 Theoretically, the absence of distal landmarks in probe trial should only impair strategies relied on landmarks (i.e. allocentric strategy or guidance strategy), leaving the egocentric strategy unimpaired. As shown in Figure 2, the performance of egocentric strategy users (including both young and older adults) did not differ between the 9th learning trial and the probe trial, whereas non-egocentric strategy users performed significantly worse in the probe trial than in the 9th learning trial, including slower navigation speed (t = 2.273, p = .029), more rotations (t = -10.382, p < .001), higher distance error (t = -9.113, p < .001) and lower percentage of successful trials (t = 6.508, p < .001). This specific impairment of non-egocentric navigation from the 9th learning trial to the probe trial highly supported the strategy discrimination of the probe trial.

Navigation performancein the learning phase
According to the results of repeated-measureANOVA (

Visuo-spatial ability in older adults
All the Pearson correlations in young adults were not significant. For older adults, the WAIS block-design score was positively correlated with navigation speed (r = .370, p = .044) and the percentage of successful learning trails (r = .417, p = .022),suggesting that better visuo-spatial abilityin older adults was correlated with faster navigation speed andhigher possibility to successfully complete the star maze.
Furthermore, the independent sample t tests revealed that egocentric older adultsscored higher on the WAIS-block design test than non-egocentric older adults, suggesting bettervisuo-spatial abilityin egocentric older adults (Table 3). However, two groups did not differ in forward digit span(t = -.785, p = .457) and backward digit span(t = .041, p = .967). These results suggest that non-egocentric older adults have a specific deficit in visuo-spatial ability. In addition, The non-egocentric older adults showed more satisfaction and confidence about their memory (MMQ1) and reported less trouble about memory during the last fortnight (MMQ2), whereas the egocentric older adults preferred to adopt different memory strategies during daily life (MMQ3, thought this result does not reach significant, the effect size is large), which may indicate a preserved ability to adopt or switch to an efficient strategy.

Discussion
In the current study, by using the virtual reality star maze task, we found older adults presented slower speed in virtual reality star maze navigation.Older adults adopting egocentric strategy performed as well as young adults, while older adults using non-egocentric strategy showed worse navigation performances than young adults.
Moreover, better visuo-spatial ability was related with better navigation performance in older adults.
The current study found a selective deficit of older adults in spatial navigation speed, which was consistent with previous studies [3,13,27]. These results revealedage-related impairments of the navigationefficiency. Previous studies also found that older adults had slower reaction times, while their error rate was comparable to young adults in spatial representation task [42] or virtual reality navigation task [43].A recent meta-analysis including 80 samples examined the age-related changes in spatial abilities. The results found that the measures of response time yielded a larger effect size than the measures of accuracy, demonstrating that speed was a critical factor in spatial tasks, for instance, mental rotation and perception tasks [44].This was coincident with current study that the star maze paradigm revealed a significant effect of aging in navigation speed, but not in other performance. This specific impairment suggests that the navigation speed isalso a sensitive factor for aging in navigation as well as in neuropsychological tests, which means the healthy older adults should be able to complete the navigation task if they are provided with adequate time 21 The impairment of navigation speed may be the consequence of speed-accuracy trade-off [15].It was established that older adults paid more attentionto accuracy and tended to sacrificespeed [45].To successfully complete the task in the current study, older adults navigated at asignificantly slower speed. Notably, even at the early stage of the learning phase, the percentage of successful older adults was comparable to the young adults. Another explanation for theslower speed in older adults in virtual reality navigation might be the consequence of a poor perception of visual motion information [46].Similar results were found in vestibular perception [47] and direction perception [48,49]. Thesedeficitsin primary perception in older adults may decrease the signal-to-noise ratios as well as the directional tuning, which may hamper their higher cognition processing [2].
Considering that the older adults showed less exposure to virtual reality and computers, it was worth noting that, in current study, the average navigation speed cross learning phasecorrelated with the experience of virtual reality (r = -0.327, p = .012)and the experience of computers significantly (r = -0.332, p = .011), which were recorded in a 3-point scale. Meanwhile, the manual dexterity may also affect the navigation performance [50], which was not recorded in our study. Therefore, future studies should consider these possibilities.
In the current study, we defined theegocentric strategy as a pure egocentric strategy ignoring all the landmarks and only relied on a series of body turns [13,51].The strategy discrimination was based on a hypothesis that the navigation performance of egocentric 22 strategy userswere not affected by the absence of distal landmarks, whereas non-egocentric strategiesthat relies heavily on the spatial relation of external landmarks were impaired in the probe trial [2,9,37]. The resultsof paired sample t tests revealed that the navigation performance of the non-egocentric group in the probe trial was significantly worse than the 9th learning trial (Figure 4), whereas these differences did not reach significance in the egocentric group. This dissociation was consistent with our hypotheses and supported the discrimination of the probe trial. We alsoexamined these differences between 9th and probe trial inegocentric older adults andnon-egocentric older adults, egocentric young adults and non-egocentric young adults separately, the results were similar and further verified the value of the star maze in differentiating strategies in both young and older adults.
Recent study suggested that the impairments in strategy use in older adults might further hinder their navigation performance [52]. For instance, recent study using virtual reality plus maze paradigm found that older adults were impaired in strategy switching [53]. Although previous studies revealed that the deterioration of navigation during aging might be specific to certain strategies [9,54], the results were inconsistent.
Some studies reported a specific impairment in allocentric navigation [13,55], whereas others suggested a decline in egocentric strategy accuracy in older adults [15].
In current study, the main effect of strategy during learning phase was significant. In other words, although both egocentric and non-egocentric strategy could be used to complete the navigation task during the learning phase [6,37], participants using the 23 egocentric strategy performed much more accurate. Specifically, the superiority effects of egocentric strategy were only found in older adults. The results showed that non-egocentric older adultscompleted task with significant more rotation and distance error, whileegocentric older adultsshowed comparable navigation accuracy to both egocentric young adults and non-egocentric young adults during the learning phase.
Before the disappearance of distal landmarks in the probe trial, both egocentric strategy and non-egocentric strategy should be validtheoretically during the entire leaning phase. As expected, the navigation performance of egocentric young adults and non-egocentric young adultsdid not show significant differences. The results demonstrated that older adults while not young adults were affected by the strategy use during performing the task when the environment provided adequate landmarks for navigation.
The selective impairment of non-egocentric navigation in older adults may be attributed to the different cognitive loads of thenavigation strategies [54]. To navigate in an egocentric way, participants were required to remember the time sequence of body turns [6]. For instance, in the current study, the time sequence was "straightleftstraightrightstraightleft -straight", which could even be simplified as "leftright -left". In contrast, non-egocentric strategy may be ambiguous and involve several strategies. Taking allocentric strategy as an example, navigation demands several cognitive functions, including encoding the spatial scene to form an aerial view, retrievingspatial imagery and route planning [56]. Previous studies have reported a greater activation of the frontal-parietal attentional control network in allocentric 24 navigation than in egocentric navigation, which indicated an additional resource recruitment of allocentric strategy [54,57]. Moreover, retrogenesis theory supports this selective impairment with another explanation that late-acquired abilities are more vulnerable to loss during aging [14]. According to this theory, egocentric strategy, as a more elementary representation, develops earlier,even before 5 years old,and is expected to be maintained during aging [15], whereas allocentric strategy is acquired by later school-age children [6]and is considered to have deficits preceding egocentric strategy [12].
In the current study, we found the WAIS block-design was positively correlated with navigation performance in older adults, suggesting that visuo-spatial ability may play an essential role in the navigation of older adults, whereas the navigation of young adults may be relatively independent of visuo-spatial ability. The difference in environment encoding may explain why the navigation of young and older adults showed different extended dependence on visuo-spatial ability [58]. For example, previous studies reported that aging reduced the signal-to-noise ratios in the motion sensitive middle temporal area [59], and the difficulties in extracting the spatial information from optic flow may further impair the spatial updating, which is vital during navigation. In addition, older adults acquired less accurate memories for the spatial relationship of landmarks to finish their navigation task [27], and were less likely to use the geometric cues [22]. 25 Moreover, we found no significant differences in MMSE and MMQ between the young and older adults, which excluded the possibility that the navigation performance impairment of older adults might be the consequence of a decline in general cognition or of amnesia. Even though, the difference of MMQ between egocentric and non-egocentric older adults might help explain the strategy use of this two groups.
These results may indicate that not all is lost in older adults, some older adults showed a preserved memory, while the others who have memory troubles may using diverse memory strategies to compensate the memory loss. In current study, the egocentric older adults may benefit from their ability to choose an appropriate strategy. Even though, these results need further verified, due to the small sample size of egocentric older adult group.
The visuo-spatial ability may play an important role in the navigation of older adults.
Specifically, in the current study, egocentric older adults showed significantly higher scores in visuo-spatial ability than non-egocentric older adults. It has been reported that the visuo-spatial ability or the visuo-spatial component of working memory played a part in encoding the environment information, which may underpin navigation performance [60]. Another real-world navigation study also revealed that participants with high visuo-spatial ability achieved higher accuracy with less time during navigation [61]. Moreover, visuo-spatial ability as well as navigation performance shows a clinical potential in diagnosis of MCI and AD [16,18,38]), which provided a direction for future aging navigation studies. 26

Limitations
There are several limitationsto the current study. First,in the current paradigm, the probe trial could not subdivide the non-egocentric strategy into different strategies (allocentric, guidance strategy etc.). Therefore, we hypothesized that all the participants kept using the same strategy during the entire experiment or, at least, they used an appropriate or an inappropriate strategy in the probe trial. In other words, we supposed that older adults who used an appropriate strategy learning the navigation task faster and better than older adults who used an inappropriate strategy.Second, the limited sample size (egocentric older adults) restricted further analysis between two strategy groups in older adults. Therefore, our findings should be treated cautiously and further evidence is warranted.

Conclusion
In the current study, older adults navigated the virtual reality star mazesignificantly more slowly than young adults but with comparable accuracy. Moreover, older adults using egocentric strategy performed as well as young adults, whereas older adults adopting non-egocentric strategy showed significant lower navigation accuracy relative to young adults. The selective impairment was related with visuo-spatial ability in older adults. The current study offers evidence that navigational deficits could be markers for cognitive decline in older adults, and emphasizes the necessity for future virtual navigational intervention for older adults.

Consent for publication
Not applicable.   whereas 0% indicated the ideal path. d. Larger successful population indicated more successful participants in a group and 100% indicated that all members succeeded in 38 reaching the destination in 90 seconds. 3 young adults were excluded in the 6th and 7th trials because they exceeded three standard deviations from the average in rotation.
Error bars indicate mean ± SEM. performed significantly more rotations than non-egocentric young adults and egocentric older adults in the leaning phase. c. Non-egocentric older adults completed the task with more detours than non-egocentric young adults and egocentric older adults during the learning phase. d. The success rate of non-egocentric older adults during the learning phase was significantly lower than for non-egocentric young adults and egocentric older adults. Error bars indicate mean ± SEM. **p < .01, ***p < .001.