Report No. MAE0820

Neuroscience Research on Human Visual Path Integration: Empirical Overview and Strategic Considerations

Jimmy Y. Zhong12

1School of Mechanical and Aerospace Engineering (MAE), Nanyang Technological University, Singapore 639798, Singapore 2 GSU/GT Center for Advanced Brain Imaging (CABI), Georgia Institute of Technology, Atlanta, GA 30318, USA Email: [email protected] Report No. MAE0820

1 Foreword 2 3 This manuscript is based on modifications and extensions of copyrighted writings of the author 4 that were previously unpublished in any book or journal. It is intended as a 5 conceptual/theoretical piece of writing and should not be regarded as a research report or article 6 of any kind. All written opinions and recommendations expressed herein belong solely to the 7 author and do not belong to any other person or organization. If deemed useful, please cite and 8 reference this paper in its original form: 9 10 Zhong, J. Y. (2021, August). (Technical Report No. MAE0820). Neuroscience research on 11 human visual path integration: Empirical overview and strategic considerations. School of 12 Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore. 13 https://doi.org/10.31234/osf.io/h6u3b

i

Report No. MAE0820

14 Abstract 15 16 Over the past two decades, many neuroimaging studies have attempted to uncover the brain 17 regions and networks involved in path integration and identify the underlying neurocognitive 18 mechanisms. Although these studies made inroads into the neural basis of path integration, they 19 have yet to offer a full disclosure of the functional specialization of the brain regions supporting 20 path integration. In this paper, I reviewed notable neuroscientific studies on visual path 21 integration in humans, identified the commonalities and discrepancies in their findings, and 22 incorporated fresh insights from recent path integration studies. Specifically, this paper presented 23 neuroscientific studies performed with virtual renditions of the triangle/path completion task and 24 addressed whether or not the hippocampus is necessary for human path integration. Based on 25 studies that showed evidence supporting and negating the involvement of the hippocampal 26 formation in path integration, this paper introduces the proposal that the use of different path 27 integration strategies may determine the extent to which the hippocampus and entorhinal cortex 28 are engaged during path integration. To this end, recent studies that investigated the impact of 29 different path integration strategies on behavioral performance and functional brain activity were 30 discussed. Methodological concerns were raised with feasible recommendations for improving 31 the experimental design of future strategy-related path integration studies, which can cover 32 cognitive neuroscience research on age-related differences in the role of the hippocampal 33 formation in path integration and Bayesian modelling of the interaction between landmark and 34 self-motion cues. The practical value of investigating different path integration strategies was 35 also discussed briefly from a biomedical perspective. 36 37 Keywords: path integration, strategy, spatial navigation, hippocampus, entorhinal cortex, 38 cognitive neuroscience

ii

Report No. MAE0820

39 1. Introduction 40 41 In everyday contexts, whenever we have a destination in mind we want to travel to, we 42 rely on our innate adeptness for navigation, as typified by our cognitive ability to reason in 43 space, plan, and execute a path to our desired location (Gallistel, 1990). Navigation generally 44 occurs in the presence of external cues that present themselves ubiquitously in the form of 45 stationary landmarks in our everyday environments but can also proceed when only internal 46 bodily cues are available (i.e., during non-visual self-locomotion, see, e.g., Klatzky et al., 1990; 47 Loomis et al., 1993). The process of estimating distance and directional changes relative to a 48 starting position and integrating such displacements with self-motion cues to compute and update 49 a homing vector is called path integration (also known as dead reckoning) [see, e.g., Gallistel, 50 1990; Mittelstaedt & Mittelstaedt, 1980, 1982; Müller & Wehner, 1988; for reviews, see Etienne 51 & Jeffery, 2004; Loomis, Klatzky, Golledge, & Philbeck, 1999; Loomis, Klatzky, & Golledge, 52 2001; Srinivasan, 2015]. 53 Path integration was first postulated to apply to humans by Darwin (1873), who recounted 54 the remarkable feat in which the natives of North Siberia were able to chart direct courses of 55 return after meandering through icy plains without relying on any visible cues at sea or in the 56 sky, concluding that: 57 58 “We must bear in mind that neither a compass, nor the north star, nor any other such sign, suffices to 59 guide a man to a particular spot through an intricate country, or through hummocky ice, when many 60 deviations from a straight course are inevitable, unless the deviations are allowed for, or a sort of ‘dead 61 reckoning’ is kept.” 62 Darwin (1873, p. 418) 63 64 Subsequent empirical studies showed that this special “dead reckoning” ability is present 65 in a wide variety of animals, encompassing insects (e.g., honey bees, Saharan desert ants), 66 spiders, birds (e.g., pigeons, geese), and mammals (e.g., golden hamsters, gerbils, dogs, humans) 67 [Gallistel, 1990; Mittelstaedt & Mittelstaedt, 1980, 1982; see Etienne & Jeffery, 2004, 68 Srinivasan, 2015, for reviews of path integration in non-human animals; see Loomis et al., 1999, 69 2001, for reviews of path integration in humans]. In the animal kingdom, the perfect example of 70 the ”path integrator” is perhaps the Saharan desert ant, Cataglyphis fortis, which is capable of

- 1 -

Report No. MAE0820

71 returning in a straight-line path back to its nest after tracking a tortuous outbound path in a 72 featureless landscape to find food (see, e.g., Müller & Wehner, 1988; Wehner & Srinivasan, 73 2003; Wehner & Wehner, 1990) [Fig. 1]. Based on the ant’s foraging and homing behavior, path 74 integration was proposed to enable the navigating animal to compute its movement changes 75 continuously along each step of its outbound journey and integrate such changes with idiothetic 76 signals (derived from the vestibular and proprioceptive systems) to update a homing vector (i.e., 77 a representation of distance and directional estimates from a point of origin) (Müller & Wehner, 78 1988). 79 To date, research on path integration continues to draw the attention of many researchers 80 in spatial cognition due to the possibilities it offers for a better understanding of the basic 81 neurocognitive processes or mechanisms that are involved in spatial perception and cognitive 82 mapping (Burgess, 2014; Hafting, Fyhn, Molden, Moser, & Moser, 2005; Moser, Kropff, & 83 Moser, 2008). 84 Fig. 1 Path integration in Cataglyphis fortis. An ant’s tortuous outbound path (from N to F) and straight homeward path (from F to N, in bold) recorded in a featureless salt pan. [Source: Fig. 1.1 in Wehner & Srinivasan (2003). Reproduced with permission.]

85 86 1.1 Assessing Path Integration in Humans 87 88 The traditional behavioral paradigm used for studying path integration in humans is the 89 path- or triangle-completion task (see, e.g., Klatzky et al., 1990; Loomis et al., 1993; Fukusima,

- 2 -

Report No. MAE0820

90 Loomis, & Da Silva, 1997; Philbeck, Klatzky, Behrmann, Loomis, & Goodridge, 2001; Sholl, 91 1989). The path completion task, originally developed by Klatzky et al. (1990), requires a 92 participant, donning blindfolds and headphones, to walk two or more straight segments with one 93 or more turns in between under the guidance of the experimenter, and to walk back (or point 94 back) to the origin on his/her initiative at the end of the outbound journey. Triangle completion 95 (Fig. 2) is the popular derivative of this path completion paradigm and refers to participants’ 96 attempts at returning to the origin (or pointing toward it) after traversing two straight segments 97 with one turn (i.e., the whole trajectory forms a triangle) [Klatzky, Loomis, Beall, Chance, & 98 Golledge, 1998; Loomis et al., 1993; Philbeck et al., 2001; Sholl, 1989]. Importantly, the 99 donning of blindfolds and headphones throughout the task blocked out visual and auditory cues 100 that could facilitate the updating of positional estimates and obliged the participant to attend to 101 idiothetic cues from vestibular and proprioceptive systems (e.g., efferent motor commands, 102 kinesthetic feedback from the musculature, acceleratory signals from the vestibular system) for a 103 moment-to-moment updating of the perceived home or target location. 104 Fig. 2 In a trial of the conventional triangle completion task, the participant traverses an outbound path (shown by dark arrows) and then attempts to walk back to the starting position with visual and auditory cues occluded (the dotted line segment shows the ideal/perfect path of return). Apart from walking responses, pointing or heading responses toward the point of origin have also been applied.

105 106 In the first extensive study that applied the triangle completion task, Loomis et al. (1993) 107 showed that sighted participants who were blindfolded performed as well as their congenitally 108 and adventitiously blind counterparts when walking back to the starting position at the end of 109 outbound travel. No group showed particularly good performance and analyses of the distance

- 3 -

Report No. MAE0820

110 and turning responses revealed systematic errors that depended on the types of path parameters. 111 On average, the participants tend to over-rotate when small turns to the origin were required and 112 to under-rotate when large turns were required (Fig. 3). Similarly, the average participant also 113 overshot and undershot small and large return distances to the origin respectively. These regular 114 patterns of errors have been found in numerous triangle completion studies conducted in both 115 real-world (e.g., Klatzky et al., 1990; Loomis et al., 1993; Sholl, 1989; Philbeck et al., 2001; 116 Wiener, Berthoz, & Wolbers, 2011) and virtual reality settings (e.g., Adamo, Briceño, Sindone, 117 Alexander, & Moffat, 2012; Arnold, Burles, Bray, Levy, & Iaria, 2014; Gramann, Müller, Eick, 118 & Schönebeck, 2005; Harris & Wolbers, 2012; Klatzky et al., 1998; Mahmood, Adamo, Briceno, 119 & Moffat, 2009; Wolbers, Wiener, Mallot, and Büchel, 2007). 120

Fig. 3 Depictions of the triangular paths used by Loomis et al. (1993) (A). X represents the origin of travel and the large dots (A, B) represent the termini of the two-legged outbound paths, each being 2, 4, or 6 m. The turn varied between 60, 90, and 120 degrees. Average performance of Loomis et al.’s (1993) participants (B). Small dots represent the centroids of the stopping points of 37 participants. The majority of those centroids did not coincide with X. [Adapted from Fig. 5.3 of Loomis et al. (1999). Reproduced with permission.] 121 122 The classical configural encoding model (Fujita et al., 1993) explains these systematic 123 errors of return as arising out of difficulties in generating a configural mental representation of 124 the outbound path, and not when participants executed their homeward responses. An alternative 125 mathematical model — the leaky path integration model (Lappe & Frenz, 2009; Lappe, Jenkin,

- 4 -

Report No. MAE0820

126 & Harris, 2007; Lappe, Stiels, Frenz, & Loomis, 2011) — proposes that the representation of 127 distance is affected by two parameters: a (i) leak rate and a (ii) gain rate. The leak rate reflects 128 the extent to which an integrated distance estimate decays over the length of locomotion whereas 129 the gain rate describes the amount of distance that is added to the integrated distance estimate 130 with each single step. Together, these two parameters can explain the underestimation and 131 overestimation of travel distances when combined with relevant distance or positional values 132 (see Lappe et al., 2007, for details). Notably, the leak rate tends to become progressively larger 133 with longer distances away from home, leading to the underestimation of return distance during 134 triangle completion (Harris & Wolbers, 2012). 135 136 2. Notable Neuroimaging Studies on Visual Path Integration in Humans 137 138 Over the past two decades, there have been a series of neuroimaging studies that 139 investigated the neural correlates of human path integration through virtual reality forms of the 140 triangle completion task (for fMRI studies, see Arnold et al., 2014; Wolbers, Wiener, Mallot, & 141 Büchel, 2007; for EEG studies, see Chiu et al., 2012; Lin, Chiu, & Gramann 2015; Gramann, 142 Müller, Schönebeck, & Debus, 2006; Gramann et al., 2010; Plank, Müller, Onton, Makeig, & 143 Gramann, 2010). Unlike the behavioral studies on human path integration that excluded the 144 availability of visual cues, participants performing virtual triangle completion in neuroimaging 145 experiments receive full exposure to visual cues in the form of optic flow at the cost of having no 146 proprioceptive or kinesthetic feedback due to the testing constraints of neuroimaging 147 experiments that prevent locomotion. Consequently, participants performing triangle completion 148 in virtual environments rely primarily on optic flow information when computing their homing 149 responses at the end of outbound paths. Therefore, the performance of path integration in virtual 150 reality (or in any desktop virtual environment) has been called visual path integration (Gramann 151 et al., 2005). 152 The common feature of all these virtual triangle completion tasks was the presentation of 153 open plains (Arnold et al., 2014; Wolbers et al., 2007) [Fig. 4] or passageways (Chiu et al., 2012; 154 Lin et al., 2015; Gramann et al., 2006, 2010; Plank et al., 2010) [Fig. 5] that lacked any salient 155 landmark or object cues. In such virtual environments, the perception of optic flow was rendered 156 through experimentally controlled translations and rotations. In a typical path integration trial,

- 5 -

Report No. MAE0820

157 the participants experienced animated/passive travel either along the outbound path of the 158 triangular trajectory (two straight segments with a turn in between, Chiu et al., 2012; Lin et al., 159 2015; Gramann et al., 2006, 2010; Plank et al., 2010; Wolbers et al., 2007) or along the entire 160 three legs/segments of the triangular trajectory (Arnold et al., 2014). The homing responses 161 generally involve: (i) pointing back to the starting position by deflecting the joystick (Wolbers et 162 al., 2007), (ii) pressing buttons to adjust 3D homing arrows (Chiu et al., 2012; Lin et al., 2015; 163 Gramann et al., 2006, 2010; Plank et al., 2010), or (iii) pressing buttons to indicate whether or 164 not they have successfully returned to their starting position (Arnold et al., 2014). As for fMRI- 165 based control trials, they varied across studies from passive travels along straight paths (i.e., 166 forward translations, Arnold et al., 2014; Gramann et al., 2006, 2010; Plank et al., 2010) to along 167 winding paths with turns that differed from the turns in the test trials (see Gramann et al., 2006, 168 2010; Lin et al., 2015; Wolbers, 2007). 169

Fig. 4 A ground-level view of the featureless virtual plain used by Wolbers et al. (2007) and a schematic diagram of the outbound paths with turns that varied systematically in increments of 30º from 30º to 120º in both clockwise and anticlockwise directions. Each of the eight outbound paths were presented five times. The second segment after the turn was varied in length to keep the animation within a suitable temporal range for functional scanning. The participants pointed back to the perceived starting position at the end of travel along each outbound path. [Adapted from Fig. 1 of Wolbers et al. (2007). Reproduced in compliance with the terms of Creative Commons Attribution License supported by The Journal of Neuroscience and the Society for Neuroscience.] 170

- 6 -

Report No. MAE0820

Fig. 5 The virtual tunnel task used by Gramann and colleagues showing: (A - C) the first-person scenes during a typical tunnel passage along a straight segment along with a schema of the allocentric and egocentric return bearings, and (D, E) the two types of homing vectors that can be specified based on adjusting the 3D homing arrows at the end of the tunnel passageway. Panels F and G show the schematics of the mental representations of the allocentric and egocentric homing vectors. [Retrieved from: http://sccn.ucsd.edu/~klaus/images/Description%20Strategy.pdf (Reproduced with permission). To download an animation clip of a sample virtual tunnel trial, go to: http://sccn.ucsd.edu/~klaus/download/tunnel_cat.mpg] 171 172 Distinct types of data analyses were applied to the respective neuroimaging studies. In the 173 first fMRI study on visual path integration, Wolbers et al. (2007) contrasted functional activity 174 between path integration and control conditions, and investigated how functional activity in 175 selected brain regions were modulated parametrically by different degrees of homeward pointing 176 responses. The functional contrast maps showed activations in the precuneus, subdivisions of the 177 intraparietal sulcus, posterior middle temporal gyrus, and superior frontal gyrus. Brain-behavior 178 correlations showed negative correlations between pointing errors and activations in the right 179 hippocampus and bilateral medial prefrontal cortex (BA9) during encoding of the outbound paths 180 across subjects. 181 As for the EEG studies on visual path integration, the modes of analyses generally 182 involved: (i) EEG current density reconstruction (Gramann et al., 2006), and (ii) EEG 183 independent component analysis (ICA) – a spatio-temporal filtering method that separates the 184 far-field EEG potentials arising from synchronized cellular assemblies into spatially fixed but 185 temporally separated processes (Chiu et al., 2012; Lin et al., 2015; Gramann et al., 2010; Plank et 186 al., 2010; for details of ICA, see Makeig, Bell, Jung, & Sejnowski, 1996; Onton & Makeig, 187 2006; Onton, Westerfield, Townsend, & Makeig, 2006). Based on these EEG analysis 188 techniques, a common set of brain regions was observed to be activated during virtual triangle 189 completion across these studies; namely, the precuneus (BA 7), the retrosplenial cortex (RSC)

- 7 -

Report No. MAE0820

190 [BA 29/30], and the middle frontal cortex (BA 6, BA 9). The precuneus was demonstrated to be 191 more activated [as assessed through EEG source density and alpha desynchronization (i.e., 192 decreased oscillation of the alpha frequency band, 8 – 13 Hz)] before and during the turning 193 phase(s) of the outbound path among participants who preferred an egocentric/body-centered 194 frame of reference for estimating their positional changes (“Turners”, Fig. 5G) than among 195 participants who preferred an allocentric/environment-centered frame of reference for estimating 196 their positional changes (“Non-turners”, Fig. 5F) [Chiu et al., 2012; Gramann et al., 2006; Lin et 197 al., 2015; Plank et al., 2010]. These findings implicated the precuneus as a key site for attending 198 to optic flow information and integrating it with egocentric positional estimates. Conversely, 199 when Non-turners were compared with Turners, higher alpha desynchronization in the right 200 precuneus/posterior parietal cortex was observed during the turning phase of an outbound route 201 (Lin et al., 2015; Gramann et al., 2010). This suggested that the right precuneus/posterior parietal 202 cortex could also be involved in directing information about egocentric directional changes to the 203 construction of an allocentric representation of the outbound path (Lin et al., 2015). 204 Similarly, alpha desynchronization in the RSC has been shown to be higher in Non- 205 Turners than in Turners before and after the turning phase, and alpha power increases in the RSC 206 have been shown to be relatively higher in Turners during the turning phase (Lin et al., 2015). 207 Crucially, Lin et al. (2015) showed that RSC event-related spectral perturbations (ERSPs) [i.e., 208 power frequencies of brain waves] during separate path integration phases covaried differently 209 with the pointing errors committed by Non-Turners. Specifically, Non-Turners’ pointing errors 210 covaried negatively with the ERSPs recorded before and after the turn but positively with the 211 ERSPs recorded during the turn. Based on these findings, Lin et al. (2015) proposed that the 212 RSC was involved in two types of cognitive processes: (i) the integration of ego-motion 213 information with allocentric reference frames (e.g., boundaries of the virtual environment) during 214 forward translations (before and after the turn), and (ii) the computation and tracking of 215 allocentric headings during turns. The second part of their proposal was supported by later 216 findings showing that the RSC played a specific role in processing rotational changes during 217 virtual motion (Chrastil, Sherrill, Hasselmo, & Stern, 2016). 218 Unlike the precuneus and RSC whose activations are influenced by the type of spatial 219 reference frame that one prefers, activation in the middle frontal cortex (BA 6 and/or BA 9) has 220 consistently been found to occur during the turning phase of the outbound path regardless of the

- 8 -

Report No. MAE0820

221 preferred type of spatial reference frame [for variation in EEG current densities, see Gramann et 222 al., 2006; for variations in theta band synchronizations (4 – 8 Hz ERSPs), see Chiu et al., 2012; 223 Gramann et al., 2010; Lin et al., 2015; Plank et al., 2010]. These findings suggested that a 224 principal role of the middle frontal cortex in visual path integration lies in monitoring changes in 225 virtual movements (Gramann et al., 2010; Plank et al., 2010). This interpretation paralleled 226 Wolbers et al.’s (2007) view that the medial prefrontal cortex is crucially involved in 227 maintaining spatial information in working memory for the subsequent computation of homing 228 responses. 229 230 3. Involvement of the Hippocampus in Path Integration: Contentious Findings from Rats 231 and Humans 232 233 The hippocampus is a key component of the medial temporal lobe (MTL) that has long 234 been seen as being functionally relevant for cognitive mapping (O’Keefe & Nadel, 1978) and the 235 strategic control of attention and memory processes in spatial navigation (Andersen, Morris, 236 Amaral, Bliss, & O'Keefe, 2007). More specifically, the hippocampus is made up of two parts: 237 the hippocampus proper, which comprises four Cornu Ammonis (CA) subfields, and the dentate 238 gyrus (Andersen et al., 2007). These two parts, together with the subiculum, presubiculum, 239 parasubiculum, and the entorhinal cortex, constitute the hippocampal formation (Amaral & 240 Lavenex, 2007). 241 242 3.1 Positive Findings 243 244 Behavioral neuroscience studies on the foraging behavior of rats have generally linked the 245 hippocampus to path integration (see, e.g., McNaughton et al., 1996; Whishaw, McKenna, & 246 Maaswinkel, 1997; Wiener, Korshuno, Garcia, & Berthoz, 1995). Notably, Whishaw et al. 247 (1997) argued that one of the specialized functions of the hippocampus pertains to coding for 248 idiothetic information (e.g., efferent signals to the musculature, afferent proprioception from the 249 muscles and joints) and channeling such information toward the optimal processing of allothetic 250 information (e.g., visual, auditory, olfactory cues) [see also Whishaw & Tomie, 1996]. 251 Specifically, the relevance of the hippocampus for path integration was demonstrated by random

- 9 -

Report No. MAE0820

252 and inaccurate homing responses of rats with fornix-fimbria lesions (Whishaw & Gorny, 1999; 253 Whishaw, Hines, & Wallace, 2001; Whishaw & Maaswinkel, 1998; Whishaw & Tomie, 1996). 254 Unlike the control rats (with intact hippocampus) who found their way back to home locations 255 under both light and dark (or blindfolded) conditions, fornix-fimbria lesioned rats returned home 256 successfully only when allothetic cues were available (e.g., in a lighted room with visual cues). 257 In addition, fornix-fimbria lesioned rats became disoriented or took suboptimal return paths 258 when restricted to idiothetic cues (Whishaw & Gorny, 1999; Whishaw et al., 2001), and 259 perseverated in returning, under both light and dark conditions, to a refuge location that was used 260 for training pruposes (Whishaw & Maaswinkel, 1998; Whishaw & Tomie, 1996). Overall, these 261 findings gave the earliest support for the pertinence of the hippocampus for path integration. 262 Ostensibly, lesions to the fornix-fimbria fibers impaired spatial learning that came under the 263 influence of idiothetic signals from the vestibular and proprioceptive systems. 264 As for human subjects, Wolbers et al. (2007) demonstrated that their visual path 265 integration pointing errors correlated negatively with activation in the right hippocampus on a 266 trial-by-trial basis (i.e., better pointing performance was associated with higher hippocampal 267 activation across trials). This negative correlation was unaffected by overall performance levels 268 and applied to both good and poor path integrators. This finding was interpreted to suggest that 269 the hippocampus was involved in integrating distance and direction signals for updating the 270 coordinates of the starting position during spatial displacements. Subsequent region-of-interest 271 (ROI) analysis performed by Chrastil et al. (2016) focusing on the hippocampus supported this 272 interpretation by showing that bilateral activation in the anterior hippocampus (in the same areas 273 as those found by Wolbers et al., 2007) was positively correlated with the magnitudes of 274 translations and rotations that were encoded accurately during virtual motion episodes. 275 Furthermore, several neuropsychological studies showed that epileptic patients with right 276 hippocampal resections demonstrated deficiencies in path integration with regard to estimating 277 (i) the distances from the starting position during blindfolded walking (Philbeck, Behrmann, 278 Levy, Potolicchio, & Caputy, 2004) and (ii) the angle-of-return (i.e., homing direction/vector) 279 during triangle completion (Worsley et al., 2001). In addition, recent research on wayfinding 280 (i.e., goal-directed navigation to places that are beyond the sensory horizon, see Wolbers & 281 Wiener, 2014, for a review) in virtual environments implicated that the hippocampus was 282 involved in: (i) a continuous tracking of distances to a goal location as perceived from the

- 10 -

Report No. MAE0820

283 egocentric/first-person perspective (Howard et al., 2014; Sherrill et al., 2013), (ii) navigating to 284 previously observed targets under the influence of salient optic flow information and top-down 285 signals from the medial frontal cortex (Sherrill et al., 2015), and (iii) flexible navigational 286 decision-making at intersecting hallways (Brown, Ross, Tobyne, & Stern, 2012). Altogether, 287 these findings support the notion that the hippocampus is a key brain region involved in path 288 integration and common spatial navigation activities. 289 290 3.2 Negative Findings 291 292 Despite these positive findings, there were several notable studies that showed findings 293 negating the relevance of the hippocampus for path integration (Alyan & McNaughton, 1999; 294 Arnold et al., 2014; Kim, Sapiurka, Clark, & Squire, 2013; Shrager, Kirwan, & Squire, 2008). 295 Alyan & McNaughton (1999) showed that hippocampectomized rats with lesions in the dorsal 296 and ventral regions of the hippocampus could perform a path integration task as well as healthy 297 control rats. Likewise, neuropsychological studies involving human subjects showed that 298 amnesic patients with lesions in the hippocampus and adjacent regions in the MTL performed as 299 well as healthy control subjects in a blindfolded return-to-origin task (Kim et al., 2013), and in a 300 pointing-back-to-the-start task at the end of an outbound path (Shrager et al., 2008). 301 Moreover, an fMRI study by Arnold et al. (2014) that required participants to estimate 302 return/homebound path distances during virtual triangle completion showcased activations over a 303 wide range of brain regions that excluded the hippocampus. The authors attributed the lack of 304 hippocampal involvement to the possibility that the hippocampus may be more relevant for 305 tracking the starting position prior to return than for tracking the entire triangular route, and that 306 hippocampal activation may be more influenced by directional estimates than distance estimates. 307 Regardless of the reason for the absence of hippocampal involvement in path integration, these 308 studies all suggested that the hippocampus is not essential for determining homing responses, 309 and that extrahippocampal areas and neural circuits involved in cognitive control, spatial 310 attention, and working memory (see Arnold et al., 2014, for details) are more likely to constitute 311 a neural system that supports path integration. 312

- 11 -

Report No. MAE0820

313 4. Re-examining the Contentious Findings over Hippocampal Involvement in Human 314 Path Integration 315 316 Due to these contentious findings over the involvement of the hippocampus in human path 317 integration, differences in methodology, framed in terms of experimental design and subject 318 characteristics, must be considered. Starting with Arnold et al.’s (2014) study, the participants 319 experienced animated motion along all three legs of a triangular path and were instructed to 320 decide whether the return path’s distance was more (or less) than its actual distance based on one 321 of two button presses. Unlike the traditional triangle completion task which involved either 322 walking or pointing back to the origin, each of which generated a continuous/parametric measure 323 of accuracy, Arnold et al.’s (2019) task involved a binary decision-making of the return path 324 distance – whether it matched (or mismatched) with their estimated return distance (Fig. 6). This 325 makes it hard to infer whether the judgment of the return vector is comparable to the spatial 326 processes involved in a mental updating of the starting position during travel along the outbound 327 path. This spatial updating mechanism in visual path integration was originally proposed by 328 Wolbers et al. (2007) to represent the function of the right hippocampus. It is likely that the 329 transformation of the original triangle completion paradigm into a task wherein the participant 330 had no control over the trajectory of the return path diminished the amount of cognitive effort 331 directed toward the spatial updating of the starting position, culminating in the absence of 332 hippocampal activation. 333

Fig. 6 (A) A first-person view of the featureless virtual plain implemented by Arnold et al. (2014) for triangle completion. (B) The participants performed estimations (E) of the return distance to the start (S) based on three categories of triangular paths reflecting an overshot, undershot, and matching of the estimated end-point in relation to the starting position. The participants pressed one of two buttons indicating whether the animated return path’s distance matched (or mismatched; either overshot or undershot) their estimated return distance.

- 12 -

Report No. MAE0820

[Adapted from Fig. 1 of Arnold et al. (2014). Reproduced in compliance with the terms of Creative Commons Attribution License supported by Frontiers Media SA.] 334 335 In the two neuropsychological studies that called into question the involvement of the 336 hippocampus in path integration (Kim et al., 2013; Shrager et al., 2008), memory-impaired 337 patients and healthy control subjects were tested under blindfolded condition using path 338 integration tasks that required walking back to the starting position after a self-directed search of 339 target objects (i.e., tiles on the floor) [Kim et al., 2013], and pointing back to the starting position 340 after traversing outbound paths with one or two turns (Shrager et al., 2008). In view of non- 341 significant group differences in performance, the authors of both studies stressed the importance 342 of working memory for carrying out the essential spatial computations during path integration 343 and proposed that the MTL may be more involved in converting information from spatial 344 perception into long-term memory than in processing information directed to path integration per 345 se. 346 However, a detailed examination of the patient profiles suggested caveats to this 347 conclusion. Although all patients experienced considerable degrees of hippocampal volume loss 348 that reflected an almost complete loss of hippocampal neurons [as emphasized by Kim et al., 349 (2013)], lesions to the adjacent entorhinal cortex or subiculum was not common to all. The 350 entorhinal cortex is a vital subregion of the hippocampal formation that was found to contain 351 grid cells with hexagonally arranged firing fields that topographically map onto the geometric 352 surface of the environment in both rats (Fyhn, Molden, Witter, Moser, & Moser, 2004; Hafting et 353 al., 2005; Moser et al., 2008) [Fig. 7A] and humans (Doeller, Barry, & Burgess, 2010; Jacobs et 354 al., 2013) [Fig. 7B]. Crucially, these grid-like firing patterns were proposed to be coincident with 355 an online computation of positional and directional estimates during events involving path 356 integration (Chen, He, Kelly, Fiete, & McNamara, 2015), wayfinding (Howard et al., 2014; 357 Spiers & Maguire, 2007) and spatial orientation (Chadwick, Jolly, Amos, Hassabis, & Spiers, 358 2015). Compared with severely amnesic patients that have more than 90% reductions in 359 hippocampal and parahippocampal volumes [two in Shrager et al.’s (2008) study and one in Kim 360 et al.’s (2013) study], three moderately amnesic patients in Shrager et al.’s (2008) study had 361 bilateral hippocampal volume reductions of 46%, 48%, and 49%, respectively, while four 362 moderately amnesic patients in Kim et al.’s (2013) study had bilateral hippocampal volume 363 reductions of 35%, 46%, 48%, and 49%. Interestingly, based on a cross-species comparison,

- 13 -

Report No. MAE0820

364 these magnitudes of volumetric reduction are comparable with those in hippocampectomized rats 365 that demonstrated similar levels of path integration performance as healthy control rats (Alyan & 366 McNaughton, 1999). The brain lesioned rats had their dentate gyri and CA3 regions spared 367 compared with the wild-type control rats. More importantly, it is worth noting that all moderately 368 amnesic patients in both Shrager et al.’s (2008) and Kim et al.’s (2013) studies did not have large 369 bilateral reduction that are greater than 50% in the volume of the parahippocampal gyrus (which 370 includes the entorhinal cortex in its rostral region; the patients had between 5% to 17% reduction 371 in the parahippocampal gyrus in comparison with the control mean). Therefore, it is very likely 372 that large parts of the entorhinal cortex in these patients remained intact, and that might have 373 enabled them to generate relatively accurate directional (Shrager et al., 2008) and Euclidean 374 distance-to-goal estimates (Kim et al , 2013). 375

- 14 -

Report No. MAE0820

Fig. 7 (A) Firing fields of a cell recorded in layer II of the dorsocaudal region of the medial entorhinal cortex (dMEC) in a rat during 30 min of running in a 2 m wide circular enclosure. ‘t‘ represents ‘tetrode’ while ‘c’ represents ‘cell’. Left panel: Locations where neuronal spikes were recorded (red), superimposed over the rat’s trajectory (black). Middle panel: Color-coded firing rate map. Red indicates peak firing while dark blue indicates zero firing. Right panel: Spatial autocorrelation of the cell’s firing activity. Color ranges from blue (r = -1) to green (r = 0) to red (r = 1). [Adapted from Fig. 1 of Hafting et al. (2005). Reproduced with permission.] (B) The activity of a cell in the left entorhinal cortex of a human subject. Left panel: Overhead view of the environment, with color representing the firing rate (in Hz) at each virtual location. Middle panel: Spatial autocorrelation of the cell’s firing activity. Peaks in the autocorrelation function determined the spacing and angle of the fitted grid, which was then used to plot the estimated grid peaks (white cross) across the entire environment. Right panel: Neuronal spike waveform. [Adapted from Fig. 2 of Jacobs et al. (2013). Reproduced with permission.] 376 377 Moreover, Shrager et al. (2008) encouraged their participants to actively maintain the 378 outbound path in mind during each trial – a strategy-inducing procedure that was not attempted 379 in any previous studies on path integration (see, e.g., Klatzky et al., 1990, 1998; Loomis et al.,

- 15 -

Report No. MAE0820

380 1993, 1998; Philbeck et al., 2001). This led to two severely amnesic patients (with an average 381 bilateral volume reduction in the hippocampus and parahippocampal gyrus that exceeded 90%) 382 and four control participants reporting attempts to continuously track their position and update it 383 relative to the starting position as they moved. The two severely amnesic patients performed as 384 well as the four control subjects when pointing back to their starting positions immediately at the 385 end of outbound travel. However, these two patients performed significantly poorer than the 386 control subjects after a period of distraction; this suggested that an intact MTL might be 387 necessary for long-term maintenance and retrieval of path integration-related spatial information. 388 On the other hand, neuropsychological studies which supported the involvement of the 389 hippocampus in human path integration (Philbeck et al., 2004; Worsley et al., 2001) recruited 390 epileptic patients that had resections of the entire entorhinal cortex and amygdala, and resections 391 of approximately half (Philbeck et al., 2004) to two-thirds (Worsley et al., 2001) of the 392 hippocampus. Apart from the removal of brain areas that exceeded the extent of lesioned sites in 393 the memory-impaired patients of Shrager et al. (2008) and Kim et al. (2013), Philbeck et al. 394 (2004) and Worsley et al. (2001) also recruited relatively more patients that counterbalanced the 395 number of control subjects. Worsley et al. (2001) tested 33 patients (16 with right temporal 396 lobectomy; 17 with left temporal lobectomy) versus 16 control participants while Philbeck et al. 397 (2004) tested 18 patients (10 with right temporal lobectomy; eight with left temporal lobectomy) 398 versus 10 control participants. In general, the findings showed that epileptic patients with right 399 temporal lobectomy were the poorest performers in the path integration tasks. Specifically, in the 400 path integration task whereby participants have to walk blindfolded to a previously seen target 401 object (i.e., a cone) placed 5 m away from the starting position, Philbeck et al. (2004) showed 402 that right temporal lobectomy patients overshot the target location to a significantly larger extent 403 than both left temporal lobectomy patients and control participants. Likewise, in a triangle 404 completion task, Worsley et al. (2001) showed that right temporal lobectomy patients overshot 405 the required return distance and exhibited bigger angles-of-return than both left temporal 406 lobectomy patients and control participants. Unlike Shrager et al.’s (2008) protocol, no 407 instructions were given to the patients about keeping the outbound path in mind as they walked. 408 Overall, both studies by Philbeck et al. (2004) and Worsley et al. (2001) were similar in: (i) 409 testing patients with resections of both the hippocampus and the entorhinal cortex, (ii) having a 410 relatively large and balanced ratio of patients and controls, and (iii) having no instructions that

- 16 -

Report No. MAE0820

411 could unexpectedly induce the use of specific strategies. Thus, one could argue that their findings 412 offered a more convincing picture of MTL’s role in path integration. 413 414 5. Clarifying the Roles of the Hippocampus and Entorhinal Cortex in Human Path 415 Integration through the Use of Different Strategies 416 417 The empirical literature discussed thus far provided pieces of evidence that both supported 418 and negated the relevance of the hippocampus for human path integration – and this begs the 419 question of whether or not activation in the hippocampus and entorhinal cortex could be induced 420 using different path integration strategies. In the context of the wider spatial navigation literature, 421 I associate the term “strategy” with the cognitive style or mental imagery technique a moving 422 agent utilizes when deciding on and travelling to a destination in mind (Zhong, 2011, 2013; 423 Zhong & Kozhevnikov, 2016; Zhong & Moffat, 2018; Wiener et al., 2011). In view of past 424 findings showing that different types of navigation strategies predisposed their users to acquire 425 different kinds of cognitive maps from the first- (egocentric) and third-person (allocentric) 426 perspectives (Zhong, 2013; Zhong & Kozhevnikov, 2016), and that a plethora of perceptual and 427 cognitive subprocesses (or “sub-strategies”) composed the statistical dimension representing a 428 particular strategy type (Zhong, 2011, 2013; Zhong & Kozhevnikov, 2016), I propose that the 429 study of different navigation strategies can offer spatial navigation researchers better clues for 430 understanding individual differences in hippocampal or MTL engagement during path 431 integration. 432 To my knowledge, there has only been three studies to date that investigated path 433 integration strategies in the context of the triangle/path completion paradigm (He & McNamara, 434 2018; Wiener, Berthoz, & Wolbers, 2011; Zhong, 2019). All three studies investigated path 435 integration in the form of two types of spatial updating strategies that differ in the computational 436 timing of the homing vector. Specifically, these two strategies refer to: a (i) continuous updating 437 strategy focusing on keeping track of the start position at all times along the outbound path (i.e., 438 moment-to-moment updating) in order to maintain a constantly updated homing vector (for more 439 details, see Wiener et al., 2011); and a (ii) configural updating strategy focusing on encoding the 440 shape of the outbound path and computing homeward responses based on the encoded mental 441 image. Based on the conceptual framework set forth by Wiener et al. (2011), continuous

- 17 -

Report No. MAE0820

442 updating is an online process whereby the homing vector is computed continuously during 443 navigation to the exclusion of a need for visualizing the shape of the outbound path, while 444 configural updating is an offline process whereby the homing vector is computed at the end of 445 the outbound path based on an encoded representation of the path. In the subsections below, the 446 two behavioral studies of Wiener et al. (2011), He and McNamara (2018) are discussed first 447 prior to the fMRI study of Zhong (2019), which was inspired by the implications derived from 448 these two earlier works. 449 450 5.1 Behavioral Findings (He & McNamara, 2018; Wiener et al., 2011) 451 452 Based on a real-world triangle completion task, Wiener et al. (2011) instructed their 453 participants (n = 15) to apply the continuous and configural updating strategies in sequence over 454 two counterbalanced blocks of outbound paths (Fig. 8A). This means that a fully within-subjects 455 design was used that same group of participants were tested in the use of both strategies. The 456 differences in homing performance arising from the use of these two strategies were exhibited in 457 terms of head orientation and corresponding path integration errors. When the participants 458 applied configural updating, they showed little head movements when traversing both segments 459 of the outbound path. By contrast, when they applied continuous updating, they were strongly 460 biased to turning their heads in the direction of the starting position as they walked along the 461 second segment of the outbound path after the turn. Regardless of the length of the outbound 462 path, the participants took longer time in attempting to return to the perceived starting position 463 when they applied configural updating than when they applied continuous updating. However, 464 path length mattered with regard to distance and angular measures. After traversing longer 465 outbound paths (mean length = 15.3 m, mean turning angle = 148°), the participants committed 466 lower magnitudes of direction errors, distance errors, and homing errors (Fig. 8B) when they 467 applied configural updating than when they applied continuous updating. The differences in the 468 commission of errors between the two strategy conditions were smaller after traversals on shorter 469 outbound paths (mean length = 8.3 m, mean turning angle = 101°) than after traversals on longer 470 outbound paths. Overall, these findings showed that the use of two different path integration 471 strategies generated contrasting patterns of path integration performance.

- 18 -

Report No. MAE0820

Fig. 8. (A) A schematic diagram of the real-world outbound paths used in the real-world triangle completion task by Wiener et al. (2011). (B) Three path integration error variables used by Wiener et al. (2011): (i) Direction error: the difference between the expected and actual angle-of-return; (ii) Distance error: the difference between the expected and actual linear distance/path of travel; and (iii) Homing error: the Euclidean distance between the starting and stopping position. [Adapted from Fig. 1 of Wiener et al. (2011). Reproduced with permission.] 472 473 To confirm the findings of Wiener et al. (2011), He and McNamara (2018) performed a 474 VR study that combined self-locomotion in both real-world and virtual environments (i.e., 475 updating of VR scenes in a head-mounted display with each forward leg movement in real 476 physical space) and extended the differential application of these two strategies to judgments of 477 relative directions (JRDs) between virtual objects arranged in a geometrically regular layout. 478 Unlike Wiener et al. (2011), who instructed each participant on both strategies, the researchers 479 tested two groups of participants, each instructed in the use of one strategy only. Configural 480 updating strategy users were instructed to “mentally draw” out a path that connected three virtual 481 posts whereas continuous strategy users were instructed to update their position and orientation 482 continuously with respect to a red virtual post appearing close to the starting position. Results 483 showed that configural updating strategy users committed significantly fewer errors when their 484 final heading directions (i.e., head orientation at the end of the outbound path when facing a third 485 virtual post) were aligned a North-to-South principal reference axis that aligned with their initial 486 heading direction than when their heading directions were misaligned with this reference axis

- 19 -

Report No. MAE0820

487 (Fig. 9). This behavioral pattern did not apply to the continuous strategy users, who 488 demonstrated comparable levels of pointing errors regardless of whether or not their final 489 heading directions were aligned with the principal reference axis. These findings suggested that 490 configural updating strategy users retrieved information about spatial locations with referral to an 491 allocentric reference frame that organized objects in a North-to-South orientation that paralleled 492 their initial heading direction. By contrast, continuous strategy users were inattentive to the 493 allocentric layout of the virtual objects and most likely depended on egocentric cues or reference 494 frames when making their pointing responses. 495 Fig. 9 A schematic diagram of the locations of the virtual posts used by He and McNamara (2018) in their immersive virtual environment. Configural updating strategy users committed significantly lower degrees of pointing errors when their final heading direction at end of outbound travel ran parallel to the North-to-South reference axis (compare black dashed line and red arrow) than when their final heading direction was misaligned with the same reference axis (compare black dashed line with blue arrow). Dashed grey lines show the pointing directions back to the first virtual post (no. 9) from the final virtual posts (nos. 8 and 12) reached by a moving participant. Note that the examples given here are for an illustration of concepts and did not reflect two actual experimental trials. [Adapted from Fig. 2 of He and McNamara (2018). Reproduced with permission. Lines and arrows are added by the author.]

496 497 5.2 Implications 498 499 In retrospect, by comparing Wiener et al.’s findings with the neuropsychological studies 500 conducted by Shrager et al. (2008) and Kim et al. (2013), it seems possible that their amnesic

- 20 -

Report No. MAE0820

501 patients might have preserved the ability to implement some form of spatial strategy resembling 502 the continuous updating strategy. As moderately amnesic patients from both studies were 503 reported to have largely intact entorhinal cortices, Zhong (2019) hypothesized that the entorhinal 504 cortex to be the key MTL region associated with continuous updating strategy use. This proposal 505 was supported by previous studies implicating that the entorhinal cortex is involved in the online 506 computation of: (i) Euclidean distances toward distal locations (Howard et al., 2014; Spiers & 507 Maguire, 2007), (ii) linear distances toward a point of origin (Jacob et al., 2017), and (iii) 508 intended geocentric directions to target objects (Chadwick et al., 2015). Most probably, these 509 online vector computations are dependent on the spatially stable firing patterns of entorhinal grid 510 cells (Fyhn et al., 2004; Jacobs et al., 2013; Gil et al., 2018; Hafting et al., 2005; Stangl et al., 511 2018). With firing fields that represent the vertices of tessellating triangles arranged in a 512 hexagonal lattice, a common spatial metric with the same scale and orientation is maintained 513 across the geometric surfaces of different environments (Hafting et al., 2005; McNaughton, 514 Battaglia, Jensen, Moser, & Moser, 2006). This uniform distribution of firing fields of grid cells 515 can be conceived as neuronal nodes in a mapping of topographical space to mediate the 516 continuous updating of positional or locational changes (McNaughton et al., 2006). Importantly, 517 the multipeaked firing pattern exhibited by grid cells does not need reciprocal inputs from the 518 hippocampus to maintain the number of firing peaks and the size or scale of the firing field, 519 implicating that spatial information about self-movements can be expressed primarily by the 520 entorhinal cortex (Fyhn et al., 2004). This might explain why Shrager et al.’s (2008) amnesic 521 patients (with largely intact entorhinal cortices) can perform as well as the control subjects 522 despite experiencing hippocampal atrophy. 523 On the other hand, MTL resections in the epileptic patients (Philbeck et al., 2004; Worsley 524 et al., 2001) might have abolished or complicated the use of a path integration strategy 525 resembling configural updating strategy. This could be seen from the finding of Shrager et al.’s 526 (2008) amnesic patients being unable to conduct homeward pointing as well as the controls after 527 a period of distraction (see Experiment 5, Shrager et al., 2008), which implicated certain deficits 528 in encoding or accessing offline representations of outbound paths. Putatively, such mental 529 representations can be formed in an allocentric format, as shown by He and McNamara (2018), 530 and it is possible that these amnesic patients encountered difficulties in the configurational 531 learning of outbound paths. As such, Zhong (2019) hypothesized the hippocampus to be the key

- 21 -

Report No. MAE0820

532 MTL region associated with configural updating strategy use. The basis for this hypothesis is 533 also supported by the classical cognitive map theory (O’Keefe & Nadel, 1978), which proposed 534 the hippocampus as the neural substrate for storing an allocentric spatial representation of the 535 environment, as well as by numerous empirical findings that posited a pivotal role of the 536 hippocampus in cognitive mapping and spatial memory formation (e.g., Bohbot, Iaria, & 537 Petrides, 2004; Iaria, Petrides, Guariglia, Ptito, & Petrides, 2007; Harris & Wolbers, 2014; 538 Schinazi, Nardi, Newcombe, Shipley, & Epstein, 2013; Spiers & Maguire, 2006; Whishaw & 539 Tomie, 1996). 540 541 5.3 fMRI Findings (Zhong, 2019) 542 543 Set on the aims of testing the hypotheses mentioned above, Zhong (2019) conducted an 544 fMRI experiment using a virtual path integration task that was programmed with animated 545 outbound paths that presented the participant with passive traversals of short routes (Fig. 10). At 546 the end of such passive travels, the participant maneuvered a joystick and attempted to find a 547 direct path back to the starting position within a short period. This phase pertained to homebound 548 travel and path integration error variables identical to those used by Wiener et al. (2011) [Fig 549 8B] were automatically recorded at the end of the homebound path. 550 Prior to the fMRI experiment, path integration practice and test/training sessions were held 551 for 50 college-aged participants in an experimental lab (Study 1, Zhong, 2019). These 552 participants were separated into two groups taught to use continuous updating and configural 553 updating strategy respectively based on detailed instructions provided by the experimenter. In 554 brief, 24 continuous updating strategy users (11 females) were taught to continuously monitor 555 their changing positions relative to the starting positions whereas 26 configural updating strategy 556 users (11 females) were taught to visualize the overall shape of the traveled path from an aerial 557 perspective. Each group of participants practiced using the learned strategy in the vicinity of the 558 lab as well as on the computer. On the computer, practice trials of the virtual path integration 559 task featured arrows on the ground that pointed out the paths of travel and an orange cone that 560 marked the starting position (Fig. 10B). These navigational cues were programmed to vanish 561 successively over the trials (for details, see Methods of Study 1, Zhong, 2019). Overall, the 562 practice task served the important purpose of familiarizing participants with the challenging

- 22 -

Report No. MAE0820

563 demands of visual path integration – ensuring that they had some experience with strategy 564 practice in the absence of navigational cues prior to formal strategy testing/training. 565 In the strategy test/training session that followed, the biological sex of the participant was 566 found to be a significant moderator of strategy use, with male configural updating strategy users 567 scoring significantly lower homing, absolute direction, and signed distance mean errors than 568 female configural updating strategy users. Strategy group differences in performance were found 569 only among the female participants, with female continuous updating strategy users scoring 570 significantly lower homing and signed distance mean errors than female configural updating 571 strategy users.

Fig. 10 (A) A schematic diagram of the animated outbound paths designed for the path integration trials in Zhong’s (2019) fMRI experiment. These outbound paths differed from those used in pre-fMRI strategy training. Simple paths (one turn each; 1a to 1d) are colored in red whereas complex paths (two turns each; 2a to 2d) are colored in blue. Paths 2a and 2b depict unidirectional complex paths (two turns in the same direction) whiles

- 23 -

Report No. MAE0820

paths 2c and 2c depict the bidirectional complex paths (two turns in opposite directions). Passive travel over the simple and complex paths each lasted 8.0 and 10.0 seconds. Homebound travel in compound trials lasted 10.0 seconds at the end of each outbound path [Source: Fig. 6D, Zhong (2019)] (B) First-person screenshots of the practice and test environments set in a virtual desert-like plain with a starry night sky. Overhead recordings of the animated outbound path and the homebound path travelled by each participant was automatically registered at the end of each trial. [Image accessible at: https://www.silc.northwestern.edu/virtual-path-integration-task/ Reproduced with permission.] 572 573 In the fMRI experiment that commenced after the completion of in-lab strategy practice 574 and testing (Study 2, Zhong, 2019), a separate set of outbound paths (Fig. 10A) differing in the 575 length of the first path segment and turning magnitudes were used. 38 participants who 576 completed in-lab strategy practice and testing were involved in this experiment. There were 19 577 participants per strategy group. Trial presentation was different from that used in pre-fMRI 578 testing such that it followed an event-related partial trial design (for details of how it was 579 modeled and used previously, see Ollinger, Shulman, & Corbetta, 2001a, 2001b; Wheeler et al., 580 2006) featuring compound trials with both outbound and homebound paths (like the regular in- 581 lab test trial) and partial trials featuring outbound paths only. The mixing of compound and 582 partial trials in a 3:1 ratio and with jittered inter-trial intervals allowed a close examination of the 583 brain regions activated during separate phases of outbound and homebound travel. 584 Region-of-interest (ROI) analysis circumscribed within the anatomical boundaries of 585 hippocampus and entorhinal cortex showed significant activation in the left entorhinal cortex 586 among continuous updating strategy users based on a linear subtraction of homebound BOLD 587 activation coefficients of simple paths (each featuring one turn in the outbound phase, Fig. 10A) 588 from those of complex paths (each featuring two turns in the outbound phase, Fig. 10A) 589 [complex - simple] (Fig. 11). Based on this contrast, marginally significant homebound 590 activations in the left hippocampus in both strategy groups were also found (Fig. 12). During the 591 homebound phase of simple paths, whole-brain analysis showed that the two groups of strategy 592 users exhibited common patterns of activations – in the left parietal cortex – and deactivations – 593 in the parts of the default mode network that comprised the medial PFC and lateral temporal 594 lobe. Brain-behavior correlational analysis further associated individual variation in path 595 integration performance with functional activity changes in the occipito-parietal and inferior 596 frontal regions, but not in the hippocampus or the entorhinal cortex. In addition, it must be noted 597 that like the pre-fMRI test results, there was no significant main effect of strategy use. This null

- 24 -

Report No. MAE0820

598 effect paralleled the absence of any significant differences in BOLD activation levels (both 599 phase-specific and contrast-related) between the two groups of strategy users. 600

Fig. 11 Significant activation in the left entorhinal cortex (28 voxels) among continuous updating strategy users based on a contrast of activations between simple and complex homebound phases [complex - simple]. Cross hairs were centered on the peak voxel of activation [x = -18, y = -18, z = -26]. The ROI-based activation map was overlaid on a high-resolution Montreal Neurological Institute (MNI) anatomical template. L = left hemisphere; R = right hemisphere. [Adapted from Fig. 15 of Zhong (2019). Image is a higher-resolution version of the original.] 601

Fig. 12 Marginally significant clusters of activation in the left hippocampus among (A) continuous updating strategy users (17 voxels) and (B) configural updating strategy users (18 voxels) based on a contrast of activations between simple and complex homebound phases [complex - simple]. Cross hairs were centered on the peak voxel of activation in each group of strategy users [continuous updating strategy: x = -24, y = -12, z = -20; configural updating strategy: x = -30, y = -12, z = -14]. The ROI-based activation map was overlaid on a high-resolution Montreal Neurological Institute (MNI) anatomical template. L = left hemisphere; R = right hemisphere. [Adapted from Fig. 16 of Zhong (2019). Image is a higher-resolution version of the original.] 602 603 5.4 More Implications 604 605 The detection of significant contrast-related activation in the left entorhinal cortex during 606 continuous updating strategy use in the homebound phase was suggested to reflect a metabolic 607 manifestation of underlying changes in the firing patterns of entorhinal grid cells in response to

- 25 -

Report No. MAE0820

608 an alteration of homing responses between simple and complex paths. Specifically, Zhong 609 (2019) proposed that attention to the demands of continuous updating strategy might have 610 induced a rescaling of grid cell firing fields or phases when switching between exposures to 611 simple and complex outbound paths, and this neurophysiological change might have manifested 612 itself in the form of greater entorhinal activation during the complex homebound phase than 613 during the simple homebound phase (cf. Barry, Hayman, Burgess, & Jeffery, 2007; Chen et al., 614 2015; Fiete, Burak, & Brookings, 2008). 615 As regards to the presence of hippocampal activations in both groups of strategy users, 616 they showed that the involvement of the hippocampus was not specific to the use of only one 617 type of path integration strategy. As these activations were not fully significant at the cluster- 618 wise threshold of p = .05, the findings were interpreted to suggest that the hippocampus harness 619 the potential of being involved in detecting variations in homing responses between simple and 620 complex paths. 621 Rather than showing the hippocampus as the lead actor in successful path integration 622 performance, the findings generally showed that extrahippocampal processes related to executive 623 functioning, attention, and perception – brought forth by functional activity changes in the 624 anterior cingulate gyrus, inferior frontal gyrus, precuneus, and cuneus (see also Zhong & Moffat, 625 2018, for other brain regions) – were of higher priority in contributing to accurate visual path 626 integration performance. 627 628 6. Future Directions 629 630 6.1 Methodological Concerns 631 632 To date and to my knowledge, Zhong’s (2019) fMRI experiment was the first 633 neuroimaging attempt at investigating the impact of two types of spatial strategies on visual path 634 integration. The presence of contrast-based activations in the left entorhinal cortex (in continuous 635 updating strategy users) and left hippocampus (marginal significant in each group of strategy 636 users) suggested that these two MTL regions may have pivotal roles to play in detecting switches 637 in homing decisions or responses after traversing outbound paths of different complexity. As the 638 original aims were to uncover whether these two MTL regions were activated during two

- 26 -

Report No. MAE0820

639 specific path integration phases (outbound and homebound), the contrast-related findings were 640 quite unexpected. To clarify the path complexity is indeed an important variable of concern for 641 entorhinal and hippocampal engagement, future fMRI studies can consider assessing the extent 642 to which activation in these two MTL regions can be modulated parametrically by paths of 643 increasing complexity. To ensure a detailed inspection of such functional changes, future studies 644 would need to involve more categories of outbound paths (> 2) that vary systematically with 645 respect to designated path parameters (e.g., number of path segments, outbound and homebound 646 path distances, turning magnitudes). 647 Another important consideration for future fMRI investigations concern the use of 648 functional connectivity analysis. The form of analysis can pinpoint the degree of temporal 649 correlations between brain voxels and ROIs of choice and can offer a more detailed “brain map” 650 than the co-activations of discrete brain regions shown by Zhong (2019). An analysis of 651 functional connectivity between the entorhinal cortex and hippocampus (chosen as seed voxels 652 or ROIs) and extrahippocampal voxels and regions can be done using Zhong’s (2019) dataset but 653 was not performed by the author in his initial round of data analysis because of two reasons: (i) 654 the verification of the study hypotheses did not require this sophisticated form of analysis, and 655 (ii) there was not a substantial number of significant phase-specific clusters of activations that 656 can uncomplicate the selection of seed coordinates. The relatively small number of significant 657 brain clusters highlighted another possibility that task-based brain activations might have been 658 attenuated by pre-fMRI strategy testing/training in the lab, which featured more trials than the 659 ones participants experienced in the fMRI scanner. Moreover, the in-lab test trials did not differ 660 drastically from the fMRI trials, so some degrees of learning and familiarity effects might have 661 “spilled over” from the in-lab test session to the fMRI session. Henceforth, it may be highly 662 advisable for future studies to avoid extensive strategy testing/training prior to neuroimaging and 663 to use sets of outbound paths that differ dramatically between the two sessions. 664 With the right improvisations to the Zhong’s (2019) experimental design, future studies 665 can explore the patterns of functional connectivity between hippocampal formation and 666 extrahippocampal regions under different conditions of path integration strategy use to give a 667 more nuanced insight into the dual roles of the entorhinal cortex and hippocampus. A recent 668 fMRI study by Izen, Chrastil, and Stern (2018) showed that functional connectivity between 669 MTL regions (i.e., posterior hippocampus, parahippocampal cortex, entorhinal cortex) and

- 27 -

Report No. MAE0820

670 resting state activations in the fronto-parietal and default mode networks increased with path 671 integration accuracy on a loop closure task (different from the triangle/path completion task). 672 The fronto-parietal and default mode networks involve wide swaths of the lateral and medial 673 prefrontal cortices, respectively, which serve key regions for executive functioning and working 674 memory processing, and hence future fMRI studies can examine the varying degrees to which 675 MTL regions are connected to them under the influence of differential path integration strategy 676 use. Contingent on the research question and experimental design, such connectivity studies can 677 be done on fMRI data collected from either resting state or task-based brain scans. 678 Together with all these fMRI-related concerns, there is another pertinent concern on the 679 environmental context under which different path integration strategies are executed. The most 680 noticeable difference between the behavioral findings of Zhong (2019) and those of Wiener et al. 681 (2011), He and McNamara (2018), was that Zhong (2019) failed to find any main performance 682 effect posed by the two strategies while the previous two studies did. Considering the fact that 683 the previous two studies involved physical walking in real-world space and that Zhong’s (2019) 684 path integration task was conducted with vision serving as the primary source of sensory 685 information in the absence of any physical walking, it is not far-fetched to speculate that 686 kinesthetic sensory information (i.e., vestibular, proprioceptive) has some roles to play when 687 implementing different types of path integration strategies. As noted by Zhong (2019) in his pre- 688 fMRI strategy training study, the absence of kinesthetic cues and proprioceptive feedback might 689 have contributed to the salient difficulties female participants experienced in carrying out an 690 effective use of the configural updating strategy relative to their male counterparts. This 691 interpretation was backed up by a recent study by Coutrot et al. (2019) that compared path 692 integration performances between real-world and virtual environments. The authors showed that 693 real-world path integration (walking involved) generated smaller sex differences, suggesting that 694 a dual combination of kinesthetic and visual information might be required for women to 695 perform as well as men during navigation. To verify this possibility in the context of path 696 integration strategy use, future behavioral studies of path integration can examine how male and 697 female users of different strategies perform under test conditions in which varying amounts of 698 kinesthetic cues are available (e.g., walking versus moving on a wheelchair versus staying 699 stationary and performing the task on screen). 700

- 28 -

Report No. MAE0820

701 702 6.2 Impact of Cognitive Aging on Visual Path Integration 703 704 So far, the research studies presented above elucidated the neurocognitive processes 705 involved in visual path integration among college-aged young adults. As research into how 706 spatial navigation ability deteriorates with advancing age is a popular topic in the aging 707 neuroscience literature (Lester, Moffat, Wiener, Barnes, & Wolbers, 2017; Zhong & Moffat, 708 2018), it is worthwhile to investigate the neurocognitive processes involved in visual path 709 integration in older adults and how these processes differ between different age groups. 710 To date, the majority of studies which highlighted age group differences in visual path 711 integration were behavioral in nature (see, e.g., Adamo et al., 2012; Allen, Kirasic, Rashotte, & 712 Haun, 2004; Harris & Wolbers, 2012; Mahmood et al., 2009). Based on assessments of visual 713 path integration through the virtual triangle completion task performed on desktop computers, 714 older adults have been shown to commit more distance (Adamo et al., 2012; Harris & Wolbers, 715 2012; Mahmood et al., 2009) and rotation errors (Adamo et al., 2012; Harris & Wolbers, 2012) 716 than younger adults when walking back to their starting positions. Specifically, for outbound 717 paths eliciting large distances and angles of return, older adults were found to travel shorter 718 distances and turn less than younger adults when returning to the starting position, with a 719 restriction in the range of homing distances being particularly prominent (Harris and Wolbers, 720 2012). This common tendency to underestimate relatively long homing distances by older adults 721 has been suggested to reflect an age-related decline in executive, working memory, and spatial 722 imagery resources allocated to path integration (Adamo et al., 2012; Allen et al., 2004). This 723 interpretation corresponds well with numerous neuroimaging studies in wayfinding that showed 724 age differences in the brain regions supporting environmental learning (Daugherty, Peng, Dahle, 725 Bender, & Raz, 2015; Head & Isom, 2010; Moffat, Elkins, & Resnick, 2006; Moffat, Kennedy, 726 Rodrigue, & Raz, 2007). For instance, Moffat et al. (2006) showed that young and older adults 727 showed noticeably different patterns of brain activations when learning the layout of a virtual 728 maze. During this navigational learning phase, when compared to young adults, older adults 729 exhibited lower levels of activations in the hippocampus, parahippocampal gyrus, medial inferior 730 parietal lobe, and retrosplenial cortex, but higher activations in the anterior cingulate gyrus and 731 medial frontal gyrus. Lower activations in the hippocampal formation were proposed to reflect

- 29 -

Report No. MAE0820

732 an age-related impairment in generating an allocentric representation of the environment while 733 higher activations in the frontal regions were proposed to reflect an age-related compensatory 734 shift in spatial memory performance toward prefrontal regions, away from medial temporal and 735 posterior regions (see also Zhong & Moffat, 2018; Reynolds et al., 2019, for recent findings 736 supporting this compensatory shift hypothesis). 737 Moreover, structural MRI studies have shown that volume reductions in the hippocampi of 738 older adults were associated with poorer performance in wayfinding (Head & Isom, 2010), place 739 learning in a virtual water maze (i.e., learning the location of a hidden platform) (Daugherty et 740 al., 2014; Moffat et al., 2007), and episodic recall of stories and name-picture associations 741 (Rodrigue, Daugherty, Haacke, & Raz, 2013). Despite these neuroimaging evidence showing the 742 negative impact of age-related changes in the hippocampus on spatial navigation among older 743 adults, there has not been many studies that associated structural or functional changes in the 744 hippocampus (or adjacent MTL regions like the entorhinal cortex) to age-related impairments in 745 path integration. To my knowledge, there has only been one fMRI study to date by Stangl et al. 746 (2018) that linked lower degrees of grid-cell-like representations (as measured by a parametric 747 variable) in the entorhinal cortices of older adults with poorer path integration performance in the 748 form of increased homing errors in path integration tasks that are both body- (i.e., physical 749 walking involved) and vision-based (done on a computer, no walking involved). Hence, future 750 studies can investigate whether an age-related decline in path integration performance can be 751 explained on the basis of neurobiological changes in the hippocampus, as well as in 752 extrahippocampal regions that are found to be relevant for path integration (e.g., precuneus, 753 retrosplenial cortex, medial prefrontal gyrus). Linking up with the discussion of path integration 754 strategies above, it shall be highly interesting to see how the use of different strategies can be 755 related to such variations in brain functions and structures in both young and older adults. 756 From a biomedical standpoint, making these efforts to investigate the neurobiological 757 changes that underlie age-related decline in path integration will offer us greater insights into the 758 brain-related factors contributing to the onset of Alzheimer’s disease (AD), considering the fact 759 that age-related decline in spatial navigation ability continues to remain as one of the most 760 noticeable trends marking the onset of AD and related types of dementia (Laczó et al., 2015; 761 Lester et al., 2017; Lithfous, Dufour, & Després, 2013). 762

- 30 -

Report No. MAE0820

763 764 765 6.3 Interaction between Landmark and Self-motion Cues 766 767 In addition to the aforementioned studies that assessed visual path integration in virtual 768 environments which are either featureless or without distinctive landmarks, some studies 769 implementing the triangle completion task in virtual environments have added landmark cues to 770 the periphery of their paths for an investigation of the interaction between idiothetic (i.e., 771 kinesthetic) and allothetic (i.e., visual) cues in the framework of Bayesian theory (Chen & 772 McNamara, 2014; Sjolund, 2014; Zhao & Warren, 2015a, 2015b). These behavioral studies 773 showed that both types of cues were combined or integrated (in the form of a combined Bayesian 774 weight) whenever one executes a homing response (either through walking or pointing back to 775 the starting position) after travelling on the outbound path. This form of cue integration were 776 found to occur under conditions when the landmarks were not shifted or shifted slightly (15°) 777 away from their original locations when one is carrying out a homing response (Chen & 778 McNamara, 2014; Sjolund, 2014). However, under situations in which the landmarks were 779 shifted by a large degree from their original location (≥ 90°) at the end of outbound path travel, 780 participants were found to conduct cue integration less and to rely more on self-motion cues 781 (representative of path integration per se) for directing themselves back to the starting position 782 (Sjolund, 2014; Zhao & Warren, 2015a). 783 Notably, Zhao & Warren (2015b) showed that after prolonged exposure to stable 784 landmarks (i.e., landmarks that remain in fixed positions) in a virtual environment, homing based 785 on reference to landmark cues (i.e., landmark-based navigation) was promoted whereas path 786 integration (based on sensitivity towards self-motion cues) was undermined. Perhaps 787 surprisingly, path integration was shown to be neither a back-up system (responsible for homing 788 in the absence of landmarks) nor an automatic reference system (responsible for detecting the 789 changes in the locations of the surrounding landmarks) [for details, see Experiments 1 and 2, 790 Zhao & Warren, 2015b]. However, when the landmarks were designed to have chronically 791 unstable locations over multiple trials and stable/unchanging locations on only a few “catch” 792 trials, homing responses were shown to be less modulated by landmark navigation and more 793 dependent on path integration-related self-motion cues (see Experiment 3, Zhao & Warren,

- 31 -

Report No. MAE0820

794 2015b). The presence of large conflicts between landmark and self-motion cues produced by 795 large and random landmark shifts suppressed the use of landmark cues and compelled a reliance 796 on self-motion cues. However, what remains unknown is the extent to which different path 797 integration strategies play in affecting homing responses over sequences of trials that presented 798 identical versus different configurations of landmark cues. To my knowledge, there has yet to be 799 a neuroimaging study that applied a virtual triangle/path completion task with landmarks 800 appearing in different locations over separate trials, and thus it will be intriguing for future 801 studies to investigate the brain regions involved in cue integration, and the relative use of 802 landmark and self-motion cues in accordance with an appropriate Bayesian model of path 803 integration. 804 805 7. Conclusion 806 807 Research on path integration has come a long way since the early days when path 808 integration was first discovered from the navigational behavior of animals (e.g., desert ants, 809 Wehner & Srinivasan, 2003; gerbils, Mittelstaedt & Mittelstaedt, 1980). In this paper, we saw 810 that virtual renditions of the triangle/path completion task, which was designed initially to assess 811 the homing behavior of humans based on idiothetic cues (Klatzky, 1990; Loomis et al., 1993), 812 has been used widely by many neuroscience researchers over the past two decades. Their 813 research highlighted the neural substrates and networks that subserve visual path integration, as 814 well as engendered skepticisms about the pertinence of the hippocampus and entorhinal cortex 815 for visual path integration. The comparisons of the studies showing support for and opposition to 816 the involvement of the MTL in path integration revealed potential complications in the 817 participant recruitment (Kim et al., 2013; Shrager et al., 2008) and testing procedures (Arnold et 818 al., 2014). These studies called into question the involvement of the MTL in path integration and 819 whether the use of different types of path integration strategies have the potential to elicit 820 different levels of activations in the hippocampus and entorhinal cortex. So far, the study by 821 Zhong (2019) was the first fMRI study that sought to clarify this question, showing that 822 significant activation in the entorhinal cortex occurs in the comparison of complex and simple 823 paths during the homebound phase of visual path integration. In the discussion of this study and 824 related works, methodological concerns were raised, and feasible steps were suggested to

- 32 -

Report No. MAE0820

825 improve the experimental design for future studies. At the same time, this endeavor to study 826 different path integration strategies was also suggested to be relevant for future cognitive 827 neuroscience research on age-related differences in the role of the hippocampal formation in path 828 integration, and Bayesian modelling of the interaction between landmark and self-motion cues. 829 This strategy-focused approach is also supported by the large pool of pre-existing 830 neuroimaging findings which showed that distinct brain regions, namely the precuneus and the 831 retrosplenial cortex, were differentially involved during path integration through the adoption of 832 different frames of reference (Chiu et al., 2012; Lin et al., 2015; Gramann, et al., 2006, 2010; 833 Plank et al., 2010; Sherrill et al., 2013). Such findings suggested the use of different spatial 834 visualization strategies for cognitive mapping and spatial memory formation. Currently, only two 835 types of spatial visualization strategies for path integration were discussed – continuous updating 836 and configural updating – and future studies can investigate other types of path integration 837 strategies, which do not have to be spatial in nature. For instance, an egocentric pace- or step- 838 counting strategy that makes use of the number of counted steps taken over an outbound path to 839 estimate the length of the homebound path can be used for visual path integration (Zhong, 2019). 840 Instead of controlling for this strategy as a potential confound (Philbeck et al., 2004; Mahmood 841 et al., 2009; Zhong, 2019), participants can be taught to use this strategy and then compared with 842 other groups of participants using other strategy types. This approach shall provide more 843 experimental data and a better understanding of how different path integration strategies compare 844 and contrast against each other with respect to behavioral performance and brain functions. 845 Moreover, from an applied biomedical perspective, understanding the behavioral and 846 neurobiological impact of a wider variety of path integration strategies can inform the design of 847 better navigational software (e.g., Sea Hero Quest, see Coutrot et al., 2019) and artificial 848 intelligence (AI) algorithms (see, e.g., Banino et al., 2018; Zhong, Goh, & Woo, 2021) for 849 assessing and training navigational skills for different population groups. For instance, achieving 850 a potential finding from future studies showing that a particular kind of path integration strategy 851 can mitigate further age-related declines in general spatial navigation ability can help to inform 852 the programming of appropriate AI algorithms and software for training older adults in the most 853 effective use of that strategy. 854 In sum, the advent of neuroimaging research on visual path integration has greatly 855 improved our understanding of its underlying brain regions and networks, yet much remains to

- 33 -

Report No. MAE0820

856 be explored about the relative involvement of MTL regions like the hippocampus and the 857 entorhinal cortex in visual path integration. By identifying the need to do so in the context of 858 path integration strategies, this paper presented an overview of the most important works in the 859 cognitive neuroscience literature on visual path integration, and introduced feasible ideas to set 860 neuroscience researchers on higher grounds for further investigations into the behavioral and 861 brain mechanisms of path integration.

- 34 -

Report No. MAE0820

862 Acknowledgements 863 864 The author thank Drs. Roberta L. Klatzky (Carnegie Mellon University), Scott D. Moffat, and 865 Mark E. Wheeler (Georgia Institute of Technology), for scholarly discussion and advice on path 866 integration research methodology.

- 35 -

Report No. MAE0820

867 References 868 869 Adamo, D. E., Briceño, E. M., Sindone, J. A., Alexander, N. B., & Moffat, S. D. (2012). Age 870 differences in virtual environment and real world path integration. Frontiers in Aging 871 Neuroscience, 4:26, 1-9. https://doi.org/10.3389/fnagi.2012.00026 872 Allen, G. L., Kirasic, K. C., Rashotte, M. A., & Haun, D. B. (2004). Aging and path integration 873 skill: Kinesthetic and vestibular contributions to wayfinding. Perception & 874 Psychophysics, 66, 170-179. https://doi.org/10.3758/BF03194870 875 Alyan, S., & McNaughton, B. L. (1999). Hippocampectomized rats are capable of homing by 876 path integration. Behavioral Neuroscience, 113, 19-31. 877 https://doi.org/10.1037/0735-7044.113.1.19 878 Amaral, D., & Lavenex, P. (2007). Hippocampal neuroanatomy. In Andersen, P., Morris, R., 879 Amaral, D., Bliss, T., O'Keefe, J. (2007). The hippocampus book (1st Ed.). New York: 880 Oxford University Press. 881 Andersen, P., Morris, R., Amaral, D. Bliss, T., & O'Keefe, J. (2007). The hippocampal 882 formation. In P. Andersen, R. Morris, D. Amaral, T. Bliss, & J. O'Keefe (Eds.). The 883 hippocampus book (1st ed.). New York: Oxford University Press. 884 Arnold, A. E. G. F., Burles, F., Bray, S., Levy, R. M., & Iaria, G. (2014). Differential neural 885 network configuration during human path integration. Frontiers in Human Neuroscience, 886 8:263. https://doi.org/10.3389/fnhum.2014.00263 887 Banino, A., Barry, C., Uria, B., Blundell, C., Lillicrap, T., Mirowski, P., ... & Kumaran, D. 888 (2018). Vector-based navigation using grid-like representations in artificial 889 agents. Nature, 557, 429-433. https://doi.org/10.1038/s41586-018-0102-6 890 Barry, C., Hayman, R., Burgess, N., & Jeffery, K. J. (2007). Experience-dependent rescaling of 891 entorhinal grids. Nature Neuroscience, 10, 682-684. https://doi.org/10.1038/nn1905 892 Bohbot, V. D., Iaria, G., & Petrides, M. (2004). Hippocampal function and spatial memory: 893 evidence from functional neuroimaging in healthy participants and performance of patients 894 with medial temporal lobe resections. Neuropsychology, 18, 418-425. 895 https://doi.org/10.1037/0894-4105.18.3.418

- 36 -

Report No. MAE0820

896 Brown, T. I., Ross, R. S., Tobyne, S. M., & Stern, C. E. (2012). Cooperative interactions 897 between hippocampal and striatal systems support flexible navigation. NeuroImage, 60, 898 1316-1330. https://doi.org/10.1016/j.neuroimage.2012.01.046 899 Burgess, N. (2014). The 2014 Nobel Prize in Physiology or Medicine: A spatial model for 900 Cognitive Neuroscience. Neuron, 84, 1120-1125. 901 https://doi.org/10.1016/j.neuron.2014.12.009 902 Chadwick, M. J., Jolly, A. E., Amos, D. P., Hassabis, D., & Spiers, H. J. (2015). A goal direction 903 signal in the human entorhinal/subicular region. Current Biology, 25, 87-92. 904 https://doi.org/10.1016/j.cub.2014.11.001 905 Chen, X., He, Q., Kelly, J. W., Fiete, I. R., & McNamara, T. P. (2015). Bias in human path 906 integration is predicted by properties of grid cells. Current Biology, 25, 1771-1776. 907 https://doi.org/10.1016/j.cub.2015.05.031 908 Chen, X., & McNamara, T. P. (2014). Bayesian cue interaction in human spatial navigation. 909 In C. Freksa, B. Nebel, M. Hegarty, & T. Barkowsky (Eds.), Spatial Cognition IX: 910 International Conference Proceedings, Bremen, Germany (pp. 147-160). Springer 911 International Publishing. 912 Chiu, T. C., Gramann, K., Ko, L. W., Duann, J. R., Jung, T. P., & Lin, C. T. (2012). Alpha 913 modulation in parietal and retrosplenial cortex correlates with navigation performance. 914 Psychophysiology, 49, 43-55. https://doi.org/10.1111/j.1469-8986.2011.01270.x 915 Chrastil, E. R., Sherrill, K. R., Hasselmo, M. E., & Stern, C. E. (2016). Which way and how far? 916 Tracking of translation and rotation information for human path integration. Human Brain 917 Mapping, 37, 3636-3655. https://doi.org/10.1002/hbm.23265 918 Coutrot, A., Schmidt, S., Coutrot, L., Pittman, J., Hong, L., Wiener, J. M., ... & Spiers, H. J. 919 (2019). Virtual navigation tested on a mobile app is predictive of real-world wayfinding 920 navigation performance. PloS ONE, 14, e0213272. 921 https://doi.org/10.1371/journal.pone.0213272 922 Darwin, C. (1873). Origin of certain instincts. Nature, 7, 417-418. Accessible at: http://darwin- 923 online.org.uk/converted/published/1873_instincts_F1760.html 924 Daugherty, A. M., Yuan, P., Dahle, C. L., Bender, A. R., Yang, Y., & Raz, N. (2015). Path 925 complexity in virtual water maze navigation: differential associations with age, sex, and

- 37 -

Report No. MAE0820

926 regional brain volume. Cerebral Cortex, 25, 3122-3131. 927 https://doi.org/10.1093/cercor/bhu107 928 Doeller, C. F., Barry, C., & Burgess, N. (2010). Evidence for grid cells in a human memory 929 network. Nature, 463, 657-661. https://doi.org/10.1038/nature08704 930 Etienne, A. S., & Jeffery, K. J. (2004). Path integration in mammals. Hippocampus, 14, 180-192. 931 https://doi.org/10.1002/hipo.10173 932 Fiete, I. R., Burak, Y., & Brookings, T. (2008). What grid cells convey about rat location. 933 Journal of Neuroscience, 28, 6858-6871. 934 https://doi.org/10.1523/JNEUROSCI.5684-07.2008 935 Fujita, N., Klatzky, R. L., Loomis, J. M., & Golledge, R. G. (1993). The Encoding‐Error Model 936 of Pathway Completion without Vision. Geographical Analysis, 25, 295-314. 937 https://doi.org/10.1111/j.1538-4632.1993.tb00300.x 938 Fukusima, S. S., Loomis, J. M., & Da Silva, J. A. (1997). Visual perception of egocentric 939 distance as assessed by triangulation. Journal of Experimental Psychology: Human 940 Perception and Performance, 23, 86-100. https://doi.org/10.1111/1467-8721.ep10772783 941 Fyhn, M., Molden, S., Witter, M. P., Moser, E. I., & Moser, M. B. (2004). Spatial representation 942 in the entorhinal cortex. Science, 305, 1258–1264. https://doi.org/10.1126/science.1099901 943 Gallistel, C. R. (1990). The organization of learning. Cambridge, MA: Bradford Books/MIT 944 Press. 945 Gramann, K., Müller, H. J., Eick, E. M., & Schönebeck, B. (2005). Evidence of separable spatial 946 representations in a virtual navigation task. Journal of Experimental Psychology: Human 947 Perception and Performance, 31, 1199-1223. https://doi.org/10.1037/0096-1523.31.6.1199 948 Gramann, K., Müller, H. J., Schönebeck, B., & Debus, G. (2006). The neural basis of ego-and 949 allocentric reference frames in spatial navigation: Evidence from spatio-temporal coupled 950 current density reconstruction. Brain Research, 1118, 116-129. 951 https://doi.org/10.1016/j.brainres.2006.08.005 952 Gramann, K., Onton, J., Riccobon, D., Mueller, H. J., Bardins, S., & Makeig, S. (2010). Human 953 brain dynamics accompanying use of egocentric and allocentric reference frames during 954 navigation. Journal of Cognitive Neuroscience, 22, 2836-2849. 955 https://doi.org/10.1162/jocn.2009.21369

- 38 -

Report No. MAE0820

956 Gil, M., Ancau, M., Schlesiger, M. I., Neitz, A., Allen, K., De Marco, R. J., and Monyer, H. 957 (2018). Impaired path integration in mice with disrupted grid cell firing. Nature 958 Neuroscience, 21, 81-91. https://doi.org/10.1038/s41593-017-0039-3 959 Harris, M. A., & Wolbers T. (2012). Ageing effects on path integration and landmark navigation. 960 Hippocampus, 22, 1770-1780. https://doi.org/10.1002/hipo.22011 961 Harris, M. A., & Wolbers, T. (2014). How age-related strategy switching deficits affect 962 wayfinding in complex environments. Neurobiology of Aging, 35, 1095-1102. 963 https://doi.org/10.1016/j.neurobiolaging.2013.10.086 964 Hafting, T., Fyhn, M., Molden, S., Moser, M. B., & Moser, E. I. (2005). Microstructure of a 965 spatial map in the entorhinal cortex. Nature, 436, 801-806. 966 https://doi.org/10.1038/nature03721 967 He, Q., & McNamara, T. P. (2018). Spatial updating strategy affects the reference frame in path 968 integration. Psychonomic Bulletin & Review, 25, 1073-1079. 969 https://doi.org/10.3758/s13423-017-1307-7 970 Head, D., & Isom, M. (2010). Age effects on wayfinding and route learning skills. Behavioural 971 Brain Research, 209, 49-58. https://doi.org/10.1016/j.bbr.2010.01.012 972 Howard, L. R., Javadi, A. H., Yu, Y., Mill, R. D., Morrison, L. C., Knight, R., ... & Spiers, H. J. 973 (2014). The hippocampus and entorhinal cortex encode the path and Euclidean distances to 974 goals during navigation. Current Biology, 24, 1331-1340. 975 https://doi.org/10.1016/j.cub.2014.05.001 976 Iaria, G., Chen, J. K., Guariglia, C., Ptito, A., & Petrides, M. (2007). Retrosplenial and 977 hippocampal brain regions in human navigation: Complementary functional contributions to 978 the formation and use of cognitive maps. European Journal of Neuroscience, 25, 890-899. 979 https://doi.org/10.1111/j.1460-9568.2007.05371.x 980 Izen, S., Chrastil, E. R., & Stern, C. E. (2018). Resting State Connectivity Between Medial 981 Temporal Lobe Regions and Intrinsic Cortical Networks Predicts Performance in a Path 982 Integration Task. Frontiers in Human Neuroscience, 12:415. 983 https://doi.org/10.3389/fnhum.2018.00415 984 Jacob, P. Y., Gordillo-Salas, M., Facchini, J., Poucet, B., Save, E., & Sargolini, F. (2017). 985 Medial entorhinal cortex and medial septum contribute to self-motion-based linear distance 986 estimation. Brain Structure and Function, 1-16. https://doi.org/10.1007/s00429-017-1368-4

- 39 -

Report No. MAE0820

987 Jacobs, J., Weidemann, C. T., Miller, J. F., Solway, A., Burke, J. F., Wei, X. X., ... & Kahana, 988 M. J. (2013). Direct recordings of grid-like neuronal activity in human spatial navigation. 989 Nature Neuroscience, 16, 1188-1190. https://doi.org/10.1038/nn.3466 990 Kim, S., Sapiurka, M., Clark, R. E., and Squire, L. R. (2013). Contrasting effects on path 991 integration after hippocampal damage in humans and rats. Proceedings of the National 992 Academy of Sciences, USA, 110, 4732-4737. https://doi.org/10.1073/pnas.1300869110 993 Klatzky, R. L., Loomis, J. M., Beall, A. C., Chance, S. S., & Golledge, R. G. (1998). Spatial 994 updating of self-position and orientation during real, imagined, and virtual locomotion. 995 Psychological Science, 9, 293-298. https://doi.org/10.1111/1467-9280.00058 996 Klatzky, R. L., Loomis, J. M., Golledge, R. G., Cicinelli, J. G., Doherty, S., & Pellegrino, J. W. 997 (1990). Acquisition of route and survey knowledge in the absence of vision. Journal of 998 Motor Behavior, 22, 19-43. https://doi.org/10.1080/00222895.1990.10735500 999 Laczó, J., Andel, R., Vyhnalek, M., Matoska, V., Kaplan, V., Nedelska, Z., ... & Hort, J. (2015). 1000 The effect of TOMM40 on spatial navigation in amnestic mild cognitive impairment. 1001 Neurobiology of Aging, 36, 2024-2033. 1002 https://doi.org/10.1016/j.neurobiolaging.2015.03.004 1003 Lithfous, S., Dufour, A., & Després, O. (2013). Spatial navigation in normal aging and the 1004 prodromal stage of Alzheimer's disease: Insights from imaging and behavioral studies. 1005 Ageing Research Reviews, 12, 201-213. https://doi.org/10.1016/j.arr.2012.04.007 1006 Lappe, M., & Frenz, H. (2009). Visual estimation of travel distance during walking. 1007 Experimental Brain Research, 199, 369-375. https://doi.org/10.1007/s00221-009-1890-6 1008 Lappe, M., Jenkin, M., & Harris, L. R. (2007). Travel distance estimation from visual motion by 1009 leaky path integration. Experimental Brain Research, 180, 35-48. 1010 https://doi.org/10.1007/s00221-006-0835-6 1011 Lappe, M., Stiels, M., Frenz, H., & Loomis, J. M. (2011). Keeping track of the distance from 1012 home by leaky integration along veering paths. Experimental Brain Research, 212, 81-89. 1013 https://doi.org/10.1007/s00221-011-2696-x 1014 Lester, A. W., Moffat, S. D., Wiener, J. M., Barnes, C. A., & Wolbers, T. (2017). The aging 1015 navigational system. Neuron, 95, 1019-1035. https://doi.org/10.1016/j.neuron.2017.06.037

- 40 -

Report No. MAE0820

1016 Lin, C. T., Chiu, T. C., & Gramann, K. (2015). EEG correlates of spatial orientation in the 1017 human retrosplenial complex. NeuroImage, 120, 123-132. 1018 https://doi.org/10.1016/j.neuroimage.2015.07.009 1019 Loomis, J. M., Klatzky, R. L., & Golledge, R. G. (2001). Navigating without vision: Basic and 1020 applied research. Optometry & Vision Science, 78, 282-289. 1021 https://doi.org/10.1097/00006324-200105000-00011 1022 Loomis, J. M., Klatzky, R. L., Golledge, R. G., Cicinelli, J. G., Pellegrino, J. W., & Fry, P. A. 1023 (1993). Nonvisual navigation by blind and sighted: assessment of path integration ability. 1024 Journal of Experimental Psychology: General, 122, 73-91. 1025 https://doi.org/10.1037/0096-3445.122.1.73 1026 Loomis, J. M., Klatzky, R. L., Golledge, R. G., & Philbeck, J. W. (1999). Human navigation by 1027 path integration. In Golledge, R. G. (Ed.), Wayfinding behavior: Cognitive mapping and 1028 other spatial processes (pp. 125-151). Baltimore: Johns Hopkins University Press. 1029 Mahmood, O., Adamo, D., Briceno, E., & Moffat, S. D. (2009). Age differences in visual path 1030 integration. Behavioural Brain Research, 205, 88-95. 1031 https://doi.org/10.1016/j.bbr.2009.08.001 1032 Makeig, S., Bell, A. J., Jung, T. P., & Sejnowski, T. J. (1996). Independent component analysis 1033 of electroencephalographic data. Advances in Neural Information Processing Systems, 8, 1034 145-151. https://doi.org/10.1007/978-1-4615-5351-9_17 1035 McNaughton, B. L., Battaglia, F. P., Jensen, O., Moser, E. I., & Moser, M. B. (2006). Path 1036 integration and the neural basis of the 'cognitive map'. Nature Reviews Neuroscience, 7, 1037 663-678. https://doi.org/10.1038/nrn1932 1038 McNaughton, B. L., Barnes, C. A., Gerrard, J. L., Gothard, K., Jung, M. W., Knierim, J. J., ... & 1039 Weaver, K. L. (1996). Deciphering the hippocampal polyglot: The hippocampus as a path 1040 integration system. Journal of Experimental Biology, 199, 173-185. Accessible at: 1041 https://www.researchgate.net/publication/14623230_Deciphering_the_hippocampal_polyglo 1042 t_The_hippocampus_as_a_path_integration_system 1043 Mittelstaedt, M. L., & Mittelstaedt, H. (1980). Homing by path integration in a mammal. 1044 Naturwissenschaften, 67, 566-567. https://doi.org/doi:10.1007/BF00450672 1045 Mittelstaedt, H., & Mittelstaedt, M. L. (1982). Homing by path integration. In F. Papi & H. G. 1046 Wallraff (Eds.), Avian navigation (pp. 290-297). Berlin Heidelberg: Springer-Verlag.

- 41 -

Report No. MAE0820

1047 Moffat, S. D., Elkins, W., & Resnick, S. M. (2006). Age differences in the neural systems 1048 supporting human allocentric spatial navigation. Neurobiology of Aging, 27, 965–972. 1049 https://doi.org/10.1016/j.neurobiolaging.2005.05.011 1050 Moffat, S. D., Kennedy, K. M., Rodrigue, K. M., & Raz, N. (2007). Extrahippocampal 1051 contributions to age differences in human spatial navigation. Cerebral Cortex, 17, 1274- 1052 1282. https://doi.org/10.1093/cercor/bhl036 1053 Moser, E. I., Kropff, E., & Moser, M. B. (2008). Place cells, grid cells, and the brain's spatial 1054 representation system. Annual Review of Neuroscience, 31, 69-89. 1055 https://doi.org/10.1146/annurev.neuro.31.061307.090723. 1056 Müller, M., & Wehner, R. (1988). Path integration in desert ants, Cataglyphis fortis. Proceedings 1057 of the National Academy of Sciences of the United States of America, 85, 5287-5290. 1058 https://doi.org/10.1098/rspb.2004.2749 1059 O'Keefe, J., & Nadel, L. (1978). The hippocampus as a cognitive map. Vol. 3. Oxford: 1060 Clarendon Press. 1061 Ollinger, J. M., Shulman, G. L., & Corbetta, M. (2001a). Separating processes within a trial in 1062 event-related functional MRI: I. The method. NeuroImage, 13, 210-217. 1063 https://doi.org/10.1006/nimg.2000.0710 1064 Ollinger, J. M., Corbetta, M., & Shulman, G. L. (2001b). Separating processes within a trial in 1065 event-related functional MRI: II. Analysis. NeuroImage, 13, 218-229. 1066 https://doi.org/10.1006/nimg.2000.0711 1067 Onton, J., & Makeig, S. (2006). Information-based modeling of event-related brain dynamics. In 1068 C. Neuper, E. Klimesch (Eds.), Progress in Brain Research, Vol. 159, pp. 99-120. 1069 https://doi.org/10.1016/S0079-6123(06)59007-7 1070 Onton, J., Westerfield, M., Townsend, J., & Makeig, S. (2006). Imaging human EEG dynamics 1071 using independent component analysis. Neuroscience and Biobehavioral Reviews, 30, 808- 1072 822. https://doi.org/10.1016/j.neubiorev.2006.06.007 1073 Philbeck, J. W., Behrmann, M., Levy, L., Potolicchio, S. J., and Caputy, A. J. (2004). Path 1074 integration deficits during linear locomotion after human medial temporal lobectomy. 1075 Journal of Cognitive Neuroscience, 16, 510-520. 1076 https://doi.org/10.1162/089892904323057254

- 42 -

Report No. MAE0820

1077 Philbeck, J. W., Klatzky, R. L., Behrmann, M., Loomis, J. M., & Goodridge, J. (2001). Active 1078 control of locomotion facilitates nonvisual navigation. Journal of Experimental Psychology: 1079 Human Perception and Performance, 27, 141-153. 1080 https://doi.org/10.1037/0096-1523.27.1.141 1081 Plank, M., Müller, H. J., Onton, J., Makeig, S., & Gramann, K. (2010). Human EEG correlates of 1082 spatial navigation within egocentric and allocentric reference frames. In C. Hölscher, T. 1083 Shipley, M. O. Belardinelli, J. Bateman, & N. Newcombe (Eds.). Spatial Cognition VII 1084 (Vol. 6222, pp. 191-206). Berlin Heidelberg: Springer. Accessible at: 1085 https://www.researchgate.net/publication/221104093_Human_EEG_Correlates_of_Spatial_ 1086 Navigation_within_Egocentric_and_Allocentric_Reference_Frames 1087 Reynolds, N. C., Zhong, J. Y., Clendinen, C. A., Moffat, S. D., & Magnusson, K. R. (2019). 1088 Age-related differences in brain activations during spatial memory formation in a well- 1089 learned virtual Morris water maze (vMWM) task. Neuroimage, 202, 116069. 1090 https://doi.org/10.1016/j.neuroimage.2019.116069 1091 Rodrigue, K. M., Daugherty, A. M., Haacke, E. M., & Raz, N. (2013). The role of hippocampal 1092 iron concentration and hippocampal volume in age-related differences in memory. Cerebral 1093 Cortex, 23, 1533-1541. https://doi.org/10.1093/cercor/bhs139 1094 Schinazi, V. R., Nardi, D., Newcombe, N. S., Shipley, T. F., & Epstein, R. A. (2013). 1095 Hippocampal size predicts rapid learning of a cognitive map in humans. Hippocampus, 23, 1096 515-528. https://doi.org/10.1002/hipo.22111 1097 Sherrill, K. R., Chrastil, E. R., Ross, R. S., Erdem, U. M., Hasselmo, M. E., & Stern, C. E. 1098 (2015). Functional connections between optic flow areas and navigationally responsive 1099 brain regions during goal-directed navigation. NeuroImage, 118, 386-396. 1100 https://doi.org/10.1016/j.neuroimage.2015.06.009 1101 Sherrill, K. R., Erdem, U. M., Ross, R. S., Brown, T. I., Hasselmo, M. E., & Stern, C. E. (2013). 1102 Hippocampus and retrosplenial cortex combine path integration signals for successful 1103 navigation. The Journal of Neuroscience, 33, 19304-19313. 1104 https://doi.org/10.1523/JNEUROSCI.1825-13.2013 1105 Sholl, M. J. (1989). The relation between horizontality and rod-and-frame and vestibular 1106 navigational performance. Journal of Experimental Psychology: Learning, Memory, and 1107 Cognition, 15, 110-125. https://doi.org/10.1037/0278-7393.15.1.110

- 43 -

Report No. MAE0820

1108 Shrager, Y., Kirwan, C. B., & Squire, L. R. (2008). Neural basis of the cognitive map: Path 1109 integration does not require hippocampus or entorhinal cortex. Proceedings of the National 1110 Academy of Sciences, USA, 105, 12034-12038. https://doi.org/10.1073/pnas.0805414105 1111 Sjolund, L. A. (2014). Cue integration and competition during navigation. (Master’s thesis, Iowa 1112 State University, Ames, IA). Retrieved from https://lib.dr.iastate.edu/etd/13682 1113 Spiers, H. J., & Maguire, E. A. (2006). Thoughts, behaviour, and brain dynamics during 1114 navigation in the real world. NeuroImage, 31, 1826-1840. 1115 https://doi.org/10.1016/j.neuroimage.2006.01.037 1116 Spiers, H. J., & Maguire, E. A. (2007). A navigational guidance system in the human brain. 1117 Hippocampus, 17, 618-626. https://doi.org/10.1002/hipo.20298 1118 Srinivasan, M. V. (2015). Where paths meet and cross: Navigation by path integration in the 1119 desert ant and the honeybee. Journal of Comparative Physiology A, 201, 533-546. 1120 https://doi.org/10.1007/s00359-015-1000-0 1121 Stangl, M., Achtzehn, J., Huber, K., Dietrich, C., Tempelmann, C., & Wolbers, T. (2018). 1122 Compromised Grid-Cell-like Representations in Old Age as a Key Mechanism to Explain 1123 Age-Related Navigational Deficits. Current Biology, 28, 1108-1115. 1124 https://doi.org/10.1016/j.cub.2018.02.038 1125 Wehner, R., & Srinivasan, M. V. (2003). Path integration in insects. Oxford: Oxford University 1126 Press. 1127 Wehner, R., & Wehner, S. (1990). Insect navigation: Use of maps or Ariadne's thread? Ethology 1128 Ecology & Evolution, 2, 27-48. https://doi.org/10.1080/08927014.1990.9525492 1129 Wheeler, M. E., Shulman, G. L., Buckner, R. L., Miezin, F. M., Velanova, K., & Petersen, S. E. 1130 (2006). Evidence for separate perceptual reactivation and search processes during 1131 remembering. Cerebral Cortex, 16, 949-959. https://doi.org/10.1093/cercor/bhj037 1132 Whishaw, I. Q., & Gorny, B. (1999). Path integration absent in scent-tracking fimbria-fornix 1133 rats: Evidence for hippocampal involvement in “sense of direction” and “sense of distance” 1134 using self-movement cues. The Journal of Neuroscience, 19, 4662-4673. 1135 https://doi.org/10.1523/JNEUROSCI.19-11-04662.1999 1136 Whishaw, I. Q., Hines, D. J., & Wallace, D. G. (2001). Dead reckoning (path integration) 1137 requires the hippocampal formation: Evidence from spontaneous exploration and spatial

- 44 -

Report No. MAE0820

1138 learning tasks in light (allothetic) and dark (idiothetic) tests. Behavioural Brain Research, 1139 127, 49-69. https://doi.org/10.1016/S0166-4328(01)00359-X 1140 Whishaw, I. Q., & Maaswinkel, H. (1998). Rats with fimbria-fornix lesions are impaired in path 1141 integration: A role for the hippocampus in “sense of direction”. The Journal of 1142 Neuroscience, 18, 3050-3058. https://doi.org/10.1523/JNEUROSCI.18-08-03050.1998 1143 Whishaw, I. Q., McKenna, J. E., & Maaswinkel, H. (1997). Hippocampal lesions and path 1144 integration. Current Opinion in Neurobiology, 7, 228-234. 1145 https://doi.org/10.1016/S0959-4388(97)80011-6 1146 Whishaw, I. Q., & Tomie, J. A. (1996). Of mice and mazes: Similarities between mice and rats 1147 on dry land but not water mazes. Physiology & Behavior, 60, 1191-1197. 1148 https://doi.org/10.1016/S0031-9384(96)00176-X 1149 Wiener, J. M., Berthoz, A., & Wolbers, T. (2011). Dissociable cognitive mechanisms underlying 1150 human path integration. Experimental Brain Research, 208, 61-71. 1151 https://doi.org/10.1007/s00221-010-2460-7 1152 Wiener, S. I., Korshunov, V. A., Garcia, R., & Berthoz, A. (1995). Inertial, substratal and 1153 landmark cue control of hippocampal CA1 place cell activity. European Journal of 1154 Neuroscience, 7, 2206-2219. https://doi.org/10.1111/j.1460-9568.1995.tb00642.x 1155 Wolbers, T., & Wiener, J. M. (2014). Challenges for identifying the neural mechanisms that 1156 support spatial navigation: The impact of spatial scale. Frontiers in Human Neuroscience, 8: 1157 571. https://doi.org/10.3389/fnhum.2014.00571 1158 Wolbers, T., Wiener, J. M., Mallot, H. A., & Büchel, C. (2007). Differential recruitment of the 1159 hippocampus, medial prefrontal cortex, and the human motion complex during path 1160 integration in humans. The Journal of Neuroscience, 27, 9408-9416. 1161 https://doi.org/10.1523/JNEUROSCI.2146-07.2007 1162 Worsley, C. L., Recce, M., Spiers, H. J., Marley, J., Polkey, C. E., & Morris, R. G. (2001). Path 1163 integration following temporal lobectomy in humans. Neuropsychologia, 39, 452-464. 1164 https://doi.org/10.1016/S0028-3932(00)00140-8 1165 Zhao, M., & Warren, W. H. (2015a). How you get there from here: Interaction of visual 1166 landmarks and path integration in human navigation. Psychological Science, 26, 915-924. 1167 https://doi.org/10.1177/0956797615574952

- 45 -

Report No. MAE0820

1168 Zhao, M., & Warren, W. H. (2015b). Environmental stability modulates the role of path 1169 integration in human navigation. Cognition, 142, 96-109. 1170 https://doi.org/10.1016/j.cognition.2015.05.008 1171 Zhong, J. Y. (2011). Individual differences in navigational strategies and mental representations 1172 of large scale environments. (Honors thesis. National University of Singapore, Singapore). 1173 Retrieved from 1174 https://www.researchgate.net/publication/303049187_Individual_differences_in_navigation 1175 al_strategies_and_mental_representations_of_large-scale_environments 1176 Zhong, J. Y. (2013). Three types of environmental representations and individual differences in 1177 spatial navigation. (Master’s thesis, National University of Singapore, Singapore). 1178 Retrieved from https://scholarbank.nus.edu.sg/handle/10635/47243 1179 Zhong, Y. J. (2019). The impact of two spatial strategies on entorhinal and hippocampal 1180 involvement in visual path integration (Doctoral dissertation, Georgia Institute of 1181 Technology, Atlanta, GA). Retrieved from https://smartech.gatech.edu/handle/1853/61594 1182 Zhong, J. Y., Goh, S. K., & Woo, C. J. (July, 2021). Heading toward trusted ATCO-AI systems: 1183 A literature review (Technical Report No. ATMRI-P1-ATP3_AI-NEURO). 1184 https://doi.org/10.13140/RG.2.2.36694.78409/1 1185 Zhong, J. Y., & Kozhevnikov, M. (2016). Relating allocentric and egocentric survey-based 1186 representations to the self-reported use of a navigation strategy of egocentric spatial 1187 updating. Journal of Environmental Psychology, 46, 154-175. 1188 https://doi.org/10.1016/j.jenvp.2016.04.007 1189 Zhong, J. Y., & Moffat, S. D. (2016). Age-related differences in associative learning of 1190 landmarks and heading directions in a virtual navigation task. Frontiers in Aging 1191 Neuroscience, 8:122, 1-11. https://doi.org/10.3389/fnagi.2016.00122 1192 Zhong, J. Y., & Moffat, S. D. (2018). Extrahippocampal contributions to age-related changes in 1193 spatial navigation ability. Frontiers in Human Neuroscience, 12:272. 1194 https://doi.org/10.3389/fnhum.2018.00272

- 46 -