Part 2:Environmental Deformations Dynamically Shift Human Spatial Memory
Mar 22, 2022
Contact: Audrey Hu Whatsapp/hp: 0086 13880143964 Email: audrey.hu@wecistanche.com
2.2 | Experiment 2: Desktop virtual environment with visual information obscured during replacement trials
At an implementation level, dynamic phase-anchoring of the grid code during environmental deformations is thought to reflect a path integration mechanism that is reset when a familiar boundary is experienced (Cheung et al., 2012; Keinath et al., 2018). Alternatively, some models posit that the effects of environmental deformations on spatial representations and memory reflect changes to self-localization estimates derived from continuous visual input from boundaries (Raudies & Hasselmo, 2015; Sheynikhovich et al., 2009). To explicitly test whether the dynamic anchoring of human spatial memory to the boundary of origin depends on continuous visual access to boundaries, we probed human spatial memory during deformations when this visual information was unavailable.
To this end, we used desktop VR to test a new cohort of participants on the locations of four nameable objects in a room. The design was similar to the first experiment (Figure 2). The experiment comprised three blocks. Each block began with 8 “collect” trials, followed by a sequence of 16 trial pairs consisting of a “replace” trial immediately followed by a “collect” trial for the replaced object. Each object was replaced four times in this replace-collect sequence, with each replaces trial starting from a position facing the middle of one of the four walls. During the first block, all trials were conducted with full visual information available (Figure S1b). In the second and third blocks, visual cues during the collect trials were visible, but visual cues during replace trials were obscured by a dense fog once the participant moved from her starting location (Figure S1b). This fog made it impossible for the participant to see anything beyond what was immediately in front of her (within 12.5 v.u., about 10% the length of the familiar environment). In most parts of the environment, this meant that only the floor was visible, which provided optic flow information but no cues to environment rescaling. To perform the task accurately under these conditions, participants needed to plan their path at the beginning of the trial before the fog appeared and to keep track of where they were while moving. The undeformed environment was used for all trials in blocks 1 and 2, and for collect trials in blocks 3. For replace trials in block 3, the environment was either stretched (n= 24 participants) or compressed (n= 24 participants; randomly assigned) by 50% along one axis during all replace trials. The floor, wall, and ceiling textures were not rescaled but were instead truncated (during compressions) or continued to tile the new space (during stretches). No differences in Block 2 accuracy (distance from the correct location, mean ± SEM; stretched: 25.87 ± 1.90 v.u.; compressed: 26.17 ± 1.52 v.u.; Wilcoxon rank-sum test, two-tailed: W = 608, p= .688) or reaction time (stretched: 15.25 ± 1.47 s; compressed: 13.67 ± 1.37 s; Wilcoxon rank-sum test, two-tailed: W = 534.5, p= .274) were observed between participants assigned to the stretched versus compressed conditions, indicating that both groups learned the task equally well.
cistanche tubulosa: improve memory
Similar to the results of our first experiment, and matching the predictions of a dynamic anchoring account, we again observed a positive shift difference for replacing trials in the stretched environment (Wilcoxon signed-rank test vs. 0, one-tailed: W = 275, p= .0002) and a negative shift difference for replacing trials in the compressed environment (Wilcoxon signed-rank test vs. 0, one-tailed: W = 72, p= .0129), with a significant difference between these deformations (Wilcoxon rank-sum test, one-tailed: W = 382, p= .00002; Figures 5 and S2b). No significant difference in a shift along undeformed dimensions was observed between stretched and compressed environments (Wilcoxon rank-sum test: W = 542, p= .3481).
As in Experiment 1, participants on average tended to face away from the boundary of origin when replacing objects (Figure S3). Due to this in combination with learned object locations being far from walls (>30v.u.) and the visibility restrictions due to the dense fog during replacement (complete occlusion at 12.5 v.u.), walls were barely visible at the time of object replacement. Indeed, during both stretched and compressed blocks all boundaries were completely obscured to the participant during much of the replacement trial after her initial movement (compressed: 49.3 ± 1.5% of trial; stretched: 60.0 ± 1.3% of trial; mean ± SEM across trials), and were not visible at the time of object replacement on the vast majority of trials (compressed: 364 of 384, 94.8%; stretched: 364 of 384, 94.8%). Thus anchoring to dominant visual boundaries at the time of object replacement cannot explain the pattern of shift differences we observe in this experiment. Moreover, comparing these results to those of Experiment 1 in which the deformations, apparatus, and object locations were comparable, the magnitude of shift differences was numerically larger on average during replacement in dense fog (Compressed: Exp. 1: −4.4697 ± 2.6455 v.u., Exp. 2: −10.9614 ± 4.0599 v.u.; Stretched: Exp. 1:9.3747 ± 3.9241 v.u., Exp. 2:20.0170 ± 4.6564 v.u.; mean ± SEM across participants). This suggests that if anything visual access to walls reduces the appearance of path-dependent shifts in object location memory, though the differences between experiments did not reach significance (Wilcoxon rank-sum test, Compressed: W = 591, p= .959; Stretched: W = 505, p= .089).
As in Experiment 1, when the data were analyzed without regard to the boundary of origin, the pattern of replacing locations during deformation blocks resembled a rescaling of the familiar object locations that qualitatively matched the rescaling of the environment (Figure S4). Interestingly, this indicates that the appearance of rescaling in the pattern of object replace locations also does not depend on visual access to boundaries at the time of object replacement, contrary to the prediction of some visually-guided models.
Together, these results thus indicate that human spatial memory is dynamically anchored to the boundary of origin in deformed environments even when participants have no visual access to boundaries at the time of object replacement.
2.3 | Experiment 3: Immersive virtual environment with full visual and vestibular information available
Finally, we tested whether human spatial memory is biased by the boundary of origin in deformed familiar environments when both vestibular and immersive visual cues are available. To this end, we had a new group of participants completes a fully immersive version of experiment 1. Participants viewed the virtual room through a stereoscopic head-mounted display (Figure S1c), and their heading and location were tracked as they physically moved about the environment. The experiment comprised two blocks. Each block began with 8 “collect” trials, followed by a sequence of 16 trial pairs consisting of a “replace” trial immediately followed by a “collect” trial for the replaced object. Each object was replaced four times in this replace-collect sequence, with each replaces trial starting from a position facing the middle of one of the four walls. During the first block, participants learned the locations of four objects as in Experiments 1 and 2, inside a 2.4 m × 2.4 m square virtual room. During the second block, the environment was either stretched (n= 24) or compressed (n= 24; randomly assigned) by 0.4 m along one axis during all replace trials, while remaining undeformed during collect trials. No differences in Block 1 accuracy (distance from correct location; stretched: 0.300 ± 0.023 m; compressed: 0.352 ± 0.026 m; Wilcoxon rank-sum test, two-tailed: W = 654, p= .177) or reaction time (stretched: 3.27 ± 0.17 s; compressed:3.27 ± 0.15 s; Wilcoxon rank-sum test, two-tailed: W = 592, p= .942) were observed between participants assigned to the stretched versus compressed conditions, indicating that both groups of participants learned the task equally well.
During the deformation block, we again observed a positive shift difference for replacing trials in the stretched environment (Wilcoxon signed-rank test vs. 0, one-tailed: W = 286, p = .00005) and a negative shift difference for replace trials in the compressed environment (Wilcoxon signed-rank test vs. 0, one-tailed: W = 88, p= .0382) matching our predictions, with a significant difference between conditions (Wilcoxon rank-sum test, one-tailed: W = 398, p= .00009; Figures 6a and S2c). No significant difference in shift along undeformed dimensions was observed between stretched and compressed conditions (Wilcoxon rank-sum test: W = 600, p= .8126).
As in Experiments 1 and 2, participants on average tended to face away from the boundary of origin when replacing objects (Figure S3). Likewise, when the data were analyzed without regard to the boundary of origin, the pattern of replacing locations during deformation blocks resembled a rescaling of the familiar object locations that qualitatively matched the rescaling of the environment (Figure S4). To further examine the relationship between rescaling and path-dependent shift dynamics, we sought to characterize each participant’s pattern of deformation block replace locations by fitting a model which
incorporated two factors: rescaling and path-dependent shift dynamics. To this end, we computed the mean square error (MSE) of deformation block replace locations relative to transformed familiar object locations across a range of rescaling and shift values (Figure 6b and S5a). Doing so revealed that participants’ patterns of responses were best explained by a hybrid model which incorporated both rescaling (Wilcoxon signed-rank test vs. 0, one-tailed, compressed: W = 378, p= 2.75e–6, stretched: W = 378, p= 2.75e–6; Wilcoxon rank-sum test, one-tailed, compressed versus stretched: W = 404, p= 4.85e–9) and path-dependent shift dynamics (Wilcoxon signed-rank test vs. 0, one-tailed, compressed: W = 38, p= 1.40e– 4, stretched: W = 345.5, p= 8.40e–5; Wilcoxon rank-sum test, one-tailed, compressed vs. stretched: W = 443, p= 2.22e–7) in the predicted directions (Figures 6c,d and S5b). We note that we were unable to perform a similar analysis in previous experiments as participants tended to stop early when replacing objects regardless of the boundary of origin (Figures 4 and 5); therefore, modeling the patterns of replacing locations in Experiments 1 and 2 would require at least one additional early stopping factor, with some dubiousness surrounding how to implement such a component.
Together, these results provide further evidence that human spatial memory is dynamically anchored to the boundary of origin during environmental deformations. We observe dynamic shifts in replacing location not only with desktop VR but also in immersive VR when full vestibular and visual cues are available. Moreover, we provide specific evidence that a hybrid account incorporating both rescaling and path-dependent shift dynamics best explains human spatial memory in deformed environments.
Buy cistanche supplement: improve memory
3 | DISCUSSION
Our results show that human spatial memory exhibits a history-dependent anchoring effect. When participants are asked to replace an object in a remembered location in a deformed environment, they choose locations biased by the distance to their boundary of origin. That is, they tend to recapitulate the distance to the boundary of origin, even though this means that they replace the object at inconsistent positions. This finding was replicated in three experimental situations–a desktop virtual environment with localizing visual information available, a desktop virtual environment with localizing visual information obscured, and an immersive virtual environment with localizing visual and vestibular information available. This triple replication indicates that the results are robust and that anchoring dynamics are resilient to variations in the perceptual information available during navigation. Shifts in human spatial memory dependent on the boundary of origin were qualitatively similar to dynamic shifts in the grid code of rats freely exploring deformed environments (Keinath et al., 2018). Together, these results suggest the existence of a common mechanism underlying the history-dependent dynamics assayed by both rodent spatial representations and human spatial memory during environmental deformations.
These results build on earlier work showing correspondences between memory-driven behavior in humans and the neural patterns observed in rodent place and grid cells but move beyond past results in several ways. It has been previously shown that when people are asked to navigate in desktop VR to the remembered locations of objects in stretched versions of familiar environments, their responses qualitatively match the stretching and bifurcations observed in hippocampal place cells (Hartley et al., 2004). Similarly, when participants in immersive VR are asked to walk to an earlier location in the absence of visual cues after navigating through a stretched or compressed environment, they exhibit biases that are consistent with the use of stretched or compressed grid cells for path integration (Chen et al., 2015). Relatedly, a recent study using immersive VR in square and trapezoidal environments found that object replacement locations and distance estimations were biased in a manner that could be predicted based on the inhomogeneities of rodent grid cells in such environments (Bellmund et al., 2020). In the current study, we tested a specific prediction: that the remembered locations of objects would undergo trajectory-dependent shifts in deformed environments. At an algorithmic level, this prediction follows from the hypothesis that the path-integrated self-localization estimate would be reset when a familiar boundary is encountered. At an implementation level, this prediction was derived from a computational model where the grid phase is dynamically anchored to the most recently-contacted boundary, possibly through input from border cells. We previously showed that many implementation-level predictions of this account were borne out in existing grid cell data sets (Keinath et al., 2018). Here we show even stronger evidence for this account, by demonstrating that these predictions are borne out in human behavior, across three experiments using both desktop and immersive VR.
These findings are important for two reasons. First, they provide additional evidence for a strong and specific correspondence between neural effects observed in rodent grid and place cells and memory effects observed in human behavior during environmental deformations. Such correspondences between cellular-level phenomena and behavior are far from guaranteed—in many cases predictions generated on the basis of subsets of neuronal representations are not realized behaviorally (Ekstrom et al., 2020; Jeffery et al., 2003; Krakauer et al., 2017; Warren, 2019; Zhao, 2018). Second, in contrast to previous studies of deformation effects, our results cannot be easily explained in terms of scaling deformations of the cognitive map. Rather, our results suggest that a primary function of place and grid cells is to integrate distances from a particular point of reference. In most ecological situations, this function would lead to an accurate and consistent representation of where the navigator is in space, and hence an accurate cognitive map. In the case of a deformed environment, however, the system must adjudicate between maintaining the coherence of the global coordinate system through scaling deformations and maintaining the fidelity of the spatial metric by shifting the map based on anchoring to some external cue (in this case the most recently encountered boundary). Previous studies have highlighted evidence for scaling deformations (Barry et al., 2007; Chen et al., 2015; Hartley et al., 2004; Munn et al., 2020; Stensola et al., 2012). Consistent with the results of these studies, we find that the average pattern of replacing locations resembles a rescaling that qualitatively matches the deformation of the environment (Chen et al., 2015; Hartley et al., 2004). However, when we break down the data based on the boundary of origin, we observe robust evidence that the spatial metric is dynamically dependent on the starting point of each trial, and that distance to this starting point is preserved to a greater degree than expected on the basis of a rescaling deformation alone. Moreover, when we specifically fit a two-factor model to our immersive VR data, we find clear evidence of concurrent rescaling and path-dependent shift effects, suggesting an integrative interaction between dynamically-anchored path integration and visual cues (Cheng, Shettleworth, Huttenlocher, & Rieser, 2007). Thus, a static coordinate transformation alone cannot completely explain the data; dynamic shifts must play a role.
cistanche supplement: improve memory
One limitation of the current study was that replacing trials initiated from each boundary always began at the same starting location. As such, starting location and boundary of origin were confounded. Thus, we cannot say for certain that the dynamic anchoring mechanism we observe anchors to boundaries. Other spatially informative non-boundary cues or experimental regularities could, in theory, also serve as dynamic anchors. Indeed, behavioral (Etienne, Boulens, Maurer, Rowe, & Siegrist, 2000; Etienne & Jeffery, 2004; Zhao & Warren, 2015) and neural (Jayakumar et al., 2019; Pérez-Escobar, Kornienko, Latuske, Kohler, & Allen, 2016; Save, Cressant, Thinus-Blanc, & Poucet, 1998) evidence indicates that punctate landmarks can reset and recalibrate path integration. Moreover, neural correlates of a vector landmark representation that could mediate such dynamics have been observed throughout the hippocampal formation (Deshmukh & Knierim, 2011, 2013; Høydal, Skytøen, Andersson, Moser, & Moser, 2019). Nevertheless, growing evidence indicates boundaries play a privileged role in shaping spatial representations and spatial memory alike (Doeller & Burgess, 2008; Doeller, King, & Burgess, 2008; Keinath et al., 2017; Weiss et al., 2017). Similarly, it remains an open question whether the dynamic biases we observe during deformations are a product of a continuously updated self-localization process, or instead reflect carrying out a previously established route. In our experiments, objects were always collected from random starting locations to minimize the likelihood that they might learn a fixed replacement trajectory. Nevertheless, participants experienced replacement trials in the familiar arrangement prior to deformation blocks, and this experience may have influenced their trajectories in subsequent replacement blocks.
The implementation-level mechanism by which environmental deformations induce shifts in human memory has not been established from empirical data. The predictions tested here were derived from a neural network model in which the phase of the grid code, thought to reflect a self-localization estimate updated via self-motion information, was reset upon encountering a familiar boundary. We previously showed that many implementation-level predictions of this model were borne out in existing rodent grid cell data sets (Keinath et al., 2018). Neuroimaging and neural recording data have established the existence of a grid cell network in humans with striking representational parallels to the networks observed in rodents (Doeller, Barry, & Burgess, 2010; He & Brown, 2019; Jacobs et al., 2013; Julian, Keinath, Frazzetta, & Epstein, 2018; Kunz et al., 2019), including the appearance of rescaling (Nadasdy et al., 2017). Thus, it is possible that dynamic shifts in the grid phase directly mediate boundary-anchored shifts in human spatial memory. Nevertheless, the purpose and representational content of the grid code, as well as its relationship to path integration, are currently a topic of much debate (Burak & Fiete, 2009; Bush et al., 2015; Dordek, Soudry, Meir, & Derdikman, 2016; McNaughton et al., 2006; Stachenfeld, Botvinick, & Gershman, 2017), and the behavioral predictions of this account could be realized through a variety of implementations which do or do not include a grid code component. Thus while a growing literature supports the correspondence of cross-species grid coding, as well as a general concordance between deformations induced by environmental geometry in both grid coding and spatial memory (Bellmund et al., 2020; Chen et al., 2015), the precise mechanistic basis of these effects remains to be resolved.
In sum, we have shown that human spatial memory in deformed environments exhibits a history-dependent anchoring effect which parallels the dynamic anchoring of grid cells to recently encountered boundaries. This anchoring is robust to a variety of visual and vestibular conditions, and cannot be accounted for by a static transformation of a navigator’s cognitive map. These results have important implications for human spatial memory and its relationship with hippocampal spatial representations and raise additional questions about the specific mechanisms which might link the two.
